The role of competition in adaptive radiation: a field study on sequentially ovipositing host-specific seed predators

Authors


Dr Laurence Després. Fax: 33 4 76 51 42 79. E-mail: laurence.despres@ujf-grenoble.fr.

Summary

  • 1We propose an alternative model to the host-shifting model of sympatric speciation in plant–insect systems. The role of competition in driving ecological adaptive radiation was evaluated in a seed predator exploiting a single host-plant species. Sympatric speciation may occur through disruptive selection on oviposition timing if this shift decreases competition among larvae feeding on seeds.
  • 2The globeflower fly Chiastocheta presents a unique case of adaptive radiation, with at least six sister species co-developing in fruits of Trollius europaeus. These species all feed on seeds, and differ in their oviposition timing, one species ovipositing in 1-day-old flowers (early species), while all the other species sequentially oviposit throughout the flower life span (late species). We evaluated the impact of conspecific and heterospecific larvae on larval installation success, and on larval fresh mass and area, for early and late species, in natural conditions.
  • 3None of the three larval traits measured was correlated with fruit size, and no fruit lost all seeds to predation, suggesting that seed availability was not a limiting factor for larval development.
  • 4Our results show strong intraspecific competition among early larvae for larval installation, and among late larvae for larval mass. By contrast, larval competition between species was weak. These results are consistent with the hypothesis that shifts in oviposition promoted rapid radiation in globeflower flies by lowering competition among larvae.

Introduction

Beyond the long-term controversy on the geographical-based classification of speciation modes (sympatric vs. allopatric), the time is coming to evaluate the relative importance of the different evolutionary mechanisms promoting reproductive isolation and speciation (Via 2001). These include natural selection, sexual selection and purely genetic mechanisms such as polyploidization, chromosomic fusion, genetic drift and invasion by a parasite responsible for genetic incompatibilities. The ecological theory of adaptive radiation stresses the role of divergent natural selection in rapid differentiation of ecological phenotypes from a single ancestor, either through the invasion of different environments (i.e. niche expansion) or through competition leading to fine resource partitioning (Schluter 2000).

In his recent revisitation of the ecological theory of adaptive radiation, Schluter (2000) concluded that most adaptive radiations are likely to result from expansion to new resources and environments rather than niche subdivision. Accordingly, in most reported cases of adaptive radiation in host–parasite and plant–insect systems, speciation was shown to originate by host shifts and adaptation to new host species (Berlocher & Feder 2002). Resulting host races by definition only occasionally compete on a common resource, and therefore competition has a marginal role in speciation through the counter-selection of hybrids, poorly suited to both hosts.

However, the coexistence of sister species within a single host plant's fruit observed in several specialized plant–seed predator interactions, including the globeflower/globeflower fly (Pellmyr 1989; Després & Jaeger 1999), the yucca/yucca moth (Pellmyr & Leebens-Mack 2000; Marr, Brock & Pellmyr 2001), the fig/fig wasp (Weiblen & Bush 2002) and the larch/larch fly interactions (Roques et al. 1995), suggests a central role of competition on a limited resource in driving ecological divergence in these systems. In these highly specialized interactions, the plant is parasitized by two or more congeneric host-specific insect species ovipositing sequentially throughout the flower/fruit lifetime, and whose larvae develop by eating a fraction of the host-plant seeds. In the yucca/yucca moth and globeflower/globeflower fly associations, the insect adults are the sole pollinators of the host plant. Resource sharing among species involves a fine timing in oviposition, particularly in the globeflower–globeflower fly interaction, where at least six species oviposit sequentially throughout the 6–10 days of a flower's lifetime (Després & Jaeger 1999). In the globeflower flies, several lines of evidence, including phylogeny and biogeography, indicate that divergence between early and late ovipositing species evolved in sympatry (Després et al. 2002a; Després, Loriot & Gaudeul 2002b) through disruptive selection on oviposition timing. If competition is indeed a key factor in driving phenotypic divergence in globeflower flies, we expect to find stronger intra than interspecific competition among larvae developing in a fruit. This prediction is based on an adaptive dynamic model of globeflower fly evolution, showing that the only condition required for evolutionary branching to occur on oviposition time was that larval competition strongly decreased with increasing difference in oviposition time (Ferdy, Després & Godelle 2002). If this is the case, early and late larvae may have little impact on each other. Alternatively, differences in timing of oviposition may give an advantage to the early ovipositing species, which benefits from about 2 days without interspecific competition at the beginning of its development (Pompanon, Pettex & Després unpublished). One would then expect asymmetric competition in favour of early species due to the presumably smaller size of the late species. Larval interactions within a fruit may include direct interference, either through aggressive behaviour, including larval exclusion, or through contact avoidance by changing mining paths when encountering (Pellmyr 1989). Competition may also occur indirectly through resource-limited competition if seeds are limiting, without involving direct larval interactions. In either case, fruit size may affect the outcome of larval competitive interactions (Marr et al. 2001).

