Science Magazine

A caffeine jolt gives bees a buzz to remember


Editor's Introduction 

Caffeine in Floral Nectar Enhances a Pollinator's Memory of Reward. Wright et al.
annotated by Tara Bracken
Many compounds produced by plants as a defense against predation are known to have interesting effects on humans. One such drug, caffeine, improves human memory and alertness, lending increased focus and a manic energy. But what is the ecological role of drugs such as caffeine? Do other organisms experience a similar memory boost when consuming caffeine? This paper investigates the effects of naturally occurring floral nectar caffeine on pollinator behavior, and finds that some flowering plants use caffeine to manipulate honey bee behavior in a way that improves the plant’s reproductive success.


Plant defense compounds occur in floral nectar, but their ecological role is not well understood. We provide evidence that plant compounds pharmacologically alter pollinator behavior by enhancing their memory of reward. Honeybees rewarded with caffeine, which occurs naturally in nectar of Coffea and Citrus species, were three times as likely to remember a learned floral scent as were honeybees rewarded with sucrose alone. Caffeine potentiated responses of mushroom body neurons involved in olfactory learning and memory by acting as an adenosine receptor antagonist. Caffeine concentrations in nectar did not exceed the bees' bitter taste threshold, implying that pollinators impose selection for nectar that is pharmacologically active but not repellent. By using a drug to enhance memories of reward, plants secure pollinator fidelity and improve reproductive success.


Many drugs commonly consumed by humans are produced by plants as a form of toxic defense against herbivores (12). Although plant-derived drugs like caffeine or nicotine are lethal in high doses (35), at low doses they have pharmacological effects on mammalian behavior. For example, low doses of caffeine are mildly rewarding and enhance cognitive performance and memory retention (6). Caffeine has been detected in low doses in the floral nectar and pollen of Citrus (7), but whether it has an ecological function is unknown.
Two caffeine-producing plant genera, Citrus and Coffea, have large floral displays with strong scents and produce more fruits and seeds when pollinated by bees (89). If caffeine confers a selective advantage when these plants interact with pollinators, we might expect it to be commonly encountered in nectar. We measured caffeine in the nectar of three species of Coffea (C. canephoraC. arabica, and C. liberica) and four species of Citrus (C. paradisiC. maximaC. sinensis, and C. reticulata) using liquid chromatography–mass spectrometry (10) (fig. S1A). When caffeine was present, its concentration ranged from 0.003 to 0.253 mM. The median caffeine concentration in both genera was not significantly different (Fig. 1A, Mann-Whitney, Z = –1.09, P = 0.272). Caffeine was more common in the nectar of C. canephora than in that of C. arabica or C. liberica (Coffea: logistic regression χ22 = 11.1, P = 0.004); it was always present in Citrus nectar. The mean total nectar sugar concentration ranged from 0.338 to 0.843 M (Fig. 1B; see fig S1B for individual sugars). Caffeine concentration in nectar did not correlate with total sugar concentration (Pearson's r = 0.063, P = 0.596).
Fig. 1.  (A) Caffeine concentration in Coffea and Citrus spp. and a cup of instant coffee. Caffeine concentration depended on species within each genus (Coffea: Kruskal-Wallis, χ22 = 28.1, P < 0.001; Citrus: Kruskal-Wallis, χ22 = 6.98, P = 0.030); C. canephora had the highest mean concentration of all species sampled. (B) The sum of the concentration of sucrose, glucose, and fructose (total nectar sugars) depended on species (one-way analysis of variance F5, 161 = 4.64, P < 0.001) and was greatest in Citrus maxima and hybrids (citron, lemons, clementines). [C. can., Coffea canephora, N = 34; C. lib., Coffea liberica, N = 31; C. arab., Coffea arabica, N = 27; C. par., Citrus paradisi and hybrids, Ncp = 17; C. max.,Citrus maxima and hybrids, N = 5; C. sin. and C. ret., Citrus sinensis and Citrus reticulata, NCS = 7, NCR = 5 (data for these two species were pooled).] Mean responses ± SE.

