Neural and behavioural responses of the pollen-specialist bee Andrena vaga to Salix odours

A majority of insects utilize food from a small subset of host plants (Bernays and Chapman, 1994). These specialist insects must be able to find their specific plants in a complex environment. An effective means of finding food sources is essential for their reproductive fitness, because the availability of food is one of the major factors limiting the development of offspring and survival of adults (Müller et al., 2006). To find appropriate food sources, insects explore various plant stimuli by using different senses, such as olfaction and vision. In many specialized associations, plant-derived volatiles are most important to guide the insects to food sources (Kessler and Baldwin, 2001; Raguso, 2008). Pollinators, for example, use floral scents to orientate towards flowers from a long distance, to discriminate between and recognize plant species, and as a landing cue or nectar guide (Raguso, 2008). Although floral volatiles were identified in many plant species (Knudsen et al., 2006), the behaviour-modulating components among them and their interplay with multimodal stimuli (e.g. floral colours) have rarely been identified (Dobson, 2006; Raguso, 2020).

Bees are amongst the most important pollinators. While many bee species, including honeybees, visit several different plants, many other bee species restrict their foraging to a subset of available resources. Depending on the habitat, between 15 and 60% of the local bee species are strictly specialist (oligolectic) and collect pollen from only few plant species belonging to one plant family or genus (Michener, 2007; O'Toole and Raw, 1991; Westrich, 2018). The collected pollen is primarily used to provision their offspring with food. The specialized bees recognize their host plants with olfactory cues alone (Schäffler et al., 2015) or with the combination of olfactory and visual cues (Burger et al., 2010). Host finding and recognition is crucial in oligolectic bees, because their larvae normally either fail to develop or experience high mortality rates on non-host pollen (Praz et al., 2008; Sedivy et al., 2008).

Oligolectic bees are very sensitive to the characteristic host-plant volatiles. At the peripheral level, odorants are detected by olfactory receptors neurons on the bees' antennae. As demonstrated by electroantennographic experiments for a megachilid bee (Chelostoma rapunculi) that is specialized on Campanula spp., antennal responses were recorded to much lower concentrations of host-specific compounds (spiroacetals) than for generalist bees that also visit Campanula flowers (Brandt et al., 2017). In the olfactory system, information from olfactory receptor neurons is transferred to the antennal lobe, the first brain structure to process olfactory information (Galizia and Menzel, 2000). Here, all receptor neurons expressing the same olfactory membrane receptor converge onto a specific glomerulus (Galizia, 2014). Using optical imaging experiments, it was recently shown that antennal lobe glomeruli of the oligolectic bee A. vaga responded to lower concentrations of 1,4-dimethoxybenzene than those of generalist honeybees (Burger et al., 2013). 1,4-Dimethoxybenzene is a common floral volatile of Salix host flowers and attractive to both bee species (Dötterl et al., 2005, 2014).

Despite these results, the host finding behaviour of A. vaga is not fully understood. Although the floral scents of different Salix species are well described (Füssel et al., 2007) and some of the compounds elicited physiological (Burger et al., 2013) or behavioural (Dötterl et al., 2005; Dötterl and Vereecken, 2010) responses in A. vaga, Salix compounds that are absent in co-flowering non-host species and attractive for A. vaga are not yet known. Studies that integrate behavioural and neurophysiological experiments in a naturalistic context are rarely performed (but see Najar-Rodriguez et al., 2010; Piñero et al., 2008; Riffell et al., 2009, 2013), but can give new insights into the foraging behaviour of A. vaga as the neural processing of floral volatiles is a central part of their behavioural responses.

The aim of this study was to determine the essential cues A. vaga bees use to discriminate between Salix host and non-host plants, and to identify physiologically and behaviourally active floral scent compounds. We tested the attractiveness of floral cues under semi-natural conditions in a flight cage. Next, we performed calcium imaging experiments in the antennal lobes and coupled the physiological setup to a gas chromatograph (GC imaging). This innovative approach, previously developed in Drosophila (Schubert et al., 2014), allowed us to analyse the elements of multi-component samples to reveal potential behaviourally active compounds for bees. We also compared the results of the GC imaging with results obtained by electroantennographic detection experiments (GC-EAD) to determine whether these approaches produce consistent results or complement each other. Finally, we compared the floral scents and colours of different Salix and non-Salix species to reveal floral cues that are common for the bees’ host plants.

Specifically, we asked: (1) What are the spectral and chemical properties of visual and olfactory displays of host and non-host inflorescences? (2) Which components of host and non-host scent bouquets elicit specific responses in the antennae and the brains of A. vaga? (3) How sensitive is A. vaga to 4-oxoisophorone, a compound of Salix scents identified in the present work as being physiologically highly active in the antennal lobes of A. vaga, as compared with the generalist honeybee (A. mellifera)? (4) What is the relative importance of olfactory and visual cues of S. cinerea hosts in the attraction of A. vaga bees, and in the discrimination against dandelion (Taraxacum officinale) and wild narcissus (Narcissus pseudonarcissus) non-hosts? And (5) is 4-oxoisophorone and a synthetic mixture with further physiologically active Salix compounds attractive for A. vaga?

