Ecology Letters (2010) 13: 643–656
Plants have evolved a range of strategies to manipulate the behaviour of their insect partners. One powerful strategy is to produce signals that already have a role in the animals’ own communication systems. To investigate to what extent the evolution of floral scents is correlated with chemical communication in insects, I analyse the occurrence, commonness, and evolutionary patterns of the 71 most common ‘floral’ volatile organic compounds (VOCs) in 96 plant families and 87 insect families. I found an overlap of 87% in VOCs produced by plants and insects. ‘Floral’ monoterpenes showed strong positive correlation in commonness between plants (both gymnosperms and angiosperms) and herbivores, whereas the commonness of ‘floral’ aromatics was positively correlated between angiosperms and both pollinators and herbivores. According to a multivariate regression analysis the commonness of ‘floral’ aromatics was best explained by their commonness in pollinators, whereas monoterpenes were best explained by herbivores. Among pollinator orders, aromatics were significantly more common in Lepidoptera than in Hymenoptera, whereas monoterpenes showed no difference among the two orders. Collectively, these patterns suggest that plants and insects converge in overall patterns of volatile production, both for attraction and defence. Monoterpenes seem to have evolved primarily for defence under selection by herbivores, whereas aromatics evolved signalling functions in angiosperms, primarily for pollinator attraction.
To survive and reproduce, plants interact with various organisms in their environment. For survival, plants must defend themselves against herbivores and pathogens, and for sexual reproduction, they often need to attract animal pollinators. Thus, plants have evolved direct and indirect defence signals, such as herbivore-induced volatile signals (Huber & Bohlmann 2002; Heil 2008; Willmer et al. 2009), and visual and olfactory floral signals for attracting pollinators (Müller 1883; Vogel 1962; Chittka & Menzel 1992; Dobson 1994; Raguso 2008). Among floral signals, floral scent stands out for its chemical complexity and variation, both among and within taxa (Raguso 2001; Dobson 2006). This chemical diversity mediates a variety of functions, such as attracting pollinators and fostering floral constancy (Dobson 1994; Raguso 2008; Wright & Schiestl 2009) and deterring herbivores and ants from consuming the plants’ reproductive structures (Willmer & Stone 1997; Theis et al. 2007; Kessler et al. 2008). Although the chemical composition of floral scent has been well studied (Kaiser 1993; Dudareva & Pichersky 2000; Dobson 2006; Knudsen et al. 2006; Pichersky et al. 2006), little is known about the evolutionary factors generating the composition and variation of this complex signal. Interestingly, some components of floral bouquets are especially common. For example, the aromatic compound benzaldehyde and the monoterpene linalool were found in over 50% of plant families investigated so far (Knudsen et al. 2006). In contrast, some compounds, such as 2,5-dialkylcyclohexan-1,3-dione (chiloglottone) have only been identified in a single plant genus (Schiestl et al. 2003; Franke et al. 2009).
Several typical floral scent compounds are also found in insects (Kullenberg 1956; Vogel 1962; Rodriguez & Levin 1976; Harborne 1992; Knudsen & Tollsten 1993; Dötterl & Vereecken in press). The 19th century naturalist Fritz Müller observed that the smell of Papilio grayi males was so strong and spicy, that ‘I carried the butterfly like a flower in my hand, to smell it from time to time’ (Müller 1878). Much like in floral scent compounds, there are striking differences in the commonness (number of taxa producing the compound) of volatile organic compounds (VOCs) in insects. Although cinnamic alcohol has only been observed in a single family of true bugs, benzaldehyde is produced by a phylogenetically diverse group of seven orders and 13 families of insects, as well as three non-insect arthropod orders. However, individual scent compounds can play different roles among insect taxa. For example, benzaldehyde is an attractant sex pheromone in males from several species of four families of Lepidoptera (Aplin & Birch 1970; Clearwater 1975; Bestmann et al. 1977; Honda 1980; Schulz et al. 1993), an aggregation pheromone in grasshoppers (Torto et al. 1994), bed bugs (Siljander et al. 2008), and ticks (Apps et al. 1988), an alarm pheromone in stingless bees (Wittmann et al. 1990), and a defence secretion in tiger beetles (Kelley & Schilling 1998), walking sticks (Bouchard et al. 1997), chrysomelid beetles (Pasteels et al. 1988), and millipedes (Duffey et al. 1977). Additionally, individual compounds can have multiple functions in the same species, depending on context and amount released. For example, the monoterpene linalool serves as sex and aggregation pheromone in solitary bees of the genus Colletes (Hefetz et al. 1979; Borg-Karlson et al. 2003) but is also released in large quantities by handled male bees (Mant et al. 2005), suggesting an additional defensive role. In general, multiple functions of individual compounds are common in chemical communication (Blum 1996).
