Evolution of oviposition strategies and speciation in the globeflower flies Chiastocheta spp. (Anthomyiidae)

Authors


Després Laboratoire de Biologie des Populations d’Altitude, CNRS-UMR 5553, Université J. Fourier, BP 53-38041 Grenoble Cedex 09, France. Tel: +33 4 76 63 56 99; fax: +33 4 76 51 42 79; e-mail: laurence.despres@ujf-grenoble.fr

Abstract

Trollius europaeus (Ranunculaceae) is involved in an intimate interaction with several species of Chiastocheta flies (Anthomyiidae) that are both seed predators and pollinators. In this paper, we analyse the oviposition strategy of the six Chiastocheta species found to coexist on T. europaeus in 19 populations from the French Alps. We show that the species are not equivalent in their oviposition behaviour: C. rotundiventris usually deposits no more than one egg per flower in first-day flowers whereas C. dentifera aggregates its eggs on fruits and thus does not contribute to pollination at all; the four remaining species deposit eggs sequentially during the flowering period from the 2nd to the 7th day. Hence, the outcomes of the interaction in terms of net seed production for the plant greatly depend on the Chiastocheta species visiting it, ranging from a mutualistic to a purely parasitic interaction. We assessed mitochondrial divergence between Chiastocheta spp. by sequencing a 1320-bp mitochondrial DNA fragment. The low divergence observed between species (0–4.15%) suggests that genus diversification took place recently. Unlike in other plant–insect systems where diversification is usually thought to be driven by cospeciation or host shifts, we propose that Chiastocheta speciation took place within the host plant. Basal separation of a particularly mutualistic species provided favourable conditions for plant specialization on this seed-parasite as a pollinator early in the evolution of the association. The parasitic species ovipositing on fruits derived from a species ovipositing on flowers. Diversification of the intermediate strategies probably occurred in relation with the Pleistocene climatic events, reproductive isolation between species being reinforced by niche partitioning for oviposition and/or sexual selection.

Introduction

Plant-seed parasitic pollinator mutualisms involve a plant pollinated by an insect whose larvae develop by eating a fraction of the seeds. When the insect is the exclusive pollinator and when its larva relies only on the plant seeds for its development, the interaction can be defined as an obligate mutualism. Few cases of such extremely specific plant-seed parasitic pollinator mutualisms have been convincingly documented: the fig–fig wasp ( Bronstein, 1992; Anstett et al., 1996 ), the yucca–yucca moth ( Addicott et al., 1990 ; Powell, 1992) and the globeflower–globeflower fly interactions ( Pellmyr, 1989; Jaeger & Després, 1998). However, depending on its oviposition behaviour, the insect can be more or less beneficial for the plant. If no pollination occurs during oviposition, for example when oviposition takes place on fruits, the insect will have a strictly detrimental effect on seed production. Such nonpollinating behaviour was observed among individuals of the pollinating species ( Aker & Udovic, 1981; Tyre & Addicott, 1993; Addicott & Tyre, 1995) or among all individuals of a species associated with the pollinating species on the same host ( Galil & Eisikowitch, 1968; Pellmyr, 1989; Bronstein, 1991; Powell, 1992; West & Herre, 1994; Addicott, 1996; Kerdelhué & Rasplus, 1996; Pellmyr et al., 1996a ; West et al., 1996 ; Bronstein & Ziv, 1997). If pollination occurs, the benefits for the plant will depend on the oviposition strategy through the pollination service provided during oviposition and the number of seeds destroyed by larvae. The pollination service provided will depend on the amount and quality of pollen deposited ( Tyre & Addicott, 1993; Pellmyr, 1997), the number of ovules remaining to be fertilized and, hence, the age of the flower visited ( Pellmyr, 1989). The number of seeds destroyed will depend on the number of eggs laid ( Addicott, 1986). Thus, studying the oviposition strategy of insect species coexisting on a plant is of primary importance in understanding the nature of the plant–insect association.

