Analysis of Pyrrolizidine Alkaloids in Queensland Honey: Using Low Temperature Chromatography to Resolve Stereoisomers and Identify Botanical Sources by UHPLC-MS/MS

Pyrrolizidine alkaloids (PAs) are a diverse group of plant secondary metabolites with known varied toxicity. Consumption of 1,2-unsaturated PAs has been linked to acute and chronic liver damage, carcinogenicity and death, in livestock and humans, making their presence in food of concern to food regulators in Australia and internationally. In this survey, honey samples sourced from markets and shops in Queensland (Australia), were analysed by high-resolution Orbitrap UHPLC-MS/MS for 30 common PAs. Relationships between the occurrence of pyrrolizidine alkaloids and the botanical origin of the honey are essential as pyrrolizidine alkaloid contamination at up to 3300 ng/g were detected. In this study, the predominant alkaloids detected were isomeric PAs, lycopsamine, indicine and intermedine, exhibiting identical MS/MS spectra, along with lesser amounts of each of their N-oxides. Crucially, chromatographic UHPLC conditions were optimised by operation at low temperature (5 °C) to resolve these key isomeric PAs. Such separation of these isomers by UHPLC, enabled the relative proportions of these PAs present in honey to be compared to alkaloid levels in suspect source plants. Overall plant pyrrolizidine alkaloid profiles were compared to those found in honey samples to help identify the most important plants responsible for honey contamination. The native Australian vines of Parsonsia spp. are proposed as a likely contributor to high levels of lycopsamine in many of the honeys surveyed. Botanical origin information such as this, gained via low temperature chromatographic resolution of isomeric PAs, will be very valuable in identifying region of origin for honey samples.


Introduction
Pyrrolizidine alkaloids (PAs) are secondary metabolites that comprise more than 600 compounds, typified by a pyrrolizidine base with one or more ester linkages. The 1,2-unsaturated PAs are toxic to animals and humans, causing acute and chronic liver and lung damage or cancer and are mainly produced by flowering plant species belonging to the families Asteracaeae (Compositae, tribes Senecioneae and Eupatorieae), Fabaceae (Crotolaria, Chromolaena, Lotonis), Apocynaceae (Echiteae) and Boraginaceae [1], estimated to represent 3% of the Earth's flowering plants [2]. Plants containing pyrrolizidine alkaloids are globally distributed and PAs provide a chemical defence for plants against herbivores. PA biosynthesis has been found to be dependent on many factors, with implications for plant/animal/insect interactions [3]. Various food products can contain toxic PAs either directly from plant origin (certain herbs, herbal medicines) or indirectly through natural transfer from floral nectar and pollen (e.g., some honey, pollen dietary supplements) or inadvertent cross-contamination (e.g., grains, herbs, teas) [4]. The Australian provisional tolerable daily intake of pyrrolizidine alkaloids is 1 µg/kg Bw/day [5], whilst the recommended European accepted intake is 150 times lower at 0.007 µg/kg Bw/day [6][7][8], although this has been recently revised [9,10].
Several studies have described the presence of toxic pyrrolizidine alkaloids (PAs) in honey produced by Apis mellifera. Internationally it has been reported that such toxins can be found in honey due to transfer by bees of pollen/nectar from certain flowers, including Heliotropium, Crotolaria, Echium and Senecio species. Concern has been raised as to the extent of contamination in Australian honey [11,12] but their presence in Queensland honey has not previously been examined. Previously, investigations into the presence of PAs in Australian honey have concentrated on the introduced pest plant Echium plantagineum L. (Paterson's curse) as the major likely PA source [11][12][13]. However, rigorous eradication and biological control programs in recent decades have decreased the prevalence of this weed in Australia [14]. Diverse PA containing plant species occur in Australian pastures and have intoxicated grazing horses, cattle, sheep or pigs and poultry fed contaminated grains [15,16] and include both native and introduced Crotalaria, Heliotopium and Senecio species [17]. Additionally, native Australian Parsonsia species grow in rainforests and on the margins of rainforest/eucalypt forest and contain PAs known to be sequestered by butterflies [18][19][20]. Parsonsia species spread from tropical and subtropical Asia to Australia and the south-west Pacific. Parsonsia straminea is native to Queensland and New South Wales [21] but there have been no field reports of livestock poisonings [17]. The distribution of these PA containing plants varies throughout the country and different alkaloids would be expected in honey from tropical/sub-tropical Queensland regions as compared to honey originating from southern temperate states. Given that honey represents a significant food source of human exposure to PAs [10], identification of plant PA sources to reduce this exposure is crucial.
In this study of market honey, samples have been identified with alkaloid profiles that appeared to be consistent with a number of PA containing plant species present within the Australian environment. This study examines the presence of PAs in a market survey of honey purchased in Queensland with the aim to assess any food safety concern for the consumer of honey and to correlate PAs identified with previously unsuspected plant sources of these alkaloids.

