UHPLC-Q-TOF analysis of pyrrolizidine alkaloids in North-Macedonian honey

ABSTRACT Honey contaminated with pyrrolizidine alkaloids (PAs) could pose a risk for human consumption, being a widely consumed food product. A fast and simple LC/MS method for the analysis of pyrrolizidine alkaloids in honey was optimised to collect occurrence data. The extraction efficiency was evaluated by a systematic study of multiple solvent mixtures and clean-up procedures. The best results for PA extraction were obtained using a formic acid/methanol mixture with subsequent clean-up by the QuEChERS method, resulting in a mean recovery range of 91.8–102%. The method validation showed satisfactory intra-day (RSD < 5.1%) and inter-day precision (RSD < 9.1%). The proposed method was applied to 14 samples. A total of six PAs and two N-oxides were detected, with levels between 89 and 8188 µg/kg. This assessment highlights the potential risk of intoxication and the need for further investigations regarding an effective quality system for manufacturers to control PAs in honey.


Introduction
Honey is one of the most appreciated and valued natural products, widely consumed by humans.Honey is also used in traditional medicine to treat a variety of clinical conditions such as bronchial asthma, throat infections, fatigue, dizziness, hepatitis, constipation, worm infestation, piles, eczema, healing of ulcers and wounds (Samarghandian et al. 2017).On the other hand, honey and other bee products may be used as good indicators of environmental pollution by toxic substances, such as heavy metals, radioactive elements, or persistent organic pollutants (Girotti et al. 2020).Moreover, they can be used for risk assessment due to the presence of natural toxic contaminants, such as certain secondary plant metabolites.The group comprising pyrrolizidine alkaloids (PAs) is noteworthy in this regard.
Over 6000 plant species have produced over 660 pyrrolizidine alkaloids (PA) and associated N-oxides (PANO), including representative species from Asteraceae, Boraginaceae, Orchidaceae, Apocynaceae, and Fabaceae (Culvenor et al. 1981;Edgar et al. 2002;Fletcher et al. 2009).From a structural point of view, they can be described as esters of necine base and necic acid (aliphatic C5 to C10 mono or dicarboxylic acids) (Figure 1) (Hungerford et al. 2019).The heterocyclic necine base can be esterified with one or two necic acids forming monoesters, open-chain diesters, and macrocyclic diesters.The focus of the researchers is targeted to 1,2-unsaturated PAs, because of their demonstrated hepatotoxicity, as well as other toxic effects (Stegelmeier et al. 2016;Casado et al. 2022b).The presence of all three structural types of PA/PANOs has been reported in the most widespread species of Boraginaceae family in North Macedonia such as Echium spp., Cynoglossum spp., Onosma spp., and Symphytum spp.The compounds lepthantine-N-oxide, echimidine, uplandicine-N-oxide, echiumine, and viridiflorine were found to be the most abundant (Matevski 2010;Stanoeva et al. 2022;Stefova et al. 2022) so there is a potential risk of these to enter the food chain.
The plants which produce PAs are easily available for the bees and in that way these compounds can be transferred to honey but also to the other bee products.According to a number of studies, pollen from these plant species serves as a nectar supply for honeybees (Edgar et al. 2002;Kast et al. 2019).This means that pollen is a significant contributor of PAs in honey (Crews et al. 1997), but more recent studies show that nectar is the main carrier of PAs from flowers to honey (Lucchetti et al. 2016).The first reports of PAs in honey were published in the late 70s and early 80s, when honey from Senecio jacobaea (USA) and Echium plantagineum (Australia) was found to contain PAs up to 3900 µg/kg and 1800 µg/kg, respectively (Deinzer et al. 1977;Culvenor et al. 1981).
Since then, studies with a larger number of samples and a broad sample selection were conducted, in order to gain insight into the general market situation.Recent studies of retail honey samples show PAs levels from 141 to 4078 µg/kg (Dübecke et al. 2011;Griffin et al. 2013;Bandini and Spisso 2022).Lycopsamine and echimidine were found to be the main cause of contamination.Samples attributed to Echium species were found to contain PAs of up to 2000 µg/kg in Australia (Beales et al. 2004) and up to 237 µg/kg in Spain (Orantes-Bermejo et al. 2013).The most recent European studies show lower PA levels in honey, ranging from 2 to 162 µg/kg in Swiss honey and up to 33 µg/kg in Italian honey (Lorena et al. 2016;Kast and Lucchetti 2019;Roncada et al. 2023).Regarding these studies, the European Food Safety Authority (EFSA 2016) conducted a risk assessment that resulted with the conclusion that exposure to PAs could present both acute and chronic effects in consumers (EFSA 2017).