Andrena vaga Panzer 1790 (Andrenidae) is a solitary bee species active from March to May. The bees are oligolectic (pollen specialist) on Salix spp. (Fig. 1). They hibernate as adults and build nests in aggregations, preferably in sandy ground. The geographic range is from middle and northern Europe to Central Asia. We collected foraging-naïve A. vaga bees at nest aggregations, consisting of hundreds to thousands of individuals, in the Botanical Garden of the University of Ulm (48.418453, 9.967323) and at a site (47.704696, 9.155909) close to Konstanz. Shortly before bees started to become active in spring, we carefully dug out males and females, placed them in boxes containing sand and stored the boxes in dark at 6°C in a fridge. We used the bees for either physiological or behavioural experiments.

For physiological experiments, we placed the boxes with bees in small flight cages (60×60×60 cm) containing a sandy substrate, and located them close to a window in the laboratory at room temperature few days before experiments started. For behavioural experiments, we placed the boxes with bees in a bigger flight cage (2×3×2 m) established at an A. vaga nesting site in the Botanical Garden of the University of Ulm. Some further individuals directly hatched within the cage. In total, approximately 50 individuals (females and males) were active when bioassays started. To supply them with food, we offered sponges saturated with inverted sugar water (Apiinvert, Südzucker AG, Mannheim/Ochsenfurt, Germany). All bees were still naïve with respect to flower foraging for our experiments.

Honeybees (Apis mellifera Linnaeus 1758) foraging on sugar water baits were collected from hives at the University of Konstanz for physiological experiments. Salix is an important food source for honeybees in early spring, when only few other plant species are flowering.

We chose different willow species (S. cinerea, S. fragilis, S. alba, S. alba×babylonica, S. caprea, S. udensis and S. purpurea) as host plants, and widespread dandelion (Taraxacum officinale aggr.) and wild narcissus (Narcissus pseudonarcissus) as non-hosts. We collected flowering stems in the Botanical Garden of the University of Ulm and in Hockgraben park, close to the University of Konstanz. The flowers are normally arranged in catkins whereby female flowers offer only nectar, and male flowers pollen and nectar. We used only male Salix flowers for the experiments. The chosen non-hosts have yellow flowers, bloom at the same time and in syntropy with Salix hosts. Though N. pseudonarcissus has a quite restricted native occurrence (www.floraweb.de), we chose this species because members of this genus are known to emit the same main compound, 1,4-dimethoxybenze, as several Salix species (Füssel et al., 2007; Knudsen et al., 2006).

All experiments complied with the current laws of the country in which they were performed: in this case, Germany.

We measured the spectral reflection of floral parts of six different Salix host (S. cinerea, S. caprea, S. alba, S. alba×babylonica, S. purpurea and S. fragilis) and of the two non-hosts (T. officinale and N. pseudonarcissus). The measured wavelengths ranged from 300 to 700 nm, which corresponds to the colour spectrum perceived by bees (Peitsch et al., 1992). We recorded the floral colours using an Ocean Optics Jaz Spectrometer (Ocean Optics, Inc., Dunedin, FL, USA). A deuterium-halogen lamp floated light over a glass cable at an angle of 45 deg. We used a black adhesive tape and barium sulphate (Merck KGaA, Darmstadt, Germany) as black and white standard, respectively. We stuck the samples on the black tape for the analyses.

In T. officinale, we measured the outer parts (corollas of ray flowers) and the centre of the inflorescences where the pollen is located. The corolla, the pseudocorolla, and the pollen were analysed in N. pseudonarcissus. In Salix, we measured only the pollen colours because pollen is the only colourful part of male willow flowers. In each species, we used the flowers/inflorescences from three different individuals for the analyses and calculated an average per species.

We used the spectral data to plot the colour loci in the colour hexagon (Chittka, 1992). Bees are UV-blue-green trichromats (Peitsch et al., 1992), and this colour diagram visualizes how they perceive the colours. We used the reflectance function of a typical green leaf as background colour (Chittka et al., 1994).

We cut Salix catkins and T. officinale blossoms, immediately transferred them to 20 ml headspace glass vials (Klaus Trott, Kriftel, Germany) filled with gaseous nitrogen (to avoid chemical oxidation), and sealed the vials with aluminium ring caps fitted with a silicon/Teflon septum as described and detailed elsewhere (Pelz et al., 2006). We pooled eight individual male Salix catkins and five Taraxacum blossoms to obtain one sample for each experimental day. Empty vials filled with nitrogen served as controls.

We used synthetic 4-oxoisophorone and 2-phenylethanol (both >99%, Sigma-Aldrich, St Louis, MO, USA) to stimulate neural responses in antennal lobes of A. vaga and A. mellifera using optical imaging. We chose 4-oxoisophorone because it is present in small amounts in the bouquets of different Salix species and consistently elicited a response in the antennal lobes of A. vaga, but was absent in the two other tested non-host species. 2-Phenylethanol is a widely distributed floral scent compound that is found in more than 50% of all flowering plant species (Knudsen et al., 2006), which is also true for T. officinale, and is often selected as a test substance representing commonly occurring floral volatiles (Brandt et al., 2017). Furthermore, the vapour pressure of both compounds is in a similar range, 0.1580 mmHg for 4-oxoisophorone and 0.0868 mmHg for 2-phenylethanol. We prepared nine different dilutions, from 10⁻² to 10⁻¹⁰ (v/v) in mineral oil (>99%, Fluka). Linalool (>97%, racemic mixture, Sigma-Aldrich), a widely distributed floral volatile (Knudsen et al., 2006), was included as a reference odorant at a dilution of 10⁻⁴, allowing normalization of the magnitude of the calcium responses across tested females. We kept 5 ml of the diluted odorants in 20 ml headspace glass vials as described above. As a control stimulus, we used a vial containing 5 ml mineral oil only.

