1. The evolution of flowering plants has undoubtedly been influenced by a pollinator’s ability to learn to associate floral signals with food. Here, we address the question of ‘why’ flowers produce scent by examining the ways in which olfactory learning by insect pollinators could influence how floral scent emission evolves in plant populations.
2. Being provided with a floral scent signal allows pollinators to learn to be specific in their foraging habits, which could, in turn, produce a selective advantage for plants if sexual reproduction is limited by the income of compatible gametes. Learning studies with honeybees predict that pollinator-mediated selection for floral scent production should favour signals which are distinctive and exhibit low variation within species because these signals are learned faster. Social bees quickly learn to associate scent with the presence of nectar, and their ability to do this is generally faster and more reliable than their ability to learn visual cues.
3. Pollinators rely on floral scent as a means of distinguishing honestly signalling flowers from deceptive ones. Furthermore, a pollinator’s sensitivity to differences in nectar rewards can bias the way that it responds to floral scent. This mechanism may select for flowers that provide olfactory signals as an honest indicator of the presence of nectar or which select against the production of a detectable scent signal when no nectar is present.
4. We expect that an important yet commonly overlooked function of floral scent is an improvement in short-term pollinator specificity which provides an advantage to both pollinator and plant over the use of a visual signal alone. This, in turn, impacts the evolution of plant mating systems via its influence on the species-specific patterns of floral visitation by pollinators.
Many plants produce flowers that are multi-sensory advertisements which lure pollinators into contact with a plant’s reproductive structures. In turn, pollinators rely both on the visual and olfactory cues provided by these advertisements to locate and identify flowers with food resources such as nectar, pollen and oils (Dobson 1994). Floral visual cues clearly make it easier for pollinators to locate floral resources, as nectar and pollen are located in or on structures which pollinators can see. The exact role of floral scent cues and their use by pollinators is less obvious, however. Floral scent is important in situations where relying on visual signals is more difficult, such as its use by pollinators that forage on night-blooming flowers (Raguso & Willis 2002; Raguso et al. 2003). Scent is also crucial for the pollination of plants which infiltrate specialized olfactory relationships (e.g. the use of female pheromones by male insects), as the scent produced is the signal of primary attraction to the pollinator (Schiestl 2005). If a visual signal was sufficient for diurnally active pollinators to learn about rewarding nectar, producing a scent could be unnecessary, yet thousands of plant species actively emit specific floral scent signals. This is in spite of the fact that producing scent is both metabolically costly (Vogel 1983) and risky, as it may attract unwanted visitors such as herbivores (Baldwin et al. 1997). Scent production as advertisement of reward must, therefore, deliver a net fitness benefit for plants.
Understanding the function of floral scent in plant–pollinator relationships has largely been addressed by examining how scent increases the attractiveness of flowers either on its own (Pichersky & Gang 2000) or in synergy with visual cues (Kunze & Gumbert 2001; Raguso & Willis 2002). We argue here that it is the ability of insect pollinators in particular to learn and remember floral scent in the context of nectar foraging plays an important role in the evolution of plant–pollinator relationships. Most insects have excellent olfactory acuity and can learn to associate odours with food. Unlike other appetitive olfactory signals, the chemical compounds emitted by flowers as scent are not directly food-related, insofar as they are not volatile representatives of the chemical composition of food, however. Instead, floral scent is a complex blend of volatile compounds produced by floral tissues (Dudareva & Pichersky 2006): it is a chemical signal whose composition is unrelated to the compounds found in nectar. The dissociation of signal from outcome produces the opportunity for dishonest signallers to invade a primarily mutualistic relationship (Dobson 1994), and presents pollinators with a dilemma: they must determine whether a floral scent is always associated with nectar. Is there evidence that the ability of insect pollinators to learn floral scents shapes the nature of floral scent in a way that keeps plants honest about the presence of nectar?
To answer the ‘why’ of floral scent production, it is necessary first to identify key factors of plant fitness influenced by the plant–pollinator mutualism. These are often dictated by the fact that a plant’s reproductive fitness is affected by its ability to produce seeds and to export pollen. Self-pollination lowers fitness as it reduces the amount of pollen available for out-crossing (i.e. pollen discounting) and leads to the production of seeds with lower fitness as a result of inbreeding depression (Barrett 2002). Thus, the pattern of floral visitation by pollinators greatly influences plant reproductive fitness: inbreeding, geitonogamy, and pollen loss are largely determined by pollinator behaviour (Chittka, Thomson & Waser 1999). For plants, the ideal pollinator transfers pollen efficiently between individuals of the same species. This is best achieved by short visits to individual flowers to avoid self-pollination and repeated visits to several flowers of different plants of a single plant species. This type of pollinator behaviour is an advantage because little or no pollen is lost, pollen is distributed to many conspecific stigmas and stigmas do not become ‘clogged’ with non-compatible pollen (Klinkhamer & Dejong 1993; Waser & Campbell 2004).
