Matthew R. E. Symonds, School of Tropical Biology, James Cook University, Townsville, Queensland 4811, Australia. Tel.: +61 7 4781 4133; fax: +61 7 4725 1570; e-mail: firstname.lastname@example.org
Aggregation pheromones are used by fruit flies of the genus Drosophila to assemble on breeding substrates, where they feed, mate and oviposit communally. These pheromones consist of species-specific blends of chemicals. Here, using a phylogenetic framework, we examine how differences among species in these pheromone blends have evolved. Theoretical predictions, genetic evidence, and previous empirical analysis of bark beetle species, suggest that aggregation pheromones do not evolve gradually, but via major, saltational shifts in chemical composition. Using pheromone data for 28 species of Drosophila we show that, unlike with bark beetles, the distribution of chemical components among species is highly congruent with their phylogeny, with closely related species being more similar in their pheromone blends than are distantly related species. This pattern is also strong within the melanogaster species group, but less so within the virilis species group. Our analysis strongly suggests that the aggregation pheromones of Drosophila exhibit a gradual, not saltational, mode of evolution. We propose that these findings reflect the function of the pheromones in the ecology of Drosophila, which does not hinge on species specificity of aggregation pheromones as signals.
Pheromones, chemicals used in intraspecific communication between organisms, are extraordinarily diverse in their chemistry among insect species (Wyatt, 2003). In the case of sex pheromones and aggregation pheromones these signals comprise particular combinations of chemical components that are often species specific (Roelofs, 1995). Although considerable efforts have been made by chemists to identify the molecular composition of these pheromones (see e.g. Stevens, 1998) we still know relatively little about the evolutionary processes that are involved in shaping pheromone blends (Vet, 1999; Wertheim et al., 2005). This is surprising given the ubiquity of pheromones, and the research interest in the evolution of signals and communication more generally (Johnstone, 1997).
At least two hypotheses have been proposed regarding the mode of evolution of pheromones. The first of these is that pheromones evolve in a gradualistic Darwinian fashion, with species accumulating small changes in proportions and structures of chemical components as they diverge (see e.g. Roelofs & Brown, 1982). An alternative hypothesis is that pheromone evolution occurs through sudden major shifts (‘saltational shifts’–Baker, 2002), because there should be strong selection against small modifications in species-specific signals (Paterson, 1985). Theoretical (Butlin & Trickett, 1997) and experimental genetic evidence (Roelofs et al., 2002) have provided support for this latter hypothesis. Furthermore, a recent comparative analysis of aggregation pheromone evolution in bark beetles (Symonds & Elgar, 2004a) showed that closely related species of bark beetles tended to be just as, or possibly more, different in the composition of their pheromones than more distantly related species in the same genus. Therefore, the authors concluded that pheromone evolution was indeed characterized by large saltational shifts. Of course, their analysis represents just one particular small group of organisms (34 species in two genera), and the generality of their results is uncertain. Here we perform the same tests as those used in Symonds & Elgar (2004a) on the aggregation pheromones produced by a different group of insects –Drosophila fruit flies.
Fruit flies use aggregation pheromones to assemble on breeding substrates, where they feed, mate and oviposit communally (Spieth, 1974; Wertheim, 2001). Males mostly produce the pheromones, but males of many species transfer aggregation pheromone to females during mating, and mated females can also use the pheromone to attract conspecifics. The function of the aggregation pheromone in the ecology of fruit flies has only been investigated for some species (Wertheim et al., 2002a,b; Rohlfs & Hoffmeister, 2003). In these species, aggregation of adults on a resource improved the survival and growth of their offspring developing on these resources (Wertheim et al., 2002a; Rohlfs & Hoffmeister, 2003), indicating a function of the pheromone in facilitating larval resource exploitation (Wertheim et al., 2002a; Wertheim, 2005). Such facilitation is not species specific, as it was shown that facilitation could be achieved by heterospecifics as well (Hodge et al., 1999). Field experiments showed that the aggregation pheromones of several species attracted both conspecifics and heterospecifics (Jaenike et al., 1992; Wertheim, 2001). Males can also attract potential mates by using the pheromone, but rely on separate nonvolatile sex pheromones for species recognition (Antony & Jallon, 1982). The aggregation pheromones of Drosophila species provide an interesting comparison with bark beetle species. Not only is the aggregation pheromone identified for a sizeable number of species (see Table 1), we also have some insight in the role of these pheromones in the ecology of fruit flies (Wertheim, 2005). Even though closely related species are frequently sympatric and co-exist on the same resources (see Table 1), the aggregation pheromones appear to function less in ensuring species specificity in fruit flies than in bark beetles, and therefore we may expect a more gradual mode of evolution. Here, by mapping pheromone components of different Drosophila species onto published phylogenies, we test whether the pheromones have evolved in a gradual or more saltational fashion. For gradual evolution, we should expect closely related species to have greater similarity in their pheromones with more distantly related species showing increasing pheromone distinctness. If aggregation pheromones in Drosophila have evolved in the same way as in bark beetles (Symonds & Elgar, 2004a) then we would expect sibling species to be just as distinct as more distantly related species and no relationship to exist between pheromonal difference and phylogenetic distance.
