Diacetin, a reliable cue and private communication channel in a specialized pollination system

The interaction between floral oil secreting plants and oil-collecting bees is one of the most specialized of all pollination mutualisms. Yet, the specific stimuli used by the bees to locate their host flowers have remained elusive. This study identifies diacetin, a volatile acetylated glycerol, as a floral signal compound shared by unrelated oil plants from around the globe. Electrophysiological measurements of antennae and behavioural assays identified diacetin as the key volatile used by oil-collecting bees to locate their host flowers. Furthermore, electrophysiological measurements indicate that only oil-collecting bees are capable of detecting diacetin. The structural and obvious biosynthetic similarity between diacetin and associated floral oils make it a reliable cue for oil-collecting bees. It is easily perceived by oil bees, but can’t be detected by other potential pollinators. Therefore, diacetin represents the first demonstrated private communication channel in a pollination system.

Occurrence of EAD-active compounds. The most widespread EAD-active compound was diacetin, which occurred in 41 of the 50 (82%) studied oil species, but in only one of the eight (12.5%) related non-oil species examined. It was present in all of the Holarctic (seven) and South African (18) oil species, as well as in 16 (73%) of the Neotropical oil species (Table 1, for complete list see Table S1). Nearly as widespread as diacetin was 2-tridecanone, which was found in 34 (68%) of the oil species and in one non-oil species. Heptanoic acid was detected in 20 (40%) and 4-hydroxy-3-methoxystyrene in 18 (36%) of the oil species, whereas the remaining EAD-active compounds occurred in less than 10 (20%) of the oil secreting species.
Behavioural experiments. In two-choice experiments conducted in the flight cage, diacetin alone attracted significantly more bees than did a negative control, but significantly less bees than did a natural floral extract (Fig. 3). However, the creation of a synthetic mixture with diacetin and four additional EAD-active compounds increased the attractiveness to the same level as the natural floral extract (see also supplementary data, Movie 1).
Bees responded differently to samples (natural extract, reduced synthetic mixtures) that were tested against the complete synthetic mixture (Fisher's exact test: P = 0.01). Removal of geranic acid or 2-tridecanone from the mixture had no effect on attractiveness to M. fulvipes females, but removal of heptanoic acid or (E)-2-dodecenal reduced the attractiveness significantly relative to the complete synthetic mixture (Fig. 3). When diacetin was removed from the mixture (together with geranic acid, see material and methods) bees were attracted only to the complete mixture (Fig. 3).