Pyrrolizidine Analysis Method Validation
Pyrrolizidine alkaloid levels in honey/plant material were quantitated by HRAM UHPLC-MS/MS analysis against 30 certified PA standards, through comparison of the precursor parent ion intensity (Table 1) to the standard curves, with squared correlation coefficients (R 2 ) typically in the range of 0.9932-0.9997. The honey analysis method was validated according to the National Association of Testing Authorities (NATA) guidance document [22]. The method was validated in blank honey, based on results for 10 spiked samples, giving Limits of Reporting (LORs) of 5 ng/g for individual PAs ( Table 2). The uncertainties given are at the 95% confidence level as required by NATA [22,23].
Note that for multiresidue analyses at these levels, a default standard uncertainty of ±25% RSD at the 95% confidence level is routinely applied by the authors and is used unless there is evidence that the actual uncertainty is greater than this value. Erucifoline, erucifoline N-oxide, jacobine N-oxide and seneciphylline N-oxide consistently gave low recoveries, resulting in high calculated standard uncertainty (% RSD) for these PAs, but which are not unusual for analyses at these levels. The EU recommends ±50% RSD unless it is demonstrated to be a bigger value [24]. The uncertainty values are calculated at the LOR because it is expected that this level will be the worst case scenario. A small peak was present in the blank for trichodesmine, explaining the higher LOD/LOR and increased uncertainty for this compound.

Alkaloid Levels Measured in Honey
Honeys purchased in supermarkets, health food shops, and from individual commercial/small-scale producers were analysed and calculated to contain pyrrolizidine alkaloids levels between <LOR (i.e., below limit of reporting) to ≈3300 ng/g of honey. Figure 1 summarises the results in a histogram, with single PA test results below the limit of reporting (5 ng/g) set equal to zero. PAs were detected in 84% of the honey samples examined (n = 465). Notably the mean total PA level of PA-positive samples (280 ng/g) was greater than the median (97 ng/g), indicating that the distribution was skewed, with a prevalence of low values (Figure 1, histogram). Whilst the prevalence of low values is reassuring, the overall distribution of total PA concentration is wide, ranging from <LOR to ≈3300 ng/g.

LC-MS/MS Separation of Alkaloids
Analysis revealed that the individual PA pattern detected by the LC-MS/MS analysis of honeys was characterised almost exclusively by lycopsamine-type PAs. In this study the lycopsamine-type PAs were represented by standards intermedine (1), indicine (2) and lycopsamine (3) (Figure 2). These diastereomeric PAs cannot be distinguished based on their MS/MS spectra [27], and Figure 3 shows the identical mass spectra obtained for standards intermedine (1), indicine (2) and lycopsamine (3) by our described HRAMS method. Given the diastereomeric nature of these alkaloids all parent MH + ions and fragment ions are identical, even with HRAMS. Separation based on retention time (RT) was therefore necessary in order to ascertain the botanical origin of PA contamination in these honeys. In most previous studies of PAs in honey, lycopsamine-type PAs were reported as the sum of unresolved stereoisomers, (including indicine (2), intermedine (1) and/or lycopsamine (3), and even the less common rinderine and echinatine) [28][29][30][31] or partially resolved stereoisomers [8,[32][33][34]. Under our initial UHPLC conditions, with a column oven temperature of 40 °C, intermedine (1) eluted separately first, but indicine (2) and lycopsamine (3) co-eluted from the Kinetex XB-C18 UHPLC column. Notably the combined indicine/lycopsamine (2/3) peak represented 75% of the alkaloids present in Queensland honey. As these two alkaloids originate from distinctly different PA plant sources, our aim was to be able to separately quantify the levels of each of these PAs in honey to enable the major plant source of PA contamination to be identified.