For these naturally occurring toxins found in honey and bee products, no legal requirements or maximum levels are set.However, certain European countries have already established PA tolerance thresholds that are allowed for use in herbal medicines and extracts and these values have been applied to the commerce of honey.In 2017, the Federal Institute of Risk Assessment (Bundesinstitut für Risikobewertung, BfR, Germany) and the UK Committee on Toxicity (COT) have recommended an exposure limit from different foods of 0.007 μg/kg body weight per day (COT 2008;BfR 2011).Recently, and in coordination with EFSA, this value was revised to 237 μg/kg body weight per day (EFSA 2017), making use of an external scientific report of EFSA (Mulder et al. 2015) which, unfortunately, excluded honey.
The diversity of chemical structures and sources means that the analyst is faced with the challenge of extracting, separating, identifying, and measuring a wide variety of PAs in very different and complex matrices, including plants, seeds, honey, pollen, body fluids, and insects (Akuamoa et al. 2023).For these reasons, MS-based methods are required for analysis.GC methods are rarely taken into consideration due to their inability to detect PANOs together with PAs, due to their thermal instability.On the other hand, LC-MS methods demonstrate adequate sensitivity (lower than 1 ppb) and selectivity by having the ability to detect both PAs and PANOs (EURL-MP-Method_002 (Version 3) 2019).One main drawback of this technique is the limited availability of commercial reference substances of diverse PAs and PANOs (Kempf et al. 2011).
Sample preparation is the crucial step in the analysis of minor constituents in a food matrix.As PAs are polar compounds with a basic nature, they are soluble in polar solvents such as methanol and acidified water.Additionally, the extraction solvent must be able to dilute the sample completely, considering its viscosity.Published methods suggest the use of only water or diluted sulphuric acid as the extraction solvent (BfR 2013; Griffin et al. 2013;Orantes-Bermejo et al. 2013;Bodi et al. 2014;Lorena et al. 2016;Letsyo et al. 2017;Roncada et al. 2023).In more recent studies, acetonitrile has been suggested, combined with a QuEChERS clean-up procedure (Casado et al. 2022a).Only one published report suggests acidified methanol as an extraction solvent (Sixto et al. 2019).Moreover, its extraction efficiency was previously confirmed for the analysis of plant material (Stefova et al. 2022).
The next important step in the sample treatment is the clean-up.As previously stated, the abundance of many compounds in honey requires that some should be removed, to prevent potential interference.Three options are suggested as the most suitable clean-up techniques: reversed-phase columns (C18), strong cation exchange columns (SCX), and QuEChERS method.It is considered that SCX columns have the advantage over C18 columns, because of strong retention of basic PAs on the sorbent, whilst elution of interferent substances (Crews et al. 2010).The QuEChERS method was originally developed for clean-up to be able to simultaneously determine multiple pesticides in a sample (Anastassiades et al. 2003).It is specifically designed to use the salting-out effect in order to extract analytes from complex matrices as well as subsequent clean-up of the extract (Martinello et al. 2017;Casado et al. 2022a).
Taking into consideration previously published data on extraction, the main goal of this study was to investigate and optimise the extraction efficiency of PAs and PANOs from honey using different solvent mixtures and clean-up techniques.Furthermore, the LC/MS n method was optimised, validated and applied to honey samples taken from various regions of the country, as in spite of confirmed toxicity and recommendation for monitoring, there are still no data available about the distribution and diversity of pyrrolizidine alkaloids in the south-east European and Balkan regions and in North Macedonia.The analytical results will enable authorities to evaluate the distribution and diversity of pyrrolizidine alkaloids in honey samples, in order to obtain indicators for risk assessment.