While generalist pollinators have the option of visiting different flower types in one floral ‘patch’, they often specialize on a short-term basis, and so, fulfil a plant’s goals. Many pollinators learn these signals to predict those flowers offering the highest quality rewards (Waser 1986; Chittka, Thomson & Waser 1999; Chittka & Thomson 2001). For a plant to enable and foster the tendency of pollinators to specialize, it must provide them with a reliable cue representing the best means of identifying its flowers. The ability of pollinators to learn floral cues, concomitantly, is likely to select for the evolution of honest, species-specific floral signals probably through stabilizing selection within plant species and disruptive selection between plant species. We argue here that olfactory cues are often the basis upon which pollinators decide to visit flowers because scent cues are easily learned and remembered by pollinators and because floral scent makes flowers more distinctive. In this review, we will also discuss the conditions under which we expect selection to favour floral scent production. In general, we expect floral scent to convey the greatest fitness advantage to a plant when it engages pollinators to visit the flowers of conspecific plants with a high frequency.
Olfactory learning in insects: are pollinators predisposed to favour odour cues?
Although field studies of pollinator learning behaviour have historically focused on a pollinator’s attention to differences in visual signals (but see Wells & Wells 1985), we argue here that floral scent is an important cue that insect pollinators learn to associate with nectar because of the dominant role that odours play in the life history strategies of insects. Throughout the animal kingdom, the co-evolution of signals and signal receivers is often influenced by the sensory capabilities of the receivers (Endler & Basolo 1998; Ten Cate & Rowe 2007). Insects depend upon olfactory signals for their most vital activities: feeding and mating. Indeed, most insects have large antennae with several types of sensilla housing a diverse array of chemical-sensing neurones which provide them with the capacity to detect and differentiate chemical compounds (Chapman 1998). The insect olfactory system in some cases has become highly specialized – for example, many moth species have large antennae with highly sensitive olfactory receptor neurones that can detect minute amounts of female pheromone (Kaissling & Kasang 1978). Some insects are also quite specific in their behavioural reactions to odours emitted by food: herbivores use plant volatile compounds to identify hosts (Raffa 2001); carrion flies use sulphide compounds emitted by bacteria to identify rotting carcases for oviposition (Stensmyr et al. 2002), and mosquitoes and blood feeding insects use CO2 coupled with other volatile metabolites such as lactones to identify hosts (Dekker et al. 2002).
Insects also have the ability to learn to associate general chemical cues with food (Papaj & Lewis 1993; De Boer & Dicke 2006; Dukas 2008). Hemimetabolous insects, such as cockroaches, crickets, and locusts, learn to associate scent with food (Dukas & Bernays 2000; Sakura & Mizunami 2001; Matsumoto & Mizunami 2004) and can even differentiate food types (e.g. protein vs. carbohydrate) using scent cues (Gadd & Raubenheimer 2000). However, some of the best examples of olfactory learning in insects are from holometabolous insects (e.g. Diptera, Lepidoptera and Hymenoptera). The fruit fly Drosophila melanogaster learns both how to use scent to find food and how to avoid odours paired with the pain caused by electric shock (Tully & Quinn 1985); as such, it has become an important model system for studying olfactory learning and memory (Keene & Waddell 2005). An insect’s brain size does not limit its ability to learn odours: insects as large as the American cockroach Periplaneta americana and as minute as Microgastrine parasitoid wasps learn (Vet & Groenewold 1990; Wackers 2004).