Production of the same compounds in only distantly related organisms such as insects and plants is unlikely related to a common ancestor. Insect sensory and behavioural ecology, however, may have had a strong impact on the signal evolution of flowers, because the adaptive value of a signal depends on how well the intended receiver detects it and responds to it (Chittka & Menzel 1992; Raguso 2001). Hermann Müller, who rigorously applied Darwin’s theory of natural selection to the evolution of flowers under selection by pollinators, assumed that ‘different groups of insects, according to their sense of taste or colour, the lengths of their tongues, their way of movement and their dexterity, have produced various odors, colours and forms of flowers’ (Müller 1883). Now, it is well established that insect sensory preferences can select for specific floral signals. For example, mimetic plants deceive their insect pollinators by imitating insect pheromones or oviposition signals (Schiestl 2005; Schiestl & Schlüter 2009). In such mimicry systems, floral signals are thought to evolve in response to the pollinators’ sensory and behavioural ecology in a one-sided or sequential way (Mant et al. 2002). On the other hand, floral signals may also influence insect chemical ecology in cases where insects sequester or collect floral compounds for their own communication or defence (Rodriguez & Levin 1976; Norden & Batra 1985; Schulz 1998; Eisner et al. 2000; Eltz et al. 2007). Besides one-sided evolution, these interactions can also lead to mutualistic or antagonistic coevolution, depending on the specificity and fitness outcome for the plant and insect (Berenbaum et al. 1986; Stowe 1988; Thompson & Cunningham 2002). Both one-sided and coevolutionary mechanisms should produce a positive correlation between the commonness of a signal in insects and plants. Compounds that are phylogenetically widespread in one group (insects) should evolve to be widespread in the other group (plants), too.
Recent progress in the identification of VOCs from flowers and insects and the abundance of these data in reviews and online databases makes it possible to test the hypothesis of congruence in volatile signals in plants and insects. Here, I analyse the overlap, phylogenetic commonness and patterns of the most common ‘floral’ scent compounds in angiosperms, gymnosperms, and insects. Specifically, I ask five questions about the evolution of ‘floral’ scent compounds in plants and insects: (1) How much overlap is there in VOCs produced by flowers and insects? (2) Is there an association between the commonness of VOCs in flowers and insects? (3) Do these associations differ depending on the insect’s relationship to plants? (4) Do patterns of VOCs differ among pollinator-insect orders? (5) How are floral and insect VOCs distributed across the phylogenies? My findings support the hypothesis that the evolution of certain floral VOCs was driven by insect chemical communication. Both herbivores and pollinators influenced floral scent evolution, but to a different extent in different compound classes.
Materials and methods
Floral VOCs from four major chemical groups [aromatics, monoterpenes, sesquiterpenes and fatty acid derivatives (FADs)] that were present in at least 15 plant families were identified from Knudsen et al. (2006) (Table 1). The occurrence of these ‘floral’ compounds in gymnosperms were identified from the Roman Kaiser internal Givaudan ‘nascents’ database. This database contains VOCs from cones, needles, and wood of the eight gymnosperm families (Cupressaceae (14 genera), Taxaceae (one genus), Sciadopityaceae (one genus), Araucariaceae (two genera), Pinaceae (seven genera), Ginkgoaceae (one genus), Cycadaceae (one genus), and Zamiaceae (two genera)). The production of the same compounds in insects was determined from the Pherobase (http://www.pherobase.com; El-Sayed 2008). To restrict the compounds to only those actually produced by the insects, an insect compound was only included in the dataset if it was labelled as pheromone (a semiochemical for intra-specific communication, such as mate finding) or allomone (a semiochemical benefiting the sender, such as a defence compound). If a compound only attracted insects, such as a kairomone or a synomone, it was excluded, because such compounds are not necessarily produced or emitted by the insects themselves. Despite the pherobase provides very good coverage of the literature on insect volatiles, data of two additional papers that were not included in the pherobase were added here, because they improve the number of insect families covered (Pasteels et al. 1988; Bestmann et al. 1993). Different isomers of compounds, such as (E)- and (Z)-β-ocimene and (+)-(−)-linalool were analysed together because they often occur together and the information of isomeric and enantiomeric composition is not always reported in the literature. I recorded the production of all compounds in plants and insects at both the family and order level.