In this paper, our objective was to determine the importance of competition in globeflower fly radiation by addressing the three following questions: what is the intensity of intra- vs. interguild competition? Is competition between early and late species asymmetric? What is the effect of resource size (number of seeds available) on the intensity of larval competition?

Materials and methods

The European globeflower Trollius europaeus L. (Ranunculaceae) is a hermaphroditic, homogamous, arctic–alpine perennial species growing in moist meadows. A plant usually produces only a single flower in a given year, more rarely two or three (Hemborg & Després 1999). Each flower contains about 30 multiovulate carpels dehiscing 3–4 weeks after the end of flowering, and about 160 stamens sequentially dehiscing throughout flower longevity (typically 5–8 days). The flowering is synchronized within populations and lasts typically 2–3 weeks. In the Alps, the plant is pollinated passively by at least six species of Chiastocheta flies (Anthomyiidae): C. rotundiventris Hennig, C. dentifera Hennig, C. inermella Zetterstedt, C. macropyga Hennig, C. setifera Hennig and C. trollii Zetterstedt. A molecular analysis showed that C. inermella is paraphyletic, so that up to eight different evolutionary lineages coexist in the Alps (Després et al. 2002a). Chiastocheta flies are the only pollinators of T. europaeus and Chiastocheta larvae feed only on T. europaeus seeds. Both male and female flies visit the globe-shaped flower where they eat nectar and pollen, and mate. Both sexes contribute to pollination (Després 2003). Females deposit one to several eggs on or between the carpels at various flower stages, depending on the species (Pellmyr 1989; Després & Jaeger 1999). Embryonic development lasts 4–7 days. Each larva eats several seeds and falls into the soil to pupate and overwinter. Seed consumption per larva in the absence of competition is similar for all species (Pompanon et al. unpublished), and was shown to decrease strongly with increasing numbers of larvae developing in the fruit (Jaeger, Pompanon & Després 2001). Seed maturation and larval development last 3–4 weeks. The early ovipositing species C. rotundiventris visits 1-day-old flowers, and typically lays a single egg per flower, at the base of an external carpel, whereas all the latter ovipositing species tend to aggregate their eggs (Després & Jaeger 1999). Shifts in oviposition time among species range from 2 days up to 1 week, and larval development time in the absence of competition does not differ strongly among species, except that the last ovipositing species appears to have a slightly shorter development time (Pompanon et al. unpublished). Therefore, there is a large overlap in larval development of the different species. Egg morphology, colour and position on the fruit differ among species (Pellmyr 1992). By contrast, there are no morphological criteria to determine larval species status. Larvae can move freely from one carpel to another, and were shown to exhibit specific mining patterns (Pellmyr 1989): the larva of the first ovipositing species (C. rotundiventris) stays in the central pith, where it can be found at the end of its development. After larval exit, black tracks in the pith indicate successful C. rotundiventris larval development. A molecular phylogenetic analysis showed that the European Chiastocheta species fall into two well-differentiated evolutionary lineages (Després & Jaeger 1999; Després et al. 2002a), diverging about 2 million years ago: one contains the early ovipositing species C. rotundiventris, the other contains all the other species. Thereafter, we will refer to Chiastocheta species as ‘early’ or ‘late’, this latter category including four late ovipositing species diverging since less than 1 million years (C. dentifera, C. inermella, C. trollii and C. setifera) that cannot be distinguished at the larval stage. C. macropyga was absent in the population studied.

