Fragrance is an ancient medium of chemical communication between flowering plants and animal pollinators (Pellmyr & Thien 1986). Pollinators use fragrance for distance orientation, approach, landing, feeding, and associative learning (Williams 1983; Metcalf 1987; Dobson 1994). In turn, scent-driven pollinator preference and constancy has been invoked as an isolating mechanism for diverse angiosperm taxa (Dodson et al. 1969), particularly among sympatric, synchronously blooming species with similar floral form, coloration, and pollination mechanisms (Knudsen 1999). In this chapter, I explore the potential for odor-driven floral evolution by reviewing the physiological and behavioral responses of pollinators to floral scent.
What is fragrance? A floral scent primer
Floral scents are mixtures of small, volatile organic compounds that vary in molecular weight, vapor pressure, polarity, and oxidation state (Knudsen et al. 1993). Diverse chemical classes of floral volatiles are surveyed comprehensively by Croteau & Karp (1991). The analytical methods used to collect and identify floral scent have improved dramatically over the past decade, and are discussed by Raguso & Pellmyr (1998) and Agelopoulos & Pickett (1998). Floral volatiles are produced by biosyn-thetic pathways, through anabolic and catabolic processes. Figure 5.1 summarizes the major biosynthetic routes to fragrance production, illustrating representative products for each pathway (Azuma et al. 1997). These multifunctional pathways also produce plant pigments, defense
Fig. s.i. The major biosynthetic routes to volatile production in plants. Pathway names are given in italics; representative metabolites and end products are capitalized; volatiles are bold-faced and capitalized. (Modified from Pare & Tumlinson 1997.)
compounds, structural components, growth, and signaling substances (Dixon & Paiva 1995). Recent progress in fragrance biosynthesis is reviewed by Dudareva et al. (1999) and Dudareva & Pichersky (2000).
Floral scent variation: defining the phenotype
Variation in fragrance chemistry is prerequisite for scent-driven floral evolution. Geographic, altitudinal, and intrapopulation variation are frequently encountered when large samples are analyzed (Dodson et al. 1969; Tollsten & Bergström 1993). Fragrance varies spatially within flowers, in tissue-specific patterns and odor gradients that convey information as nectar or pollen guides (Lex 1954; von Aufsess 1960; Dobson 1987), and can be emitted from nearly any floral tissues, from surface epidermal cells to glandular trichomes or multicellular osmophores (Vogel 1963; Stern et al. 1987). Finally, fragrance composition and emission rates may vary temporally according to circadian rhythms and post-pollination changes (Dudareva et al. 1999).
The assumption of simple Mendelian inheritance of fragrance chemistry, combined with blend-specific behavioral responses by euglossine bees, prompted speculation that fragrance mutants could give rise to floral isolation and sympatric speciation in Neotropical orchids (Dressler 1968; Hills et al. 1972). Chemical analyses of interspecific Fi hybrids of Anthurium (Araceae: Kuanprasert et al. 1998) and Cycnoches (Orchidaceae: Gregg 1983) yielded hybrid scent profiles that were qualitatively additive and quantitatively intermediate between parental phenotypes, suggesting polygenic inheritance or at least codominance. In the first balanced genetic analysis of fragrance, terpenoids segregated as dominant traits in interspecific Clarkia (Onagraceae) hybrids, but quantitative variation in emission rates was not correlated with floral morphology or environmental factors (Raguso & Pichersky 1999). This pattern may reflect the independent assortment of regulatory elements, substrate fluxes, or ultracellular factors (Curry 1987; Skubatz et al. 1996). Intraspecific crosses between lines of C. breweri that were polymorphic for methyleugenol yielded similar results, suggesting that biosynthetic enzyme activity alone does not explain phenotypic variation in fragrance (Wang & Pichersky 1999). Temperature, relative humidity, photoperiod, and edaphic conditions also contribute to plasticity in fragrance emissions (Hansted et al. 1994); "norm of reaction" studies with isolated genotypes would greatly contribute to defining fragrance phenotypes. Quantitative trait locus analysis of fragrance variation in a model plant system with a genetic map (e.g., Petunia) may be the best way to dissect phenotypic variation in floral scent (cf. Rieseberg & Noyes 1998).