Table 1. Aggregation pheromone components for the 28 species of Drosophila used in the analyses. The description of main breeding sites and geographical distribution is provided to indicate which species are likely to co-exist.
We collated from the literature information on the chemical constituents of the aggregation pheromone blends of 28 species of Drosophila. These species represent three subgenera (Dorsilopha, Drosophila and Sophophora) and eight species groups (busckii, immigrans, quinaria, repleta, testacea, virilis, melanogaster and obscura). We entered all the chemical components (23 in total) into a data matrix as characters and recorded their presence or absence for each species. Note that we only recorded the presence of a component if that chemical was both produced by the species and was a functionally active component in the pheromone blend (i.e. it actually promoted aggregation). The pheromone data and references are listed in Table 1. One advantage, incidentally, of this data set is that all of the pheromone identifications and analyses were conducted by the same research group, with pheromone components being quantified using very similar methods and then applied appropriately in behavioural response assays. As with any literature-based comparative analysis, though, we are reliant on the quality of the data available. The chemical identification of aggregation pheromones may be incomplete for some species, especially regarding minor components of the pheromone blend (e.g. Bartelt et al., 1985a). This could result in slight underestimation of the differences between pheromones. However, we see no reason why this would cause a systematic bias in our results. Moreover, the strong and consistent behavioural response of each species to the synthetic blend of major components of their aggregation pheromone indicates that these blends do reflect the essential components of that pheromone.
We related the chemical information on pheromone components to phylogenetic information for the Drosophila species. Because the results of comparative analyses can depend on the phylogeny used, we performed all analyses twice using two different phylogenetic topologies collated together from recent sources in the literature (see Table 2). Neither phylogeny may be fully correct, but by examining if and how the results vary according to phylogeny we can determine the robustness of our conclusions. Figure 1 presents the first of these phylogenetic topologies (phylogeny A) complete with a diagram illustrating the chemical components expressed by each species in the analysis.
Table 2. Details of the two phylogenies used in the analyses. Figure numbers refer to figures in the cited references.
*For Drosophila borealis the phylogenetic position of the eastern population was used.
†The melanogaster subgroup includes the species D. melanogaster, D. simulans, D. mauritiana and D. yakuba.
We conducted two series of tests of how the chemical components related to phylogeny. These tests, described below, are the same as those used in the earlier analysis of bark beetle pheromones by Symonds & Elgar (2004a), to enable direct comparison of results.
Degree of congruence with phylogeny
We assessed the extent to which the chemical components were grouped on the phylogeny using three measures. First we counted the number of steps (s, the number of changes in the presence/absence of chemical components) along the phylogeny. Second, we assessed the degree of homoplasy – the Homoplasy Index (HI) used by Cognato et al. (1997), and equal to 1 – the Consistency Index (Kluge & Farris, 1969), or 1 − m/s, where m is the minimum possible number of steps. Third, we calculated the degree of independence of the data from the phylogeny calculated as 1 – the Retention Index (Björklund, 1997), or 1 − (g − s)/(g − m), where g is the maximum possible number of steps for a given phylogeny. When calculating these estimates we used a pruned phylogeny (as shown in Fig. 1) consisting only of the species for which we had data.
In order to test whether these three estimates of phylogenetic congruence were significantly different from that produced by a random distribution of characters we utilized the Shuffle option of MacClade 4 (Maddison & Maddison, 2001) to randomly shuffle the character data among the species, whilst holding the phylogeny constant. After shuffling, the three measures were recalculated, this process being repeated 500 times to generate a null distribution for each measure. The values obtained from the actual data set were then compared to this null distribution (see Björklund, 1997).