As the GC-EAD system was not equipped with a headspace unit, we trapped the odours from inflorescences/flowers of the various Salix hosts (S. cinerea N=13, S. caprea N=4, S. aurita N=5, S. udensis N=7, S. purpurea N=4) and the two non-hosts (T. officinale N=14, N. pseudonarcissus N=7) using dynamic headspace methods. We cut the flowering/inflorescence stems and placed them in water after cutting. Samples were collected from four flowering stems (approximately 50 flowering catkins) of each Salix species and approximately 30 blossoms and flowers of T. officinale and N. pseudonarcissus, respectively. The cut flowering stems were enclosed within polyester oven bags (Toppits^®, Minden, Germany; size: 40×30 cm). We trapped the volatiles for 4 h during daytime in an adsorbent tube using a membrane pump (G12/01 EB, Rietschle Thomas, Puchheim, Germany) at a flow rate of 100 ml min⁻¹. The adsorbent tubes were filled with 10 mg SuperQ (mesh 80/100, Alltech Associates Inc., Deerfield, IL, USA), fixed in the tubes with glass wool. We eluted trapped volatiles with 100 µl of dichloromethane (99.9%, Merck, Darmstadt, Germany) and stored the samples at −20°C. Samples without plant material were taken as blank controls and contaminants were excluded from the analyses.

For behavioural experiments, we used a synthetic mixture of physiologically active compounds that resembled the composition of S. cinerea scent samples in quantitative and proportional parameters, and 4-oxoisophorone alone. The compounds of the synthetic mixture were diluted in dichloromethane (99.9%, Merck) to obtain stock solutions and then added to the synthetic mixture according to Table S1. In a next step, this solution was further diluted 1:100, and 50 µl of the final mixture were used in bioassays. The amount of each compound was in the range found in natural S. cinerea scent samples (N=7 headspace samples) and verified in GC-MS analyses. The 4-oxoisophorone test solution contained 50 µg of 4-oxoisophorone (>98%, Sigma-Aldrich) dissolved in 50 µl dichloromethane.

Neuronal responses in the antennal lobes of A. vaga females and A. mellifera foragers were recorded using calcium imaging experiments (Galizia and Vetter, 2005; Najar-Rodriguez et al., 2010). Calcium imaging allows quantification of changes in intracellular calcium concentrations as a measure of odorant-evoked neuronal activity (Galizia and Menzel, 2001). We recorded responses of A. vaga and A. mellifera to dilutions of both 4-oxoisophorone (A. vaga N=13, A. mellifera N=11) and 2-phenylethanol (N=11 in both species). In A. vaga, we additionally recorded responses to natural occurring volatiles of host and non-host inflorescences/flowers, either to the unseparated headspace of the samples (S. cinerea N=8 recordings, S. fragilis N=3, S. alba×babylonica N=4, T. officinale N=7) or to the components of the headspace separated by a gas chromatograph (S. cinerea N=11, S. fragilis N=4, S. alba×babylonica N=3, T. officinale N=5).

We mounted individual A. vaga or A. mellifera bees on a custom-made Plexiglas^® stage, fixed them with wax and dissected the brains using the technique as described by Galizia and Vetter (2005) and adapted for A. vaga by Burger et al. (2013). We stained the antennal lobes for 60 min by bath-applying Calcium Green 5N AM (Molecular Probes, Invitrogen, Carlsbad, CA, USA) dissolved in saline with Pluronic and DMSO (dimethylsulfoxide). This procedure leads to signals that combine the activities of several different types of neurons and possibly also glial cells. In honeybees, it has been shown that the prevailing signal magnitude is a good estimate of sensory neuron responses (Galizia and Vetter, 2005). After staining, we removed the excess dye and washed the brain in saline (in mmol l⁻¹: 130 NaCl, 6 KCl, 4 MgCl₂, 5 CaCl₂, 160 sucrose, 25 d-glucose, 10 Hepes-free acid, pH 6.7, 500 mOsmol). We carefully removed the mouthparts of A. vaga to prevent movement artefacts, which was not necessary in A. mellifera. We covered the exposed antennal lobes with a thin layer of transparent two-component silicon (KwikSil, World Precision Instruments, Sarasota, FL, USA). Within 10–30 min, we placed the bees under an upright microscope (Olympus BX50WI, Hamburg, Germany) with a 20× water-immersion objective (NA=0.95, Olympus XLUM Plan FI).