Multidimensional odor space; the natural distribution of floral scent
Distinct human-defined fragrance types traditionally have been grouped with specific pollinator classes (Knuth 1906; Vogel 1954; Faegri & van der Pijl 1979). However, no scheme relates fragrances to pollinator olfactory perception in the way that the color hexagon plots floral "visual space" vis-à-vis trichromatic visual perception (Chittka 1992; Menzel & Shmida 1993). Fragrances defy categorization along any single axis of physical properties. Various ordination techniques, such as principal components (Tollsten 1993) and multidimensional scaling (Dobson et al. 1997) are useful in visualizing differences in fragrance chemistry as multidimensional entities with characteristic locations in "odor space."
Does fragrance chemistry really covary with distinct pollinator guilds or foraging modes (Armbruster 1990).? Combined chemical analyses and bioassays have identified putative products of convergent evolution for pollinator attraction. For example, carvone oxide is unique to flowers pollinated by male euglossine bees, such as orchids, aroids, and Dalechampia euphorbs (Whitten et al. 1986). Knudsen & Tollsten (1993, 1995) used standard analytical methods to test the fit between fragrance chemistry and "pollinator syndromes" by analyzing fragrances from Neotropical hawkmoth- and bat-pollinated plant guilds. Hawkmoth-pollinated flowers produced "strong, sweet" scents that shared distinctive compound classes (oxygenated terpenoids, nitrogenous oximes), while the "fermented, garlicky" odors of bat-pollinated flowers shared sulfides rarely found in other angiosperms (see Winter and von Helversen, this volume). Independent surveys by Bestmann et al. (1997) and Miyake et al. (1998) confirmed these conclusions. However, interspecific fragrance variation was greater than expected, and compounds common to diurnal, generalist-pollinated flowers were present in nearly all taxa. These findings suggest that plants in pollinator guilds converge upon certain core constituents as required attractants while maintaining species-specific blends. This implies both innate and learned pollinator responses. Alternatively, fragrance may evolve in response to more than one pollinator class (Knudsen et al. 1999), and may be constrained by phy-logenetic differences between guild members (Armbruster 1997) or selective pressures by enemies (Baldwin et al. 1997).
Most fragrance compounds appear to be too homoplaseous to be phy-logenetically informative above the genus level (Dobson et al. 1997). Two recent studies (Azuma et al. 1999; Williams & Whitten 1999) used improved methods to re-evaluate earlier chemotaxonomic surveys of Magnolia trees and Stanhopea orchids. Fragrance data suggested multiple origins for the genus Michelia (Azuma et al. 1997), while chloroplast DNA analyses supported a monophyletic Michelia nested within Magnolia, with the loss of monoterpenoid- and shikimate-derived fragrance compounds in one species. Similarly, the addition of fragrance information to nuclear sequence data reduced the phylogenetic resolution (retention index) among subclades of Stanhopea (Williams & Whitten 1999). Clearly, fragrance data are most useful in identifying convergence, drift, and biogeo-graphic patterns when mapped onto independently derived molecular phylogenies. However, rare synapomorphic compounds, when combined with other data sets, may help resolve closely related species clusters (Gerlach & Schill 1989). The application of biosynthetic pathway information to step-matrix coding of fragrance compounds as independent, multi-state characters (Barkman 2001) is the best way to combine such data.
How do pollinators detect and perceive fragrances?
The structure of fragrance plumes is influenced by physical and environmental factors (Murlis et al. 1992). Odors are less ephemeral and less informative than visual or auditory cues, as scent trails can be followed to their sources from a distance, but convey only superficial information about species identity and patch size (Bradbury & Vehrencamp 1998). Pollinators responding to fragrance plumes face three challenges: signal detection over a range of concentrations; signal differentiation from a noisy olfactory background; and information coding and retrieval (Masson & Mustaparta 1990; Hildebrand & Shepherd 1997). Exciting progress has been made in vertebrate olfactory research (Mori et al. 1999), but the absence of bat, lemur, and bird studies limits my discussion to insect olfaction.