A total of 169 young fruits were selected on their egg content and bagged in one large T. europaeus population in the French Alps (Col du Lautaret, 2000 m asl) in July 2000, at the end of flowering/ovipositing period. The fine nylon mesh bags prevent subsequent ovipositions and full-grown larvae and mature seeds from dropping to the ground. Fruits were categorized depending on their egg load as follows: category 1 = fruits with a single early egg (n = 32), category 2 = fruits with two early eggs (n = 21), category 3 = fruits with late eggs only (n = 18), category 4 = fruits with a single early egg plus late eggs (n = 46) and category 5 = fruits with two early eggs plus late eggs (n = 52). Because the population studied was heavily parasitized (about 10 eggs per fruit on average), all flowers had at least one early egg, and sometimes two, a situation observed rarely in typically less parasitized populations (Després & Jaeger 1999), so that we had to remove early eggs before hatching to constitute category 3. Our experimental design and the constitution of the five categories reflect the differences in oviposition patterns among early and late species in natural conditions: in order to evaluate the intensity of intraspecific competition, we compared many singly developing vs. two codeveloping early larvae, while we compared fruits containing one to 10 codeveloping late larvae. Fruits were checked daily during larval development, and collected at first larval exit and/or first carpel dehiscence. Collected fruits were dissected and each larva immediately weighted to the nearest 0·1 mg (fresh mass). Larval area in ventral position was measured to the nearest 0·01 mm2 by image analysis (software Optimas 5TM) using an optical binocular connected to a PC computer via a video camera. Larval installation success (LIS) was estimated as the ratio between the number of larvae found in the fruit and the number of eggs counted at bagging. Depending on the larval installation success, some fruits categorized first in categories 2 or 5 were reclassified in categories 1, 3 or 4 for larval measurements (mass and area). The number of fruits per category after exclusion of the 32 fruits that did not harbour a larva, and reclassification of fruits according to their larval content was 33, 3, 25, 51 and 16, respectively, for the five categories. Some fruits with ambiguous larval content were excluded from the analysis.

When intra- or interspecific competition was evidenced for a larval trait, relative competition intensity was estimated as the average percentage decrease in larval trait value due to competition [i.e. (larval trait without competition–larval trait with competition)/larval trait without competition].

Fruit seed production before predation was estimated by counting seed production for five undamaged carpels, and by multiplying the average number of seeds per carpel by total carpel number (Jaeger & Després 1998). Seeds remaining intact after first larval exit (undamaged seeds) were counted, and the number of seeds destroyed by developing larvae was estimated as seed number before predation minus undamaged seed number.

Data were first checked for their normality and homogeneity of variances (Kolmogorov–Smirnov and Levene tests). These criteria were satisfied for larval mass and area, and a one-way analysis of variance (anova) was performed with category considered as a fixed effect; Bonferroni's correction for multiple means comparisons used. For non-normally distributed data (LIS, egg, larval and seed numbers), non-parametric rank tests (Kruskal–Wallis test on medians, Mann–Whitney U-test for pairwise median comparisons, Spearman's correlations) were performed. Analyses were carried out using SAS statistical programs (version 8·0).

Results

Out of 169 fruits, a total of 1090 eggs were identified (224 and 866 for early and late species, respectively) and 458 living larvae were measured and weighted (119 and 339 for early and late, respectively). Twenty-six additional larvae were found dead inside or outside the fruits, representing only 5% of the total larval number. These dead larvae were taken into account for the larval installation success calculation; they were measured, but were not weighed because of water loss. Egg and larval numbers per fruit ranged from one to 28 and from zero to 13, respectively. None of the collected fruits lost all its seeds to predation. The mean percentage of seeds consumed by larvae was 55% (range 0–93%). First larval exit usually corresponded to the beginning of carpel dehiscence, with only 26% of fruits being nondehiscent at first larval exit.

