Male flowers of Aconitum compensate for toxic pollen with increased floral signals and rewards for pollinators

Many plants require animal pollinators for successful reproduction; these plants provide pollinator resources in pollen and nectar (rewards) and attract pollinators by specific cues (signals). In a seeming contradiction, some plants produce toxins such as alkaloids in their pollen and nectar, protecting their resources from ineffective pollinators. We investigated signals and rewards in the toxic, protandrous bee-pollinated plant Aconitum napellus, hypothesizing that male-phase flower reproductive success is pollinator-limited, which should favour higher levels of signals (odours) and rewards (nectar and pollen) compared with female-phase flowers. Furthermore, we expected insect visitors to forage only for nectar, due to the toxicity of pollen. We demonstrated that male-phase flowers emitted more volatile molecules and produced higher volumes of nectar than female-phase flowers. Alkaloids in pollen functioned as chemical defences, and were more diverse and more concentrated compared to the alkaloids in nectar. Visitors actively collected little pollen for larval food but consumed more of the less-toxic nectar. Toxic pollen remaining on the bee bodies promoted pollen transfer efficiency, facilitating pollination.

reduced male fitness and to inbreeding depression 2,13 . Thus, when more pollinators visit the plant, attracted by large arrays of inflorescences, but they end up probing fewer flowers because flowers are robbed, geitonogamy actually decreases and this can favour outcrossing 9,13,14 .
Many species with hermaphroditic flowers are dichogamous; that is, their pollen is ready at a different time than their stigmas. Nectar composition and volume usually vary between sexual phases in dichogamous plants 15,16 . Nectar volume, composition, and accessibility influence visitor behaviour. However, the effects of dichogamy on pollinator attractiveness and pollen transfer, and subsequent reproductive success remain unclear 2 . Showy floral displays are thought to evolve as a means to attract pollinators and thus increase mating opportunities for male-phase flowers 2,[17][18][19] . This supposes that more pollinator visits means increased male fertility, but only a few pollinator visits will saturate female flowers 18,19 . Thus, if the number of pollinator visits limits male fitness, greater rewards and more effective pollinator signals should be favoured in male-phase flowers 19,20 . Furthermore, since male-and female-phase flowers provide pollinators with different resources (i.e. pollen and nectar, respectively), sexually dimorphic morphological, visual, or olfactory signals allow the pollinators to determine, as it were, what is 'on the menu' , and choose their flower according to their needs for nectar or pollen [19][20][21] . Therefore, sexually dimorphic male-and female-phase flowers can direct pollinator movements from male to female flowers, thus making more opportunities for fertilization 20 .
Studies that examine the balance between pollinator attraction (signals) and rewards as a function of dichogamous sexual phase are rare, as are studies that focus on male reproductive success in toxic plant species. Here, we examined reproductive fitness in male-and female-phase flowers in the protandrous hermaphrodite Aconitum napellus spp. lusitanicum Rouy (Common monkshood, Ranunculaceae, Fig. 1). In contrast to previous studies that examined only one aspect of plant-pollinator interactions (toxicity, robbing, signals, etc.), we combined observations in natural populations (pollinator behaviour) with experimental approaches (detection of toxic compounds) and fine chemical analyses of signals and rewards.
Aconitum spp. are bee-pollinated plants 4,8,9 . Inflorescences of A. napellus spp. lusitanicum are composed of numerous (up to 50) large and zygomorphic flowers (Fig. 1a-c) 21 . The flowers are strongly protandrous and exhibit a longer male phase (5-6 days) than a female phase (1-2 days) 9 , similar to other Aconitum species 22-24 . In each inflorescence, the flowers at the bottom open first; these female-phase flowers open before the younger male-phase flowers at the top of the inflorescence.