LC-MS/MS Separation of Alkaloids
Analysis revealed that the individual PA pattern detected by the LC-MS/MS analysis of honeys was characterised almost exclusively by lycopsamine-type PAs. In this study the lycopsamine-type PAs were represented by standards intermedine (1), indicine (2) and lycopsamine (3) (Figure 2). These diastereomeric PAs cannot be distinguished based on their MS/MS spectra [26], and Figure 3 shows the identical mass spectra obtained for standards intermedine (1), indicine (2) and lycopsamine (3) by our described HRAMS method. Given the diastereomeric nature of these alkaloids all parent MH + ions and fragment ions are identical, even with HRAMS. Separation based on retention time (RT) was therefore necessary in order to ascertain the botanical origin of PA contamination in these honeys. In most previous studies of PAs in honey, lycopsamine-type PAs were reported as the sum of unresolved stereoisomers, (including indicine (2), intermedine (1) and/or lycopsamine (3), and even the less common rinderine and echinatine) [27][28][29][30] or partially resolved stereoisomers [8,[31][32][33]. Under our initial UHPLC conditions, with a column oven temperature of 40 • C, intermedine (1) eluted separately first, but indicine (2) and lycopsamine (3) co-eluted from the Kinetex XB-C18 UHPLC column. Notably the combined indicine/lycopsamine (2/3) peak represented 75% of the alkaloids present in Queensland honey. As these two alkaloids originate from distinctly different PA plant sources, our aim was to be able to separately quantify the levels of each of these PAs in honey to enable the major plant source of PA contamination to be identified.  Ultimately, separation of indicine/lycopsamine (2/3) was achieved by simply adjusting the column temperature to 5 °C. A more complicated 'multiple heart-cutting two dimensional chromatography' method has previously been reported for the resolution of multiple PA isomer pairs [35], but in our hands the simple gradient elution at 5 °C was sufficient to achieve our desired resolution of indicine/lycopsamine (2/3). Under these conditions, of the 30 PAs and PA-NOs all were resolved based on retention time or mass fragmentation of the MS/MS except for intermedine Noxide (4) and indicine N-oxide (5) which displayed identical RT and MS/MS ( Figure 4). In plants where these N-oxides (4) and (5) are prevalent, the N-oxides could be distinguished by reduction to the corresponding parent alkaloid (2/3) which were resolved by RT under the described conditions.   Ultimately, separation of indicine/lycopsamine (2/3) was achieved by simply adjusting the column temperature to 5 °C. A more complicated 'multiple heart-cutting two dimensional chromatography' method has previously been reported for the resolution of multiple PA isomer pairs [35], but in our hands the simple gradient elution at 5 °C was sufficient to achieve our desired resolution of indicine/lycopsamine (2/3). Under these conditions, of the 30 PAs and PA-NOs all were resolved based on retention time or mass fragmentation of the MS/MS except for intermedine Noxide (4) and indicine N-oxide (5) which displayed identical RT and MS/MS ( Figure 4). In plants where these N-oxides (4) and (5) are prevalent, the N-oxides could be distinguished by reduction to the corresponding parent alkaloid (2/3) which were resolved by RT under the described conditions. Ultimately, separation of indicine/lycopsamine (2/3) was achieved by simply adjusting the column temperature to 5 • C. A more complicated 'multiple heart-cutting two dimensional chromatography' method has previously been reported for the resolution of multiple PA isomer pairs [34], but in our hands the simple gradient elution at 5 • C was sufficient to achieve our desired resolution of indicine/lycopsamine (2/3). Under these conditions, of the 30 PAs and PA-NOs all were resolved based on retention time or mass fragmentation of the MS/MS except for intermedine N-oxide (4) and indicine N-oxide (5) which displayed identical RT and MS/MS ( Figure 4). In plants where these N-oxides (4) and (5) are prevalent, the N-oxides could be distinguished by reduction to the corresponding parent alkaloid (2/3) which were resolved by RT under the described conditions.
In individual honeys, lycopsamine (3) was detected at up to ≈3100 ng/g, indicine (2) at up to 1700 ng/g, with the highest total PA content in any individual honey of ≈3300 ng/g which contained mainly a mixture of lycopsamine (3) and lycopsamine N-oxide (6).
In individual honeys, lycopsamine (3) was detected at up to ≈3100 ng/g, indicine (2) at up to 1700 ng/g, with the highest total PA content in any individual honey of ≈3300 ng/g which contained mainly a mixture of lycopsamine (3) and lycopsamine N-oxide (6).   (2) and lycopsamine N-oxide (6). In honeys where lycopsamine (3) and its N-oxide (6) were abundant these were generally the dominant PAs (>90% of PAs detected).   Similarly, in honeys where indicine (2) and its N-oxide (5) were abundant these were generally the dominant PAs (>68% of PAs detected). In order to explain the relative predominance of these diastereomeric PAs in different honeys, it was clear that we had to identify two main and distinctly different PA plant sources.

Plant Sources of Indicine (2) in Honey
An examination of the locally abundant weed Heliotropium amplexicaule (Blue heliotrope) by our LC-MS/MS method revealed that indicine (2) and indicine N-oxide (5) were the predominant pyrrolizidine alkaloids in this plant, and examination of more minor components including the newly identified helioamplexine (8) provided a unique fingerprint in the HRAM LC-MS/MS profile [26]. Interrogation of the pyrrolizidine alkaloid profile from market honey samples with high amounts of indicine (2), demonstrated that there was strong correlation between the honey PA profile and the H. amplexicaule plant alkaloid profile. The presence of both major and minor H. amplexicaule alkaloids in this honey provided strong evidence that this plant represented the floral source for this alkaloid contamination [26].