The globeflower Trollius europaeus is widespread in northern Europe (Scandinavia and Russia), and restricted to mountains (above 700 m) in southern Europe. The plant is found in moist meadows and usually produces a single yellow globe-shaped flower. Previous studies showed that the interaction between T. europaeus and Chiastocheta flies (Anthomyiidae) is an obligate mutualism in Scandinavia ( Pellmyr, 1989) and in the Alps ( Jaeger & Després, 1998): Chiastocheta larvae feed only on Trollius seeds and flies are the only pollinators of flowers. In Europe, eight Chiastocheta species, including two vicariant species, were described based on male genitalia and egg morphology ( Collin, 1954; Hennig, 1976; Michelsen, 1985; Pellmyr, 1992): C. rotundiventris (northern vicariant: C. abruptiventris), C. dentifera, C. trollii, C. inermella (eastern vicariant: C. lophota), C. macropyga and C. setifera. Species names are follow Pellmyr (1992). Only four species were found in globeflower populations in Scandinavia (C. abruptiventris, C. trollii, C. inermella and C. dentifera) and were shown to have contrasted oviposition strategies, visiting the flower at different developmental stages and laying more or less eggs per visit ( Pellmyr, 1989; Johannesen & Loeschcke, 1996a).

Diversification in plant–insect systems is usually thought to be either the result of strict cospeciation or frequent host shifts followed by specialization ( Ehrlich & Raven, 1964; Thompson, 1989; Chenuil & McKey, 1996; Köpf et al., 1998 ). Accordingly, in fig–fig wasp and yucca–yucca moth systems, one plant species is associated with one pollinating insect although a few exceptions have been reported ( Powell, 1992; Michaloud et al., 1996 ; Kerdelhuéet al., 1997 ). The globeflower–globeflower fly system is unique in that it involves several Chiastocheta species coexisting on one plant species, raising the question of the evolution and maintenance of so many species within the same host plant.

In this paper, we studied coexistence and oviposition strategies of the six Chiastocheta spp. we found on T. europaeus in the Alps. To characterize the nature of the association between the host plant and the different species (mutualism or parasitism), we analysed the rates of pollination achieved throughout flower longevity. To determine how the various oviposition strategies have arisen during the course of evolution, we inferred the first molecular phylogeny of the Chiastocheta spp. found on T. europaeus. Evolution of the different oviposition strategies and diversification of the European Chiastocheta spp. are discussed in the light of this molecular phylogeny.

Materials and methods

Species frequency and oviposition characteristics

The relative frequency of each Chiastocheta species was indirectly estimated in 19 populations from different parts of the French Alps by counting eggs laid on 37–132 fruits per population at the end of the flowering period in 1996. Populations ranged from 700 to 2400 m elevation and were 10–300 km apart. Flowering of populations lasted 3–5 weeks, and the mean individual flower duration was estimated to be 7 days ( Jaeger & Després, 1998). Chiastocheta eggs were easy to count on or between the carpels. Even when the larva leaves the egg, the empty eggshell remains tightly attached to the carpel walls. Egg characteristics like its size, colour and position on the carpels allow it to be assigned to one of the six species ( Pellmyr, 1992), only C. trollii and C. setifera eggs being difficult to discriminate without an optical microscope. For each species, the mean number of eggs per fruit (M) and its variance (V) in each population were estimated, and a coefficient of dispersion (CD V/M) was determined. The coefficient of dispersion indicates how eggs are distributed among flowers. If CD = 1, eggs are randomly distributed in the population (Poisson distribution), if CD < 1 the distribution is uniform and if CD > 1 the distribution is aggregated.

To determine the age of flowers visited by ovipositing females, 350 flowers were tagged on their first day of flowering and 50 of them were bagged each following day from day 1 to day 7 in three populations (Cherlieu, Som and Lac) in 1997. We thus defined seven groups of flowers, corresponding to flowers available for pollination and oviposition during 1, 2, … or 7 days. The mean number of eggs laid per flower was calculated for each group of flowers and for each Chiastocheta species. However, eggs from C. trollii and C. setifera were not separated in this analysis.

In addition, to analyse evolution of pollination rates throughout flower longevity, the mean seed set per flower was calculated for each group of flowers. On each flower, the seed set was estimated as the number of seeds produced per carpel (before predation) divided by the mean number of ovules per carpel in the population as described by Jaeger & Després (1998). Fifty other unmanipulated fruits were sampled in each population as a control (full natural pollination).