Bees usually begin to forage at the bottom of the inflorescence and proceed toward the top 11 . Due to the longer male phase, floral display is male-biased 9 . The upper, helmet-shaped petaliferous sepal protects two spurred nectariferous petals (Fig. 1d); as they crawl into the helmet, pollinators go over the stigma and the anthers 9,11 . Most of the pollinators visiting flowers are bumblebees (up to 95% of all insect visitors) and honeybees (18-54%), but the number of each kind of visitors changes in different years and locations 8,9,25 . Insects make two types of visits to A. napellus flowers: 'legitimate' visits in which the visitor contacts the reproductive organs allowing the pollination of the flower, and 'illegitimate' visits in which the insect robs the nectar and does not participate to pollination (Fig. 1e,f) 9,11 . Long-tongued bumblebees, Bombus hortorum (tongue length of about 12 mm) and B. pascuorum (tongue length of about 8 mm), are the main pollinators, whereas short-tongued (tongue length of about 6 mm) bumblebees (Bombus terrestris and B. lucorum) and honeybees (Apis mellifera) are mainly considered robbers of nectar 8,9,25 . The proportion of robbed flowers varies from 2 to 23% according to the population and the year 9,25 . Fruit set in natural populations remains high (90-93%), showing that female-phase flowers continue to attract pollinators despite nectar robbery 9 .
As with nearly all species in the Ranunculaceae, A. napellus is poisonous. All Aconitum species contain diterpenoid aconitine-like alkaloids; these toxic alkaloids affect the neural and cardiovascular systems in insects and mammals 26,27 . The main alkaloids in Aconitum plants are aconitine and its N-methyl homolog mesaconitine 26 , although variations in composition and concentration have been observed among species and in different parts of the same plant 4,8 .
In this study, we compared floral signals (colour, size, and odour) and floral rewards (nectar and pollen quantity, composition, and toxicity) between male-and female-phase flowers. We hypothesized (i) that male reproductive success is more pollinator-limited than female reproductive success and that floral signals and rewards should be enhanced in male-phase flowers as a result and (ii) that at the insect level, visitors should prefer to forage flowers only for nectar due to the toxicity of the pollen.
We addressed the following questions. (i) Do floral signals (size, shape, and odour) differ between male and female phases? (ii) Do floral reward quantities and compositions differ between male and female phases? (iii) Do pollen and nectar rewards differ in total and profile of alkaloids? (iv) Do insect visitors detect alkaloid toxicity? (v) Do insect visitors differ in their behaviour and pollination efficiency?

Results
Do floral signals differ between male and female phases. Male-phase flowers produced a mean of 0.83 mg of pollen, from 40 to 53 stamens, corresponding to a minimum of 180,000 pollen grains per flower. Some pollen grains were still present during the female phase. Flower morphology did not significantly differ between male and female phases ( Table 1, Fig. 2a). Tepal colours (spectrometry) and UV reflectance also did not differ ( Fig. 2c-l). Fragrance analyses revealed complex chromatographic profiles with many well-resolved peaks (Fig. 3). Benzene ethanol and trans-β-ocimene (122 ± 112 and 104 ± 41 ng h −1 , respectively) were detected within the volatile organic compounds (VOCs) from male-phase flowers but were found only in minute amounts (benzene ethanol, 13 ± 13 ng h −1 , F 1,6 = 3.7382, P = 0.1014) or not at all (trans-β-ocimene < limit of detection [LOD], F 1,6 = 25.73, P = 0.0023) in the female-phase flowers.