Relationship between pheromone differences and phylogenetic distance
We compared all the possible pairs of species for which we had data and counted the number of phenotypic (pheromonal) differences between them (i.e. a binary squared Euclidean distance measure – the number of chemicals that were absent in one species but present in the other, and vice versa). We did not count shared similarities because they may have resulted from shared ancestry, and thus might not represent independent evolution. The pheromonal difference matrix was then related to a phylogenetic distance matrix generated by counting the number of nodes in the phylogeny that separates each species pair (sibling species have a phylogenetic distance of one, and so on). For this calculation we used the full phylogenies (inclusive of species for which we had no data on pheromones) available from the original sources (see Table 2) in order to get a better estimate of the true degree of phylogenetic distance between species. We then calculated a mean amount of phenotypic difference for each level of phylogenetic distance and tested whether the two variables were correlated. Because the elements of the two dissimilarity matrices are not independent bivariate observations, as required by correlation theory, we calculated the significance levels for these correlations using Mantel tests (see Sokal & Rohlf, 1995). The rows and columns of the distance matrix were randomly perturbed and the correlation coefficient recalculated 999 times, comparing the observed coefficient against this reference distribution. All Mantel tests were performed using the computer package GenAlEx (Peakall & Smouse, 2001).
Congruence with phylogeny
Using all three measures [number of steps, HI, degree of phylogenetic independence (PI)] our analyses show that aggregation pheromone components are highly congruent with phylogeny, and are significantly more congruent in distribution than expected by chance. These results stand when either phylogeny is used as the basis for analysis (see Table 3). In fact, closer inspection of the individual chemicals and their distribution on the phylogenies reveals no spectacularly incongruent characters (see Fig. 1). The chemical that displays the least congruence with phylogeny is (Z)-11-octadecenyl acetate (sometimes known as cis-vaccenyl acetate or cVA) – one of the more common components of Drosophila aggregation pheromones. This chemical is expressed as part of the aggregation pheromone of all of the melanogaster group species for which we have data except D. bipectinata and D. malerkotliana, as well as the closely related species groups of quinaria and repleta, and the species (one each) in our analyses representing the obscura and immigrans groups. There are other behaviourally active chemical components that are only expressed by members of particular species groups. For example, the heneicosenes are only produced by species of the virilis group, indicating a strong phylogenetic component to their expression.
Table 3. Measures of congruence with phylogeny of the pheromone-component data for Drosophila species, comparing the real data with the expected mean result from randomly perturbed data on the same phylogeny (results from using two different phylogenies are shown).
Number of steps
The number of changes in pheromone components along the phylogeny (Number of steps), homoplasy index (HI) and degree of phylogenetic independence (PI) were calculated from phylogenies consisting only of the 28 species in our analysis. All differences were highly statistically significant (P < 0.001) as assessed using two-tailed Z-tests.
67.08 ± 2.01
0.66 ± 0.01
0.88 ± 0.04
66.10 ± 1.95
0.65 ± 0.01
0.86 ± 0.05
Relationship between pheromonal differences and phylogenetic distance
Generally speaking, there appears to be a positive relationship between the amount of pheromonal difference and the phylogenetic distance between species. In other words, as species become more phylogenetically diverged they also tend to exhibit greater differences in the make-up of their aggregation pheromones. Figure 2 illustrates the relationship (for both the analyses using different phylogenies), showing how the mean number of pheromone component differences relates to the phylogenetic distance. Statistical analysis reveals that across all the species in our analysis there is a correlation between pheromonal difference and phylogenetic distance, with the degree of significance varying according to the phylogeny used (Mantel test r = 0.126; P = 0.078 using phylogeny A, r = 0.166; P < 0.01 using phylogeny B). In fact, it seems likely that these statistics mask an even stronger relationship among closely related species. The graphs show something of a saturation effect, with a general tendency towards increasing pheromone difference for species up to a phylogenetic distance of 10–12, with no obvious increase thereafter. These results would tend to indicate that there is gradual evolution of aggregation pheromones in Drosophila.
The tests of congruence of the data with phylogeny indicate that pheromone components may be restricted to certain groups or lineages of Drosophila. It is possible then that the observed positive correlation between pheromonal differences and phylogenetic distance may in part be due to the larger pheromonal differences between groups than within groups. When we break down the results to examine the pattern within two of the groups (melanogaster and virilis) we find differing patterns of association between the two variables (see Fig. 3). Within the melanogaster group of species there remains a strong correlation between pheromone difference and phylogenetic distance (Mantel test r = 0.788; P < 0.02 using phylogeny A, r = 0.758; P < 0.02 using phylogeny B). However, within the virilis group no such pattern exists (r = 0.087; n.s. using phylogeny A, r = 0.057; n.s. using phylogeny B).