The fact that odours are commonly associated with food may have driven the evolution of the neural structures necessary for facilitating not only the olfactory sensory capacities of insects, but also the integration of gustatory and olfactory information in the areas of the brain necessary for performing food-related learning tasks. While most, if not all, insects can learn odours, generalist pollinators such as social bees are renowned for their olfactory learning abilities, and indeed, are among the best studied insects for this reason (Menzel 1999; Giurfa 2007). Furthermore, even pollinators with clear biases towards the scents of specific flowers (e.g. hawkmoths) readily learn to associate other odours with food rewards and change their foraging preferences accordingly (Riffell et al. 2008). Importantly, pollinators such as bees learn odours and remember them for longer than visual cues (Menzel 1985, 1991; Gould 1996; Kunze & Gumbert 2001). Using social bees (i.e. honeybees and bumblebees) as examples, we argue here that a plant’s emission of scent as a means of advertising floral rewards and a pollinator’s attendance to scent signals provide fitness advantages to both plant and pollinator which exceed those resulting from the use of visual signals alone.
Floral scent: a cue for learning about nectar rewards
The fact that many insect pollinators exhibit advanced olfactory learning abilities strongly suggests that reliance on olfactory cues for identifying rewarding nectar and pollen occurs not only as a result of an insect’s general trait of possessing the sensory and neural capacity for olfactory learning: learning odours must also convey a fitness advantage. In general, associative learning allows animals to adapt to local environmental conditions – most, if not all, animals have the ability to learn. Interestingly, in a survey conducted by Angermeier (1984) on the speed with which well-studied animals form learned associations, the honeybee (a generalist pollinator) learned to perform the simple task of associating an olfactory cue with food at the fastest measured rate of any animal. Honeybees learn to associate scent with sucrose solution after a single learning trial (Friedrich, Thomas & Muller 2004); robust, long-term olfactory memory formation for the learned association occurs after only three trials (Menzel 1999). This implies that from a plant’s perspective flowers which emit scent are more likely to secure the attention of pollinators, an important trait which could provide a plant with a competitive advantage. That the memory for scent lasts several days is also likely to improve the chance that a pollinator will visit flowers of the same species later in its foraging history. Honeybees can also learn to avoid odours that are not paired with reward, though the rate at which this learning occurs is much slower (Wright et al. 2005a). Importantly, many bee species can learn to attend to dilute or atypical scent cues such as the footprints left by previous floral visitors, in order to avoid visiting flowers recently depleted of nectar (Giurfa 1993; Stout & Goulson 2001).
Visual cues are obviously essential for orientation towards floral food sources, but whether or not a pollinator chooses to alight or probe for nectar may depend on whether a learned scent signal is present (Raguso & Willis 2005) and whether the scent signal is ‘correct’ (Kunze & Gumbert 2001). For example, honeybees visit flowers more frequently and longer when the flower bears a scent they have previously associated with food (Fig. 1). Honeybees learn scent faster than visual cues and form stronger associations of scent with reward than visual cues (Menzel 1985, 1991; for a review, see Gould 1996; Hori et al. 2006); in fact, when honeybees are trained in a laboratory, olfactory cues interfere with visual learning to such an extent that animals with intact antennae have considerable difficulty in learning to perform the proboscis extension response towards a visual cue associated with food (Kuwubara 1957; Hori et al. 2006; Hempel de Ibarra et al. 2009). Though free-flying honeybees exhibit better performance in visual learning tasks than restrained honeybees, the ability to acquire the association is still slower than that produced during olfactory learning (Menzel 1985, 1991; Gould 1996; Kunze & Gumbert 2001; Giurfa 2004).
Floral scents learned in other contexts also influence the decision-making of foraging pollinators. Floral nectar sometimes contains scent compounds as well as nutritional ones (Raguso 2004) and returning foragers can themselves become scented via contact with floral surfaces (Molet, Chittka & Raine 2008). Honeybees have the ability to learn to associate odour with food while sharing nectar via trophallaxis with nestmates (Gil & De Marco 2005), and bumblebees also learn scents acquired by foragers as a result of contact chemoreception (Molet, Chittka & Raine 2008). The fact that scent can be transferred from the foraging site to a social bee’s colony makes floral scent potentially advantageous for both pollinator and plant: whereas a visual signal is something that can only be perceived by an individual forager, scent can be physically transferred between social bees as a means of communicating good floral resources within a hive (Gil & Farina 2003). Furthermore, scent compounds within nectar are often concentrated in food stores and wax comb, perfuming a honeybees’ nest; this scent then influences how bees forage, as they are more likely to visit a flower or feeder with the same scent (Von Frisch 1967; Reinhard, Srinivasan & Zhang 2004; Farina et al. 2007; Molet, Chittka & Raine 2008). The fact that social bees use scent to communicate information about profitable floral resources is also a particular advantage to a plant species because it could potentially mean the recruitment of an entire colony of nest mates.