This figure shows the amounts of caffeine and three different types of sugars in nectar from Coffea and Citrus plants. Knowing the amount of these compounds present in these flowering plant species allows the authors to design the subsequent experiments in a way that accurately reflects how honey bees and plants interact in nature.

Nuts and bolts

In both panels, the colored bars represent the mean values measured for each species. The error bars represent the standard error of the mean, which tells you how the actual caffeine or sugar concentrations measured for each sample are distributed around the calculated mean value.

The exact number of measurements taken per species is indicated in the figure legend as “N=”, so Coffea canephora is indicated as having N=34, meaning the authors measured nectar caffeine and sugar concentrations of 34 individual Coffea canephora plants.

The samples collected from Citrus sinensis and Citrus reticulata were pooled, meaning they were mixed together and sampled simultaneously. Though their reasoning is not given in the materials and methods section of the supplemental materials, the authors likely did this because they were unable to collect large enough volumes of nectar from these plants to conduct the assay.

Caffeine and sugar concentrations were measured using a technique called liquid chromatography-mass spectrometry (LC-MS). LC-MS first separates individual chemicals present in a sample (via liquid chromatography) and then identifies what those chemicals are by measuring the masses of the particles in those separated chemicals (via mass spectrometry). For more details on how LC-MS works, see the Glossary section of the SitC annotations.


Panel A shows the amount of caffeine present in a cup of instant coffee, three species of Coffea, and four species of Citrus. Coffea canephora exhibited the highest nectar caffeine concentration at approximately 0.25 mM. Despite the relatively high caffeine concentration of C. canephora, the authors report in the figure legend that the average concentration across all Coffea and all Citrus species was not significantly different.


Panel B shows the amount of sugar present in three species of Coffea and four species of Citrus. The Σ (the uppercase form of the Greek letter “sigma”) on the y-axis title is a mathematic symbol that indicates “sum of.” Thus, “Σ of sucrose, glucose, fructose (M)” means the measurements presented are the combined sum of the nectar molar concentrations of the sugars sucrose, glucose, and fructose. The authors referred to this sum in the text as “total nectar sugar concentration.” The authors present the data for each individual sugar alone in supplemental figure 1B.

Citrus maxima had the highest total nectar sugar concentration. The authors found that the total sugar concentrations in panel B did not correlate with caffeine concentrations presented in panel A. This means that you cannot predict the amount of sugar in a plant’s nectar based on the amount of caffeine present and vice versa.

We hypothesized that caffeine could affect the learning and memory of foraging pollinators. To test this, we trained individual honeybees to associate floral scent with 0.7 M sucrose and seven different concentrations of caffeine and tested their olfactory memory. Using a method for classical conditioning of feeding responses (proboscis extension reflex) (11), we trained bees for six trials with 30 s between each pairing of odor with reward. This intertrial interval approximated the rate of floral visitation exhibited by honeybees foraging from multiple flowers on a single Citrus tree (see methods). The presence of low doses of caffeine in reward had a weak effect on the rate of learning (Fig. 2A), but it had a profound effect on long-term memory. When rewarded with solutions containing nectar levels of caffeine, three times as many bees remembered the conditioned scent 24 hours later and responded as if it predicted reward (Fig. 2B, logistic regression, χ72 = 41.9, P < 0.001). Twice as many bees remembered it 72 hours later (Fig. 2C). This improvement in memory performance was not due to a general increase in olfactory sensitivity resulting from caffeine consumption (fig. S2A). Indeed, the effect of caffeine on long-term olfactory memory in bees was greater than that produced by high concentrations of sucrose when the same experimental methods were used (e.g., 2.0 M, fig. S2B).
Fig. 2.  (A) The rate of learning of bees conditioned with an odor stimulus paired with a 0.7 M sucrose reward containing caffeine. The rate of learning was slightly greater for the bees fed caffeine in reward during conditioning (logistic regression, χ12 = 4.85, P = 0.028). N ≥ 79 for all groups. (B) Memory recall test for odors at 10 min (white bars) or 24 hours (red bars) after bees had been trained as in (A). Bright red bars indicate that the response at 24 hours was significantly different from the control (0.7 M sucrose) (least-squares contrasts P < 0.05); dark red bars were not significantly different. Nectar levels of caffeine are indicated by hatching. N > 79 for each group. (C) Bees fed 0.1 mM caffeine in sucrose (orange bars) were more likely to remember the conditioned odor than sucrose alone (white bars) (logistic regression, χ12 = 9.04, P < 0.003) at 24 hours and 72 hours after conditioning. N = 40 per group.