Molecular phylogeny

Total DNA was extracted from adult male flies from C. dentifera (n = 3), C. inermella (n = 4), C. macropyga (n = 2), C. rotundiventris (n = 2), C. setifera (n = 3) and C. trollii (n = 2) sampled at various places in the Alps, and from C. abruptiventris collected in Sweden (n = 2) and Russia (n = 1). The only European species not sequenced in this study, C. lophota, is a closely related eastern vicariant of C. inermella, and according to Michelsen (1985), these two species are morphologically ‘very similar’. We also extracted DNA from the onion fly Delia brassica (Anthomyiidae) as an outgroup. Flies were preserved in 70% ethanol before DNA extraction. Total DNA was extracted from the head of the fly using the Chelex procedure ( Walsh et al., 1991 ). A total of 1320 bp were sequenced containing most of the COI (3′ end, 772 bp) and COII (5′ end, 548 bp) mitochondrial genes. The two fragments were separately amplified by PCR using direct primers COI-2171 (5′ TTGATTTTTTGGT CAYCCNGAAGT 3′) and tRNAleu-3023 (5′ GA TTAGTGCAATGGATTTAGCTC 3′), and reverse primers tRNAleu-3048 (5′ TGGAGCTTAAATCCATTGCAC 3′) and COII-3683 (5′ CCRCAAATTTCTGAACATTGACC 3′) derived from several insect sequence comparisons (fruitfly, bee, mosquito) found in nucleotide databases. Primer numbers refer to Drosophila yakuba mitochondrial sequence ( Clary & Wolstenholme, 1985). The amplification primers were also used for sequencing of the two DNA strands and sequences were run on an ABI PRISMTM DNA sequencer (Perkin Elmer). Sequences were aligned and analysed using MEGA (version 1.01, Kumar et al., 1993 ) and phylogenetic reconstruction inferred by unweighted parsimony (MP), neighbour joining (NJ) and maximum likelihood (ML) analyses using PAUP* (test version 4.0d64 written by David L. Swofford). Robustness of the MP and NJ trees was tested by performing 1000 bootstrap replicates.

Results

Fly species relative frequency in natural populations

Six Chiastocheta spp. were found on T. europaeus in the French Alps: C. rotundiventris, C. dentifera, C. inermella, C. macropyga, C. setifera and C. trollii. Most populations sampled contained five or six species, and only one population (Tête) had a single species (Fig. 1). No population was found without Chiastocheta flies. Among the 1102 flowers checked for eggs, only 9% did not harbour Chiastocheta eggs, 27% had eggs from one species, and 64% of the flowers had eggs from 2–5 species (35% from two species, 19% from three species, 8% from four species, 2% from five species). Thus, not only do several Chiastocheta spp. coexist in each of the populations except one, but they also coexist within a single flower.

Figure 1 Relative frequency of the six Chiastocheta species in 19 populations of the French Alps estimated from eggs laid on 37–1.

Figure 1 Relative frequency of the six Chiastocheta species in 19 populations of the French Alps estimated from eggs laid on 37–1.

32 fruits sampled per population.

Oviposition characteristics of the different species

C. rotundiventris had a CD under 1 in all populations: females avoid laying more than one egg per flower (Table 1). All the other species had a CD greater than 1. C. dentifera had the most aggregated distribution with CD ranging from 1.83 to 13.42. In all species except C. rotundiventris, there was an aggregation of the eggs resulting from multiple oviposition by females in the flowers.

Table 1.   Dispersion of eggs among flowers in the 19 populations studied in the French Alps. n = sample size, M = mean, SD = standard deviation, V= variance, CD= coefficient of dispersion (= V/M). Thumbnail image of

The age of flowers visited varies among Chiastocheta species as shown by the oviposition patterns (Fig. 2). The mean number of C. rotundiventris eggs laid per flower increases until the second day and is stable from the second to the last day of flowering: most of the eggs are laid on the first-day and a few on the second day of flowering. According to Pellmyr (1989), C. trollii typically starts to lay eggs on the 5th day of flowering. The two-step increase pattern we observed for C. trollii and C. setifera eggs deposition (eggs were pooled together) suggests that C. setifera lays (or begins to lay) its eggs as early as the second day. C. inermella eggs are found from the 5th day of flowering and most of them are laid on the 7th day of flowering. C. macropyga lays eggs from the 3rd to the 6th day of flowering. Finally, C. dentifera starts laying eggs on the 6th day of flowering, thus just before the end of flowering. However, this species mostly lays eggs after the end of flowering, as described by Pellmyr (1989). No other Chiastocheta species were observed to oviposit on fruits.