|Compounds||Angiosperm families||Gymnosperm families||All insect families||Pollinator families||Herbivore families||Insects with no association to plants||Ant genera|
|Nerolidol (E&Z )||26||4||5||1||3||2||0|
Insect families found to produce any of the ‘floral’ compounds (n = 87) were divided into four plant-association groups: pollinator families (n = 25), herbivore families (n = 37), insect families with no association to plants (n = 39), and ant genera (n = 46; see Table 2 for details). Because some pollinators are also herbivores, such as the Lepidoptera, the individual groups do not sum up to the total of insect families. In a more detailed analysis for all aromatics included in Knudsen et al. (2006) and the pherobase (440 compounds), both the distribution in angiosperms, insects, or non-insect arthropods and the total number of insect orders and plant families producing the compound were recorded.
|Order||Family||Pollinator (yes/no)||Herbivore (yes/no)|
To test for linear associations between variables, Pearson product–moment correlations were calculated between the phylogenetic commonness (number of families producing a given compound) of the compounds in angiosperm families (n = 88), gymnosperm families (n = 8) and in insect families (n = 87). To test for statistical differences of correlation coefficients among angiosperms and gymnosperms, Steiger’s Z was calculated. This measure is recommended for dependent (correlated) correlation coefficients and was used here since one variable (insect families) was the same within the comparisons (Meng et al. 1992). Calculations were done using a calculator program provided by Dr Calvin P. Garbin (Department of Psychology, University of Nebraska; http://www-class.unl.edu/psycrs/statpage/comp.html). An independent samples t-test was used to compare the commonness of ‘floral’ and ‘non-floral’ aromatic compounds in insect orders (total n = 168). In addition, to investigate the relative importance of different independent variables, on the dependent variable ‘number of angiosperm families producing the compounds’, multivariate regression analysis was done. As independent variables, ‘numbers of insect families producing the compounds’ was used, split into pollinator families, herbivore families, insects with no association to plants, and ant genera. To assess the contribution of individual independent variables to the overall model, the F-probability was used (threshold for entry was set to 0.05, for removal to 0.1) to remove in a stepwise manner the variables with the smallest explanatory contribution to the overall model. To compare the commonness of ‘floral’ aromatics and monoterpenes among pollinator orders with sufficient families present in the dataset (Hymenoptera, Lepidoptera), Mann–Whitney U-tests were calculated for each compound group. If not otherwise specified, all analyses were done using spss 16 (SPSS Inc., Chicago, IL, USA, 2007).
To correct for phylogenetic bias, a pre-analysis was run to detect any phylogenetic structure in the dataset. For each compound, the most parsimonious number of gains and losses (parsimony steps) in a plant or insect phylogeny was compared to the parsimony steps in a series of randomly modified phylogenetic trees. This method assumes that if the occurrence of compounds in families is at least partly due to common ancestry, fewer parsimony steps are expected than when trees are randomly reshuffled. To test this assumption, the number of parsimony steps was calculated for each compound using a phylogeny of the seed plants (gymnosperms and angiosperms) at the family level (Davies et al. 2004) and an insect phylogeny at the order level (Gullan & Cranston 2000). Subsequently, a distribution of parsimony steps was calculated for 100 simulated trees by randomly reshuffling terminal taxa of the original phylogeny. It was then assessed whether the number of parsimony steps for the original phylogeny was within confidence intervals (5% level) calculated for the distribution of parsimony steps from the simulated trees. For compounds that showed significantly less parsimony steps, phylogenetic structure was assumed, and as a consequence, the ‘number of independent origins’ of the compounds was used in all analyses instead of the ‘total number of families’ producing the compounds.
To detect any sampling bias in the dataset, i.e. to assess whether families of plants and insects were evenly distributed among orders, commonness was analysed at both the family and order levels where possible. In addition, correlation between commonness on the family and order level was calculated, with significant correlation indicating a more or less even distribution. To assess the evolutionary distribution of scent compounds, the flower and insect volatiles were mapped onto their respective phylogenies at the order level. A plant phylogeny from the literature was used based on morphological and molecular datasets (Judd et al. 2002). Only plant orders producing at least one of the 71 analysed compounds were included in the phylogeny. An insect phylogeny based on molecular and morphological characters was used (Gullan & Cranston 2000). Compounds that were produced by both flowers and insects (n = 63) were mapped onto the insect phylogeny. Mesquite 2.7 was used for all phylogenetic analyses (Maddison & Maddison 2009). The ancestral state of character evolution was assessed by parsimony.