The authors hypothesized that caffeine present in floral nectar would affect pollinator learning and memory. The figure presents the results of experiments designed to test this hypothesis, in which the authors trained honey bees to associate a specific floral scent with a reward (in this case, sugar) with or without consuming caffeine.

The data show that caffeine has a only mild, but significant, effect on honey bees’ rates of learning a new scent— that is, the honey bees spent about as much time learning the scent with or without caffeine—but it had a profound effect on a honey bee’s ability to actually remember that new scent in the long run.

Nuts and bolts

Panel A represents the rate of learning across subsequent trials as a line graph, with each line representing a different “treatment group.” That is, each line represents bees given sucrose containing different concentrations of caffeine, ranging from 0 M to 109 M caffeine/sucrose solutions.

As with the caffeine concentrations presented in Figure 1, panels B and C show the magnitude of the memory response as bars, with the error bars representing the standard error of the mean. Additionally, significance was indicated in panel B by bright red bars. Dark red/brown bars indicate the results were not significantly different from the sucrose-only control group.

The bees in these experiments were exposed to a specific scent at the same time that they were presented with their reward (sucrose) so that they could learn to associate that scent with reward. Caffeine was added to the sucrose in different concentrations to see how consuming caffeine affected the bees’ ability to associate that scent with reward. After being conditioned, the bees were also presented with an unrelated scent to see whether their memory of reward was associated specifically to the scent they had learned; if the bees responded to the unrelated scent in the same way that they reacted to the scent they had learned, then the response would be nonspecific.

To learn more about how bees are conditioned, see the Glossary section of the SitC annotations.

Rate of learning

Panel A shows that although the effect wasn’t great, adding caffeine to the reward caused the bees to learn to associate the floral scent with reward slightly, but significantly, more quickly than the bees that received no caffeine.

Long-term memory

Together, Panels B and C indicate that bees that received caffeine during associative learning were better able to remember scents for longer periods of time.

Panel B shows that compared with sucrose alone, the addition of caffeine concentrations ranging from 10-3 M to 10-7 M improved the bees’ ability to remember the scent 24 hours after training. The two lowest concentrations of caffeine, 10-8 and 10-9 M, did not have a significant effect on long-term memory. Panel C shows that compared with the sucrose-only group (white bars), twice as many bees that had received 10-4 M (0.1 mM) caffeine in their reward (orange bars) were able to remember the learned scent 72 hours after training.