Discussion
Our data demonstrate that flowers of most of the studied oil species around the world emit the fatty acid derivative diacetin. This compound elicits strong antennal responses in oil bees from different floristic regions and continents, but it does not elicit antennal responses in related non-oil bees. This suggests an olfactory adaptation in oil bees to this uncommon compound. Diacetin is a key signal in the Lysimachia-Macropis pollination system, but other compounds can also add to the attractiveness of a scent blend. Overall, our data suggest that diacetin is a private communication channel and honest signal in the oil flower/oil bee pollination system. Diacetin, only recently described as a floral compound 39 , occurs as a floral scent constituent in most (82%) of the oil plant species tested, regardless of floristic region (Holarctic, Neotropical, Afrotemperate region) or plant lineage (Asparagales, Malpighiales, Ericales, Lamiales). These findings strongly suggest, therefore, that the production of diacetin in oil flowers has evolved independently several times, in accordance with the independent evolution of oil secretion in these flowers 18,40 .
In contrast to the widespread occurrence of diacetin in oil plants, we did not find diacetin in related non-oil species with one exception. In the non-oil secreting Lysimachia thyrsiflora, the sister species of  Table S1).  oil secreting and diacetin emitting L. vulgaris 40 , this compound was detected in flower extracts and more recently also in headspace samples. The presence of diacetin in this non-oil species may be the result of a recent switch away from pollination by oil bees yet with retention of the ability to produce small quantities of floral oil and diacetin due to relaxed selection against its production 41,42 . Diacetin has not been found in dynamic headspace collections from several oil species 25,43 , even though we identified it in solvent extracts of flowers of these same species. This suggests that diacetin is present only in small and hard to detect amounts in floral headspace samples (see also below). Interestingly, diacetin has not been identified in studies focusing on the chemistry of the floral oils 15,31 . We attribute this to its smaller size and higher volatility compared to the target non-volatile oils, and a methodology that did not allow its detection. The amount of diacetin available in the samples was quite small compared to the oils and this small amount may have been lost in the process of evaporating the "oil samples" to dryness.
The basic structure of floral oils (i.e. acylglycerols) is similar for the oil species found around the world and resembles that of the volatile compound diacetin as well as some lipids in plant tissues 31 . As exemplified by L. punctata, major compounds in the floral oil are 1-[(3R)-acetoxystearoyl]-2-acetylglycerol and 1-[(3R)-acetoxystearoyl]-3-acetylglycerol and both of these compounds are composed of a glycerol esterified with one acetic acid and with one substituted long-chain fatty acid (Fig. 4). Structural similarities of these two compounds with 1,2-and 1,3-diacetin are evident. It can be assumed that metabolic pathways or enzymes utilized, such as those involved in ester formation of glycerol with fatty acids 44 (specifically 3-hydroxy/3-acetoxy fatty acids) or acetic acid, are to some extent identical for this group of lipids and for diacetin production (Fig. 4).
Since acetylation of glycerol or the backbone of the hydroxylated long chain fatty acids is almost universal in "non-volatile" floral oils 30,31 , it can be hypothesized that diacetin might be present in all oils of this type, whereas it may not be present in oils made up of other types of lipids (e.g. free fatty acids, classical triglycerides or wax esters, terpenoids). Indeed, we found diacetin in all plants having oils congruent with these criteria with the exception of Momordica (Cucurbitaceae) and Bunchosia (Malpighiaceae) species. Diacetin was also missing from Nierembergia (Solanaceae) species, but their oils do not consist of acetylated glycerols (Table S2). The common occurrence of diacetin with 'acetylated' floral oils supports the idea that these compounds are derived from the same metabolic pathway or, at least partially, rely on the same enzymatic endowment. Even if diacetin evolved initially as a by-product of oil synthesis and was subsequently co-opted by oil bees as an "index signal" 28,29 , it still represents a reliable cue for bees looking for floral oils. Parallels can be drawn to a communication system between male and female rattlebox moths (Utetheisa ornatrix). Here, the males use a volatile derivative of a larger defence compound, to indicate to the female the quality of the sequestered defence compounds that they pass along to the female during mating 45 .