Whether or not odour affects learning and memory in other insect pollinators to the same extent it does honeybees is not yet well understood, though it is possible that odours also play an equally important role in garnering their attention. Moths can also learn to associate odours with food rewards (Daly, Durtschi & Smith 2001; Skiri et al. 2005), and for them, scent is a crucial aspect of the multimodal signal. If the odour signal is not present, even though the visual floral signal is correct, the hawkmoth Manduca sexta will not attempt to probe a flower for nectar (Raguso & Willis 2002). On the other hand, the butterfly Vanessa indica appears to prioritize colour over floral scent when choosing to forage on flowers (Omura & Honda 2005). Even though colour is important for this species, the addition of a scent cue to a flower model dramatically increases the probability that V. indica will try to visit any flowers, including those with the preferred colour.
Olfaction: discrimination and specificity of foraging
Floral scent improves a pollinator’s ability to distinguish similar flowers, not just by scent, but also by providing multimodal information (i.e. a visual and an olfactory cue) which can be used to identify flowers. Like other animals, bumblebees are better at discriminating objects when they are presented with multimodal cues such as a visual cue paired with an olfactory cue (Gegear & Laverty 2001, 2005; Kunze & Gumbert 2001; Kulahci, Dornhaus & Papaj 2008). The improvement in performance above what a pollinator can achieve using only an olfactory cue, however, is small in comparison with the benefits of adding scent to a visual cue (Gegear & Laverty 2001, 2005; Kunze et al., 2001; Kulahci, Dornhaus & Papaj 2008). Additionally, when bumblebees are given the opportunity to forage on several possible types of artificial flowers, they exhibit a greater degree of floral constancy if the artificial flowers differ with respect to olfactory rather than visual cues (Wells & Wells 1985; Gegear & Laverty 2001). Floral scent, therefore, also allows a pollinator to be more specific as it can reject flowers with similar signals but different rewards. This ultimately provides a plant with an advantage as pollinators are more likely to exhibit floral constancy when multimodal cues are available (Gegear & Laverty 2001, 2001, 2005) and, therefore, be more efficient pollen vectors.
Furthermore, floral scent itself may be the local cue used to determine whether or not a pollinator actually chooses to alight, and, therefore, come into contact with a flower’s stamens and stigma (Raguso & Willis 2002). Early work by Manning (1956) reported that the addition of ‘rose’ scent to dandelion flowers (Taraxacum officinale) caused foraging bumblebees to reject these flowers immediately after alighting prior to probing for nectar, suggesting that they detected the change in scent cue at close range and then rejected the flower. In a similar experiment, we added a single scent compound (1-hexanol) to the pistils of flowers of borage, alkanet or grapefruit and recorded the behaviour of floral-constant, foraging bee species (honeybees and solitary bees) (Fig. 2). The bees observed in this study were foraging on other flowers of the same species within a patch and had presumably learned to associate these floral traits (e.g. for borage: blue, radial symmetry, no scent) with nectar. The flowers were also protected from floral visitors by fine mesh netting for 1 h prior to the addition of scent. In most cases, the bees rejected flowers with scent added in flight and chose not to alight, even though nothing about the visual signal had been altered. Two of the three flower types in this study (borage and alkanet) did not have an odour (as perceived by the human nose); in this case, a bee’s rejection of scent-modified flowers indicates that scent was not simply ignored by the bees in favour of the recognized visual stimulus, as the visual stimulus was unaltered. Furthermore, the addition of 1-hexanol to grapefruit flowers which have a strong scent also caused bees to reject scent-modified flowers. Bees do not instinctually avoid 1-hexanol; it is, in fact, found in the floral scents of many plant species (Knudsen et al. 2006) and commonly used in olfactory learning experiments with honeybees (Fig. 3; Wright, Kottcamp & Thomson 2008). Thus, these data support the idea that a flower’s olfactory signal may be the basis upon which many pollinators, especially social bees, decide to alight and probe for nectar (Gould 1996; Kunze et al. 2001; Raguso & Willis 2002).