Antennae are the primary olfactory organs; their shape, size, and receptor organization reflect functional, phylogenetic, and biophysical tradeoffs (Chapman 1998). "Stereo" olfaction allows insects to detect spatial differences in scent concentration, permitting them to navigate within odor plumes (Mafra-Neto & Carde 1994) and to use intrafloral scent guides (Lindauer & Martin 1963; Lunau 1991). Antennae are studded with microscopic sensilla (knobs, plates, or pits) that house sensory neurons (Chapman 1982). Olfactory sensilla have numerous pores that permit airborne odorants to pass to the lymph, where they contact general odorant-binding proteins (GOBPs) before reaching dendritic receptors (Steinbrecht et al. 1995; Vogt 1995). The binding of an odor molecule-GOBP complex to membrane-bound acceptors releases a signal-transduction cascade that opens ion channels, increases membrane calcium conductance, and generates a receptor potential (Krieger & Breer 1999). This electrochemical impulse conveys olfactory information to processing centers in the antennal lobes (Kaissling 1974). Odor-degrading enzymes inactivate odor ligands and remove them from acceptor sites (Vogt et al. 1990), a process that determines the kinetics of receptor recovery after stimulation (Kaissling 1986) and limits the input rate of olfactory information (Dusenbery 1992).
Do insect antennae preferentially amplify or filter specific fragrance compounds.? Electroantennogram (EAG) recordings represent the sum of all receptor potentials evoked by an odor (White 1991). EAG differences reflect variation in receptor number or timing of response, but can also indicate uncorrected volatility differences among test stimulants (Todd & Baker 1993). Most EAG studies have measured herbivore responses to host-plant odors (e.g., Visser 1979), with a small subset of studies on pollinators and floral fragrances (Topazzini et al. 1990; Zhu et al. 1993). The fragmentary EAG literature suggests that most insects detect most plant volatiles at physiologically relevant concentrations, but the olfactory physiology of many pollinators remains unstudied.
A related technique, gas chromatographic-electroantennographic detection (GC-EAD), measures antennal responses to individual scent components as they elute from a gas chromatograph column (Thiery et al. 1990). When combined with behavioral assays, GC-EAD helps dissect complex fragrances. For example, Priesner (1973) surveyed the EAG responses of males from 50 species of Hymenoptera to fragrances from 18 Ophrys orchid taxa. These plants mimic the sex pheromones of female bees, and are pollinated through the copulatory efforts of aroused males (Kullenberg & Bergstrom 1976; Paulus & Gack 1990). EAG magnitudes were highly correlated with orchid-pollinator specificity, but stimula tion with individual Ophrys volátiles rarely elicited comparable results. With GC-EAD, Schiestl et al. (1999) identified a series of electrophysiologi-cally active alkanes present in comparable ratios in female bee cuticular extracts and in Ophrys fragrance. These compounds, not the more abundant "floral" terpenoids, elicited male attraction and copulation in field trials. The addition of conditioned proboscis extension (CPE) to GC-EAD identified the most salient components of blends to which tethered honeybees had been trained (Le Metayer et al. 1997). Honda et al. (1998) found proboscis extension responses to be more accurate than EAG magnitude in predicting the feeding choices of naive pierid butterflies among artificial, scented flowers. Thus, although EAGs confirm that an insect detects odors at relevant concentrations, they do not always indicate behavioral relevance (Gabel et al. 1992).