Plant Sources of Lycopsamine (3) in Honey
We similarly sought to understand the source of lycopsamine (3) (and its N-oxide (6)), the major PA observed in Queensland honey. Examination of the PAs co-occurring with lycospamine (3) and lycopsamine N-oxide (6), in the source plant would enable us to establish a unique floral PA fingerprint that could be correlated with PAs observed in honey. In past studies, Echium plantagineum L. (Paterson's curse) has been named as the source of lycopsamine (3) in Australian honey [36], despite the fact lycopsamine (3) is usually only a minor alkaloid in Echium spp. [11,28,37,38]. In fact, a previous European study noted the presence of high amounts of lycopsamine (3) (607 ng/g) Similarly, in honeys where indicine (2) and its N-oxide (5) were abundant these were generally the dominant PAs (>68% of PAs detected). In order to explain the relative predominance of these diastereomeric PAs in different honeys, it was clear that we had to identify two main and distinctly different PA plant sources.

Plant Sources of Indicine (2) in Honey
An examination of the locally abundant weed Heliotropium amplexicaule (Blue heliotrope) by our LC-MS/MS method revealed that indicine (2) and indicine N-oxide (5) were the predominant pyrrolizidine alkaloids in this plant, and examination of more minor components including the newly identified helioamplexine (8) provided a unique fingerprint in the HRAM LC-MS/MS profile [25]. Interrogation of the pyrrolizidine alkaloid profile from market honey samples with high amounts of indicine (2), demonstrated that there was strong correlation between the honey PA profile and the H. amplexicaule plant alkaloid profile. The presence of both major and minor H. amplexicaule alkaloids in this honey provided strong evidence that this plant represented the floral source for this alkaloid contamination [25].

Plant Sources of Lycopsamine (3) in Honey
We similarly sought to understand the source of lycopsamine (3) (and its N-oxide (6)), the major PA observed in Queensland honey. Examination of the PAs co-occurring with lycospamine (3) and lycopsamine N-oxide (6), in the source plant would enable us to establish a unique floral PA fingerprint that could be correlated with PAs observed in honey. In past studies, Echium plantagineum L. (Paterson's curse) has been named as the source of lycopsamine (3) in Australian honey [35], despite the fact lycopsamine (3) is usually only a minor alkaloid in Echium spp. [11,27,36,37]. In fact, a previous European study noted the presence of high amounts of lycopsamine (3) (607 ng/g) compared to low amounts of echimidine (7) (15 ng/g) in imported Australian honeys, and postulated an unknown plant source as a possible interpretation [27]. Indeed our analysis of E. plantagineum revealed that after Zn reduction echimidine (7) and echiumine were the dominant PAs, with both lycopsamine (3) and intermedine (1) present in much lower quantities. Clearly E. plantagineum is not the major source of lycopsamine (3) seen in our Queensland honeys, which is also consistent with the more temperate distribution of this species within Australia [38]. Other species/genera known internationally to contain lycopsamine (3) (and intermedine (1)) include Anchusa off., Borago off., Lithospermum spp., and Symphytum spp., and Eupatorium spp. [39], and are generally not geographically distributed within Australia [40]. They can logically be excluded as potential lycopsamine (3) floral sources.
When considering PA species which are known to be prevalent in Queensland, both Ageratum and Aminscka spp. have been reported to contain lycopsamine (3). Ageratum conyzoides for example has been reported to contain lycopsamine (3) and echinatine [41,42] or lycopsamine (3) and 3 -O-acetyllycopsamine [43]. A targeted screen by Avula reported lycopsamine (3) and its N-oxide (6) as the two major PAs, together with minor amounts of dihydrolycopsamine, dihydrolycopsamine N-oxide and echinatine [1,44]. The closely related Ageratum houstonianum is locally abundant in Queensland, and our analysis of Zn reduced plant extract revealed the predominance of retrohoustine, heliohoustine and tentatively echinatine (ratio 2.7:1.7:1 respectively), with much lower amounts of lycopsamine (3) and intermedine (1) (data not shown). This result is consistent with analysis of this same species from Mexico that showed that lycopsamine (3) was not the predominant pyrrolizidine alkaloid present with three other pyrrolizidine alkaloids (retrohoustine, heliohoustine and isoretrohoustine) isolated in greater amounts than lycopsamine (3) [45]. Lycopsamine (3) and intermedine (1) have also been identified in Amscinckia spp. [46], with NMR analysis revealing the relative proportion of intermedine (1) to lycopsamine (3) varied from roughly 2:1 to 1:2 in A. intermedia, A. hispida, and A. lycopsoides. Amsinckia spp. are however regionally controlled as noxious weeds in Australia, and not likely to be a widely abundant PA sources in Queensland. The invasive aquatic weed Gymnocoronis spilanthoides has been recently been shown [47] to contain predominantly lycopsamine (3) followed by intermedine (1), however, this species is also controlled by government eradication programs. None of these plant species matched either the predominant lycopsamine (3) profile observed in our Queensland sourced honey or the regional abundance of plant species.
Historically lycopsamine (3) was identified in the hair pencil of Australian danaid butterflies in Queensland in a region where Amsinckia plants are rare [48]. An examination of the native vines Parsonsia straminea (family Apocynaceae) and Parsonsia eucalyptophylla, by these authors revealed the presence of lycopsamine (3) and intermedine/indicine (1 or 2), and acetyl derivatives. As native Parsonsia species occur widely in Queensland this species was deduced as the source of lycopsamine (3) in danaid butterflies [49]. Lycopsamine-type PAs have been identified in a number of species in Apocynaceae [50].
Interestingly, in a study of butterfly food plants, a comparison of Parsonsia straminea flowers revealed the ratio of lycopsamine N-oxide (6) to intermedine N-oxide (4) to other alkaloids of 98:1:1. By contrast, Ageratum sp. gave a predominance of two M+ 269 isomers compared to lycopsamine (3) (45:48:1) [51]. Evidently, lycopsamine (3) and intermedine (1) and their N-oxides are present in a wide variety of plant species, but we sought to identify an origin for the almost exclusive predominance of lycopsamine (3) (and its N-oxide (6)) and these literature reports of Parsonsia provided the best clue.