Finally, using floral scent cues to identify rewarding flowers at close range, rather than relying on visual cues, may allow a pollinator to react in a manner that is both fast and accurate. In generalist pollination systems, many types of pollinator may visit a plant species’ flowers, producing a situation where pollinators are competing for access to nectar resources. Pollinators may be forced to quickly decide about whether to alight on a flower or not, placing them in a situation where they must trade-off speed for accuracy (Chittka et al. 2003; Chittka & Spaethe 2007; Chittka, Skorupski & Raine 2009). Restrained honeybees are able to accurately discriminate and respond to strong floral scents within 0·43 s of perceiving them, and while discrimination of very faint floral scents takes longer (0·57 s) (Wright, Carlton & Smith, 2009), this is still much faster than the measured response times of free-flying bumblebees asked to discriminate flowers on the basis of colour (fastest: ∼ 6 s; most accurate: over 9 s) (Chitka et al. 2003). Though the speed-accuracy trade-off has not been measured using olfactory cues with free-flying bees, the large difference in response times in the two tasks still suggests that bees are faced with less of a speed-accuracy trade-off when using scent cues than when using visual ones. Whether or not bees are both faster and more accurate using both visual and olfactory cues has not yet been tested.
How do pollinators influence the evolution of floral scent?
Floral scent is clearly an important cue that pollinators learn, but as signals go, it is difficult to produce reliably. Scent compounds are emitted into the air surrounding a flower and dispersed as a plume of molecules the properties of which depend on many physical variables such as temperature and air velocity (Crimaldi, Wiley & Koseff 2002). In addition, the complex blend comprising a floral scent changes dramatically over a diurnal course of emission (Helsper et al. 1998; Raguso et al. 2003; Dudareva & Pichersky 2000, 2006; for a review, see Wright & Thomson 2005). In fact, both the quantities of each compound and the overall intensity of the scent vary diurnally (Helsper et al. 1998). Many factors influence this complexity and variation, not all of which are related to pollinator-mediated selection. Pre-adaptations can influence the nature of odour bouquets emitted by plants (Armbruster et al. 1997; Schiestl & Cozzolino 2008). The number and types of compounds emitted depends on plant genotype, varieties or populations (Kim et al. 2000; Wright et al. 2005a) and also on environmental conditions such as light intensity and temperature (Jakobsen & Olsen 1994). Furthermore, enzymes producing scent compounds are sometimes not specific to the production of floral volatiles, as exemplified in Arabidopsis thaliana, where only two genes encode the enzymes responsible for the complex blend of sesquiterpenes emitted by the flowers (Tholl et al. 2005). Such circumstances make it difficult to predict exactly which compounds will be emitted and in what proportions; the volatiles that comprise a blend may, instead, be mere by-products of biosynthesis and not the target of selection. Thus, the production of a reliable floral scent signal by a population of plants could be constrained by factors unrelated to communication with pollinators.
On the other hand, a pollinator’s ability to learn floral scent is affected by the reliability of the scent signal. The rate of learning, olfactory memory formation and the rate of generalization towards other odours all depend upon signal reliability (Wright & Smith 2004; Wright & Thomson 2005; Wright, Kottcamp & Thomson 2008). What a pollinator learns about a scent signal is also influenced by the quality and quantity of the reward associated with odour (Drezner-Levy & Shafir 2007; Wright, Choudhary & Bentley 2009). The fact that producing a reliable signal is both difficult but potentially important again begs the question: why do plants produce floral scent? Here, we argue that using scent as a signal retains a selective advantage for honest signallers competing for the attentions of insect pollinators. We expect that insect pollinator-mediated selection for plants that honestly signal the presence of food rewards reduces variability in signal production within plant populations, but increases species-distinctiveness of floral scent. We also predict that this selection is not independent of nectar production.
Selection for floral scents with low variation
Although both adaptive and non-adaptive processes shape the nature of floral scent, several lines of evidence including those from macroevolutionary studies which compare the scent bouquets of several plant lineages suggest a primary role for pollinator-driven selection. Although the differences in the chemical composition of floral scents are generally greater between species than within species (Wright et al. 2005a; Dobson 2006), unrelated plant species may also convergently evolve floral scents with common volatile compounds as a result of selection by a specific pollinator clientele (Andersson et al. 2002; Fenster et al. 2004; Dobson 2006). A good example of this is the floral scent of night-blooming, hawkmoth-pollinated plants which emit a suite of common compounds (Raguso et al. 2003). In certain orchid species, a subset of the volatile compounds emitted in the floral scent elicit responses in an insect pollinator’s olfactory receptor neurones; these same compounds also exhibit less variation in concentration than the rest of the volatiles comprising the orchid’s scent (Huber et al. 2005; Mant, Peakall & Schiestl 2005; Salzmann et al. 2007b). In contrast to selection for the production of specific compounds, flowers that are primarily bird pollinated often produce little or no scent (Knudsen et al. 2004). Such correlations in floral scent emission and pollinator responses suggest that pollinators exert stabilizing selection on the production of specific scent compounds, reducing variation both in composition and quantity in plant populations.