Images were acquired with a CCD camera (Imago QE, T.I.L.L. Photonics, Gräfelfing, Germany). An 8×8 binning on chip was applied to a spatial sampling rate of 1.57×1.57 µm pixel⁻¹. Excitation light was 470 nm (Polychrome V, intensity 10%) and emission was filtered by a LP505 emission filter (Carl Zeiss GmbH, Hamburg, Germany). The exposure time was between 30 and 100 ms per image, depending on the basal fluorescence values of the individual females. Optical recordings consisted of 100 frames for each stimulus at a frequency of 4 Hz.

A computer-controlled autosampler (Combi PAL, CTC Analytics AG, Zwingen, Switzerland) applied 1 ml of odorant-loaded headspace from the 20 ml vials that contained inflorescences/flowers at 1 ml s⁻¹ during the recordings of brain activity. The application started at 6 s after the recording had started (frame 24).

The system consisted of a gas chromatograph (Thermo Finnigan Trace GC Ultra, San Jose, CA, USA) equipped with a flame ionization detector and coupled to the described imaging set-up. The autosampler injected 2 ml of the odour-loaded headspace from the 20 ml vials in splitless mode (injection temperature 200°C) at an oven temperature of 40°C followed by opening the split vent after 1 min and heating the oven at a rate of 7°C min⁻¹ to 200°C (held for 1 min). The carrier gas was helium with a constant flow of 1.5 ml min⁻¹. We used an Optima 5 MS column (length 30 m, inner diameter 0.25 mm, film thickness 0.25 µm, Macherey-Nagel, Düren, Germany) with an installed four-arm flow splitter (OSS-2, SGE, UK) that split the column at the end into two pieces of deactivated capillary (length 70 cm, ID 0.32 mm) leading to the detector and to a heated (temperature 200°C) transfer line (length 40 cm, Axel Semrau, Sprockhövel, Germany), respectively. Makeup gas (nitrogen, 30 ml min⁻¹) was introduced through the fourth arm of the splitter. The separated volatiles were transferred via the transfer line to the outside of the gas chromatograph. The column finally ended in a Teflon tube (inner diameter 2 mm, air stream 1.3 m s⁻¹) directed to the antennae of an insect prepared for antennal lobe imaging.

In GC-imaging experiments, antennal lobe responses were recorded during GC runs using three different temporal sequences. Each sequence consisted of five individual optical recording blocks consisting of 340 frames that were recorded at a frequency of 4 Hz. A break of 35 s between the blocks allowed for recovery from photobleaching. The alternation of relatively short recordings and breaks resulted in an increased longevity of the prepared animal and allowed recordings of clearly recognizable odour responses. The different temporal sequences varied in their offset time (4.0, 4.5 and 5.0 min after sample injection) to cover the complete GC run with recordings. The sequences were started one after another in varying orders and the number per animal depended on its longevity.

We analysed the imaging data using custom-written programs in IDL (Research Systems, Inc., Boulder, CO, USA) and in R (https://www.r-project.org/). Raw data were corrected for lateral movement artefacts using anatomical landmarks. Then, data were corrected for fluorescence intensity decay due to bleaching (Galizia and Vetter, 2005) and filtered using a spatial median filter with a size of 5 pixels to reduce noise. An unsharp mask filter was applied to reduce scattered light produced by strongly activated glomeruli on neighbouring non-responding areas within the antennal lobes (Galizia and Vetter, 2005). The relative calcium change was then calculated for each frame as relative changes in fluorescence (ΔF/F). The background fluorescence (F) was defined as the average fluorescence of frames 4–23 for every pixel.

For the false-colour images, we used the difference between the fluorescence of frames 29–31 and 59–61. Thus, each pixel was assigned a value that was then translated into a colour. To analyse the responses to the odorant dilutions, time traces for glomeruli were calculated from squares with a side length of 11 pixels (corresponds to 18 µm) and always well within each identified glomerulus. We could not always exclude some cross-talk of neighbouring glomeruli owing to scattered light. Because the anatomical borders of glomeruli were not visible with this staining, identification and localization of glomeruli was based on their physiological responses, and their relative position. In the lack of an anatomical reference atlas for A. vaga, we labelled identified glomeruli with letters, allowing for their comparison across individuals. The nomenclature of the analysed glomeruli is the same as in Burger et al. (2013). To ensure that the brain area was comparable across individuals, first, the animal position was adapted in the microscope view using the outline of the brain for orientation (roundish form and tissue of the antennal lobes easily recognizable in contrast to the dark head capsule in the background), and one antennal lobe was positioned in the centre of the recording area. For fine tuning of the position and to check whether the antennal lobe was responding (i.e. not injured during dissection), we ran a control sequence including the reference odour and the solvent control. The response pattern of the reference odour, in particular, was useful to determine the location of the target glomeruli. Then, we recorded a variety of odorants in the same animals, e.g. natural headspaces of different plant species, odorants used for the dose–response assay and different synthetic analogues, to verify the identity of compounds and response patterns found in GC imaging.