Pyrrolizidine Alkaloids Determined in Parsonsia Vines
Local Parsonsia straminea (Qld Herbarium ID AQ522465) was collected and re-examined for PA content using our described HRAM LC-MS/MS method. The plant pyrrolizidine alkaloids were present primarily as the N-oxides (96% in the leaves and stems, 99% in the pods, 93% in the nectar and 80% in the pollen). The plant pyrrolizidine alkaloids were analysed with and without reduction by Zn to enable comparison with the honey alkaloids (primarily free alkaloids) as previously observed [25,52]. The SCX SPE methodology was previously demonstrated to be suitable for plant extracts [36]. The investigations aimed to determine for the first time whether and to what extent PAs found in honey are sourced from Parsonsia straminea (or closely-related Parsonsia species, a number of which are widespread in coastal regions of eastern Australia [53]). High resolution accurate mass (HRAM) data, combined with RT comparison with pyrrolizidine alkaloids standards enabled identification of the major pyrrolizidine alkaloids in P. straminea (Table 3). In the P. straminea nectar, the ratio of lycopsamine (3) and its N-oxide (6) to intermedine (1) and its N-oxide (4) was >45-50:1, in the flowers it was 78:1, in anthers/pollen >50:1, in the pods it was >50:1, whilst in the leaves,~3:1.
Minor peaks after reduction were tentatively identified by analysis of the HRAM data (Table 4) and corresponded to tessellatine (9) or isomer (a C7 isomer, found 300.1801, calculated for C 15  . The corresponding N-oxides were found in the non-reduced plant extract. Tessellatine (9) has the same necic acid as lycopsamine (3) but is esterified at the C7 necine position rather than C9 as seen in lycopsamine (3). The C7 esterification is evidenced in the predominant (base peak) fragment ion m/z 156.1019 (calculated for C 8 56]. Similarly the two helioamplexine isomers had identical MS to that seen in helioamplexine (8) (the C-6 homoanalogue of indicine) [25], and these components present in P. straminea which did not co-elute with helioamplexine were deduced to be the corresponding C-6 homoanalogues of lycopsamine and intermedine.   1 and 3, 4 and 6. A 3 -glucopyranosyl 2,3-dihydro-1H-pyrrolizin-1-one derivative has previously been reported from Cynoglossum gansuense [57]. Additionally, five isomeric components with MH + 286.1649 were also detected in P. straminea reduced extracts, with MS data consistent with these being desmethyl analogues of lycopsamine, i.e ideamine A (10) isomers (four esterified at C9, one at C7). Ideamine A N-oxide has previously been found in insects feeding on Parsonsia laevigata leaves [58,59]. Tessellatine (9), 3 -O-acetyl-and 7-O-acetyllycopsamine/intermedine and their N-oxides have been previously identified in Amsinckia or Cryptantha species [54,55,60,61]. To positively identify the PAs in lycopsamine-rich honey samples as originating from Parsonsia straminea, we sought to find some of these same minor PA components of this plant in honey.