Our analyses show that the mode of aggregation pheromone evolution in Drosophila is considerably different from that in bark beetles. In the beetle genera Dendroctonus and Ips, which were the subjects of a previous study (Symonds & Elgar, 2004a), pheromone chemical components were found to be incongruent with phylogeny, being distributed randomly across the phylogeny. In Drosophila, aggregation pheromones appear to be strongly congruent with phylogeny. To some extent, different species groups use different ‘palettes’ of chemicals. A cursory look at Table 1 shows, for example, that the virilis group species share many similar properties in their pheromones (e.g. note the emission of ethyl tiglate by most species) and can be distinguished pheromonally from other species in our study (e.g. the virilis group do not use acetates as part of their aggregation pheromones, unlike most other species). The congruence measurements that were calculated for bark beetles (Table 2 in Symonds & Elgar, 2004a) and fruit flies (Table 3, this study) are strikingly different. Furthermore, across all Drosophila species in our analysis we found an increase in pheromonal difference as species become more phylogenetically distinct, whereas there was no relationship between pheromonal differences and phylogenetic distance for the bark beetle species (Symonds & Elgar, 2004a). These results suggest a gradual mode of evolution in Drosophila aggregation pheromones.
Symonds & Elgar (2004a) compared their results for bark beetles with those of computer simulation models of pheromone evolution. They allowed binary characters to ‘evolve’ randomly along a phylogeny. They created three possible modes of evolution – a minor (i.e. gradual) change model (where the probability of change in state of a character following a bifurcation was low), a major change model (where the probability of change was high) and a model that represented a degree of change that was intermediate to the other two. The predicted outcomes of the three models are shown in Fig. 4. The numerical values are irrelevant here (being based on hypothetical evolutionary models of change in eight characters), but the shape of the resultant models is important. The first thing to note is that each model predicts that as phylogenetic distance between species becomes greater, the amount of phenotypic (pheromonal) difference between them approaches the same constant. In other words, each model tends towards an asymptote that is equal to the overall mean amount of difference between the species. The reason for this saturation effect is that once this level of difference between species has been reached any random changes in character states are just as likely to make the species more phenotypically similar as more different, hence the mean phenotypic difference between species will remain the same. In these models, that level is equal to n/2 characters and cannot exceed that value, but in the real world, where there may be constraints on the evolution of characters (some lineages may not be able to evolve particular pheromone components), the level is likely to be less than n/2. It is encouraging that this pattern is displayed in the Drosophila data also (see Fig. 2), with a general increase in pheromonal difference roughly until the overall mean difference (4.3) is reached, after which the relationship levels off. The results for Drosophila would probably conform best to a pattern somewhere in between the ‘minor change’, and the ‘intermediate change’ model. It would appear that at least some of the species have undergone gradual changes in pheromone composition, and saltational change is certainly not the rule in the Drosophila system. This is in contrast to the bark beetles, where the major change model seemed to be a better approximation of the data (Symonds & Elgar, 2004a).
We propose a very straightforward explanation for the observation that aggregation pheromone evolution in Drosophila species is characterized by a more gradual mode of evolution than is the case with bark beetles. In bark beetles, although there is some cross attraction between species (see e.g. Cane et al., 1990), aggregation pheromones act to a considerable extent as sex pheromones in order to attract potential mates. Therefore there is greater species specificity in the signal that is being sent out, in order to prevent mismating or hybridization. In contrast, in Drosophila, aggregation pheromones do not appear to contribute to species specificity in sexual interactions. Aggregations of adults often comprise several species (Spieth, 1974; Wertheim, 2001), and both sexes rely on other cues during courtship for species recognition, including separate nonvolatile sex pheromones (Antony & Jallon, 1982). In fact, the aggregation pheromones of almost all studied Drosophila species were found to attract multiple species (references in Table 1). Moreover, mated females also emit aggregation pheromone, and the main function for them appears unrelated to sexual interactions (Wertheim et al., 2002a,b; Wertheim, 2005). Their use of the pheromone results in aggregated oviposition, which enhances the survival and growth of their offspring (Wertheim et al., 2002a,b). Similar benefits of sharing an oviposition substrate were also found among species (Hodge et al., 1999), which can explain the lack of species specificity in the behavioural responses of fruit flies to aggregation pheromones. Because other visual, chemical and acoustic cues are used in mate assessment in fruit flies (Spieth, 1974), the aggregation pheromone is not under strong selection for species specificity. We would argue that the more gradual mode of evolution of aggregation pheromone reflects the function of the pheromone in fruit flies.