General overlap between plants and insects
Seventy-one floral volatiles were identified in at least 15 angiosperm families and were included in the analyses (Table 1). Overall, these compounds were produced by 35 orders and 96 families of plants. Of the 71 compounds, 16 were aromatics, 28 monoterpenes, 11 sesquiterpenes, and 16 FADs. Of these floral volatiles, at least one insect family produced 63 (89%) as a pheromone or an allomone. These 63 compounds were distributed among 13 insect orders and 87 families (Table 2). The overlap between floral scent and insect VOCs was largest in the FADs (93%), followed by monoterpenes (92%), sesquiterpenes (90%), and aromatics (75%). Several of these compounds were also produced by non-insect arthropod orders (four orders and 10 families), where the overlap was 43% for FADs, 31% monoterpenes, 13% aromatics and 9% sesquiterpenes.
Among all aromatic compounds known from plants and insects (440 compounds), 62% were found only in plants, 25% only in insects, and only 12% in both insects and plants. Of the 168 compounds produced by insects, 54 were also produced by flowers. These shared compounds were significantly more common in insects than those compounds produced only by insects (mean no. of orders ± SEM, only insects: 1.09 ± 0.04, insects and flowers: mean: 2.18 ± 0.23; t57.93 = −4.71; P < 0.001).
Association between commonness of volatiles in plants and insects
A significant positive correlation (r = 0.462, P < 0.001) was found between commonness of compounds in insects and angiosperms (Table 3). This correlation was robust and remained significant when only the 63 compounds occurring both in insects and plants were included (r = 0.408, P < 0.01) and when the three most common plant compounds (β-ocimene, limonene, and linalool) were excluded (r = 0.358 P < 0.01). To assess sampling bias, the correlation for all compounds was also calculated at the order level, with similar results (r = 0.559 P < 0.001). When the four chemical groups were analysed separately, significant correlations between insects and angiosperms were found for aromatics (r = 0.725 P < 0.01) and monoterpenes (r = 0.594 P < 0.01), but not for sesquiterpenes (r = 0.700 P < 0.05; not significant after Bonferroni correction) and FADs (r = −0.02 P > 0.05; Table 3). Further analysis thus focused on aromatics and monoterpenes.
|Groups||All compounds (n = 71)||Aromatics (n = 16)||Monoterpenes (n = 28)|
|Angiosperm families (n = 88)||Gymnosp. families||Angiosperm families||Gymnosp. families||Angiosperm families||Gymnosp. families|
|Gymnosperm families (n = 8)||0.611***||0.671**||0.672***|
|All insect families (n = 87)||0.468***||0.353**||0.725**||0.291||0.606**||0.458*|
|Pollinator families (n = 25)||0.359**||0.101||0.742**||0.227||0.332||0.072|
|Herbivore families (n = 37)||0.590***||0.503***||0.693**||0.238||0.738***||0.599**|
|Insects with no association to plants (n = 39)||0.351**||0.325**||0.622*||0.278||0.550**||0.553**|
|Ant genera (n = 46)||0.053||−0.089||0.408||−0.128||0.471*||0.478*|
|Groups||Sesquiterpenes (n = 11)||FAD (n = 16)|
|Angiosperm families||Gymnosp. families||Angiosperm families||Gymnosp. families|
|All insect families||0.700*||0.418||−0.019||0.191|
|No association to plant||0.639*||0.513||−0.141||0.166|
When insects were split into functional groups, commonness of ‘floral’ aromatics was significantly positively correlated between angiosperms and herbivores as well as pollinators, but not between ants or insects with no association to plants (Table 3). For the monoterpenes, significant correlations were found between angiosperms and herbivores, and insects with no association to plants, but not between plants and pollinators (Table 3). When analysed at the plant-order level, correlations were similar, suggesting no sampling bias on family and order level (data not shown). Multivariate regression analysis with stepwise removal of variables suggested that the distribution of floral monoterpenes in angiosperms was best explained by VOC production in herbivorous insects, whereas floral aromatics in angiosperms are best explained by VOCs of pollinator groups (Table 4; Fig. 1).