We normalized the data based on the reference odorant measurements within each individual animal (R package GloDataMix). A Friedman's test with post hoc analysis (R package Agricole) was performed to test for differences between odorant responses and responses to the solvent control. The response to 4-oxoisophorone in A. vaga was analysed for two glomeruli, the glomerulus with highest sensitivity and the one with the highest maximum response. For all other treatments, only one glomerulus was chosen (representing the glomerulus with both the highest sensitivity and the highest maximum response)

For the false-colour images of chromatographically separated compounds, measurements were analysed interactively. If a glomerulus responded to a compound, a change in fluorescence was detected that corresponded to an odour peak in the GC run. The response was visualised in false-coded pictures by calculating the ΔF/F between three frames around the maximum and minimum response detected in the time traces. Only if a clear glomerular response pattern was seen that was reproducible in at least three runs was the compound considered to be physiologically active. Active compounds were identified in chemical analyses. The identity of the compounds was confirmed by a comparison of retention times and/or response patterns on all used systems with synthetic analogues.

We recorded antennal responses in A. vaga females to floral scents of the Salix hosts (S. cinerea N=6 recordings, S. caprea N=3, S. udensis N=3, S. purpurea N=3, S. purpurea N=3) and the two non-hosts (T. officinale N=6, N. pseudonarcissus N=7) using GC-EAD. In the electroantennographic recordings, a stimulus elicits increased neural activity in olfactory receptor neurons (recorded as voltage drop) located on the insect antennae, the peripheral olfactory organ (Wibe, 2004).

The system consisted of a gas chromatograph (HP 6890, Hewlett-Packard, Agilent Technologies, USA) equipped with a flame ionization detector and coupled to an EAD setup (heated transfer line, two-channel USB acquisition controller, Syntech, Hilversum, The Netherlands). We injected 2 µl of the samples in splitless mode (injection temperature 310°C) at an oven temperature of 50°C followed by opening the split vent after 1 min and heating the oven at a rate of 10°C min⁻¹ to 220°C (held for 5 min). The carrier gas was hydrogen with a constant flow of 2 ml min⁻¹. We used a DB-5ms column (length 30 m, inner diameter 0.25 mm, film thickness 0.25 µm, J&W, USA) with an installed four-arm flow splitter (GRAPHPACK 3D/2, Gerstel, Mühlheim, Germany) that split the column at the end into two pieces of deactivated capillary (length 50 cm, ID 0.32 mm) leading to the detector and to the EAD setup, respectively. Makeup gas (nitrogen, 30 ml min⁻¹) was introduced through the fourth arm of the splitter. The outlet of the EAD ended in a cleaned and humidified airflow (glass tube, inner diameter: 7 mm, air stream 100 ml min⁻¹) directed to a prepared antenna.

For the EAD, we cut an antenna of A. vaga females at the base and the tip and mounted it between two glass capillaries filled with insect Ringer’s solution (in mmol l⁻¹: 137 NaCl, 5.4 KCl, 3.6 CaCl₂). Gold wires connected the capillaries for a closed electric circuit. We considered a compound to be EAD-active when we recorded a response in at least three runs per species. We identified the physiologically active compounds and determined their absolute and relative amounts in chemical analyses.

We identified the physiologically active compounds and determined their absolute and relative amounts using gas chromatography coupled to mass spectrometry (GC-MS). The equipment consisted of a gas chromatograph (7890B GC system, Agilent Technologies, USA) coupled to a mass spectrometer (Agilent 5977A mass selective detector). An autosampler (Gerstel MAS Modular Analytical Systems Controller C506, Gerstel, Mühlheim a. d. Ruhr, Germany) injected 1 µl of the samples in splitless mode into a Gerstel cooled injection system (KAS 4, Gerstel). The injection system consisted of a septumless head with an injection temperature of 10°C. The initial temperature increased at a rate of 12°C s⁻¹ to 300°C. The oven program started at 50°C followed by opening the split vent after 1 min and heating the oven at a rate of 10°C min⁻¹ to 310°C (held for 33 min). The carrier gas was helium with a constant flow of 1.5 ml min⁻¹. We used a DB-5ms column (length 30 m, inner diameter 0.25 mm, film thickness 0.25 µm, J&W). The MS interface had a temperature of 250°C and the ion source 230°C. Mass spectra were taken at 70 eV (in EI mode) from m/z 35 to 450.

We identified the compounds based on their mass spectra using multiple references from the NIST11 library and on Kovats retention indices published for Salix odours (Füssel et al., 2007). If authentic standards were available, we additionally confirmed the identification of individual components by comparison of both mass spectrum and GC retention data. Active compounds were assigned to GC-MS runs by comparing the elution sequence and Kovats retention indices. Amounts of the compounds were calculated using AMDIS 2.71 (Automated Mass Spectral Deconvolution and Identification System) and determined as absolute amounts based on an internal standard (dodecane, 1 µg per sample). We analysed the differences in the scent bouquets between species using Primer 6.1.15 (Clarke and Gorley, 2006). We calculated the relative amounts of single components with respect to the total amount in a sample and square-root transformed the data. Based on pairwise Bray–Curtis similarities, we visualized the similarities and dissimilarities among the samples using non-metric multidimensional scaling and performed an ANOSIM (9999 permutations). We evaluated the contribution of single substances to the observed dissimilarity between hosts and non-hosts using a SIMPER analysis.

We conducted 11 two-choice bioassays with foraging-naïve bees in a flight cage to test innate behavioural responses to natural visual and natural olfactory floral cues and synthetic odour samples. We performed tests only on sunny days between 09:30 am and 14:00 h, when the activity of bees was high. Prior to each behavioural experiment, we removed the sugar water supply.