Do floral resources differ between floral phases. Nectar volumes significantly differed between male and female phases, with nectar production peaking on the third or fourth day after anthesis and decreasing continuously thereafter (Fig. 2b). As a result, flowers produced more nectar during the male phase than during the female phase ( Table 1). The total sugar concentration and the proportions of sugars did not significantly differ between male and female phases. Sucrose largely dominated the nectar composition (around 80%) with sucrose/ hexose ratios higher than 4. www.nature.com/scientificreports www.nature.com/scientificreports/ Do pollen and nectar differ in total alkaloids and profiles. Liquid chromatography-mass spectrometry (LC/MS) analysis revealed that the pollen and nectar contained unique alkaloid profiles ( Table 2). Pollen had more diverse alkaloids than nectar: we analysed 37 different alkaloids and identified 22 in pollen but only 4 in nectar. Aconitine was one of the major alkaloids found in pollen ( Table 2). Aconitine concentration in pollen was 232 µg g −1 and the concentration in nectar was lower than 0.005 µg·g −1 . In pollen, the main three alkaloids were aconitine, 6-O-acetylacosepticine, and napelline. The concentration of 6-O-acetylacosepticine and napelline was about 7 to 9% of the concentration of aconitine. The other detected alkaloids represented less than 6% of the content of aconitine. Nectar contained only three alkaloids among those present in pollen ( Table 2). Concentrations of aconitine and napelline were similar in nectar. Concentrations of 6-demethyldelphatine and 18-demethylpubescenine (not detected in pollen) were about 2 and 16 times lower than aconitine contents, respectively. www.nature.com/scientificreports www.nature.com/scientificreports/ Do insect visitors detect alkaloid toxicity. Most B. pascuorum individuals tested rejected artificial solutions with aconitine ( Fig. 4a). All together, the percentage of bees that consumed sucrose solution was significantly higher than the percentage that consumed sucrose + aconitine solution (X 2 = 10.806, df = 1; P = 0.001). Consumption rates and contact time with aconitine-laced solutions, when consumed, were slightly lower than for the pure sucrose solution mainly for B. pascuorum (Fig. 4b,c). Long-tongued B. pascuorum individuals were more deterred by aconitine than individuals from short-tongued species (A. mellifera and B. terrestris); the combined deterrence index reached 0.40 for A. mellifera, 0.64 for B. terrestris, and 0.80 for B. pascuorum.

Do insect visitors differ in their behaviour and pollination efficiency.
In natural populations, the main insect visitors differed in their visitation behaviour (Table 3). A. mellifera individuals were abundant visitors with 77 individuals out of 183 observed visitors and only visited few flowers per inflorescence (total of 291 visited flowers) for nectar (legitimately (24%) or by base working robbery, sliding between the tepals (76%)) whatever the sexual phase of visited flowers (always more male phase flowers in all populations and years). Short-tongued B. terrestris individuals also only visited flowers for nectar (only three individuals, legitimately or illegitimately by piercing a hole in the upper tepal). B. hortorum and B. pascuorum individuals (long-tongued, total of 102 individuals out of 183 visitors) typically made legitimate visits (99%) to many flowers per plant (total of 1065 visited flowers). Nevertheless, they collected much less pollen (1%) than nectar (99%). Pollen was rarely actively collected and packaged in the corbiculae (pollen loads). Since insects rarely visited flowers for pollen collection,   Table 2. Alkaloid compounds detected in pollen and nectar. m/z = mass charge ratio. RT = Retention time. N = 3. Assignment based on MS data, fragmentation and presence in other Aconitum species. The content of identified alkaloids is expressed as a relative value compared to aconitine (232 µg g −1 in pollen, 5 10 −3 µg g −1 in nectar, considered as 100%). nd = not detected.   www.nature.com/scientificreports www.nature.com/scientificreports/ we obtained only 23 pollen loads from B. hortorum (all years and populations combined) and 2 from B. pascuorum. These pollen loads contained relatively few pollen grains (mean 370 ± 115 per load). A. napellus pollen represented 56.3 ± 12.6% of the pollen collected.
Most insects had A. napellus pollen grains on their bodies (Fig. 5). Efficient pollination requires that pollen be present mainly on the lower parts of the visitor's body (thoracic and abdominal sternum) to promote contacts with stigmas. With the exception of one B. terrestris individual, all other insects had aconite pollen on their bodies, available for pollinating flowers. Honeybees that legitimately visited flowers transported similar quantities of aconite pollen to that of the long-tongued B. hortorum.

Discussion
Our results revealed a clear sexual difference in floral scent production, with one alcohol, benzene ethanol, and one monoterpene, trans-β-ocimene, produced exclusively or at higher rates in male-than in female-phase flowers. It is the first time than odour detection was tried on Aconitum species. Odours specific to male-phase flowers related to pollen or anther odours have been reported in other species 28 . The two particular components we identified have also been detected in other floral scent compositions 21,29-31 with β-ocimene detected in 71% and benzene ethanol in 54% of the 991 studied species 32 . In dioecious species, more volatiles have been detected in male flowers compared to female flowers [18][19][20][21]30,33 . However, most (62%) of species reviewed by Ashman 18 presented a similar composition of volatiles in both floral sexes. Nevertheless, we were not able to confirm the attractiveness of benzene-ethanol for the three insect visitor species tested using an olfactometer, as individuals did not preferentially choose the odour-marked channel 34 .