Variation in floral scent from flower to flower affects not only what pollinators learn about a floral scent signal (Wright & Smith 2004; Wright, Kottcamp & Thomson 2008; Wright, Choudhary & Bentley 2009); it also affects the speed at which they learn floral scents, and potentially, their use of floral scent to recognize flowers. If, for example, the proportion of individual compounds in a complex scent varies in concentration from trial to trial during a simple olfactory conditioning task, honeybees learn to associate an olfactory signal with reward at a slower rate than if there is little or no variation (Fig. 3). Likewise, if floral scents exhibit low variation, honeybees are much more specific about the way that they respond to new odours (Wright & Smith 2004; Wright, Kottcamp & Thomson 2008). In particular, if they experience variation in the scent signal but not in the reward associated with scent, they are much more likely to respond to a novel odour than if they experienced no variation (Wright & Smith 2004b; Wright, Kottcamp & Thomson 2008). We expect, therefore, that because variation affects not only the speed of learning but also pollinator specificity (and, therefore, floral constancy) that plants which offer rewards to pollinators may be under selection to reduce variation in floral scent emission.
Distinct scents are better advertisements
Another important factor affecting pollinator behaviour is the distinctiveness of floral signals relative to the signals of other co-flowering species (Chittka, Thomson & Waser 1999). Like other animals, pollinators must by default generalize from learned events to new situations, as no two stimuli are ever experienced in exactly the same way (Pavlov 1927; Shepard 1987). While foraging, pollinators are often forced to visit several flowers to meet their nutritional needs. Indeed, if one single flower provided a pollinator’s complete food requirements, it would have no motivation to visit other flowers and so would fail to accomplish the plant’s goal: to outcross (Klinkhamer & Dejong 1993). Thus, a pollinator is driven to visit many flowers, and on each visit to a new flower, it must compare the signals of recently visited flowers to those of new flowers. Constancy to any particular flower type, therefore, depends in part upon the extent to which a pollinator generalizes a learned association towards novel floral signals. Olfactory generalization, in turn, depends on the perceptual similarity of odours (Daly, Durtschi & Smith 2001; Wright & Smith 2004), their concentration or intensity (Wright & Smith 2004; Wright, Thomson & Smith 2005b) and the amount of variation experienced during olfactory learning (Wright, Kottcamp & Thomson 2008). If plants are competing for the attentions of pollinators with the goal of obtaining visitors that are frequent and constant, possessing a distinctive signal provides a selective advantage because it is easier for a pollinator both to distinguish this signal from the scents of other co-flowering species, causing it to generalize less and maintain floral specificity (Chittka, Thomson & Waser 1999).
If scent signals mediate pollinator foraging specificity, then selection should favour plants with distinctive scents as plant are often limited in their reproductive success by the arrival of compatible pollen (Johnson 2006). Scent composition may vary less in a plant population than scent quantity as closely related flowering plants tend to produce scent signals whose main source of variation is the proportion of the compounds present (Dobson et al. 1997; Schiestl & Ayasse 2002; Raguso et al. 2003; Gaskett, Conti & Schiestl 2005; Wright et al. 2005a; Knudsen et al. 2006). In fact, it is possible that in some cases, compounds that a pollinator may have previously evolved the capacity to detect could in turn influence selection on a specific suite of odour compounds, such that these compounds are present with lower variation than others in a floral bouquet (Huber et al. 2005; Mant, Peakall & Schiestl 2005). Selection should also be stronger in species-rich communities with presumably greater competition for pollinator services and larger differences in scent; if this is the case, then a reproductive character displacement scenario (Armbruster & Muchala 2008) could account for the evolution of species-specific scent bouquets. This prediction is supported by experimental evidence from Silene plants, in which species-specific odour bouquets enhance intra-specific pollen flow, and can thus foster the deposition of compatible pollen as well as reproductive isolation (Wälti et al. 2008).