Honey samples
Fourteen commercially available honey samples were analysed.The choice of samples was meant to reflect varieties of geographical origin, plant origin, type of retailers, and year of production (Table 1).All samples were purchased between 2019 and 2022.Before analysis, samples were homogenised at 40°C in water bath, so as to avoid crystallisation, placed into a polypropylene tube with a screw cap, and stored at 4°C until analysis.
For clean-up of the extracts three techniques were used: strong cation exchange columns (500 mg Bond Elut SCX) and reversed-phase columns (500 mg Bond Elut C18), both obtained from Agilent (Santa Clara, CA, USA) and QuEChERS (EN method 15662, PSA tube, Sigma-Aldrich, St. Louis, MO, USA).
For optimisation of the extraction procedure, portions of 5 g (±0.01) honey of sample M13 were weighted in a plastic 50 mL centrifugation tube and 1 mL of working standard solution containing eight PAs and three PANOs (Table 3) with concentrations of 1-3 mg/kg were added.Samples were dissolved in 25 mL of solvent mixture at ambient temperature by shaking for 10 min in a Stuart SSL2 orbital shaker (Cole-Parmer, St Neots, Cambridgeshire, UK) and then centrifuged at 3500 rpm for 10 min (Hettich EBA 200, Boston, MA, USA) and a clean-up step was performed.
Strong cation exchange cartridges were preconditioned with 3 mL of methanol and 3 mL of the extraction solvent.An aliquot of 3 mL of the honey extract was loaded onto the column and washed with 3 mL of water.Afterwards, the columns were dried in a vacuum and eluted with 6 mL (2 × 3 mL) of 5% ammonia in methanol.The final volume of the extract was 10 mL.The same procedure was carried out for clean-up with reversed-phase cartridges, where methanol was used as an eluent.
The QuEChERS method was carried out by transferring 10 mL of the supernatant to a 50 mL polypropylene tube with a screw cap.A mixture of salts: 4 g MgSO 4 , 1 g sodium citrate dihydrate, 0.5 g disodium hydrogen citrate sesquihydrate, and 1 g NaCl, was added.The mixture was vortexed for 5 min and then centrifuged at 3000 rpm for 10 min.The supernatant was transferred to a polypropylene tube containing 150 mg PSA and 900 mg MgSO 4 (Supel QuE PSA Tube, 15 mL, 150 mg Supelclean PSA, 900 mg MgSO 4 ).
[f] Sample produced by small family apiaries.

FOOD ADDITIVES & CONTAMINANTS: PART B
Analyses were performed using an Agilent 1290 Infinity II UHPLC (Agilent, Santa Clara, CA, USA) coupled to an Agilent 6550 Series Accurate-Mass Quadrupole Time-of-Flight (Q-TOF).Separation was carried out on a 150 × 2.1 mm, particle size 2.7 µm Poroshell 120 EC-C18 Discovery column (Sigma-Aldrich, Darmstadt, Germany).The column temperature was maintained at 40°C, and a gradient elution was performed using a flow rate of 0.25 mL/min.Two eluents: A (pH 3.15) and B (pH 4.88), were used.Eluent A was prepared with 100% water containing 1% formic acid and 5 mmol/L ammonium formate.Eluent B was prepared with 100% methanol containing 1% formic acid and 5 mmol/L ammonium formate.Elution profile started with 5% B for 0.5 min, linear increase to 50% to 7 min, from 7 to 7.5 min linear increase up to 80% B, increase to 100% to 7.6 min and held to 9 min and finally from 9 to 15 min back to 5% B, to re-equilibrate the column.The injection volume was 1 µL, and the autosampler needle was washed with 100% methanol between injections to eliminate carryover.The electrospray ionisation was performed in positive mode in the mass range m/z 100-800.Collision-induced dissociation was performed using collision energy (CE) of 20 V. The capillary potential was 3000 V, the drying gas flow was 15 L/min, and the temperature was 200°C.The sheath gas flow was 11 L/min, and the temperature was 250°C.Nitrogen was used as a drying and sheath gas.