|Compound class||Independent variable||d.f.||Standardized coefficient||P||R2|
In the gymnosperms, there was no significant correlation between aromatics in plant and insect families (Table 3). However, the patterns in monoterpenes of gymnosperms were very similar to angiosperms, with significant correlations between plants and herbivores and insects with no association to plants, but no significant correlation between plants and pollinators or plants and ants. Sesquiterpenes and FADs of gymnosperms showed no significant correlations (Table 3). Significant differences in correlation coefficients between angiosperms and gymnosperms were found in pollinator families (all compounds, aromatics, sesquiterpenes), all insect families (aromatics), herbivore families (aromatics), ant genera (aromatics, sesquiterpenes). For monoterpenes, none of the correlation coefficient was significantly different among the two plant groups (Table 3).
Differences in fragrances produced by pollinator insects
Twenty-five ‘floral’ aromatics and monoterpenes were produced by pollinators of the orders Hymenoptera, Lepidoptera, Coleoptera, Diptera and Thysanoptera (Table 1). Lepidoptera (all families) produced aromatics more frequently than Hymenoptera families (bees only) (U = 14.5, P = 0.048), however, there was no significant difference in the production of monoterpenes (U = 132, P = 0.657). On a purely descriptive level, no aromatics were found in Thysanoptera and Diptera. In bees, common semiochemicals included the monoterpenes neral (occurring in six families), geranial (five families), sulcatone and geraniol (each three families). In the Lepidoptera, the aromatic phenylacetaldehyde and the monoterpenes linalool and β-ocimene (both isomers) occurred in four families, and the aromatics benzaldehyde and 2-phenylethanol and the monoterpenes limonene and sulcatone each occurred in three families.
Phylogenetic bias and distribution of compounds
For five plant compounds (cinnamic alcohol, (E,E)-farnesol, δ-cadinene, α-humulene, and β-cubebene) and one insect compound (heptadecane), significant phylogenetic structure was detected. Thus, for those compounds, ‘number of independent origins’ was used for all analyses instead of ‘numbers of families producing the compounds’. For heptadecane, because of the lack of an insect phylogeny on family level, the change was only implemented at the order level. Family and order data in plants and insects were strongly correlated in all compound classes: aromatics, plants: r = 0.947 P < 0.001, insects: r = 0.963, P < 0.001; monoterpenes, plants: r = 0.950, P < 0.001, insects: r = 0.864, P < 0.001; sesquiterpenes, plants: r = 0.727 P < 0.05, insects: r = 0.987, P < 0.001; FAD, plants: r = 0.749 P < 0.01, insects: r = 0.632, P < 0.01. This result suggested little or no sampling bias in the data, as phylogenetic commonness on family level is linked to phylogenetic commonness on order level.
Mapping the presence and absence of ‘floral’ scent compounds on a seed plant phylogeny suggested an ancestral state for five aromatics, 11 monoterpenes, two sesquiterpenes, and three FADs (Table 5; Fig. 2). Equivocal reconstructions, in which the ancestral state remained unresolved, were found for three aromatics, four monoterpenes, three sesquiterpenes, and five FADs. Given that less than a quarter of all angiosperm families have been investigated for floral scent, for these compounds with equivocal reconstructions the ancestral state is likely. All compounds except the aromatics phenylacetaldehyde and methyl anthranilate, the monoterpenes geranial and ocimenol, the FADs 3-hydroxy-2-butanone, decane, and heptadecane also occurred in gymnosperms, at least in trace amounts.