We performed the experiments in 2015 and repeated them in 2016. The recorded responses of both years were summed. The number of responding bees was similar among years but differed between experiments, and was dependent on the time of day, weather conditions, numbers of experiments per day and date of the season. The results of the first year were confirmed in the second year, except one result (combined olfactory and visual cues versus olfactory cues of S. cinerea) that changed from a tentative to a significant trend. To ensure that an individual bee was counted only once in a specific two-choice test, we caught the responding bees. An individual bee, however, may have participated in different two-choice tests. The responses in each choice experiment were compared using an exact binomial test (www.graphpad.com).

We used an experimental set-up analogous to Burger et al. (2010) to test the attractiveness of coupled and decoupled olfactory and visual cues of the S. cinerea host, and T. officinale and N. pseudonarcissus non-hosts, for A. vaga. We cut flowering branches/inflorescences and placed them into different Plexiglas^® cylinders. All cylinders had a height of 39 cm and a diameter of 9.5 cm. For testing of decoupled olfactory cues, we used a grey cylinder with small holes. The plant volatiles were pumped out of this cylinder using a membrane pump with an air flow of 1 l min⁻¹ (G12/01 EB, Rietschle Thomas, Puchheim, Germany). For testing of decoupled visual cues, we used a transparent cylinder without holes. For offering a combination of both cues, we used a transparent cylinder with holes. Here again, the volatiles were pumped out with the help of a membrane pump.

In each experiment, we presented two cylinders at a distance of 30 and 15 cm above ground. Bees were observed for 30 min, during which we exchanged the position of the cylinders after 15 min to control for a potential side bias. Attractive responses of bees were recorded as approaches and landings. We tested decoupled olfactory and visual cues of S. cinerea host flowers each against an empty control and against the corresponding cue of the non-hosts T. officinale and N. pseudonarcissus. We also tested the combined visual and olfactory cues of S. cinerea against an empty control and against the decoupled olfactory cues of S. cinerea.

In two-choice experiments, we tested both the synthetic mixture of physiologically active compounds (Table S1) and 4-oxoisophorone alone against a solvent control, and the synthetic mixture also against a solvent S. cinerea headspace sample. We applied 50 µl of a sample and of a solvent control (dichloromethane) on rectangular sponge strips (35×7.5×1.5 mm; Kettenbach GmbH & Co. KG, Eschenburg, Germany). We presented the impregnated filter papers at a distance of 30 cm on dissecting needles that were fixed on a horizontal metal bar. This set-up was mounted on a plastic stick that was stuck in the soil and presented at a height of 1.25 m. We exchanged the position of the stick in the flight cage at least four times in each experiment. We reported the behavioural responses of bees for 30 min as approaches and landings.

The pollen of all studied Salix hosts and the two non-hosts was bee green, but the colour loci of host and non-host pollen did not group together (Fig. 2A). The T. officinale corollas were bee UV-green and clearly different in the colouration for bees compared with Salix. The paracorolla and corolla of N. pseudonarcissus were at the transition of bee green and bee blue-green and closely positioned to the Salix loci but the flowers can be probably distinguished from Salix catkins by bees on the colour pattern formed by the corolla and paracorolla. Taken together, host and non-host colour displays are discernible by the bee visual system.

We chemically analysed host and non-host flower odorants, and tested them for physiological responses in A. vaga. We identified a total of 37 active compounds in all physiological experiments among the different Salix species, T. officinale and N. pseudonarcissus (Table S2). Two compounds were classified as aliphatic compounds, 16 as terpenes, nine as benzenoids and phenylpropanoids, two as nitrogen-containing compounds and seven had an unknown identity. The multidimensional scaling based on the Bray–Curtis index showed species-specific scent bouquets (ANOSIM R=0.83, P<0.001; Fig. 2B) and a separation between the bouquets of the Salix species and the non-hosts T. officinale and N. pseudonarcissus (ANOSIM pairwise comparisons between host and non-host species: range of R-values between 0.89 and 1, P<0.01). The compounds benzaldehyde, (E)-β-ocimene and 1,4-dimethoxybenzene were most responsible for the dissimilarity between host and non-hosts (SIMPER, average dissimilarity between host and non-hosts 72%).

In the calcium imaging experiments, the antennal lobes of A. vaga females showed clearly distinct response patterns to unseparated natural samples of different Salix species and T. officinale (Fig. 3). The responding antennal lobe regions were spatially structured and corresponded to patterns consisting of individual glomeruli. If glomeruli were strongly activated, the response of neighbouring glomeruli were sometimes visually overlapping (cluster of glomerulus a, h and b; Fig. 3). In these cases, the individuality of glomeruli was verified in response patterns to separated odorants in GC-imaging experiments (Fig. 4). Eight different glomeruli responded to the tested samples and had a comparable topology across females (Fig. 3C). Thereof, three to five glomeruli responded strongly to the different Salix species (Fig. 3A) and only one glomerulus showed a clear response to T. officinale (Fig. 3B). Glomeruli b, e and k responded across all tested Salix species but not to T. officinale or blank controls.