Method validation
Method validation was based on European Commission's (2014) guideline SANCO/12571/2013 for confirmatory analytical methods in terms of linearity range, matrix effect, limit of detection, limit of quantification, intraday precision, and inter-day precision.Accuracy was assessed by recovery experiments that included analysis of fortified blank samples at three concentration levels 50, 100, and 200 µg/kg.Linearity was estimated by analysing spiked honey samples in the concentration range from 5 to 200 µg/kg.Linear regression analysis was performed by plotting the respective peak areas against the concentration of each analyte in micrograms per kilogram.Linearity was expressed in terms of the correlation coefficient r and the coefficient of determination R 2 .
For evaluating the matrix effect two calibration curves were constructed: (i) a calibration curve in pure solvent of the analyte and (ii) a matrix-matched calibration curve.Matrix effect was assessed by calculating the ratio of the slope of the matrix-matched calibration curve to the slope of the calibration curve obtained in solvent.Values lower than 0% indicate signal suppression, whereas values greater than 0% indicate signal enhancement by the matrix.
For the evaluation of intraday precision, a standard mixture of PA/PANOs (50 µg/kg) was measured six times in 1 day.The intraday precision was expressed   [a] Numbers in bold denotes the most abundant peak (quantification ion).
as a relative standard deviation.For inter-day precision, three matrix-matched concentrations of PA/PANOs were prepared (50, 100, and 200 µg/kg).All the samples were analysed in triplicate over 2 days and relative standard deviations were calculated.
For the limit of detection (LOD) and the limit of quantitation (LOQ), a series of standard solutions were prepared in a honey as a matrix.The limit of detection was determined to be the lowest concentration of analytes that was not necessarily quantifiable but was distinguishable from zero (signal-to-noise ratio ≥ 3).The limit of quantitation was considered to be the lowest concentration at which an acceptable precision (signalto-noise ratio ≥ 10 and RSD ≤ 20%) could be achieved.

Statistical analysis
Statistical analysis of the data was performed using Excel 2019 for calculations of calibration curves, mean, standard deviation, recovery, and matrix effect.Samples were analysed in triplicate, and a one-way analysis of variance (ANOVA) was performed using STATISTICA, version 7. The Newman-Keuls post hoc test (at p < .05)was used to determine the significant differences between the results obtained for the individual compounds with different extraction solvents/purification.Principal component analysis was performed using the software TANAGRA 1.4.28 (Lyon, France).