How broadly tuned are olfactory receptors to plant volatiles? Unlike whole-antennal studies, receptor-neuron EAGs from individual sensilla have helped define olfactory sensitivity and tuning. The high specificity of "labeled-line" pheromone receptors (Hansson 1995) is not the norm for plant odor receptors (Smith & Getz 1994), since many cell types respond to compounds sharing structural or functional similarities (Anderson et al. 1996; Wibe et al. 1997). Sensillar division of labor was observed in Kaib's (1974) study of two size-classes of sensilla basiconica in Calliphora blowflies, one of which responds to "floral" terpenoids and the other to aliphatic carrion odors. In other insects, diverse receptor neurons may respond to the same odor with different sensitivities, specificities, or electrophysiological properties (Dickens et al. 1993). Vareschi (1971) mapped the responses of honeybee sensilla placodea to hundreds of odors, revealing classes of olfactory neurons covering the entire tuning spectrum. Such variability may indicate multiple acceptor classes in the dendritic membrane, or reduced binding stringency by the acceptor (Kafka 1987).
Odor coding, processing, and perception
In insects, olfactory information is first processed in the antennal lobe, where local interneurons connect distinct glomeruli, and projection neurons link them to the lateral protocerebrum and mushroom bodies, which are the regions of the insect brain associated with olfactory conditioning, sensory integration, and memory retrieval (Hammer 1993; Liu et al. 1999; Menzel, this volume). Vertebrate olfactory bulbs also include glomeruli that process inputs from receptors that share genetic identity and tuning specificity (Buck 1996). These glomeruli communicate with granule cells (the next level of odor processing) and neighboring glomeruli via dendro-dendritic synapses. Yokoi et al. (1995) demonstrated that synaptic inputs from flanking glomeruli modified mitral/tufted cell responses to aliphatic aldehydes in the target glomerulus, resulting in excitatory coding of 6-8 carbon compounds but inhibitory coding of longer- or shorter-chain aldehydes. Tuning specificity through lateral inhibition enhances perceptual contrast of visual signals (Bradbury & Vehrencamp 1998) and provides a mechanism for odor discrimination above the receptor level. Similar mechanisms have been invoked for pher-omone processing in Manduca hawkmoths (Hildebrand 1995) and may be a general property of insect olfaction (Sachse et al. 1999).
If an odor is encoded by the activity of several glomeruli, each of which encodes several odors, then the global patterns of glomerular response to scent compounds should constitute distinct epitope maps (Buck 1996). Galizia et al. (1997) used calcium-sensitive dyes to visualize glomerular processing of scent compounds varying in carbon chain length and oxidation state. The resulting spatial activity patterns were characteristic for each odor tested (Joerges et al. 1997). Principal component analysis identified chain length as the most important variable; no glomeruli responded to functional groups independent of carbon skeleton (Sachse et al. 1999). This is the first depiction, albeit incomplete, of a non-pheromonal odor code in insects. Glomerular activity increases when an odor is paired with nectar but decreases when it goes unrewarded (Faber et al. 1999), suggesting that the same odor is now perceived as a different entity. However, spatial representations alone underestimate the richness of olfactory processing. Laurent et al. (1999) proposed that individual odors are coded not only by the ensemble of responding neurons and glomeruli, but also by the time course and the electrophysiological properties - spike frequencies, magnitudes, and firing patterns - of such a response. Honeybee projection neurons conditioned to a specific odor are "tuned" to show synchronized oscillations (Stopfer et al. 1997); exposure to neural inhibitors desynchronizes the oscillations, such that bees no longer distinguish between similar odors.
Honeybees learn odors faster than colors (Koltermann 1969), distinguish between thousands of subtly different odors (Vareschi 1971), learn faster when odors are "floral" (e.g., terpenoid or aromatic; Kriston 1971), and learn to associate floral scent, size, shape, color, and tactile cues with rewards in decreasing hierarchical rank (Menzel & Müller 1996). Most studies have measured neuroethological responses to single compounds, yet most fragrances are blends. Are blends coded as the sum of their components or as unique perceptual entities.? Smith & Cobey (1994) found that pretraining with one odor diminishes the salience of a subsequent odor stimulus when both are presented together, suggesting that bees perceive individual blend components. When a visual cue is used as the pretraining stimulus, it does not block responses to the odor component of mixed visual-odor stimulus blends; this indicates a distinct processing mechanism for multi-modal signals (Gerber & Smith 1998). Finally, Hartlieb & Hansson (1999) used CPE to test whether female sex phero-mone could be paired with nectar as a conditioned stimulus (CS) for male noctuid moths. Moths learned single pheromone and floral compounds equally well, but full pheromone blends impaired learning. The authors suggested that differential odor processing and learning was a consequence of pheromone blend perception within male-specific labeled-line glomeruli. While understanding pollinator perception of truly complex fragrances remains a distant goal, the tools required to dissect neuroetho-logical responses to simple blends are now available.