GC-imaging experiments revealed distinct response patterns to 10 odour components in S. cinerea (Fig. 4) and to 13 odour components in all Salix species (Table 1). The chemical identity of each compound that elicited a response was determined based on the odorant-specific response pattern even though some compounds co-eluted. Common responses among all analysed Salix species were recorded to β-myrcene, (E)-β-ocimene, linalool, 4-oxoisophorone, veratrole, 1,4-dimethoxybenzene and (Z)-3-hexenyl 2-methylbutanoate (the latter probably emitted by green parts of the flowering stems). Salix fragilis and S. alba evoked additional responses to limonene, pinocarvone and verbenone, and S. cinerea to benzyl nitrile, octanoic acid and (E,E)-α-farnesene. The responses to T. officinale were mainly to (E)-β-ocimene and other monoterpenes, which all elicited responses in Salix species as well. We detected no calcium response to benzaldehyde, the major component in the GC runs.

Based on the GC-imaging results, Salix-specific responses of glomeruli b, e and k (Fig. 3) could be attributed to 1,4-dimethoxybenzene, veratrole and 4-oxoisophorone, respectively. Importantly, the responses to 4-oxoisophorone and veratrole were strong (mean±s.e.m. 4-oxoisophorone 1.29±0.07% ΔF/F magnitude in calcium, N=10), although the compounds were emitted in much smaller amounts (1.27±0.14% of total scent amount of physiologically active compounds) compared with 1,4-dimethoxybenzene (main compound 74.36±2.08% of total scent, with a mean response of 1.10±0.12% ΔF/F in the calcium run, N=12) and (E)-β-ocimene (17.89±1.57% of total scent, with 0.92±0.11% ΔF/F in the calcium run, N=5). Glomerulus k also responded to verbenone in S. fragilis and S. alba×babylonica.

In the GC-EAD analyses, we recorded 13 responses to S. cinerea (without responses to contaminants), which corresponded to 20 compounds (Fig. 4). The antennal responses could not always be related to a single substance because some of the compounds co-eluted. In comparison with GC imaging, we recorded additional responses to (Z)-3-hexenol (most probably emitted by green parts), limonene, α-pinene, (E)-4,8-dimethyl-1,3,7-nonatriene, 2-phenylethanol, allo-ocimene, lilac aldehyde A, methyl salicylate, lilac alcohol B+C, indole and germacrene D, but not to β-myrcene. The other responses corresponded to those found in GC imaging.

We recorded dose–responses for 4-oxoisophorone and 2-phenylethanol in A. vaga and A. mellifera, in order to test their sensitivity to characteristic host and non-host compounds. We found that the lowest concentrations of 4-oxoisophorone and 2-phenylethanol that evoked a significantly higher response than the solvent control differed between the two bee species. Glomerulus k in A. vaga responded to 4-oxoisophorone already at the 10⁻⁷ dilution (Friedman's test, critical value=82.34, P<0.01, N=13; Fig. 5A,C) and glomerulus c to the concentrations 10⁻⁴, 10⁻³ and 10⁻² (Friedman's test, critical value=71.47, P<0.001, N=13; Fig. 5A,D); however, the highest response was stronger in glomerulus c as compared with glomerulus k (2.45±0.31% ΔF/F as compared to 4.78±0.52% ΔF/F). The magnitude of the response to 4-oxoisophorone in glomerulus k in the GC-imaging runs (1.29±0.07% ΔF/F) was comparable to the response of the 10⁻⁶ dilution (1.24±0.14% ΔF/F, non-normalized data). 2-Phenylethanol evoked a significant response in glomerulus m at the three highest concentrations of 10⁻⁴, 10⁻³ and 10⁻² only (Friedman's test, critical value=63.89, P<0.01, N=11; Fig. 5A) in A. vaga. In order to elicit responses, much higher odorant concentrations were needed in the honey bee: a glomerulus responding to 2-phenylethanol was only found at the 10⁻⁴ dilution and higher concentrations (Friedman's test, critical value=64.45, P<0.05, N=11; Fig. 5B), and 4-oxoisophorone only at the two highest concentrations of 10⁻³ and 10⁻² (Friedman's test, critical value=47.31, P<0.05, N=11; Fig. 5B) in A. mellifera.

The combination of olfactory and visual floral cues of flowering S. cinerea stems was significantly more attractive than an empty control (Fig. 6). In contrast, the bees showed no preference when testing visual cues alone against an empty control or against the visual cues of the non-hosts T. officinale and N. pseudonarcissus. When testing olfactory floral cues, the bees significantly preferred their host over an empty control and the non-hosts T. officinale and N. pseudonarcissus. However, the combination of both cues of S. cinerea were significantly more attractive than the olfactory cues alone.

A synthetic mixture of the physiologically active compounds of S. cinerea was significantly more attractive than a solvent control, and had the same attractiveness as a solvent headspace sample collected from flowering S. cinerea stems (Fig. 6). Similar to the synthetic mixture of compounds, the bees preferred 4-oxoisophorone over a solvent control.