|Ancestral||Aromatics: benzaldehyde, benzyl acetate, benzyl alcohol, methyl benzoate, methyl salicylate|
Monoterpenes: limonene, β-myrcene, sulcatone, linalool, β-ocimene (both isomers), p-cymene, γ-terpinene, 1,8-cineole, α-pinene, β-pinene, linalool oxide
Sesqiterpenes: α-humulene, β-caryophyllene
FADs: nonanal, decanal, tetradecane
|Equivocal (ancestral state remained unresolved)||Aromatics: 2-phenylethanol, 2-phenylethyl acetate, methyl eugenol |
Monoterpenes: geranial, geraniol, terpinolene, 3-carene
Sesquiterpenes: α-farnesene, germacrene D, caryphyllene oxide
FADs: (Z)-3-hexenyl acetate, hexanol, (Z)-3-hexenol, octanal, pentadecane
Myrcene, sulcatone, geranial, neral, linalool, nerol, α-pinene, β-pinene,
Hexanol, hexanal, nonanal, tridecane, tetradecane, pentadecane, hexadecane, heptadecane
Phylogenetic distribution of insect compounds
Mapping the presence and absence of the compounds onto an insect phylogeny revealed an ancestral state for benzaldehyde and limonene. Both compounds were also present in non-insect arthropods supporting their ancient origin. Equivocal reconstructions were found for one aromatic, eight monoterpenes, one sesquiterpene, and eight FADs (Table 5). Similar to plants, equivocal reconstructions make ancestral state likely, because of the low percentage of insect families as yet investigated for pheromones. For all other compounds, the mapping suggested multiple independent evolutionary origins. None of the VOCs occurred exclusively in pollinator families.
An overlap between floral fragrance and insect volatile chemistry has been known for some time (Kullenberg 1956; Vogel 1962; Rodriguez & Levin 1976; Pellmyr & Thien 1986; Harborne 1992; Knudsen & Tollsten 1993; Valterova et al. 2007; Dötterl & Vereecken in press), but has mostly been interpreted as coincidental and not adaptive (Harborne 1992; Knudsen & Gershenzon 2006; but see Rodriguez & Levin 1976). Here, I present results that challenge the neutral evolution of floral and insect VOCs. I show an extensive chemical overlap in the most common VOCs, and significant positive correlations between the commonness of insect and flower volatiles. These patterns suggest that plants use the insects’ own chemical language to influence their behaviour, or insects use ‘floral’ compounds for chemical communication (Rodriguez & Levin 1976). It is well known that some flowers produce sex pheromones to attract insects (Schiestl 2005), whereas others emit alarm pheromones or defence compounds to deter herbivores or ants that prevent pollinators from visiting the flowers (Rodriguez & Levin 1976; Harborne 1992; Willmer & Stone 1997; Willmer et al. 2009). Yet another plant produces an insect alarm pheromone to attract predators for pollination, that usually utilize alarm pheromones for prey location (Brodmann et al. 2009). These examples suggest adaptive evolution, since the plant gains a fitness benefit from mimicking chemical signals that are important for the insects’ own reproduction and survival.
What causes the association between plant and insect volatiles?
My results infer that the adaptive nature of a match between floral and insect VOCs may be more pervasive than previously thought. Although an association between VOCs in plants and insects could also evolve neutrally, several lines of evidence suggest that adaptive processes caused the observed overlap and correlations. First, patterns of associations differed among insect and plant groups, and these patterns correspond to known insect-plant interactions. For example, gymnosperms are primarily wind pollinated and no correlations between their volatiles and pollinators’ VOCs were found. In contrast, the volatiles from primarily insect-pollinated angiosperms were correlated to pollinator VOCs. This difference was not a result of the smaller sample size in gymnosperms, because both gymnosperm and angiosperm volatiles were correlated to herbivore VOCs. The correlation with herbivores makes sense, as herbivore deterrence is a task common to all plants. Additionally, patterns of associations differed among compound classes, suggesting that different volatile classes have different functions. For aromatics, correlations were strongest between angiosperms and pollinators. Aromatics are often a major part of the bouquet for pollinator attraction, particularly in butterflies and moths (Knudsen & Tollsten 1993; Raguso & Pichersky 1995; Andersson et al. 2002; Huber et al. 2005; Dobson 2006; Raguso et al. 2006). For monoterpenes, both gymnosperms and angiosperms showed strongest associations with herbivores. Monoterpenes were the only compounds with a (marginally non-significant) correlation between flower and ant VOCs. Monoterpenes have pronounced roles in defence against herbivores (Rodriguez & Levin 1976; Huber & Bohlmann 2002; Heil 2008) and may also be involved in deterring ants from flowers (Junker & Bluthgen 2008). However, monoterpenes are also important for pollinator attraction, especially in bees (Dötterl & Vereecken in press). The lack of association between pollinators and floral monoterpenes suggests that either (1) the pollinators’ chemical communication does not change in response to floral monoterpenes, or (2) pollinators do not drive the evolution of monoterpenes in flowers. Individual compounds can have double functions, particularly those occurring both in flowers and vegetative tissue (Knudsen & Gershenzon 2006). If volatile chemistry of flowers and vegetative organs is correlated, the signature of herbivore-driven selection in the leaves should also be apparent in flowers. This herbivore-driven selection may override pollinator-driven selection on floral compounds (Strauss & Whitall 2006). This scenario may hold for monoterpenes, which often occur in both flowers and leaves (Azuma et al. 1999; Knudsen & Gershenzon 2006; Heil 2008). Floral monoterpenes also show pronounced correlations to insects that have no association with plants. This may be explained by the fact that patterns of VOCs in this group of insects is strongly correlated with VOCs of herbivores (r = 0.614 P = 0.001), but less so with VOCs of pollinators (r = 0.405 P = 0.033). In contrast to aromatics and monoterpenes, sesquiterpenes and FADs showed almost no correlations between plant and insect volatiles. A small sample size for sesquiterpenes may have limited the power to detect significant correlations despite high correlation coefficients with some insect groups (Table 3).