Our data show that diacetin is widespread among oil species and a good candidate for use by oil bees around the world as a reliable cue for locating oil rewards. They also indicate that diacetin represents a private communication channel between oil plants and oil bees. In our electrophysiological measurements, diacetin elicited antennal responses in melittid bees from both Europe (M. fulvipes) and South Africa (R. neliana). It also elicits responses in another European Macropis species, M. europaea Warncke, and two additional South African Rediviva species, R. brunnea Whitehead & Steiner, and R. pallidula Whitehead & Steiner (Dötterl and Steiner, unpublished data). Diacetin did not, however, elicit significant antennal responses in the closely related non-oil melittid bee (M. haemorrhoidalis) or the honey bee (A. mellifera, Apidae). This difference in antennal response to diacetin between oil and non-oil bees demonstrates that the oil bees have specific olfactory adaptations in the periphery of the olfactory circuit to detect diacetin. This adaptation functions most likely at the level of the olfactory receptors or the olfactory binding proteins 11,46,47 , but additional adaptation in the brain (e.g. processing) cannot be excluded. Such adaptations towards volatile signals of host plants have not been described for any other pollinators and our next step will be to test whether oil bees belonging to the Apidae exhibit a similar positive response to diacetin.
Our bioassays with M. fulvipes and the EAD-active scent compounds of its host plant L. punctata point towards a key function of diacetin in host plant location. The presence of diacetin alone was sufficient to attract Macropis bees. Two other EAD-active compounds (heptanoic acid, (E)-2-dodecenal) were also behaviourally active. However, a mixture containing these two and two additional EAD-active compounds (2-tridecanone, geranic acid) but lacking diacetin did not attract bees when tested against a synthetic mixture that contained all compounds (Fig. 3).
Trace amounts of diacetin were found as a contaminant in our synthetic geranic acid sample and, therefore, we had to exclude geranic acid from our mixture in order to obtain a diacetin-free sample for the choice tests. These trace amounts proved sufficient to elicit behavioural responses in Macropis, because a synthetic mixture without diacetin but with geranic acid attracted Macropis bees (Schäffler, unpublished data). When removing only geranic acid from the complete synthetic mixture the bees did not discriminate between the depleted and the complete mixture, demonstrating not only that geranic acid has no influence on bee behaviour, but also that the absence of trace amounts of diacetin (when higher amounts are still present) did not influence the choice of bees. Overall, we conclude that diacetin and not geranic acid was responsible for the loss of attractiveness relative to the complete scent mixture when excluding both substances from the complete mixture. This confirms that diacetin is a key compound in attracting Macropis.
In addition to diacetin, heptanoic acid and (E)-2-dodecenal are used by Macropis bees for locating oil flowers. Heptanoic acid was detected in about 20 oil species in three floristic regions, and recently in a few oil and non-oil species 43,48 . The only other reported instance of a biological function for the compound is as a kairomone in an insect host-parasite communication system 49 . Even rarer is the floral scent compound (E)-2-dodecenal that was found in three floral oil secreting Lysimachia species. Until now, this compound was known from only a few South African oil orchids 43 and from two species without floral oils 50,51 , and from a millipede where it acts as an insect deterrent 52 . In contrast, the EAD-active compound 2-tridecanone is very widespread among our oil species studied, among a large number of oil orchids of South Africa 43 , and among several non-oil species 48,50,51 , yet it did not influence the attractiveness of the synthetic mixture. This compound, known as a repellent for insects 53 including generalized bee pollinators 54 , could act as a floral filter 39,43 at least in the Macropis-Lysimachia pollination system to reduce visitation rates from inappropriate visitors that might remove pollen without providing adequate pollination.
Interestingly, while diacetin is very widespread among oil plants, the plants emit additional scent compounds, several of which are not widespread and do not occur in more than one or a few of the species studied 39,43,51 . There is a high overall variation in floral scent among oil plants, which is true for species within floristic regions and even for species pollinated by the same oil bee species (Holarctic: 39 , South Africa: 43 ) as well as among floristic regions. These findings lead us to believe that diacetin is a reliable volatile marker for 'non-volatile' fatty oils throughout the world, whereas the emission of other compounds, like geranic acid, 3,5-dimethoxytoluene, or (E)-2-dodecenal, may be important for allowing bees to discriminate among co-blooming species. Scents distinguishable among plant species are known to promote effective pollen transfer within species and species integrity through flower constancy of pollinators 55  However, we did not find diacetin in all of the floral oil species suggesting that they may emit diacetin in amounts too low for detection or that diacetin is not used as a signal by these plants. If the latter is true, compounds other than diacetin may occasionally be important as pollinator attractants. Since oil production has evolved independently in several families and some plants produce floral oils structurally dissimilar to diacetin 31 , it would not be altogether surprising if some oil species used different signals.