Optimization of the extraction procedure
Honey is a mixture of several major and a wide variety of minor compounds, from which isolation of PAs can be complicated.A valid method for their analysis therefore requires a sample preparation procedure that includes selection of an efficient extraction and an effective clean-up procedure.Based on the published methods for PA analysis, there are several extraction procedures, depending on the variation of these two aspects (BfR 2013; Griffin et al. 2013;Bodi et al. 2014;Valese et al. 2016).However, most of the published experimental procedures have been carried out without evaluation of the impact of solvent(s) on the extraction efficiency of individual compounds and the possibility of their degradation.In this study, the effects of the extraction solvent were systematically investigated using methanol (M), formic acid/methanol (MF), hydrochloric acid/methanol (MH), sulphuric acid/ methanol (MS), deionised water (W), formic acid/ water (WF), hydrochloric acid/water (WH), and sulphuric acid/water (WS) on portions of sample M13 spiked with 11 different standards of PAs and PANOs (Table 2).The extraction with sulphuric acid in water as recommended by BfR ( 2013) was also tested.The structures of all standard compounds are given in Figure S1.
Three different clean-up techniques were tested: strong cation exchange columns (SCX), reversed-phase columns with C18 packing (C18), and QuEChERS.Reversed-phase C18 columns (C18) were used most often, following the recommendation by BfR ( 2013), but there are also useful applications of the QuEChERS method as a simple and fast method for clean-up (Martinello et al. 2017).The extraction efficiency for 15 different extracts was studied by analyzing three replicate samples spiked with working standards containing PAs and PANOs at concentration levels of 1-3 mg/kg.Blank samples were also analysed.The results for the recovery of all tested samples are provided in Table S1.
In order to evaluate the significance of the obtained variables, i.e. extraction solvent and clean-up type of PAs, and select the most efficient extraction procedure, principal component analysis (PCA) was applied to the data set and four principal components were obtained.The first factor (PC1), which explained 82.8% of the variance, was linked to the clean-up type, whereas the second principal component, which explained 11.4% of the total variance, was related to the choice of extraction mixture.These factors affected all compounds, including PAs, but also the corresponding N-oxides.The principal component score plot and correlation scatter plot of the variables with PC1 and PC2 are presented in Figure 2. The eigen values and factor loadings from the PCA are provided in Table S2.
The obtained results from the analysis of the different extracts show that the choice of extraction solvent is of high importance.In general, lower extraction efficiency was achieved with the use of aqueous solvents compared to methanol containing mixtures.Additionally, it has been demonstrated that adding acid to the extraction medium increases the recovery rates of all compounds.This is due to the enhanced ionisation of PAs in lower pH media by protonation of the nitrogen atom.The type of acid (sulphuric, formic, or hydrochloric) in the extraction mixture was not found to be significant when using methanol and for water, using sulphuric acid or just pure water gave highest recoveries.
All the optimisation results can be arranged in three groups, with the use of clean-up techniques as criteria.For the clean-up of the first group of extracts C18 columns were used.Obtained results clearly show that this sorbent is not suitable for retaining PAs since the highest recoveries were around 50%.This suggests that the standard BfR protocol (BfR 2013) could be upgraded by modifying the clean-up procedure.In this case, the choice of solvent seems to have no effect.The weak retention of the basic PAs is a result of the polarity of these compounds and can potentially compromise the analysis.
The second group of extracts is prepared with cleanup using SCX columns.Major differences in PAs concentrations in water and methanolic extracts were observed.All aqueous solvents resulted in low and highly variable recoveries and the presence of acid seemed to additionally lower the recovery of PAs.The use of methanol mixtures resulted in higher recoveries, with the highest values obtained with formic acid/ methanol.The presence of acid in the extraction medium significantly contributed to better extraction efficiency, due to better ionisation of the target compounds and their transfer to the extract.The data also show that using methanol results in better extraction efficiency for N-oxides compared to free bases of PAs.Hence, we can conclude that SCX sorbent is able to strongly retain PАs/PANOs, which provides a clear advantage over C18 sorbent.
The third group of extracts is cleaned-up with the QuEChERS method.The best recovery rates were obtained with the use of pure methanol (up to 120%) and formic acid/methanol (up to 101%) as extraction solvents.The presence of hydrochloric and sulphuric acid results in slightly lower extraction efficiency, which may be due to decomposition of PAs by hydrolysis of the ester bonds.Тhe presence of formic acid does not have that effect as a weak acid.
After optimisation of the extraction efficiency of various solvents and clean-up techniques, the combination of formic acid/methanol and QuEChERS was found to be the most suited and feasible to be applied in future investigations.

Method validation
The accuracy of the method was confirmed by the recoveries calculated using spiked blank sample at 50, 100, and 200 µg/kg.The resulting chromatogram is shown in Figure 3.The recovery was determined by dividing the experimentally obtained value with the spiked concentration, resulting in a mean recovery across all concentration levels ranging from 91.8% to 102%, which fulfils the SANCO criteria.
The accurate quantitation of PAs in complex matrices such as honey can be compromised by significant matrix effect, which affects the ionisation of different compounds, causing signal suppression or enhancement.The obtained results for matrix effects ranged from +3.2% to +11.1% for PAs and −44.2% to +7.4% for PANOs.For most analytes the results showed matrix effects which did not exceed ± 20%.The most significant matrix effect was observed for retrorsine Noxide (−44.2%),indicating a strong signal suppression.On the other hand, an enhancement of the instrumental signal was evident for 10 compounds.The highest was for europine (+11.1%), while the rest showed signal suppression in the range −44.2% to −4.1%.
The linearity of the method was evaluated with a 5point calibration curve over a range of 5 to 200 µg/kg.Analysis of regression equation parameters showed excellent linearity with determination coefficient, R 2 , greater than 0.99 for all PAs (Table S3).The precision of the method was evaluated as intraday (RSD r ) and inter-day precision (RSD R ).Satisfactory intraday precision was obtained, as values for the relative standard deviation were below 5.1%.The results for the inter-day precision revealed RSD acceptable values for 50 µg/kg, 100 µg/kg, and 200 µg/kg, ranging from 1.4% to 9.1%, 0.1% to 5.2%, and 0.2% to 8.9%, respectively.The obtained results for all the substances are reported in Table 4.  3).
LOD and LOQ obtained in this study ranged from 3 to 14 µg/kg and 11 to 45 µg/kg, respectively, which is slightly higher than the literature reports (BfR 2013; Martinello et al. 2017).Given the fact that the maximum level of PAs in pollen and pollen products (e.g.honey) in the European Union should not exceed 500 µg/kg (European Commission 2023), the obtained LOQ values, being 10 to 50 times lower, are satisfactory.