Nectar production differed between the floral sexual phases in our species, with 4.3-fold more nectar produced during the male phase than the female phase. Male-biased nectar production was also documented in Aconitum carmichaelii, whereas nectar production was female-biased in other protandrous species 15,22,[35][36][37] . Male-biased nectar production is common in hermaphrodites, suggesting that this secretion pattern has evolved to ensure more visits in male-phase flowers 15  Floral sexual phase had no impact on the total sugar concentrations or on the relative proportions of sugars. The nectar of A. napellus ssp. lusitanicum predominantly contains sucrose (83%). Sucrose-rich or sucrose-dominant nectars with high total sugar concentrations have been detected in other long-tongued bee-pollinated Aconitum species 16 . The prevalence of sucrose is a common characteristic of nectars hidden within deep flowers 16,20 .
Such high-quality nectar in large volumes should attract bees 39 . The bottom of the inflorescence has more female flowers and thus more nectar compared with the top of the inflorescence. This change in nectar availability may improve plant fitness 20,39 , as pollinatorys may avoid the female-phase flowers and concentrate on the more-rewarding male-phase flowers. This may benefit the plant because in protandrous species, more ovules become available over the flowering season, as less pollen is available 16,40 . With a male-biased display of flowers, A. napellus should preferentially attract pollinators that disperse pollen grains and contribute to male fitness 13,23 . Deposition of pollen on female-phase flowers is maintained by the pattern of upward foraging commonly exhibited in bumblebees on vertical inflorescences 14,23,36,41 .
We detected 37 different diterpenoid alkaloids in the floral resources of A. napellus, with more diverse profiles and higher concentrations in pollen than in nectar. Total concentration (more than 200 µg g −1 ) in pollen was in the same range as in the roots (A.-L. Gauthier, unpublished data). Similar alkaloids have been detected in other Aconitum species 8,26,27,42 . Pollen (around 4 10 −3 µg g −1 ) also had significantly higher concentrations of alkaloids than leaves or flowers, and nectar had the lowest concentration of alkaloids in A. septentrionale 4 . In the few published comparisons of alkaloid levels between nectar and pollen, concentrations in nectar tended to be lower than in pollen, with concentrations in pollen equal to or exceeding that of other organs such as leaves 43 .
Measuring the toxicity of A. napellus floral rewards may require measuring the survival of growing larvae or individual bees. However, we can estimate that toxicity level is high based on our results and published data. First, we found that A. napellus ssp. lusitanicum had a high concentration of aconitine-like alkaloids in pollen (232 µg g −1 ); this exceeds the ED50 of all the alkaloids that Detzel and Wink tested 44 for A. mellifera. Second, the major alkaloid in pollen was the highly toxic alkaloid aconitine 26,27 . Alkaloid concentrations in pollen of A. napellus are thus potentially lethal or sublethal 45 .
Our gustatory tests demonstrated that some bee species were more sensitive to alkaloids than others. Nectar-robbing short-tongued bees (B. terrestris and A. mellifera) were less deterred by the same concentrations of aconitine than B. pascuorum. Indeed, B. terrestris has higher sensitivity to alkaloids compared with other plant compounds and for most compounds, these bees can only taste compounds present at greater than 10 ppm [45][46][47] . Low levels of toxic alkaloids in nectar do not affect bumblebee foraging behaviour 46 . Detection thresholds for other Bombus species have not been studied. Indeed, extrapolating our experimental data to field conditions, insect visits ceased at alkaloid concentrations up to 200 µg g −1 , contrary to the results of Barlow et al. 8 . Bees act as if they were balancing economic gains (sugar collection) against toxicity 48 . For example, Barlow et al. 8  Pollen toxicity may be considered a chemical defence that helps plants decrease herbivory and excessive harvesting of pollen 43,44 . As alkaloids at high concentrations could deter, harm, or kill visitors 44,49 , they may deter non-pollinating insects; this may prevent pollen from being wasted and thus improve pollen transfer among plants 43,50 . This protective role has been proposed for different genera in the Ranunculaceae 4,51,52 . Pollen collected for larvae food and packaged in corbicula is lost for pollination. When insects detect the toxicity of pollen and do not actively collect pollen or dislike grooming, they leave more pollen grains on the body parts that make contact with stigmas, thus ensuring efficient pollination 52 . All the observed visitors carried Aconitum pollen on all parts of their bodies, confirming their effective role in pollination.