Whatever the category, larval installation success was significantly higher for early than for late species (53·8% vs. 40·4%, respectively, P= 0·001, Table 1). There was a significant effect of fruit category on LIS for early species (Kruskal–Wallis test, H= 21·7, d.f. = 3, P > 0·001), whereas category had no effect on LIS for late species (H = 4·2, d.f. = 2, P= 0·12). LIS for early species was significantly higher without conspecific competition (category 1: 62·5% vs. category 2: 38%, P= 0·032, see Fig. 1), whereas the presence of late species had no effect on LIS of early species (62·5% vs. 78%, P= 0·147). Therefore, the significant difference of LIS for early species in category 4 vs. 5 (78% vs. 46%, respectively, P < 0·001) was due only to the presence of conspecifics. LIS for early larva decreased by about 40% in the presence of a conspecific. LIS for late species was not influenced by the presence of early species (category 3 vs. categories 4 and 5, Fig. 1), and was not correlated with conspecific egg load (Spearman's correlation rs = −0·085, P= 0·4).

Table 1.  Summary of measured fruits and larval traits for all species, and for early and late species
 All speciesEarlyLate
  1. Mean (± SE). LIS: larval installation success; LFM: larval fresh mass; LA: larval area.

LIS (%)               –  53·8 (42)  40·4 (30)
LFM (mg)    2·4 (0·9)    3·2 (0·8)    2·1 (0·8)
Larval length (mm)    3·62 (0·88)    3·92 (0·98)    3·48 (0·79)
Larval width (mm)    1·18 (0·65)    1·23 (0·22)    1·16 (0·74)
LA (mm2)    3·32 (1·10)    3·81 (1·42)    3·09 (0·85)
Carpel number  37·7 (12·3)  35·3 (10·2)  38·7 (12·8)
Seed number per carpel    9·0 (2·2)    8·6 (2·3)    9·0 (2·2)
Seed number before predation330·8 (118·9)304·9 (128·6)340·7 (114·0)
Seed number after predation145·4 (77·9)140·4 (81·3)146·6 (76·6)
Seed eaten per larva  36·2 (24·2)  46·4 (32·8)  33·0 (19·5)
Figure 1.

Mean (± SE) (a) larval installation success, (b) larval fresh mass and (c) larval area per category for early and late species. (Effect of category on LISearly: Krustal–Wallis H3,132 = 18·98, P < 0·0001; LISlate: H2,100 = 4·22, P= 0·12; LFM: anovaF4,450 = 19·61, P < 0·0001; LA: F4,424 = 47·24, P < 0·0001). Significant differences among means are indicated by different letters.

Larval fresh mass was correlated negatively with the number of larvae developing within a fruit (rs = −0·339, P < 0·001). Early larvae were significantly heavier than late larvae (3·2 vs. 2·1 mg, respectively, P < 0·001, Table 1). Early larva developing alone tended to be heavier than in presence of conspecific (Fig. 1) but this difference was not significant, due to the very low larval installation success in category 2: only three fruits harboured only two early larvae, and in all these three fruits, one of the two larvae was much larger than the other (Fig. 2). In the presence of late species, early larvae were significantly lighter (category 4 and 5 vs. category 1). Early larval mass decreased by about 16% in presence of conspecific, and by only 8·7% in presence of heterospecific larvae, indicating stronger intra- than interspecific competition. The presence of early larvae had no influence on late larval mass (categories 4 and 5 vs. category 3). Late larval mass was correlated negatively with the number of late larvae developing in the fruit (rs = −0·28, P= 0·008).

Figure 2.

Individual mass of early larvae at first larva exit in the presence of a conspecific larva and in the absence of late larvae (category 2: only three of 21 fruits selected first with two early eggs finally harboured two early larvae).