Caffeine's influence on cognition in mammals is in part mediated by its action as an adenosine receptor antagonist (6). In the hippocampal CA2 region, inhibition of adenosine receptors by caffeine induces long-term potentiation (12), a key mechanism of memory formation (13). The Kenyon cells (KCs) in mushroom bodies of the insect brain are similar in function to hippocampal neurons: They integrate sensory input during associative learning, exhibit long-term potentiation, and are involved in memory formation (1416). To determine whether nectar-caffeine doses affect mushroom body function, we made whole-KC recordings in the intact honeybee brain. Caffeine (100 μM) evoked a small increase in the holding current (IM) and depolarized KC membrane potential (VM) toward the action potential firing threshold, by increasing nicotinic acetylcholine receptor (nAChR) activation (Fig. 3, A to D). To determine whether the observed effects of caffeine were due to interactions with adenosine receptors, we applied the adenosine receptor antagonist DPCPX and observed that it similarly increased IM and depolarized VM, but to a lesser extent (Fig. 3, E and F). Both caffeine and DPCPX affected KC response kinetics evoked by brief, local application of ACh, increasing the activation rate and slowing the decay (Fig. 3, G and H). Our data show that caffeine modulates cholinergic input via a postsynaptic action, but could act via presynaptic adenosine receptors to potentiate ACh release (17). The resulting increase in KC excitability should lead to an increased probability of action potential firing in response to sensory stimulation (18), thereby facilitating the induction of associative synaptic plasticity in KCs (19). The enhanced activation of KCs may also facilitate plasticity at synapses with mushroom body extrinsic neurons (20), which exhibit spike-timing–dependent plasticity (21). In this way, a "memory trace" could be formed for the odor associated with reward during and after conditioning (2223).
Fig. 3.  The effect of caffeine on Kenyon cells. (A and B) Example traces from a KC in intact honeybee brain recorded under voltage-clamp [(A), VH = –73 mV) and current-clamp [(B), at resting VM), showing the increase in IM and depolarization evoked by bath application of caffeine (100 μM) and subsequent reversal by the nAChR antagonist d-TC (500 μM). (C and D) Mean data showing the reversal by d-TC (500 μM) of the effect of caffeine (Caff; 100 μM) on IM [(C); N = 6, t5 = 4.03, P = 0.010; t5 = 4.07, P = 0.010] and VM [(D); N = 6, t5 = 34.1, P < 0.001; t5 = 12.0, P < 0.001]. (E and F) Comparison of the mean effects of caffeine and DPCPX on IM [(E); Caff N = 10, t9 = 3.84, P = 0.004; DPCPX N = 6, t5 = 4.04, P = 0.010] and VM [(F) Caff N = 6, t5 = 34.1, P < 0.001; DPCPX N = 6, t5= 3.39, P = 0.019]. (G and H) Example traces [(G); rising phase shown on an expanded time scale below] and mean data [(H); rate of rise N = 6, t5 = 2.20, P = 0.079; τdecay N = 9, t8 = 3.54, P = 0.008] showing that DPCPX (100 nM) and caffeine (100 μM) slowed the decay and, in six of nine KCs, potentiated the fast component of the response evoked by exogenous ACh. (Student's paired t test used in all comparisons.) Mean responses ± SE.

The authors wanted to see what response caffeine induced in Kenyon cells (KCs), the cell type in a bee’s brain that is most similar hippocampal neurons in mammals—the cell type used for associative learning and memory formation.

The results of this experiments revealed that caffeine activates nicotinic acetylcholine receptors in the KCs, making them more likely to fire strongly in response to a sensory stimulus (like, for example, a floral odor). This makes it easier for these cells to help form memories associated with these odors and to learn to associate receiving sugar with that odor.

Nuts and bolts

In order to do see what the bees’ mushroom body neurons were doing, the authors took recordings of KC activity using a technique called “whole-cell patch-clamp electrophysiology.” For a detailed explanation of how this technique works, see the Author’s Experiment section of the SitC annotations.

Panels A and B show the recordings of an individual bee, rather than the whole group. Papers often show results of what they call a “representative sample”—an individual sample that shows the most common result, and so can be said to represent the group as a whole. This is useful for presenting data that are difficult or impossible to consolidate into averages for a whole group; for example, these individual spectra of nerve activity or tissue sections stained to show damage or protein expression patterns.

Panels A and B show one representative sample, but C and D show the mean data across the entire group for three critical time points: when the cells were at rest (before anything was added to them), when caffeine was added, and then when a nicotinic acetylcholine blocker was added. The bars represent standard error of the mean, which tells you how the individual data measured for each sample are distributed around the calculated mean value. An asterisk (*) indicates two measurements are statistically significantly different from each other.

The colored bars in panels E and F show the mean effect of DPCPX, a nicotinic acetylcholine receptor antagonist, on holding current. An asterisk (*) indicates two measurements are significantly different.