Honey PA Profiles Linked to P. straminea
The detection of minor alkaloids in Parsonsia straminea provides a distinctive PA fingerprint in its HRAM LC-MS/MS profile, albeit in minor quantities compared to the major alkaloid lycopsamine (3). By comparison with the PA profile observed in market honey samples, there is clear evidence that this plant species is being used as a honey floral source by bees (Figure 7). Honey samples such as H-PA#146 and H-PA#157 were independently purchased. When these honey samples were analysed against the 30 PA standards in our screen (Table 1), only the major alkaloid lycopsamine (3) and lesser intermedine (1) (and their N-oxides) were detected (Table 3). Characteristic major/minor components present in certain honeys (Table 4) included in addition to lycospamine and intermedine, the helioamplexine isomers at RT 8. Honey samples such as H-PA#146 and H-PA#157 were independently purchased. When these honey samples were analysed against the 30 PA standards in our screen (Table 1), only the major alkaloid lycopsamine (3) and lesser intermedine (1) (and their N-oxides) were detected (Table 3). Characteristic major/minor components present in certain honeys (Table 4) included in addition to lycospamine and intermedine, the helioamplexine isomers at RT 8.20, 8.44 and 9.38 min ( Figure 8) and putative 3 -O-acetylintermedine (8.91 min) and 3 -O-acetyllycospamine (9.64 min). The tentatively assigned 3 -O-glucosylintermedine and 3 -O-glucosyllycopsamine were also identified in these honey samples with the MS 2 spectra observed identical to that found in the plant pods and nectar. Non-toxic dihydrolycopsamine isomers were also identified in the plant and honey. Due to the low levels of these minor PAs in the plant, they were seen most readily in honey samples highest in lycopsamine (3)

Plant Origins of PAs in Honeys Surveyed
Of the 30 PA standards utilised in our survey, fifteen PAs (50%) were not detected in any of the market honey samples ( Table 5). As shown in Table 5, based on profiles of alkaloids identified, most of the honey PAs were likely sourced from Parsonsia straminea or Heliotropium amplexicaule, with honey containing Parsonsia alkaloids being dominant in lycopsamine (3) (up to 3100 ng/g) and honey containing Heliotropium amplexicaule alkaloids dominant in indicine (2) (up to 1700 ng/g). PAs sourced from Echium plantagineum were much lower, with the dominant PA detected being echimidine (7) (up to 260 ng/g) in agreement with previous studies [11,63]. Even lower levels of PAs from Heliotropium europaeum (containing lasiocarpine, heliotrine and europine [11,63] (and their N-oxides)) and Senecio species (most likely Senecio madagascariensis) [64] were detected (Table 5). Of course, many of the honey samples are ascribed by their label to particular non-PA producing floral sources, so the observation of PAs in these honeys is a product of either the natural foraging of bees on different available plants, or the blending of honeys in the packaging process. This coforaging/blending is also evident in honey samples that show co-occurrence of pyrrolizidine alkaloids from multiple floral sources, for example, honeys containing indicine (2) (from H. amplexicaule) and lycopsamine (3) (likely from P. straminea due to lack of the dominant PA echimidine (7) as present in E. plantagineum). Both these sets of PAs were present in significant levels in H-PA#11, 32, 216, 630 and 642. Geographically both the low-growing heliotrope, H. amplexicaule, and the arboreal vine, P. straminea, can co-occur in sub-tropical coastal regions of Queensland [54,65], so the co-occurrence of their respective alkaloids in honey would seem logical if both plants are visited by foraging bees within the same landscape. The high abundance of alkaloids from these quite different plant species in honey suggests that both are attractive to foraging bees, and where possible both species should be avoided when siting honey hives. It is apparent that the 'standard set' for PA/PANO testing of honeys varies depending on the natural flora of the region, as well as the cultivated plants present. In this study erucifoline, jacobine, monocrotaline, senciphylline, or their corresponding N-oxides and senkirkine or trichodesmine were not found in the honey tested, which is a considerably different result to those found recently in Schleswig-Holstein region of Germany [29].

Plant Origins of PAs in Honeys Surveyed
Of the 30 PA standards utilised in our survey, fifteen PAs (50%) were not detected in any of the market honey samples ( Table 5). As shown in Table 5, based on profiles of alkaloids identified, most of the honey PAs were likely sourced from Parsonsia straminea or Heliotropium amplexicaule, with honey containing Parsonsia alkaloids being dominant in lycopsamine (3) (up to 3100 ng/g) and honey containing Heliotropium amplexicaule alkaloids dominant in indicine (2) (up to 1700 ng/g). PAs sourced from Echium plantagineum were much lower, with the dominant PA detected being echimidine (7) (up to 260 ng/g) in agreement with previous studies [11,62]. Even lower levels of PAs from Heliotropium europaeum (containing lasiocarpine, heliotrine and europine [11,62] (and their N-oxides)) and Senecio species (most likely Senecio madagascariensis) [63] were detected (Table 5). Of course, many of the honey samples are ascribed by their label to particular non-PA producing floral sources, so the observation of PAs in these honeys is a product of either the natural foraging of bees on different available plants, or the blending of honeys in the packaging process. This co-foraging/blending is also evident in honey samples that show co-occurrence of pyrrolizidine alkaloids from multiple floral sources, for example, honeys containing indicine (2) (from H. amplexicaule) and lycopsamine (3) (likely from P. straminea due to lack of the dominant PA echimidine (7) as present in E. plantagineum). Both these sets of PAs were present in significant levels in H-PA#11, 32, 216, 630 and 642. Geographically both the low-growing heliotrope, H. amplexicaule, and the arboreal vine, P. straminea, can co-occur in sub-tropical coastal regions of Queensland [53,64], so the co-occurrence of their respective alkaloids in honey would seem logical if both plants are visited by foraging bees within the same landscape. The high abundance of alkaloids from these quite different plant species in honey suggests that both are attractive to foraging bees, and where possible both species should be avoided when siting honey hives. It is apparent that the 'standard set' for PA/PANO testing of honeys varies depending on the natural flora of the region, as well as the cultivated plants present. In this study erucifoline, jacobine, monocrotaline, senciphylline, or their corresponding N-oxides and senkirkine or trichodesmine were not found in the honey tested, which is a considerably different result to those found recently in Schleswig-Holstein region of Germany [28].
a Not resolved-indicine N-oxide and intermedine N-oxide co-eluted. b,c Intermedine and intermedine N-oxide are present in multiple plants, and prominent in both Parsonsia straminea and Heliotropium amplexicaule. d Helioamplexine was quantified using heliotrine standard curve. e Helioamplexine N-oxide was quantified using heliotrine N-oxide standard curve. f Lycopsamine observed in honey containing echimidine not necessarily attributed solely to E. plantagineum, but of the 93 honey samples containing echimidine, the concentration of lycopsamine was lower than echimidine in 76 honeys, which is consistent with the relative amounts observed in E. plantagineum. In the other 17 honeys which contain echimidine, it is likely that there is more than one source of lycopsamine. g Of the 76 honeys in which lycopsamine was at a lower level than echimidine, 37 honey samples also contained intermedine.