How do flower visitors respond to fragrance?
The mechanisms of fragrance production and perception are meaningful to plant evolution only if they evoke discriminatory pollinator behavior. Such behavior can take many forms and act at different spatial scales. For example, von Knoll (1925) sandwiched Lonicera flowers between glass, observed Hyles hawkmoths probing the flattened flowers, and concluded that close orientation and feeding were entirely visual. However, the experimental arena was bathed in fragrance, which elicits visually guided foraging in nocturnal hawkmoths (Brantjes 1978). Additionally, studies of the same pollinator performed in different contexts can reveal unexpected behavioral flexibility. Glossophaga bats use fragrance and echoloca-tion to find flowers, but learn to feed from feeders lacking these properties (Winter & von Helversen, this volume). Studies exploring how diet, physiological state, and experience modify pollinator foraging behavior are sorely needed (Simpson & Raubenheimer 1996). Recent work on scarab beetles reveals a hidden bounty of foraging modes and evolutionary patterns. Scent-driven pollination of thermogenic, nocturnal "trap flowers" by dynastine scarabs is well documented in Neotropical forests (e.g., Gottsberger & Silberbauer-Gottsberger 1991), but guilds of scentless, brightly colored, diurnal flowers are pollinated exclusively by visually foraging ruteline scarabs in South Africa (Picker & Midgley 1996; Steiner 1998) and by glaphyrid beetles in Israel (Dafni et al. 1990). The diversity of foraging strategies among monkey beetles alone (Donaldson et al. 1990) is comparable to that of the order Lepidoptera (Weiss, this volume).
Nested visual signals (display, landing, and nectar guides) are appreciated as components of complex floral phenotypes (Sprengel 1793; Waser & Price 1985), but fragrance blends also serve multiple roles during flower visitation. For example, noctuid moths visiting Silene flowers use fragrance as a distance attractant (Brantjes 1978), a landing cue and nectar guide (Haynes et al. 1991), and a conditioned stimulus (Fan et al. 1997). The recent discovery of "sweet" terpenoids beneath the stench of Sauromatum guttatum aroids and Phallus impudicus fungi (Borg-Karlson 1994; Skubatz et al. 1996) begs a behavioral and phylogenetic reassessment of sapromyo-phily (Kite & Hetterschieid 1997).
Foraging strategies, pollen movement, and behavioral predictions
Bronstein (1995) provided a model for examining evolutionary conflicts of interest between animal foraging and plant reproductive strategies. Her "pollinator landscapes" define a contingency table in which specialist or generalist flower visitors interact with synchronously or asynchronously blooming plants (Fig. 5.2). The foraging strategies of strict mutu-alists (e.g., fig wasps), sexual dupes (mate-seeking males and ovipositing females), territorial defenders (hummingbirds), central place foragers (many bees, bats), and vagile nomads (most moths) can be condensed into three categories: (1) trapliners; (2) density-dependent visual foragers; and (3) sexual foragers.
The archetypal trapliner is a long-lived animal that learns and revisits landmarks in the absence of obvious floral cues (Feinsinger 1983). Chittka et al. (1995) have shown that honeybees form traplines by sequential retrieval of landmark memories without odor cues or mental maps. The heliconiine butterflies and hermit hummingbirds that trapline Psiguria flowers readily feed from artificial, scentless flowers (Swihart & Swihart 1970; Murawski & Gilbert 1986), but the role of pollen odors in heliconiine foraging deserves closer scrutiny. Similarly, there is no evidence for odor-dependent nectar foraging or place learning by female euglossine bees
Was this article helpful?