In contrast to its pollen, A. napellus nectar has a low concentration of alkaloids and should be suitable for a large range of foragers 4,44 . Insect species mainly visited A. napellus for nectar 12,24 . Moreover, several individuals from species considered as usual robbers were observed legitimately visiting flowers. The legitimate visits to female-phase flowers likely would be enough to ensure the observed high pollination rate and high fruit set (90-93%).
Both mechanisms, the deterrence effect of pollen toxicity and high production of sugar-rich nectar work together to ensure the reproductive success of this specialised protandrous plant species, mainly by high male fitness.  A. napellus (A. mellifera, B. pascuorum, and B. terrestris) were used in trials of flower attractiveness and gustatory preference. A. mellifera and B. pascuorum individuals (10 per test and treatment) were collected in the wild around the university campus (minimum of four locations per species) in the morning prior to tests. As B. terrestris individuals cannot be easily distinguished in the field (similar morphology for several species), we used bumblebees from four hives (Biobest, Westerloo, Belgium) placed in the vicinity of the university buildings to obtain individuals for experimentation. www.nature.com/scientificreports www.nature.com/scientificreports/ weighed. Pollen counts were performed at Lille-1 University (Evo-Eco-Paleo Unit). We used a particle counter [CASY, model TT (Innovatis, Bielefeld, Germany)] to estimate the number of pollen grains in these samples. The day before pollen counts, samples were placed in a tube heater until most of the ethanol evaporated. We then forced anther dehiscence by placing samples in an oven at 56 °C overnight. We added 1 mL of distilled water to each pollen sample. Tubes were then sonicated and vortexed to separate pollen grains from the anthers and each other. Pollen solution (200 µL) was then diluted in 5 mL of isotonic measurement buffer (CASY ® ton). The particle counter analysed three 400-µL samples from the pollen suspension, providing total particle count in this 1200 µL volume, as well as counts for 400 size classes ranging from 0.125 to 150 µm. Comparisons with blank samples allowed us to identify peaks of particles, with size ranging from 20 to 30 µm, which corresponds to pollen grains. The number of pollen grains per anther was estimated by multiplying the values provided by the particle counter by the dilution ratio (i.e. (1000/200) × (5000 + 200)/1200).

Insect material. The main insect visitor species of
Flower colour and morphology. We measured floral traits (corolla length, corolla diameter, floral colour, and UV reflectance) on 10 flowers per phase (from 10 different individuals). Helmet width, corolla length and diameter were measured with a calliper. We measured tepal colour reflectance spectrum using a fibre optic spectrometer (Avaspec-ULS2048) and a xenon-pulsed light source (AvaLight-XE, Avantes, Apeldoorn, The Netherlands). For UV reflectance measurements, flowers were fixed on a black background and photographed with a Nikon D40 (Coastal Optical 60 mm 1:4 UV-VIS-IR ApoMacro; Filtre X-Nite 330).

Floral volatile collection.
We sampled volatile organic compounds (VOCs) from intact male-and female-phase flowers (N = 4 for each phase). Each flower was individually inserted in a glass bulb (50-mm length and 30-mm diameter), with a 17-mm-diameter opening. To prevent air contamination, one Teflon tube filled with active charcoal was connected to one end of the glass bulb and two glass Tenax TA tubes (6 mm × 4 mm, 7 inches; 60-80 mesh, Gerstel, Sigma-Aldrich, Overijse, Belgium) were connected to the other end. The Tenax tubes were connected to calibrated air pumps (Gilair plus, Sensidyn, St. Petersburg, USA) at a flow rate of 50 mL·min −1 for 180 min (between 11:00 and 14:00 h). Lab conditions ranged between 21.5 and 23.5 °C with 45 to 55% RH.