Conclusion
Diacetin occurs in several floral oil plants around the world and is detected by oil bees from at least two continents. It allows the Holarctic Macropis oil bee, and probably other oil bees, to rapidly and efficiently locate their oil secreting host plants. Our data for Macropis and Lysimachia suggest that diacetin represents the first demonstrated private communication channel between a pollinator and its host plant. Notably, diacetin satisfies the two requirements of a private communication channel: 1) it is an uncommon compound and 2) it can be detected by its specialized and specific pollinators, but apparently not by other potential pollinators in the environment. We cannot rule out the possibility that one or more of the thousands of non-oil bees that we didn't test may be able to detect diacetin, yet, there seems little selective value in evolving or retaining such an ability outside of an oil flower/oil bee relationship. Dated phylogenies show that Lysimachia and Macropis are of similar age making it plausible that they coevolved from the onset 18 . Thus, the described fine-tuned adaptation towards diacetin in the sensory apparatus of the bees and the chemical profile of the host plants may be the result of coevolutionary processes. The obvious sharing of the biosynthetic production by diacetin and floral oils, at least those with acetylation, make diacetin an ideal and reliable cue for oil bees.

Materials and Methods
Bee study species. The oil bee Macropis fulvipes (Fab.) (Melittidae, Melittinae) is distributed in Europe and, like other Macropis species, is specialized on the oil secreting flowers of Lysimachia species (Primulaceae) 15,56 . Fatty floral oils and pollen of these plants are the only food collected by adult females for the offspring. Adult males and females feed on pollen of Lysimachia and females use the oil to line the brood cells 15,19 . Individuals used for behavioural tests were from a flight cage population 19 (see below) and Lysimachia-naive, while those used for electrophysiological measurements (see below) were from a natural population in the Ecological Botanical Garden of the University of Bayreuth (EBG) and likely Lysimachia-experienced.
Rediviva (Melittidae, Melittinae) oil bees are closely related to Macropis, occur in Southern Africa, and also collect floral oils as food for the offspring 21 . Rediviva neliana Cock. is widespread in the summer rainfall area 57 . Specimens for electrophysiological measurements (see below) were collected in the Witsieshoek region of the Drakensberg while visiting oil or nectar/pollen plants.
Melitta bees which occur in the Holarctic and in Africa are from the same subfamily as Macropis and Rediviva, i.e. Melittinae, but species do not collect floral oils. Melitta haemorrhoidalis (Fab.) is distributed in Europe and is specialized on pollen of Campanula species 56 . Specimens for electrophysiological measurements were collected from natural populations in the EBG and served as a phylogenetic control.
The non-oil honey bee, Apis mellifera L., originally native to Europe and Africa, now occurs throughout the world and, in contrast to the other bee species used, belongs to the Apidae. It is among the most generalist bees and therefore is expected to have the capacity to detect a large array of scent compounds. Individuals used for electroantennographic measurements were collected in the EBG from established hives.

Plant material and volatile collection.
Floral scents for chemical analyses were collected from 58 plant species (50 oil and 8 non-oil) from different geographic regions and phylogenetically disparate plant families and genera (supplementary data, Table S1). Samples of four of these oil species were additionally used for electrophysiological analyses, and samples of L. punctata were additionally used for bioassays. Samples were either collected from plants growing in the natural habitat or from material collected in different greenhouses (supplementary data, Table S1). Flowers were removed from the plants using clean forceps and extracted for one minute in 2-3 ml pentane (p.a., 99%, Grüssing, Germany). The 106 obtained samples were subsequently filtered with silanized glass wool (Supelco) to remove particles and concentrated by evaporation under a gentle stream of nitrogen to a volume of 0.5 ml. The solvent extracts of leaves were used as negative controls.