Further evidence for an adaptive origin of the overlap in plant and insect volatiles comes from a qualitative comparison of VOCs produced by pollinators and the flowers visited by them. For example, moth- and butterfly-pollinated flowers show some convergence in fragrance that is often characterized by aromatics (e.g. phenylacetaldehyde, benzaldehyde, 2-phenylethanol, benzyl alcohol) and monoterpenes (linalool, (E)-β-ocimene) (Knudsen & Tollsten 1993; Andersson et al. 2002; Dobson 2006). In my analysis, Lepidoptera families produced significantly more aromatics than bees, and were particularly rich in some of the compounds typical for Lepidoptera-pollinated plants, especially in phenylacetaldehyde, benzaldehyde, 2-phenylethanol, linalool, (Z and E)-β-ocimene, limonene, and sulcatone (6-methyl-5-hepten-2-one). Although bee-pollinated flowers are more variable in their scent (Dobson 2006), monoterpenes are often the dominant compounds, congruent to the primary production of monoterpenes in bees. Thus, in addition to correlations in phylogenetic commonness, there was a trend for congruence in fragrance patterns between certain pollination syndromes and the chemical signalling of the pollinators. Unfortunately, little data was available for beetle and fly pollinators, making it as yet impossible to investigate similar convergent fragrance patterns in other pollination syndromes.
When and in what context did the associations evolve?
My analysis offers support that insect and plant volatile production evolved associations under mutual or one-sided selection. It is interesting to ask whether these associations evolved in the angiosperms or were already present in the gymnosperms. Since major gymnosperm lineages date back to the Triassic and well precede the major radiation of angiosperms in the mid Cretaceous (Labandeira et al. 2007), primary functions of floral angiosperm compounds can be inferred from their associations with gymnosperms.
In monoterpenes, the strong correlation between herbivores and both gymnosperms and angiosperms suggests that these compounds evolved primarily for herbivore deterrence. Because this association is found in both the earlier diverging gymnosperms and the newly radiating angiosperms, pollinator attraction is likely a secondary function in monoterpenes. This hypothesis is supported by the suggested presence of many monoterpenes in the ancestor of all seed plants, long before the radiation of flowering plants. Importantly, the monoterpenes were not an evolutionary novelty associated with the origin and evolution of flowers. Many floral scent compounds are not exclusively produced by flowers and are also found in other plant organs (Azuma et al. 1999; Knudsen & Gershenzon 2006), serving multiple communication functions between plants and insects. Defending reproductive structures has been suggested earlier as a primary function of VOCs in flowers and fungi (Pellmyr & Thien 1986; Schiestl et al. 2006), and my analysis supports this view, but only for monoterpenes.
For the aromatics, the observed patterns suggest a different evolutionary scenario. Although most aromatics are produced by gymnosperms (Table 1) and many aromatics have an evolutionary origin in the ancestor of the seed plants (Table 5), correlations were only found between angiosperms and insect aromatics. This finding suggests that despite aromatics were produced by gymnosperms, the signalling function of aromatics evolved later in plant evolution with flowers in the angiosperms for pollinator attraction. In support of this finding, the multivariate regression showed that pollinator VOCs were the most important variable explaining patterns of plant aromatics. Herbivore VOCs, on the other hand, had the strongest explanatory power for patterns of monoterpenes in plants.