Panel G shows a representative trace when DPCPX was added to a KC. The trace on the bottom half of the panel is simply a blown-up view of the first few centimeters of the trace above it—this is the rising phase of the action potential. For a full explanation of action potentials and how neurons fire, see the Glossary section of the SitC annotations.

Panel H shows the mean data of the experiment whose representative trace was presented in panel G. The left y-axis shows the rate of the rising phase and the right y-axis shows the time of decay. The error bars represent standard error of the mean and an asterisk (*) indicates two measurements are significantly different.

Caffeine jolt

Panels A through D show that when added to KCs, caffeine increases nicotinic acetylcholine receptor activation, leading to a depolarized KC membrane potential toward the action potential firing threshold. This means that caffeine can trigger the neurons to start firing, which is required to form memories.

Who's your receptor?

Caffeine is already known to bind to nicotinic acetylcholine receptors and trigger a response, but the authors of this study didn’t want to make any assumptions. To confirm that caffeine was, in fact, using these receptors to elicit the responses seen in panels A through D, they added a compound that prevents caffeine from binding to these receptors a activating them.

Panels E through H show that when this compound, called DPCPX, is added to KCs, caffeine doesn’t induce the effect seen in panels A through D. However, when caffeine is added without DPCPX, that same push of the neurons toward the action potential firing threshold is seen again. These data indicate that caffeine does, in fact, activate nicotinic acetylcholine receptors.

Put another way, if caffeine wasn’t acting through these nicotinic acetylcholine receptors, then why would this effect have disappeared once we blocked these receptors? The most reasonable explanation for the results seen is that caffeine-induced activation of nicotinic acetylcholine receptors is what leads to the depolarization of the KC membrane potential seen in panels A through D.

Caffeine is bitter tasting to mammals and is both toxic (24) and repellent to honeybees at high concentrations (25,26). If bees can detect caffeine, they might learn to avoid flowers offering nectar containing it (27). We found that honeybees were deterred from drinking sucrose solutions containing caffeine at concentrations greater than 1 mM (Fig. 4); they also have neurons that detect caffeine in sensilla on their mouthparts (fig. S3). However, nectar concentrations did not exceed 0.3 mM (0.058 mg/ml), even though levels of caffeine in vegetative and seed tissues of Coffea have been reported to be as great as 24 mg/ml (28). This implies that pollinators drive selection toward concentrations of caffeine that are not repellent but still pharmacologically active.
Fig. 4.  Bees are more likely to reject sucrose solutions containing caffeine at concentrations greater than 1 mM (logistic regression, χ42 = 23.4, P < 0.001; for 0.7 and 1.0 M, 1 mM caffeine versus sucrose post hoc, P < 0.05; for 0.3 M, 100 mM caffeine versus sucrose post hoc, P < 0.05). Bees were less likely to drink 0.3 M sucrose (pale pink diamonds) than 0.7 M (pink circles) or 1.0 M solutions (red triangles) (logistic regression, χ22 = 8.69, P = 0.013). Mean responses ± SE. N0.3M = 29, N0.7M = 100, N1.0M = 20.

The authors were interested to see how bees would respond to the nectar levels of caffeine determined in Figure 1. They found that the bees’ bitter taste threshold, and therefore the concentrations of caffeine bees find repellent, are three times higher than those found in the nectar of the flowering plants tested in this paper.

Nuts and bolts

This figure presents the probability that a bee would drink a 0.4-microliter drop of solution on the y-axis and increasing amounts of caffeine on the x-axis. The data points show the average probability for all bees fed the same concentration of sucrose (either 0.3 M, 0.7 M, or 1 M) at that concentration of caffeine. The bars represent the standard error of the mean, which tells you how the individual probabilities measured for each bee are distributed around the calculated mean value.

Bitter taste threshold

This figure shows that bees are more likely to drink a solution if it contains more sucrose. However, once caffeine concentrations reach 1 mM, bees are very unlikely to drink even a high concentration sucrose solution; they are deterred by larger amounts of caffeine present in their reward.