Early larvae were larger than late larvae (3·91 vs. 3·09 mm2, P < 0·001, Table 1). Early larvae were significantly smaller in presence of conspecific or of late larvae (5·36 mm2 vs. 3·48 mm2 and 3·45 mm2, respectively, Fig. 1). Late larvae tended to be smaller in the presence of early larvae, but this difference was significant only for category 5 (i.e. when late larvae were in presence of two early larvae). Late larval area was independent from the number of late larvae developing in the fruit (rs = 0·05, P= 0·65).

effect of resource size on the intensity of larval competition

The total number of seeds produced by a flower before predation (mean ± SE: 330 ± 119) increased positively with carpel number (Table 2). The number of eggs per fruit was significantly correlated with carpel and seed number: large fruits had large egg loads. The number of seeds per carpel (mean ± SE: 9 ± 2) was independent of the number of eggs; that is, pollination efficiency was independent from oviposition. Large fruits harboured more larvae and lost more seeds to predation. However, they also liberated more undamaged seeds. None of the three larval traits measured (LIS, larval mass and area) was correlated with carpel or seed number available before predation (Table 2). Larval mass was correlated positively with the number of seeds eaten per larva, which was correlated positively with fruit size, but negatively with the number of larvae developing in the fruit.

Table 2.  Effect of resource size (seed number before predation) on egg and larval numbers, on destroyed and undamaged seed numbers, and on three larval traits: installation success (LIS), fresh mass (LFM) and area (LA), for early and late larvae
 No. of fruitsSpearman's rs
  • ***

    P < 0·001,

  • **

    P < 0·01.

Carpel number × seed number1190·693***
Egg number × seed number1190·449***
Egg number × seed number/carpel1190·153
Larval number × seed number  830·302***
Seed destroyed × seed number1010·722***
Seed undamaged × seed number1010·708***
Seed destroyed × larval number  840·498***
Seed destroyed/larva × seed number  840·24***
LISearly × seed number119 −0·009
LISlate × seed number1000·203
LFM × seed number119 −0·05
LFMearly × seed number103 −0·094
LFMlate × seed number  92 −0·002
LA × seed number119 −0·02
LAearly × seed number103 −0·19
LAlate × seed number  920·07
LFM × seed destroyed/larva1190·27***
LFMearly × seed destroyed/larva1030·26***
LFMlate × seed destroyed/larva  920·18**
LA × seed destroyed/larva1190·18***
LAearly × seed destroyed/larva1030·28***
LAlate × seed destroyed/larva  920·09

Discussion

intra- vs. interspecific competition

The strongest evidence for competition came from early larvae in the presence of conspecifics. Flowers naturally infected by more than one early egg were difficult to find, indicating that early ovipositing females usually avoid laying more than one egg per fruit. Because Chiastocheta density in the study population was unusually high, some flowers contained more than one early egg, and an intensive search effort allowed us to select a total of 73 flowers with two early eggs. Among them, only 18 finally harboured two early larvae. This is an indication of strong intraspecific exclusion. In the rare cases where two early larvae did develop within a single fruit, one was much smaller and lighter than the other. This suggests contest competition among early larvae. The specificity of the resource exploited by early larva (i.e. the central pith of the fruit) may explain exclusive competition. In this case, living space rather than seed availability seems to be the limiting factor to intraspecific larval coexistence. By contrast, LIS among late larvae was independent of the number of late larvae in the fruit, but both mass and area of larvae decreased with increasing larval number. This study therefore evidences intraspecies competition among early larvae mainly for larval installation, and intraspecific competition among late larvae for larval development.

The presence of late species had no effect on LIS of early species, and the presence of early species did not affect LIS of late species. The two species, therefore, did not interact for LIS. By contrast, late larvae had a negative effect on early larval mass and area, while early species had no effect on late larval mass or area. This finding is opposed to our ‘first come, first served’ hypothesis, which assumes that early larvae should monopolize the resource and therefore suffer less from interspecific competition compared to late larvae. Although we found indeed that early larvae were larger and heavier overall than late larvae, this size difference did not result in a higher competitive ability. In a field study of Chiastocheta larval development in absence of competition, Pompanon et al. (unpublished) also found that late species tended to be smaller than early species, despite the fact that all species consumed the same amount of seeds per capita.