Figure 2 Chronology of oviposition of the six Chiastocheta spp. Eggs of each species were counted on seven groups of 50 flowers after 1, 2.

Figure 2 Chronology of oviposition of the six Chiastocheta spp. Eggs of each species were counted on seven groups of 50 flowers after 1, 2.

, … 7 days of natural pollination in three populations (Cherlieu, Lac and Som). The first day eggs are found indicates the age of the flower when oviposition begins. If the average egg number on flowers the following days is increasing, oviposition continues for several days, whereas if it is stable, oviposition takes place mainly on the first day. Eggs from C. trollii and C. setifera which are difficult to tell apart were pooled.

Seed set as a function of flower age

There were significant effects of population ( ANOVAF2,856= 36.06, = 0.0001), age of flower (F7,856 = 51.23, P = 0.0001) and population*age of flower (F14,856 = 2.83, < 0.001) on seed set. In all populations, there was a gradual increase in seed set throughout all flower lifespan (Fig. 3). This increase ceased to be significant after 4 days of natural pollination at Som, 5 days at Lac and 6 days at Cherlieu (Duncan’s test, P > 0.05). This indicates that pollination is gradual over time and not fully achieved before days 4–6 (Fig. 3).

Figure 3.

  Patterns of pollination during flower life. Seed set was determined as the average number of seeds per carpel divided by the average number of ovules per carpel per group of flowers naturally pollinated during 1, 2, … 7 days in three populations (Cherlieu, Lac and Som). The control group are unmanipulated flowers (full seed set). Vertical bars are standard errors. The level of pollination is not different in groups 6, 7 and control (Duncan’s test, P > 0.05), indicating that pollination is fully achieved on day 6.

Molecular phylogeny

Among a total of n = 19 individuals sequenced, we found 12 different haplotypes. The two vicariant species C. rotundiventris (n = 2) and C. abruptiventris (n = 3) shared the same haplotype. All the C. trollii individuals sequenced (n = 3) shared the same haplotype as one individual from C. setifera (n = 3 individuals/2 haplotypes). C. dentifera (n = 3 individuals/3 haplotypes) and C. inermella (n = 4 individuals/4 haplotypes) were the more variable species, with intraspecific divergence ranging from 0.23 to 1.23%. A total of 80 of the 1320 nucleotide sites were variable within the genus Chiastocheta (6.1%), and 38 sites were informative (2.9%). The translation of the nucleotide sequence resulted in a 440 amino acid sequence, and only eight amino acid positions were variable within the 12 haplotypes. Most of the nucleotide substitutions were transitions (Table 2) and, as generally found in insect mtDNA, there was an important A + T bias in base composition (A: 32.9%, T: 39.5%, G: 13.3%, C: 14.2%), especially at the third codon position where 82.9% of nucleotides were A or T. We corrected pairwise distances with the option ‘maximum likelihood’ using the observed base frequencies and TS/TV ratio ( Felsenstein, 1984). Corrected distances among Chiastocheta species range from 0 to 4.15% (Table 2).

Table 2.   Divergences in COI-COII nucleotide sequences (1320 bp) between the 12 Chiastocheta haplotypes. Corrected maximum likelihood distances are given below the diagonal. The pairwise transition/transversion ratio is given above the diagonal. C. de = C. dentifera, C. in = C. inermella, C. ma = C. macropyga, C. ab + ro = C. abruptiventris + C. rotundiventris, C. se + tr = C. setifera + C. trollii. Number in parentheses indicates the number of individuals sequenced sharing the same haplotype. Intraspecific divergences are in italics. Thumbnail image of

Two most parsimonious trees were found (branch-and-bound search, length: 190 steps, CI: 0.847, RI: 0.73). The ML tree was not different from the NJ tree and congruent with the MP consensus tree (Fig. 4). C. rotundiventris and its northern vicariant C. abruptiventris have an ancestral position, then the parasitic species C. dentifera arise, followed by the four intermediate species. All these three nodes were supported by high bootstrap values in the NJ and MP analyses. Rooting the tree with different outgroups (Delia sp., Drosophila sp. or Aedes sp.) did not change its topology. Phylogenetic relationships within the cluster inermella/macropyga/trollii/setifera remain poorly resolved, owing to the lack of nucleotide variability (less than 1.23%), which is comparable to intraspecific variability found in the two more variable species, C. dentifera (0.23–0.92%) and C. inermella (0.46–1.23%). C. inermella appeared to be paraphyletic, and constraining its monophyly by using MacClade (version 3, Maddison & Maddison, 1992) resulted in six additional steps in the MP tree.