Two aspects of our results suggest some caution is needed regarding this interpretation of the data. First, in contrast to the similarity that would be expected under a gradual mode of evolution, there are obvious cases where closely related species are very different. For example, D. virilis and D. lummei, which may be sibling species (Spicer, 1992; as used in our Phylogeny B), show five differences in their pheromone structure, where the mean difference between all species as a whole is 4.3.
Second, we found no relationship between pheromone differences and phylogenetic distance within the virilis group of species, a finding similar to the results within genera of bark beetles (Symonds & Elgar, 2004a). This might suggest the occurrence of sudden, not gradual, shifts in chemical composition at speciation events. However, these results contrast with those from the melanogaster group, which strongly suggest that gradual evolution has taken place within that group. Within the melanogaster species group, the indications are that there is very little selection acting to produce changes in the aggregation pheromone blend at speciation events.
There is no obvious evolutionary explanation for the difference in results between these two species groups. Differences in the rate of speciation in the two groups are unlikely to be a cause as both groups originated at approximately the same time (about 11 mya for the virilis group: Spicer & Bell, 2002; and about 12.8 mya for the melanogaster group apart from D. ananassae which has a much earlier origin: Tamura et al., 2004). Likewise, the ecology of aggregation pheromone use for members of both groups appears rather similar. In both, males and females are attracted in approximately equal ratios to breeding substrates, and cross-attraction with other species is frequent. At this stage in this analysis, there are not enough species from different groups whose aggregation pheromones are known, to satisfactorily judge whether gradual or saltational evolution best describes the pattern at this taxonomic scale. Clearly, investigation of pheromone composition of more Drosophila species in the same and other groups would help to resolve this issue.
If there were evidence for nongradual evolution in aggregation pheromones, as might be the case within the virilis species group, it could imply that the aggregation pheromone is under selection to provide species specific signals. An obvious factor that may be of importance for the specificity of blends is ecological and geographic overlap of closely related species. If closely related species are sympatric and rely on the aggregation pheromone for maintaining mating barriers, we should expect a greater divergence in species-specific signals and characters as a result of reinforcement for species isolation, and to prevent hybridization (Dobzhansky, 1940; Noor, 1999). Several virilis group species co-exist on the same food sources (slime fluxes and decaying bark, see Table 1) and hybrids have been found in nature (Throckmorton, 1982), which supports the need for ecological isolation mechanisms. Classic studies of Drosophila species by Coyne & Orr (1989,1997) have indeed shown that sympatric species tend to show greater adaptations for reproductive isolation than allopatric species, which might suggest that pheromones, too, would show similar patterns. However, strong behavioural responses to the aggregation pheromones of other species in the virilis species group were frequent (Bartelt et al., 1986,1988). This means that even though the composition of the blend may be species specific, the behavioural responses are not. Additionally, species of the melanogaster group also occur sympatrically. Furthermore, bark beetle pheromones, surprisingly, do not appear to show any effect of species overlap, either in range, host-plant use or both (Symonds & Elgar, 2004b). This finding would tend to argue against any effects of sympatry on aggregation pheromone differentiation.
We conclude that, as a whole, there is no ubiquitous mode of aggregation pheromone evolution across animal species. Bark beetles appear to exhibit a mode characterized by saltational shifts in aggregation pheromone composition at speciation events, whereas Drosophila pheromones appear to exhibit more gradual evolution. We suggest that our results support the prediction (Paterson, 1985; Butlin & Trickett, 1997) that the species specificity of pheromones determines their mode of evolution. We would therefore predict that, in contrast to their aggregation pheromones, the sex pheromones of Drosophila should exhibit a saltational mode of evolution. Because the function of aggregation pheromones in the ecology of many insect species (bark beetles aside) does not hinge on species specificity of the pheromones as signals, and cross-attraction between species is indeed frequent (Wertheim et al., 2005), we predict that a gradual mode of evolution may be the more commonly found pattern for these chemical signals.
We thank Louise Vet, Marcel Dicke, Therésa Jones and two reviewers for their comments on this manuscript.