Another way to make scent distinctive could be to produce unique volatile compounds (e.g. novel compounds or enantiomers) or highly complex floral scents. The production of chemically distinct floral scents might be less costly for plants because volatile compounds can be produced from a variety of different enzymatic pathways (Dudareva & Pichersky, 2000). In addition, when odours are complex, they often take on perceptual properties which are unrelated to those produced by their component compounds (Jinks & Laing 2001). While this has been studied in human subjects, considerably less is known about the way that odour signal complexity influences how a pollinator perceives floral scent, and only a few studies have shown that pollinators can perceive differences in complex blends of compounds (Grison-Pige et al. 2001; Wright, Skinner & Smith 2002; Wright et al. 2005a).
Odour signals, nectar rewards and learned olfactory biases
One aspect of the role of olfactory learning on the evolution of floral scent that we have hitherto ignored is the importance of the reward in dictating what the pollinator learns. Like scent signals, both nectar quantity and quality vary within plant populations (Herrera, Perez & Alonso 2006) and between plant species (Petanidou 2005). Some of the factors affecting nectar production are a plant’s water balance and nutrient status (Pacini, Nepi & Vesprini 2003), time of day (Matile 2006), the plant’s genotype (Witt et al. 1999) and damage caused by herbivores (Kessler & Halischke 2009). We expect pollinators to be sensitive to differences in floral rewards; indeed, both nectar quality and quantity influence pollinator behaviour (Cnaani, Thomson & Papaj 2006). For example, bumblebees (Bombus terrestris) are less floral constant when they experience floral signals associated with a low frequency of encountering rewards (Fontaine, Collin & Dajoz 2008). Furthermore, just as bees can learn to associate scent with reward, they can also learn to avoid scents associated with no reward (Greggers & Menzel 1993; Wright, Skinner & Smith 2002; Wright et al. 2005a). Thus, having a floral scent could potentially detract from a plant’s fitness if it did not produce sufficient nectar.
Differences in nectar quality and quantity experienced in association with slight variation in floral signals affect the way that bees generalize towards similar, but novel floral signals (Lynn, Cnaani & Papaj 2005; Wright, Kottcamp & Thomson 2008; Wright, Choudhary & Bentley 2009). In particular, this experience causes honeybees to be more specific about the way they respond to floral scents. For example, if two closely related odours are associated with distinctly different outcomes, honeybees respond more to odours associated with high quality rewards and less to those associated with poor quality rewards (Wright, Kottcamp & Thomson 2008; Wright, Choudhary & Bentley 2009). Indeed, this form of learning can produce a bias in the way that honeybees respond to scents, such that they respond more towards new scents they have not encountered which are perceptually similar to the scent associated with high quality nectar (Wright, Choudhary & Bentley 2009). This kind of bias depends on how great the differences in the ‘outcomes’ associated with floral scent are: when the differences in nectar quality are large, such as a situation in which one scent is associated with a high quality reward containing the amino acid, proline, and another, similar scent is associated with nectar-containing salt (i.e. NaCl), a large bias is formed (Wright, Choudhary & Bentley 2009). On the other hand, if two scents are associated with a small difference in nectar quality, then only a slight change in the way that a honeybee responds to new scents occurs (Wright, Choudhary & Bentley 2009).
Such generalization biases are likely to influence the evolution of scent signals in plant populations. Any bias towards odours associated with high quality signals would predispose a pollinator to visit flowers with floral scents similar to a highly rewarding plant. Likewise, scent signals must be sufficiently different from those of other flowers emitting scent but offering no nectar or poor quality nectar, as pollinators will otherwise avoid all scents similar to these dishonest signallers. Thus, even by emitting a scent a plant ‘makes a rod for its own back’, as it must continue to produce not only a strong, distinct scent signal, but also to produce sufficient quantities of high quality nectar. If it fails to do this, a plant risks causing its pollinators to learn to avoid its signals and any flowers like it, biasing pollinators away from the floral signals it produces.
Floral scent signals: an honest indicator of reward?
Encountering flowers with no nectar is common for pollinators, as many plants are facultatively or obligately rewardless (Renner 2006). Pollinators also compete for access to the same floral resources (Fontaine et al. 2008) potentially causing a high rate of nectar depletion. In fact, ‘food deceptive’ flowers (i.e. those that never offer a reward to pollinators) are an excellent means of testing hypotheses regarding the way that floral signals have been selected for by a pollinator’s ability to learn odours (Schiestl & Schlüter 2009). Deceptive plants are dependent upon pollinators that visit their flowers incidentally, either because they cannot discriminate them from their usual, specific food plants (i.e. Batesian mimicry) or because they generalize what they have learned about the signals of previously visited flowers and ‘mistakenly’ visit deceptive flowers that produce similar signals (i.e. generalized mimicry; Schiestl 2005). If the rate of this mistaken visitation is sufficient to facilitate out-crossing in plants, these plants could benefit from pollination services without being required to pay for those services by offering a reward (Johnson 2006).