We analysed the physiological responses to host and non-host odorants and found that many of them elicited physiological responses, indicating that A. vaga has the appropriate complement of olfactory receptors to detect these substances. The bees responded to both minor and major compounds of the scent bouquet of Salix hosts in our physiological experiments. Among those, we found that 4-oxoisophorone elicited particularly strong responses in the antennal lobes, and was present in all analysed Salix specimens. Interestingly, despite eliciting among the strongest responses of all chemical constituents, it was only a minor compound within the blend. It appears that the olfactory system of A. vaga bees is highly tuned to this compound at a very low concentration.

Indeed, our dose–response measurements confirmed this assumption, and showed that, in contrast, generalist honeybees were less sensitive. A high sensitivity in A. vaga was previously also shown for another behaviourally active Salix component, 1,4-dimethoxybenzene (Burger et al., 2013). Both 4-oxoisophorone and 1,4-dimethoxybenzene are present in the odour of many willow species (Füssel et al., 2007). Therefore, we hypothesize that oligolectic A. vaga bees could use the common presence of these compounds in Salix spp. to specifically recognize their hosts and discriminate them from non-hosts. Indeed, in contrast to 1,4-dimethoxybenzene, 4-oxoisophorone was absent in the non-host plants used here, and according to Knudsen et al. (2006) it is not widespread among floral scents. However, there are recent studies that describe 4-oxoisophorone as floral scent component in a broader range of species (e.g. El-Sayed et al., 2018; Lukas et al., 2019, 2020). Future studies should test how A. vaga bees behaviourally respond to 1,4-dimethoxybenzene and 4-oxoisophorone when they are offered simultaneously, and to which extent these two compounds may be synergistically involved in host recognition.

We detected responses to more compounds in the antennal GC-EAD measurements (12 responses/20 compounds in S. cinerea) than in the glomerular GC-imaging patterns (9 responses/10 compounds). One reason was that the samples for GC-EADs were collected for several hours from flowering stems and were thereby enriched, in contrast to few flowers enclosed in a headspace glass used for GC imaging. We also analysed more samples in GC-EADs because more runs can be performed in the same time than for GC imaging. Thus, at first sight, GC-EAD appears to be more sensitive. However, when compounds co-elute in GC-EADs, the antennal responses cannot be assigned to a specific compound, and responses to minor compounds can be hidden or overseen. Specifically, in previous GC-EAD measurements, 4-oxoisophorone co-eluted with other compounds, and it remained unclear to which compounds the antennae responded (Dötterl et al., 2005; Jürgens et al., 2014). A high sensitivity in A. vaga towards this compound was first revealed in this study using optical imaging. Using synthetic analogues, the assigned activity of a compound can be proven in both methods, but imaging provides, besides a quantitative signal (in the form of a peak), a qualitative one consisting of the odorant-specific glomerular pattern. Conceptually, the information of glomerular patterns is analogous to a mass spectrum in GC-MS runs used to reveal the identity of a compound, compared with GC-FID runs that only give chromatogram peaks. Because every glomerulus collects information from a uniform family of olfactory receptors, imaging data also reveal odour–response spectra of these receptors. This indicates, in our current data, that A. vaga has at least two separate olfactory receptors for 4-oxoisophorone (Fig. 5): one with very high sensitivity (low-concentration mode, glomerulus k), and one with high saturation levels (high-concentration mode, glomerulus c). However, imaging experiments do not monitor all glomeruli of the antennal lobe, but only the subset in the superficial focal plane; while receptor responses can be separated, not all can be measured. In contrast, antennograms measure all receptors, but they cannot be separated. We conclude that both physiological methods have advantages and limits. When used together, they are complementary tools when searching for potential attractants within complex scent bouquets.

Our behavioural experiments revealed that the olfactory cues of S. cinerea flowering stems were highly attractive for A. vaga bees and that the bees preferred their host plants when those were offered against T. officinale or N. pseudonarcissus non-hosts. In contrast, decoupled visual cues were not attractive and not preferred over those of non-hosts, although the spectral analysis revealed that the bees should be able to distinguish host and non-host inflorescences. However, visual cues appear to be relevant for A. vaga when present in combination with olfactory cues. This confirms results in other bee species, which also demonstrated that the combination of olfactory and visual cues is more attractive both for specialist (Burger et al., 2010; Dötterl and Vereecken, 2010) and generalist bees (Dötterl et al., 2014; Rachersberger et al., 2019). In line with the physiological experiments, we demonstrated that 4-oxoisophorone was behaviourally active using synthetic analogues and therefore contributes to the attractiveness of Salix scent cues.

Andrena vaga bees rely mainly on olfactory cues while searching for Salix host flowers. Physiological experiments combined with bioassays allowed us to describe characteristic Salix compounds that enable them to effectively locate food sources. In particular, we showed that 4-oxoisophorone is a low-concentration–high-efficiency compound in the olfactory bouquet of natural Salix scents, behaviourally active for A. vaga and detected with high sensitivity. This might represent an evolutionary adaptation in bees that strongly depend on willows to collect pollen and therefore can profit from sensory adaptations to effectively find host flowers.

We gratefully acknowledge Hanna Schnell, who performed calcium imaging experiments with Apis mellifera, Daniel Münch, who provided statistical tools, Michael Kaminski for the bee photograph, and two anonymous reviewers for their valuable comments.