Evolutionary patterns of floral signalling to pollinators
Two evolutionary scenarios may account for the patterns of associations between floral and insects aromatics, namely coevolution and one-sided evolution. Coevolution between pollinators and flowers makes the assumption that (1) ‘floral’ compounds have primarily evolved in pollinators, and (2) scent production in plants and pollinators evolved at approximately the same time. However, none of the compounds was produced exclusively by pollinator groups. Additionally, the insects’ use of ‘floral’ aromatics precedes the evolution of flowers, because aromatics are widespread in both early (e.g. Hemiptera) and late (e.g. Lepidoptera) diverging insect orders (Grimaldi & Engel 2005). Some floral aromatics are even used in non-insect arthropods. As an alternative, one-sided evolution predicts that one partner has evolved similarities to a pre-existing pattern in the other partner. One-sided evolution can principally occur in insects and plants, however a scenario where insects evolve onto patterns of plants seems unlikely as it assumes, similar to coevolution, that ‘floral’ VOCs have evolved primarily in pollinator insects, which is not supported by my dataset. In the reverse scenario, floral signals evolved to match pre-existing sensory signals of their pollinators during the relatively recent radiation of angiosperms. This ‘pre-existing bias’ hypothesis thus assumes that (1) floral scent is associated with insect volatiles in both pollinator and non-pollinator groups, since their commonness should not be influenced by an association with flowers, and (2) insect volatiles should be evolutionarily older than floral scent. In support of the pre-existing bias hypothesis, floral aromatics were correlated with aromatics from pollinators and herbivores and marginally correlated with insects with no association to plants. Additionally, ‘floral’ aromatics appear to have originated in arthropods well before the angiosperm radiation. However, to confirm that chemical signalling within pollinator orders preceded the evolution of floral scent compounds, this second assumption should be tested with a more detailed analysis at the family or genus level.
Pre-existing bias may be an important selective force for the evolution of mutualisms (Edwards & Yu 2007). In addition to scent, it could also have influenced the evolution of other floral signals, such as colour. For example, the widespread convergent evolution of red colouration in bird-pollinated flowers may result from better detection of red in birds (Rodriguez-Girones & Santamaria 2004), possibly because red serves as an intraspecific communication signal in birds (Herrera et al. 2008). Invoking pre-existing bias as the driver of floral scent evolution suggests that plants exploited sensory or neuronal preferences of their pollinators outside the context of sexual selection (Endler & Basolo 1998; Edwards et al. 2007) by mimicking volatiles with established receptors and perceptual preferences. In a mutualistic relationship, the operators (in this case, the pollinators) are selected to respond to a mimetic signal, even if it is not identical to the model (Stowe 1988). Therefore, this scenario does not predict an exact one-to-one match between insect and floral volatile chemistry. Instead, it suggests a loose overlap with floral scents that contain both insect-like and plant-specific components. This pattern is prevalent in floral bouquets.
Conclusions and future directions
My analyses suggest that many ‘floral’ scent compounds evolved earlier than flowers. For monoterpenes, plant defence may be the primary role, secondarily adapting for pollinator attraction. In contrast, floral aromatics seem to have evolved with the primary signalling role of pollinator attraction. Pre-existing preferences of pollinators may have played a key role generating similar patterns of commonness in volatiles in angiosperms and pollinator groups. To test these hypotheses further, it will be necessary to combine multiple approaches. Comparative studies should focus on floral scent analyses in plant groups with well resolved phylogenies at the genus or species level. These phylogenies can be incorporated in studies on pollinator visitation. The pollinators’ pheromone chemistry and innate preferences for VOCs should also be investigated. Analysing the associations at a finer scale would clarify the magnitude and patterns of evolutionary responses in chemical signalling in both plants and insects. Additionally, experimental studies should investigate how insects respond to flowers artificially enhanced with pheromone components and how this influences plant fitness (Valterova et al. 2007). A more complete understanding of the evolution of floral signals in response to pollinators should shed light on evolutionary processes mediating the massive radiation of angiosperms.
I would like to sincerely thank Roman Kaiser for allowing me to use his extensive database on scent compounds. Further, I am grateful to Scott Armbruster, Monika Hilker, Eran Pichersky, Rob Raguso, as well as four anonymous referees who provided helpful comments on earlier versions of this manuscript. Wendy Grus helped to improve the language. Further I would like to thank Alexandre Antonelli, Benny Bytebier, and Reto Nyffeler for fruitful discussion as well as Edward Connor and Philipp Schlüter for editorial help. The author is funded by the Swiss National Science Foundation (SNF; project no. 31003A_125340).