Trapped VOCs were then thermally desorbed on an Agilent 6800 gas chromatograph fitted with a Gerstel TDU thermodesorption unit and CIS cryocooler maintained at −50 °C. Immediately after thermodesorption, the CIS unit T° was programmed from −50 °C to 300 °C at 12 °C sec −1 . The GLC apparatus was coupled to an Agilent 7975 C mass spectrometer. Analytical conditions were as follows: split mode injection (with split ratio of 50:1, helium at 1.6 mL min −1 as carrier gas), VF-Max MS 30 m x 250 µm (df = 0.25 µm) column (Agilent). The temperature programme allowing the best separation (preliminary tests not shown) was fixed as follows: 40 °C (5 min) then to 75 °C at 4 °C min −1 and to 115 °C (3 °C min −1 ) and finally to 250 °C at 13 °C min −1 . MS conditions were as follows: EI mode at 70 eV, source T° = 250 °C, MS quadrupole at 200 °C, and scanned mass range 35 to 500 amu.
The recorded chromatograms were systematically interpreted to search for VOCs specific to male or female phases. When detected, the molecules were identified according to computed databases (Wiley 250.L and PAL 600). The identifications were corroborated based on the retention data of pure molecules commercially available. They were also quantified relative to highly pure internal standard (limonene, 1 µL methanol solution at 0.67 µg µL 1 automatically added to each desorption tube before the thermal process).
Ambient air was also collected as a control to identify background contaminants. To prevent a possible breakthrough, a second glass cartridge loaded with the same adsorbent (Tenax TA) was systematically placed in tandem. The analysis did not reveal any traces of the molecules of interest.
Alkaloid collection and analyses. Anthers were collected from 50 flowers (from 10 individuals), dried and then crushed on a sieve (90 µm) to collect pollen grains. Three replicates of 15 mg of pollen were sampled for alkaloids. Samples were frozen (−80 °C) for 2 h and then freeze-dried. Dry samples were ground to a fine powder with liquid nitrogen, and weighed amounts of powder were placed in a 1.5-mL microcentrifuge tube (Eppendorf, Hamburg, Germany). Alkaloids were extracted using a tissue homogenizer (VWR Ultrasonic Cleaner) for 10 min in 100 µL of extraction buffer (70% aqueous methanol and 0.5% formic acid). Following centrifugation at 12,000 rpm for 5 min (Centrifuge 5430 R), the supernatant was transferred to a 1.5-mL microcentrifuge tube. After drying in a SpeedVac, alkaloids were resuspended in 200 µL of LC-MS-grade methanol and filtered with a 0.45-µm PTFE syringe filter (Whatman).
Nectar was collected in 5-μL glass capillary tubes (Hirschmann Laborgeräte, Eberstadt, Germany) from 50 flowers (from 10 individuals). Three replicates of 15 µL of nectar were dried in a SpeedVac and resuspended in 200 µL of LC-MS-grade methanol before filtration with a 0.45-µm PTFE syringe filter (Whatman).
Chemical analyses were performed using an LC-MS/MS system consisting of a Thermo Accela pump, autosampler, photodiode array detector, and a Thermo Scientific LTQ orbitrap XL mass detector (MASSMET, Louvain Drug Research Institute, Woluwe, Belgium). A Phenomenex Gemini C18 column was used (2 mm × 150 mm, packed with 3-µm particles). A binary solvent system with a flow rate of 300 µL min −1 was used: solvent A, HPLC-grade water (0.1% formic acid); and solvent B, LC-MS-grade methanol (0 min: 95% A, 10-12 min: 0% A, 13-17 min: 95% A). A 10-µL volume was injected and the column temperature was set at 30 °C. High-resolution MS was performed with an electrospray ionisation source in positive mode. The following inlet conditions were applied: capillary temperature, 275 °C; capillary voltage, 13 V; tube lens, 140 V; sheath gas flow, 30 a.u.; and auxiliary gas flow, 10 u.a. Data acquisition and processing were performed with Xcalibur software. This method covered the mass range of all alkaloids (MW from 329 to 673). Alkaloids from pollen and nectar were detected and tentatively identified by high-resolution MS. Fragmentation obtained by collision-induced dissociation helped in the identification. Alkaloid concentrations were calculated based on an aconitine standard curve in MeOH and expressed as "aconitine-equivalent". Analysed alkaloids were selected based on their presence in other Aconitum species 4 www.nature.com/scientificreports www.nature.com/scientificreports/ Nectar production and sugar composition. Nectar was collected in 5-μL glass capillary tubes (Hirschmann Laborgerate, Eberstadt, Germany) from flowers of the plants transferred in the growth chambers at the University campus (no insect visits). We estimated the nectar production trend during the day at the beginning of the anthesis by sampling 2 flowers from 12 individuals every 2 hours during 10 hours. We also measured the total nectar production per flower (2 flowers from 12 individuals) for the same flowers every 2 days during the entire (9 days) flower life span, at the same time each day, 2 hours after the light was switched on. Total production for male and female phases was estimated from sampling at the end of the male (5 days after anthesis) and female (8 days) phases.