Honey samples
The optimised and validated procedure was applied to a total of 14 honey samples: eight of these were commercially available in supermarkets and six were purchased from family beekeepers (Table 1).All tested samples were positive for the presence of PA/PANOs (Figure 4).In total, eight different PAs were detected, six of which were 1,2-unsaturated and two were 1,2-saturated PA/ PANOs.The identification of PAs and PANOs was conducted by a combined assessment of their retention times, the expected molecular ion adducts ([M+H] + or [M+NH 4 ] + ) and their fragmentation patterns.All identified peaks were confirmed using MS/MS spectra, either by a comparison with the MS/MS spectra of an available reference substance or by observation of characteristic fragments and comparison to published data (Table 5).Among the detected compounds, two were identified as N-oxides which is in accordance with publications of Betteridge et al. (2005) and Orantes-Bermejo et al. (2013).
The vast diversity of chemical structures of PA/ PANOs combined with the limited availability of standard substances leads to the necessity of using a semiquantitative approach (Sixto et al. 2019).In this study, the individual compounds were quantified using a commercial standard substance with analogous structure pattern as the given compound.For the purpose of quantification, a calibration curve in the range of 5 µg/kg to 10 mg/ kg was constructed and divided into two sections.The overall contamination levels ranged from 89 to 8188 µg/ kg for total PA/PANOs.The highest level was from trachelantamine N-oxide, which was most frequently found, with the range 42-8179 µg/kg.Viridiflorine N-   5).
oxide and 7-O-acetyllycopsamine/intermedine and echinatine were previously detected in Boraginaceae plants from the territory of North Macedonia, where only echinatine was found in both commercial and family produced samples (Stefova et al. 2022).Based on the available published data, echiuplatine was not previously reported in honey samples and was found in two samples: acacia and a meadow/acacia mix.
A literature review shows that most of the samples contain PA/PANOs at a level that is among the highest ever measured: up to 3900 µg/kg in Australia and the United States of America (Culvenor et al. 1981;Betteridge et al. 2005;Dübecke et al. 2011;Zhu et al. 2018) and up to 1500 µg/kg in Africa (Letsyo et al. 2017).The obtained results showed higher levels of contamination compared to data from several parts of Europe (Griffin et al. 2013;Kowalczyk and Kwiatek 2018).Samples obtained from small-scale beekeepers were found to have higher levels of total PA/PANOs content (above 2000 µg/kg), except one sample with 89 µg/kg.This can be a result of the broad distribution of PA producing plants, especially Echium vulgare species, in the regions of production of the given honey samples, as reported previously (Stefova et al. 2022).On the other hand, the store-bought samples contained total PA/ PANOs in the range between 719 and 3395 µg/kg.The complete data are given in Table 6.

Conclusions
Several previous studies have proposed methods for the analysis of PAs in honey using liquid chromatography coupled to mass spectrometry after extraction and clean-up procedure.A systematic study of various methods was carried out focusing on comparison of multiple extraction solvent mixtures and clean-up procedures, including new recommended combinations.An efficient method has been developed and evaluated using formic acid/methanol as an extraction solvent followed by one step clean-up by the QuEChERS method.The validated method was applied to a number of commercial and individually produced samples of honey from North Macedonia.The results reveal a relatively high content of PAs/PANOs, highlighting a potential risk for public health.For this reason, further investigations are needed to establish a quality control system for apiarists in order to reduce the content of these natural toxins in honey.
express gratitude to CEEPUS M-RO-0010-2223-165681 "Teaching and Learning Bioanalysis" Network for the mobility grants that made possible the North Macedonian-Croatian cooperation in this paper.

Figure 2 .
Figure 2. (a) Principal component analysis score plot and (b) correlation scatterplot of the variables with PC1 and PC2 based on recovery (codes in Table2).

Table 1 .
Characteristics of the analysed honey samples.

Table 2 .
Composition of extraction mixtures (v/v/%) combined with clean-up techniques.

Table 3 .
MS parameters of reference PAs/PANOs used for optimisation and validation.

Table 4 .
Validation parameters for all 15 PAs and PANOs.