Our data show that plant-produced alkaloids like caffeine have a role in addition to defense: They can pharmacologically manipulate a pollinator's behavior. When bees and other pollinators learn to associate floral scent with food while foraging (29), they are more likely to visit flowers bearing the same scent signals. Such behavior increases their foraging efficiency (30) while concomitantly leading to more effective pollination (3132). Our experiments suggest that by affecting a pollinator's memory, plants reap the reproductive benefits arising from enhanced pollinator fidelity.
Supplementary Materials
Materials and Methods
Supplementary Text
Figs. S1 to S3
References (3336)
References and Notes
  1. R. J. Sullivan, E. H. Hagen, P. Hammerstein, Revealing the paradox of drug reward in human evolution. Proc. Biol. Sci. 275, 1231 (2008).
  2. J. B. Harborne, Introduction to Ecological Biochemistry (Academic Press, London, ed. 4, 1993).
  3. R. G. Hollingsworth, J. W. Armstrong, E. Campbell, Pest control: Caffeine as a repellent for slugs and snails. Nature 417, 915 (2002).
  4. M. Okamoto, T. Kita, H. Okuda, T. Tanaka, T. Nakashima, Effects of aging on acute toxicity of nicotine in rats. Pharmacol. Toxicol. 75, 1 (1994).
  5. T. Rudolph, K. Knudsen, A case of fatal caffeine poisoning. Acta Anaesthesiol. Scand. 54, 521 (2010).
  6. A. Nehlig, Are we dependent upon coffee and caffeine? A review on human and animal data. Neurosci. Biobehav. Rev. 23, 563 (1999).
  7. J. A. Kretschmar, T. W. Baumann, Caffeine in Citrus flowers. Phytochemistry 52, 19 (1999).
  8. T. H. Ricketts, G. C. Daily, P. R. Ehrlich, C. D. Michener, Economic value of tropical forest to coffee production. Proc. Natl. Acad. Sci. U.S.A. 101, 12579 (2004).
  9. D. W. Roubik, Tropical Agriculture: The value of bees to the coffee harvest. Nature 417, 708 (2002).
  10. Materials and methods are available in the supplementary materials in Science Online.
  11. M. E. Bitterman, R. Menzel, A. Fietz, S. Schafer, J. Comp. Psychol. 97, 107–119 (1983).
  12. S. B. Simons, D. A. Caruana, M. L. Zhao, S. M. Dudek, Caffeine-induced synaptic potentiation in hippocampal CA2 neurons. Nat. Neurosci. 15, 23 (2012).
  13. R. G. M. Morris, E. Anderson, G. S. Lynch, M. Baudry, Selective impairment of learning and blockade of long-term potentiation by an N-methyl-D-aspartate receptor antagonist, AP5. Nature 319, 774 (1986).
  14. M. Heisenberg, Mushroom body memoir: From maps to models. Nat. Rev. Neurosci. 4, 266 (2003).
  15. S. Oleskevich, J. D. Clements, M. V. Srinivasan, Long-term synaptic plasticity in the honeybee. J. Neurophysiol. 78, 528 (1997).
  16. N. J. Strausfeld, L. Hansen, Y. S. Li, R. S. Gomez, K. Ito, Evolution, discovery, and interpretations of arthropod mushroom bodies. Learn. Mem. 5, 11 (1998).
  17. B. Sperlágh, E. S. Vizi, The role of extracellular adenosine in chemical neurotransmission in the hippocampus and basal ganglia: Pharmacological and clinical aspects. Curr. Top. Med. Chem. 11, 1034 (2011).
  18. J. Perez-Orive et al., Oscillations and sparsening of odor representations in the mushroom body. Science 297, 359 (2002).
  19. P. Szyszka, A. Galkin, R. Menzel, Associative and non-associative plasticity in kenyon cells of the honeybee mushroom body. Front. Syst. Neurosci. 2, 3 (2008).
  20. R. Menzel, G. Manz, Neural plasticity of mushroom body-extrinsic neurons in the honeybee brain. J. Exp. Biol. 208, 4317 (2005).