Nectar samples were stored at −80 °C until chemical analyses were performed. After derivation, sugar identification and quantification were performed by GC-MS (Thermo Trace 1310 /Thermo ISQ-QD). A Restek Rxi-5Sil MS (30 m x 0.25 mm ID, 0.25 µm df) column was used. The flow rate was 1.0 mL min −1 (helium). The split ratios injected were 1/10 for the hexoses and 1/100 for sucrose. The following inlet conditions were applied: oven temperature, 105 °C for 4 min, then to 280 °C at 15 °C min −1 (for 20 min) and injector temperature of 250 °C.The total sugar content of nectar per flower (mg) was then calculated as volume of nectar (µL) x total sugar concentration (mg/µL). Pollen carrying capacity. Thirty individuals per species were killed with ethyl acetate, and individually stocked. Pollen was removed from the different insect body parts with small cubes of gelatine-glycerine 53 . Pollen grains concentrated in the corbicula were removed and not included in analyses. All the pollen grains were counted by light microscopy.

Insect behaviour.
Pollen collection purity. We sampled pollen loads from bumblebees visiting A. napellus in the natural populations. In the lab, pollen loads were acetolysed 53 . At least 400 randomly chosen pollen grains of each of the 25 pollen loads were identified and counted under light microscopy.

Gustatory responses of insect visitors.
We followed the protocol of Ma et al. 54 for aconitine detection.
When caught, individual bees were starved for 3 hours in a plastic vial (7 cm long, 2.8 cm inner diameter) at room temperature and in complete darkness in the lab. We transferred each bee to a holding tube, a 15-mL centrifuge tube with a 4-mm hole drilled at the tip and a piece of steel mesh fixed inside. After a pre-test phase with a droplet of sucrose, we presented the test solution in a 100-µL glass capillary tube and gently squeezed the tubing to maintain the solution at the tip of the tube. Any insect with no proboscis extension after 5 minutes was removed from the test. Test phases were 2 minutes long, and volume of the solution before and after the test were recorded with a calliper.
Tests were performed with three visitor species (A. mellifera, B. pascuorum, and B. terrestris). Ten individuals per insect species were tested with each solution. Sucrose (430 µg mg −1 ) and aconitine (232 µg g −1 ) concentrations corresponded to those found in our species. Three solutions were presented: deionized water alone (control), pure sucrose solution, and sucrose plus aconitine solution.
For all tests, a deterrence index was calculated for each subject from the choice to drink, the volume of solution consumed, and the time of contact with the solution as follows: ID = 1-(choice for sucrose/choice for sucrose + aconitine) × (volume consumed of sucrose/volume consumed of sucrose + aconitine) × (time of contact with sucrose solution/time of contact with sucrose + aconitine solution).
Statistical analyses. The normality of the data was estimated using Shapiro-Wilk tests, and homoscedasticity was verified using Levene's tests. When normality and homoscedasticity requirements were fulfilled, analyses of variance (ANOVAs) were performed (floral phase effect on floral volatiles). Otherwise, nonparametric tests were performed using Wilcoxon test for comparison of two conditions (floral phase effect on flower morphology and nectar parameters) and Kruskall-Wallis test for comparison of more than two conditions (insect and solution effect on consumed volume and contact duration with the solution in gustatory experiments, insect effect on the visitation behaviour) and chi-square tests were performed to compare the percentages of individuals in gustatory