Figure 4.

  Phylogenetic relationships between the six Chiastocheta species (12 haplotypes) based on 1320 mtDNA nucleotide sites. C. setifera and C. trollii share the same sequence. MP, NJ and ML analyses gave the same tree topology. Tree was rooted with Delia sp. Bootstrap percentage (1000 replicates) for nodes > 60% are indicated in italic above (NJ tree) or below (MP tree) the branches. Oviposition strategies are mapped on the tree. The most parsimonious state for the Chiastocheta ancestor is a flower-ovipositing strategy.

Discussion

Species coexistence and oviposition strategies

Five or six Chiastocheta species were present in most populations sampled. This result confirms previous studies concerning coexistence of Chiastocheta spp. in T. europaeus populations ( Pellmyr, 1989; Johannesen & Loeschcke, 1996a). In Scandinavia, only four species were described: Pellmyr (1989) found all coexisting in different meadows along a river in Finland and Johannesen & Loeschcke (1996a) found the four species coexisting at 12 sites out of 18 in Denmark. In the Alps, six species were found to coexist in most populations (this study). Moreover, we found that 2/3 of the flowers harboured eggs from more than one species. This indicates that larvae of different species do interact within flower heads most of the time. In another study we showed that competition occurs between larvae for seed consumption (Jaeger et al., unpublished data). So far, no general mechanism has been demonstrated to promote coexistence of the Chiastocheta species ( Johannesen & Loeschcke, 1996a) and the hypothesis of niche partitioning by specific stereotyped larval mining patterns ( Pellmyr, 1989) remains to be tested.

The six Chiastocheta spp. can be characterized by their ovipositing behaviour (distribution of eggs among flowers and age of flowers visited) as being more or less beneficial to the host plant. C. rotundiventris rarely deposits more than one egg per flower and visits only young flowers (mostly first-day flowers) that are more likely to be unpollinated. It is worth noticing that pollination levels obtained at the end of the first day of flowering represent a significant part of total seed production (30% at Cherlieu and Lac, 50% at Som), reflecting the important role played by C. rotundiventris in globeflower pollination. At the opposite, C. dentifera tends to highly aggregate its eggs and was shown to oviposit mostly on fruits, thus after pollination took place. This parasite species was found in almost all populations and represents the highest number of eggs laid, presumably leading to the highest predation costs in terms of seeds consumed among all Chiastocheta species. The four other species have intermediate strategies, aggregating eggs in 2–7-day-old flowers, thus potentially contributing to pollination as they visit flowers and not fruits. Under the pollination conditions observed in this study (i.e. full pollination not achieved before days 4–6 depending on population), pollen deposition during early oviposition of C. setifera and C. macropyga females is more likely to lead to seed production than during late oviposition of C. inermella and C. trollii females.

Evolution of oviposition strategies in Chiastocheta spp.

Three clades are strongly supported by high bootstrap values whatever the phylogenetic method used to build the tree: the first clade groups the two vicariant species C. rotundiventris and C. abruptiventris, the second clade includes all the C. dentifera haplotypes and the third clade groups the four species with intermediate oviposition strategies.

The mitochondrial phylogeny obtained strongly supports the basal position of C. rotundiventris and its northern vicariant C. abruptiventris. These two species could not be distinguished in this study: they shared the same haplotype, but also the same egg morphology and oviposition strategy. We could not find any morphological differences on the male genitalia of individuals collected in Sweden and Russia and those collected in the Alps: they all had the morphological traits specific to C. rotundiventris as described and drawn by Hennig (1976). Pellmyr (1989) called the species he found in Finland C. sp. aff. rotundiventris and only in his 1992 paper did he accept C. abruptiventris as a valid species, without giving any specific diagnostic character. The validity of C. abruptiventris as a distinct species can be questioned as no morphological description of this species has been published so far. The haplotype common to C. rotundiventris and C. abruptiventris was well differentiated from all the other haplotypes with divergences ranging from 3.65 to 4.15% (Table 2). In a study based on allozyme electrophoresis, Johannesen & Loeschcke (1996b) established interspecific Nei’s genetic distance between the four species they found in Denmark: C. abruptiventris was the most distinct species and C. dentifera, C. inermella and C. trollii were grouped in one cluster. The present mtDNA rooted phylogenetic tree with C. rotundiventris and C. abruptiventris as basal species confirms this result.