  21. S. Cassenaer, G. Laurent, Conditional modulation of spike-timing-dependent plasticity for olfactory learning. Nature 482, 47 (2012).
  22. D. S. Galili, A. Lüdke, C. G. Galizia, P. Szyszka, H. Tanimoto, Olfactory trace conditioning in Drosophila. J. Neurosci. 31, 7240 (2011).
  23. P. Szyszka et al., Mind the gap: Olfactory trace conditioning in honeybees. J. Neurosci. 31, 7229 (2011).
  24. Detzel, M. Wink, Attraction, deterrance, or intoxication of bees (Apis mellifera) by plant allelochemicals. Chemoecology 4, 8 (1993).
  25. J. A. Mustard, L. Dews, A. Brugato, K. Dey, G. A. Wright, Consumption of an acute dose of caffeine reduces acquisition but not memory in the honey bee. Behav. Brain Res. 232, 217 (2012).
  26. N. Singaravelan, G. Nee'man, M. Inbar, I. Izhaki, Feeding responses of free-flying honeybees to secondary compounds mimicking floral nectars. J. Chem. Ecol. 31, 2791 (2005).
  27. G. A. Wright et al., Parallel reinforcement pathways for conditioned food aversions in the honeybee. Curr. Biol. 20, 2234 (2010).
  28. H. Ashihara, A. Crozier, Caffeine: A well known but little mentioned compound in plant science. Trends Plant Sci. 6, 407 (2001).
  29. G. A. Wright, F. P. Schiestl, The evolution of floral scent: The influence of olfactory learning by insect pollinators on the honest signalling of floral rewards. Funct. Ecol. 23, 841 (2009).
  30. L. Chittka, A. Gumbert, J. Kunze , Foraging dynamics of bumble bees: Correlates of movements within and between plant species. Behav. Ecol. 8, 239 (1997).
  31. R. Hopkins, M. D. Rausher, Pollinator-mediated selection on flower color allele drives reinforcement. Science 335, 1090 (2012).
  32. W. E. Kunin, Sex and the Single Mustard - Population-Density and Pollinator Behavior Effects on Seed-Set. Ecology 74, 2145 (1993).
  33. R. Menzel, G. Manz, R. Menzel, U. Greggers, Massed and spaced learning in honeybees: The role of CS, US, the intertrial interval, and the test interval. Learn. Mem. 8, 198 (2001).
  34. G. A. Wright, B. H. Smith, Variation in complex olfactory stimuli and its influence on odour recognition. Philos. R. Soc. Lond. B Biol. 271, 147 (2004).
  35. A. Husch, M. Paehler, D. Fusca, L. Paeger, P. Kloppenburg, Distinct electrophysiological properties in subtypes of nonspiking olfactory local interneurons correlate with their cell type-specific Ca2+ current profiles. J. Neurophysiol. 102, 2834 (2009).
  36. S. Oleskevich, Cholinergic synaptic transmission in insect mushroom bodies in vitro. J. Neurophysiol. 82, 1091 (1999). Abstract/FREE Full Text
  37. Acknowledgments: We thank staff at Centro Agronomico Tropicale de Investigacion et Ensenanza in Costa Rica for access to the Coffea collections and at Technological Education Institute of Crete for access to Citrus orchards; M. Thomson, K. Smith, F. Marion-Poll, and A. Popescu for help with the data collection; M. Thompson for beekeeping; and J. Harvey and C. Connolly for project support. This work was funded in part by the Linnean Society of London and by a UK government Insect Pollinators Initiative grant BB/I000968/1 to G.A.W. and a separate grant to C. Connolly (BB/1000313/1). Methods and additional data are available in the online supplementary materials. All data are archived on the Natural Environment Research Council Environmental Information Data Centre.


PDF Paper Download