Honey as a Dietary Source of Pyrrolizidine Alkaloids
Major supermarket honeys by comparison represent blended honeys from diverse locations, some of which attributed the specific floral source and in general contained only low levels of PAs. It has been observed previously that blended retail honeys had a lower PA content, but that PAs were present in more samples [65]. In this study, for supermarket honeys (n = 129), PAs were detected in 84% of honeys, and showed highest total PA levels of 1400 ng/g. For supermarket honeys, the mean total PA level of PA-positive samples was 120 ng/g and the median level was 61 ng/g. Certain small producer honeys displayed the highest levels of pyrrolizidine alkaloids, with the PA content dependent on the location and attractiveness of PA containing plants to foraging honey bees. Paradoxically, even though analysed PA content of small producer honeys range from <LOR to an alarming 3000 ng/g, if equal amounts of each of these 205 small producer honeys were blended, the hypothetical resultant mixed honey would have a PA content of only 240 ng/g (i.e., the average PA content of all of these 465 honeys).
It has been observed previously in South American honeys that raw honeys showed greatest variety due to the availability of PA containing plants near to hives [65].
The cumulative toxicity of the 1,2-unsaturated PAs have been demonstrated in animal studies and genotoxicities/tumorigenicities were induced by hepatic metabolism of PAs [66]. Consequently, provisional tolerable daily intakes (PTDI) have been recommended to control the human consumption of PAs [5,10,39,67].
Using the Australian FSANZ provisional tolerable daily intake (PTDI) of 1 µg/kg BW/day, 0% of honeys tested (total n = 465) exceeded the limit for a 70 kg adult consuming 20 g of honey per day, but 19% of honeys tested exceeded the limit for a 15 kg child consuming 50 g of honey per day. Applying the lowest recommended PTDI (EFSA, COT, BfR) of 0.007 µg/kg BW/day, 63% of honeys tested exceeded the limit for a 70 kg adult consuming 20 g of honey per day and 84% of honeys tested exceeded the limit for a 15 kg child consuming 50 g of honey per day.
The PA content of honey samples varies with geographical location and climate, determined by the type and distribution of PA containing plants and by the propensity for bees to forage on these plants [65,68]. Lycopsamine (3) and intermedine (1) are present in many PA-producing plants, with the knowledge of the plants distributed in Australia and the ratio to other PAs present, it is likely that Parsonsia straminea is a major contributor to the high PA levels observed in certain honeys in this study. Of course, it is possible that there is more than one PA source of lycopsamine (3), with a small portion of lycopsamine contamination of honey potentially originating from Echium plantagineum and Ageratum houstonianium. Also, there are likely other PA containing plants that have not been considered. It is also possible that not all PAs present in honey have been identified by comparison with standards and by analysis of the top MSMS. Despite the observation that of the PAs tested in experimental rats, lycopsamine (3) induced the lowest levels of liver DNA adducts (formed from PA derived reactive pyrrolic metabolites), PA containing plants are the most common poisonous/carcinogenic plants affecting livestock, wildlife and humans [69]. Beekeepers are advised to avoid these known plant genera around the hive/apiary as much as possible to reduce PA contamination in honey.