The level of seeds available had no impact on larval mass or area, thus the decrease in larval mass and area with increasing larval number was not due to food limitation, but rather to a decreasing success in getting access to available seeds with increasing larval crowding in the fruit. This is evidenced by a decrease in the number of seeds eaten per larva with increasing larval number. This decrease in per capita seed consumption presumably results from less efficient searching/feeding activity due to interference among larvae changing their path when encountering, rather than direct aggressive fights, as larval mortality was low. It may also result from increasing release of chemical growth inhibitor by larvae (Moore & Whitacre 1972) or by the damaged plant (Schultz 1988).

consequences of larval competition patterns on the stability of the globeflowerglobeflower fly pollination mutualism

In addition to being the sole Trollius specialist seed predator, Chiastocheta spp. also represent the sole pollinators of the European globeflower. The competition patterns observed for early and late ovipositing species have major consequences for pollination efficiency of the different species. Early ovipositing flies visit young, unpollinated flowers, and deposit typically only one egg per flower, because of very strong intraspecific larval exclusion. They visit many young flowers to lay all their eggs, therefore providing a valuable pollination service. Conversely, the late ovipositing species C. dentifera oviposits several eggs on fruits, without contributing to pollination. Therefore, late species may be viewed as exploiters of the mutualism, or ‘cheaters’, and their presence could be a destabilizing factor of the plant–insect interaction (Pellmyr, Leebens-Mack & Huth 1996). It has been argued that for the interaction to remain evolutionarily stable, cheaters should suffer more from competition than mutualists. This prediction failed to be verified in the Yucca moths (Marr et al. 2001) and in the globeflower flies (present study). However, unlike in other described plant–seed parasite pollinator mutualisms, both male and female Chiastocheta pollinate globeflowers (Després 2003), so that the negative effect of late ovipositing females is counterbalanced by the positive effect of males. Therefore, in the globeflower–globeflower fly interaction, both early and late species act as pollinators.

None of the fruits analysed lost all seeds to predation. Maximum seed predation did not exceed 93%, whereas up to 13 larvae were found developing within a single fruit. Therefore, even at high larval densities, at least some seeds were spared from larval predation. This is a condition for the stability of the plant–insect pollination mutualism (Jaeger et al. 2001). The concordance observed between first larval exit and carpel dehiscence suggests that the length of larval development is constrained strongly by seed maturation duration. At high larval densities, low larval mass and area at the end of development are likely to have a negative effect on pupa and adult fitness (Hess, Abrahamson & Brown 1996), for both early and late species. Such density-dependent responses to larval crowding are likely to be a key factor in stabilizing Chiastocheta population dynamics (Varley, Gradwell & Hassell 1973), and ultimately in stabilizing the plant–insect interaction outcome. Indeed, despite the extremely high Chiastocheta larval density observed in our study sample, only about half of the seeds were lost to pollinator breeding, while the other half were preserved for plant reproduction. This outcome of the globeflower–globeflower fly interaction is found throughout the range of globeflower in Europe, despite highly variable Chiastocheta densities and ecological conditions from one locality to another, and from year to year (Jaeger & Després 1998; Hemborg & Després 1999; Jaeger, Till-Bottraud & Després 2000).

Another important result of this study is the finding that despite being more heavily parasitized, large globeflowers ultimately disperse more undamaged seeds than smaller globeflowers. This observation, together with the fact that Trollius species not parasitized by Chiastocheta have consistently lower carpel numbers than T. europaeus (Pellmyr 1992), suggests that an increase in carpel number is selected in globeflowers under the predation pressure imposed by Chiastocheta larvae. The fact that Chiastocheta females adjust the number of eggs laid to the flower size suggests an ongoing arms-race between the plant and the insect, the plant investing more in seed production to counterbalance the seed loss to pollinator larvae, and the pollinator responding by laying more eggs on larger flowers.

why are there so many globeflower fly species?