As predicted by studies of olfactory learning in honeybees, food deceptive systems do not rely on the presence of scents which are shared between rewarding and non-rewarding flowers to facilitate pollinator visitation (Schiestl 2005): such flowers are mainly scentless (Jersakova & Johnson 2006), or emit weak and highly variable scents (Galizia et al. 2004; Salzmann, Cozzolino & Schiestl 2007a). The observations that scent is not commonly found in rewardless species and that bees attend carefully to scent signals could indicate that scent is used by pollinators as an honest indicator of the presence of nectar. In this case, pollination by deceit may depend on the fact that a pollinator’s ability to distinguish flowers is not as accurate with only a visual signal and, so, it is more likely that a pollinator will ‘mistakenly’ visit flowers possessing similar visual signals to those of rewarding flowers. Furthermore, in a seminal study regarding the evolution of deceptive signals, Dawkins & Guildford (1991) proposed that deceptive signals arise in situations where accurate decision-making is either costly or time-consuming. If, as we proposed before, pollinators are under constraints with respect to accuracy and speed, we expect that they are faced with a greater trade-off when forced to rely on visual cues alone to identify flowers than when relying on scent cues. Thus, nectarless plants would be more likely to receive mistaken visits if they use visual signals to deceive pollinators rather than scent, and would, therefore, achieve higher fitness.
In an important study on the use of multimodal signals in floral constancy by Kunze & Gumbert (2001), bumblebees had the greatest difficulty distinguishing flowers with similar scents, even if the flowers could be distinguished visually. The authors predicted that floral mimics should either imitate the scent bouquets of a specific model species (i.e. Batesian mimicry) or evolve to be functionally scentless in order to prevent pollinators from learning to avoid scents associated with the absence of reward. To test whether this is found in a natural setting, the floral scent production of two closely related orchid species, one of which possessed rewarding flowers and the other which possessed nectarless flowers was examined by Salzmann et al. (2007b). The total amount of scent produced by the nectarless orchids was lower and the overall variation in scent composition was much higher in the deceptive species than in the rewarding species (Salzmann et al. 2007b). Furthermore, the fitness of the plants, measured as the sum of flowers pollinated, exhibited a downward trend when scent was added to deceptive species, suggesting that plants without nectar are at a disadvantage if they possess a recognizable scent signal. Floral scent and its association with the presence of nectar in both of these studies appears to make it more difficult for dishonest plants – plants that do not offer nectar to pollinators – to infiltrate the olfactory signaller–receiver relationship because pollinators can rapidly update what they know about the signal and its associated outcome as a result of learning.
The presence of a floral scent signal clearly affects pollinator behaviour and, therefore, is likely to be a crucial part of floral signalling involved in plant reproductive success. Pollinator olfactory sensory acuity and olfactory learning are an important part of the selective environment that shapes the evolution of floral signals through their impact on plant fitness, but our understanding of the exact means by which olfactory learning shapes the expression of this floral trait remains relatively poor. It will be important in future studies to test exactly how floral scent complexity affects a pollinator’s ability to discriminate between flowers and to test how information about nectar reward quality influences what is learned about floral scent. Furthermore, testing how much variation in floral scent naturally occurs in plant populations relative to plant mating strategies (see Ashman, this special feature) may elucidate how floral scent has evolved. In particular, studies of this kind should focus on providing explicit examples of scent polymorphism within species and its influence on pollinator behaviour and pollen deposition. By measuring variation in scent production in plant populations and by using techniques which measure the hypothetical (using fluorescent dye) or realized (using molecular markers) forms of gene flow (see Whitehead & Peakall, this special feature), it should be possible to predict how changes in floral scent affect pollinator behaviour in specific ecological contexts. In particular, the recent advance in molecular gene silencing techniques in planta also offers exciting possibilities for field-based experiments of changes in floral scent and its influence on pollinator behaviour (Kessler, Gase & Baldwin 2008).
The authors thank Robert Raguso and two anonymous referees for comments on the manuscript. Ev Christ and Guido Hüni collected the data presented in Fig. 1. The data presented in Fig. 2 were collected by two teams of students at Newcastle University’s field course to TEI, Heraklion, Crete, during 2008 and 2009.