Conclusions
The HRAM LC-MS/MS method for pyrrolizidine alkaloid analysis described here enables the ready resolution of isomeric alkaloids of the lycopsamine-type. The described simple adjustment of column conditions to a lower temperature was effective in resolving the problematic pairs of indicine/lycopsamine alkaloids present in Australian honey. This resolution has enabled us to identify Parsonsia vines as a previously unsuspected source of PA contamination in Australian honey. Low temperature chromatographic resolution may have as yet unexplored application in resolving other similar diastereomeric pyrrolizidine alkaloid isomers, of which there are many within the known pyrrolizidine alkaloids, many of which do not have commercially available standards.

Chemicals and Solvents
In total, 30 pyrrolizidine alkaloid standards were utilized in a high resolution accurate mass (HRAM) LC-MS/MS screen.

Honey Samples
Honey samples (465 in total) were purchased between September 2016 and December 2017 directly from Queensland supermarkets, fruit shops, local markets, and producers.

Honey Alkaloid Extraction
Honey samples (1 g) were dissolved in aqueous H 2 SO 4 (0.05 M, 10 mL) centrifuged and the supernatant applied to preconditioned Agilent SPE Bond Elut 100 mg LRC-SCX columns (Agilent Technologies, Folsom, CA, USA). SPE cartridges were washed with water (10 mL) and methanol (10 mL), and pyrrolizidine alkaloids were then eluted with 3% ammonia in methanol (3 mL). The eluate was evaporated to dryness under nitrogen, and the residue reconstituted in 5% methanol in water (1 mL) for HRAM LC-MS/MS analysis.

Honey Method Validation
The validation of the method was conducted according to the National Association of Testing Authorities (NATA) guidance document [22]. The method was validated in 3 blank honeys, and based on results for 10 spiked samples at a spiking level of 5 ng/g (Table 2), 10 blank samples and 10 non-extracted spike samples and the recoveries determined. Limit of detection (LOD) was calculated as 3s. Limit of quantitation was calculated as 9s. Limit of reporting was set at the levels the samples were spiked, also the level of the lowest standard used for the calibration curve. The uncertainties given are at the 95% confidence level as required by the NATA [22,23]. Replicate samples were prepared for every tenth honey test sample to assess reproducibility. The difference between replicate samples (coefficient of variance %) was typically 0.12-6.7%. High samples were diluted to levels within the calibration curve and re-run. SPE wash steps and further elutions with 3% ammonia in methanol (3 mL) were analysed for residual PAs and the extraction was found to be exhaustive. Table 2 shows good recoveries for most PAs.

Plant Source
Parsonsia straminea was collected from a suburban area in the south of Brisbane and was taxonomically identified by the Queensland Herbarium, with a voucher specimen (AQ522465) incorporated into their collection. The Parsonsia straminea foliage sample was a collection of stems and leaves, and was freeze dried, milled and stored frozen prior to analysis. Pods were collected separately and freeze-dried, milled and frozen. Flowers were sampled as both intact flowers (freeze-dried, milled and frozen) or utilized to provide nectar and pollen separately. Nectar was separated from flowers using a microcap capillary, and anthers and pollen were separated from other flower plants with tweezers and desiccated. electrospray ionisation (ESI) with a spray voltage of 3500 V and a vaporiser temperature of 400 • C. MS analysis run with arbitrary pressures of sheath gas 48, aux gas 11, sweep gas 2, spray voltage 3.5 kV, capillary temperature of 320 • C, auxiliary gas heater at 350 • C and used full scan/dd-MS 2 mode. Full scans were conducted at a resolution of 70,000 FWHM (at m/z 200), with an AGC target of 1.00 × 10 6 . The maximum time of accumulating ions per scan event was 10 ms with a scan range of 75-1125 m/z. Data dependent acquisition (dd-MS 2 ) was conducted at a resolution of 17,500 with an AGC target of 1.00 × 10 6 . The maximum time of accumulating ions per scan event was 50 ms. Normalized collision energy (nce) was set to 50% and an isolation window of 1.0 m/z was utilized. Dynamic exclusion was set to 3 s preventing subsequent triggering of the same ion in data dependent scans. A maximum of 5 most abundant precursors could be selected for dd-MS 2 per scan event.
Pyrrolizidine alkaloid levels in honey/plant material were quantitated against certified PA standards, with calibration curves obtained for each of the 30 pyrrolizidine alkaloid standards injected at 5, 10, 20, 50 and 100 and 200 ppb (in duplicate/triplicate). Honey or plant extracts were analysed by HRAM LC-MS/MS to detect pyrrolizidine alkaloids and their N-oxides by matching of retention time with the corresponding standard and identified by their precursor parent ion (M+H + ) and confirmed by the detection of product ions (Table 1). The identity of these and further alkaloids was assigned by use of the high resolution accurate mass data provided by the Q Exactive mass spectrometer, enabling the determination of elemental composition of parent and fragment ions (Tables 3 and 4).

Conflicts of Interest:
The authors declare no conflict of interest.