Maturing fruits represent a patchily distributed and temporally variable resource for seed predators. Typically, fruits are parasitized by a succession of specialists belonging to different families or orders (Elton 1966). Compared to other described plant–seed predators systems, the number of sister species coexisting on globeflowers is amazingly high. This difference probably results from the fact that globeflower flies are the sole seed predators of globeflower whereas, for example, larch seeds are parasitized by a phylogenetically diverse community of insects, in addition to two to three coexisting Strobilomyia sister species (Turgeon, Roques & Groot 1994). The taxonomically depauperate fauna exploiting globeflower seeds results presumably from the accumulation of chemical repellents against herbivores in this plant genus: considerable amounts of ferulic and sinapic acids, flavones and saponins were observed in the Trollius genus (Jensen 1995). Chiastocheta's ancestors found a way to overcome these chemical defences and invaded a competitor-free niche, a favourable context for adaptive radiation to take place (‘ecological opportunity’, Schluter 2000). The Chiastocheta sister group is unknown, but several Anthomyiid genera feed on maturing inflorescences of various plant taxa (Pettersson 1992; Brody & Waser 1995; Roques et al. 1995; Abbott 2002), so that this feeding habit was not a novelty in Chiastocheta. The key innovation for Chiastocheta larva was specialization on Trollius seeds, a previously unexploited niche.

In its search for a suitable oviposition site, the Chiastocheta female may use both attracting and deterring physical or chemical stimuli to discriminate among different host-plant stages (Kostal 1993). For example, young and old globeflowers differ in the diameter of their globe-shaped corolla and in sepals tightness, which may serve as visual cues for females. The complex fruit structure involves the evolution of specific organs to gain access to different laying sites. This is notably the case in yucca moths, where deep and shallow ovipositing species differ in their ovipositor structure (Pellmyr & Leebens-Mack 2000) and in globeflower flies, where the fruit ovipositing species, C. dentifera, has a distinctly long ovipositor allowing the female to lay eggs between maturing carpels, a site not available in young flowers (Pellmyr 1992). Differences in ovipositor lengths were also reported in sympatric fig wasps laying at different phenological stages of the fig (Kerdelhué & Rasplus 1996; Weiblen & Bush 2002). Therefore, different fruit structures at different phenological stages imply different laying sites availability, and different constraints for the emerging larva to enter into the fruit. Floral odour may also be used for long-range evaluation of the host-plant phenological stage (Irwin & Dorsett 2002). Contact chemoreception is most important for the choice of oviposition site after the female has alighted upon a suitable plant. Deterring compounds affecting oviposition can either be produced by the plant itself, or by previous visitors to the plant, through pheromone deposition (Huth & Pellmyr 1999; de Jong & Stadler 2001). All these pathways towards specialization for oviposition on different floral stages and their incidence on larval feeding behaviour and physiology remain to be studied in detail in the globeflower fly, but they imply disruptive selection against hybrids, which are poorly suited to exploit any floral stage. Such postzygotic selection is likely to be reinforced by prezygotic sexual selection. Indeed, in addition to oviposition shifting, all the globeflower fly species are easily recognized by strongly differentiated male genitalia, a character known to be under sexual selection in all insects, and more particularly in dipterans (True et al. 1997; Arnqvist 1998). Another possible reinforcement is assortative mating, if species laying eggs at different floral stages also mate in flowers at different stages.

In conclusion, although so far adaptive radiation in plant–insect systems was thought to result from cospeciation and/or host shifts followed by disruptive selection on hybrids, adaptive radiation within a single host through shifts in oviposition timing and specialization on different floral/fruit phenological stages could be an underexplored path for speciation. Such oviposition shift may have occurred in scuttle flies (Diptera: Phoridae) ovipositing on fresh and faded flowers of Aristolochia (Disney & Sakai, 2001), and in drosophilids ovipositing in sequence on rotten fruits. More generally, this pathway toward sympatric speciation should be evaluated in host–parasite systems. For example, oxyurid sister species were observed to coexist in turtle and cockroach guts (Schad 1963; Adamson & Noble 1993). In that case, variation in resource quality is both temporal and spatial: during digestion process, aliments transiting throughout the gut offer a coarse-grained resource to be shared among coexisting specialized species.

Acknowledgements

We thank Florian Lacombe, François Pompanon and Philippe Gayral for field assistance, and Steve Jordan for helpful comments on the manuscript.

Ancillary