Gas Chromatography with Electroantennographic Detection (GC-EAD). Both sexes of M. ful-
vipes were used because we did not want to use too many females from the small populations of Macropis and did not find differences in antennal responses between sexes in previous analyses 26 . Such potential differences were not expected to occur as both sexes visit Lysimachia flowers. Similar to females, males feed on pollen of the flowers after hatching, and throughout their life search for females on the flowers 19,20 . Antennae were tested using scent samples of four different oil species from three different plant orders (Ericales, Lamiales, Asparagales) and two different continents (Europe, Africa). By using this approach, compounds could be identified that are widespread among oil plants (phylogenetically independent) and potentially important in the oil flower/oil bee pollination system. Five Lysimachia punctata flower extracts (from different plants) were tested on antennae of 7 male and 6 female bees (one antenna Scientific RepoRts | 5:12779 | DOi: 10.1038/srep12779 per bee). Additionally, one flower extract of L. congestiflora Hemsl. and one of Diascia integerrima E.Mey. ex Benth. were tested on the antennae from two different males (one antenna per bee), and the flower extract of Corycium dracomontanum Parkman & Schelpe was tested on one male antenna.

Electroantennography (EAG).
For the EAG tests we used five antennae from M. fulvipes (all female), six antennae from R. neliana (five males, one female), five antennae from Melitta haemorrhoidalis (all female), and nine antennae from honey bee workers as described above to measure dose-response curves for diacetin (diluted in acetone to four concentrations, 10 −2 , 10 −3 , 10 −4 , and 10 −5 ; v/v). Antennae of M. haemorrhoidalis were only tested on the two highest concentrations. Both female and male antennae of Rediviva were used, because we found in GC-EAD analyses (unpublished data) that both sexes responded similarly to diacetin.
As a positive control we used linalool (10 −2 in acetone), a compound widespread among plants pollinated by bees 32 , and as negative control we used acetone.
To test whether different bee species responded differently to the dilution series of diacetin, data were analysed using a repeated measurement ANOVA (STATISTICA v. 7.1.; www.statsoft.com) with individual bees as subject for repeated measures and the different dilutions and bee species as categorical factors. Tukey was used as post hoc test. Responses of Melitta were excluded from these analyses as only two of the four diacetin dilutions were tested in this species. Instead, we tested for a dilution effect in Melitta using a paired t-test (STATISTICA). A paired t-test was also used to test for differences in responses to acetone and the 10 −2 dilution of diacetin in each species. For more detailed information, see Supplemental Experimental Procedures.

Chemical Analyses.
To identify the EAD-active compounds in the four species used for GC-EAD measurements, 1 μ l of the flower extracts was analysed on a Varian Saturn 2000 mass spectrometer coupled to a Varian 3800 gas chromatograph fitted with a 1079 injector (Varian Inc., Palo Alto, CA, USA). Additionally, we analysed samples of the 50 oil and eight non-oil species available by GC-MS for the presence of the EAD-active compounds (for further information, see Supplemental Experimental Procedures).
The absolute amount of synthetic compounds in the 10 μ l extract offered to the bees during the bioassay was equivalent to the quantity of compounds found in extracts of 100 flowers (few flowering stems) of L. punctata (2 μ g heptanoic acid, 4 μ g geranic acid, 2 μ g (E)-2-dodecenal, 4 μ g 2-tridecanone, and 0.3 μ g diacetin).
Bioassays. Behavioural assays were needed in this study, because electroantennographically active substances do not necessarily elicit behavioural responses in insects 58 . Two-choice bioassays in the flight cage tested the importance of EAD-active floral volatiles of L. punctata for host plant location by Lysimachia-naïve M. fulvipes females. Lysimachia-naïve bees were used to study the innate basis of host plant location. Naïve bees were not trained before the experiments, did not know the test scenario before the testing started, and were not rewarded when responding (see also Supplemental Experimental Procedures). Diacetin was tested against an acetone negative control and against a natural flower extract of L. punctata (positive control; from 100 flowers). We further tested a natural extract against the completely synthetic (5 EAD-active compounds) mixture, as well as the complete synthetic mixture against incomplete synthetic mixtures from which one of the components was omitted. To obtain a mixture without diacetin, we additionally had to eliminate geranic acid as GC-MS analyses revealed trace amounts of diacetin (0.24 ng in 4 μ g geranic acid) as a contaminant in synthetic geranic acid.