In the present study, despite the high number of sites sequenced (1320 bp), the molecular divergence observed is low (0–4.15%), particularly among species with intermediate strategies (0–1.23%). As a comparison, pairwise sequence divergences ranged from 0.1 to 17% between 16 species of the genus Greya (Lepidoptera, COI-COII, Brown et al., 1994 ), from 0.1 to 13.1% between seven species of the genus Phratora (Coleoptera, COI, Köpf et al., 1998 ), and from 2 to 4.8% between six species of the Drosophila affinis subgroup, the most recent radiation in the D. obscura group (Diptera, COII, Beckenbach et al., 1993 ). In this latter study, authors gave a rough estimation of 1.6–3.5 Ma for this radiation. As Chiastocheta spp. show divergences of the same order of magnitude, and under the assumption that the evolutionary rate of mtDNA is not totally different between these two Diptera genera, the diversification of Chiastocheta spp. may have begun in the late Pliocene/early Pleistocene (about 2 Ma). According to Doroszewska (1974), T. europaeus came to Europe from Asia through the Balkans during the Miocene (23.5–5.3 Ma). Other taxa are thought to have migrated from Asia to Europe at this period (e.g. Dactylis spp., Stebbins & Zohary, 1959). The low molecular divergence found among Chiastocheta spp. associated with T. europaeus suggests that diversification of these species is much more recent than Trollius migration to Europe and that it did take place within the host plant. An alternative hypothesis is that speciation within the Chiastocheta genus occurred before or throughout radiation of a plant host clade including four Trollius Asiatic species and T. europaeus ( Pellmyr, 1992). Further molecular phylogenetic studies including Asiatic Chiastocheta species sequences would be necessary to test whether the European and Asiatic species form two separate clades, or are clustered together.

Given the phylogeny obtained, the most parsimonious hypothesis is that the ancestor of European Chiastocheta spp. did lay eggs on flowers (Fig. 4). This is consistent with the fact that most if not all other Chiastocheta species described so far lay eggs during flowering ( Pellmyr, 1992). From this flower-ovipositing ancestor, there was (1) separation of an oviposition strategy on young flowers, (2) separation of an oviposition strategy on fruits and (3) diversification of the four intermediate strategies.

At the beginning of its development, a larva of C. rotundiventris uses the basal part of the flower where there is no room for several larvae. This may explain why females of this species lay no more than one egg/flower in young flowers, which are more likely to be unoccupied. We do not know whether the use of receptacle is an ancestral or a derived trait in Chiastocheta. However, adoption or abandonment of the use of receptacle is likely to have profound effects on the evolution of the oviposition strategy. Plant specialization on a seed parasitic-pollinator requires external conditions like rarity of copollinators ( Thompson & Pellmyr, 1992) and also particular traits of the insect like local host specificity, pollination during oviposition and nonexcessive seed destruction ( Pellmyr et al., 1996b ). T. europaeus was associated early with a particularly mutualistic species, suggesting that favourable conditions for increasing plant specialization were present early in the association history. Specialization ultimately led to an obligate association: T. europaeus has a totally closed corolla which is unique in its genus, excluding all visitors but Chiastocheta.

The parasitic species ovipositing on fruits arose from a mutualistic species ovipositing on flowers. Similarly, cheaters ovipositing on fruits rather than on flowers were found in several yucca species ( Addicott, 1996; Pellmyr et al., 1996a ) and molecular data show that they derive from pollinators ( Pellmyr et al., 1996a ). It has been argued that a cheater strategy evolving within an active pollinator species should invade because it is less costly ( Pellmyr & Huth, 1994; Pellmyr et al., 1996a ), leading to reciprocal extinction, so that sympatric speciation within the same host-plant could not occur. However, pollination by Chiastocheta is presumably passive (no special pollen transport structures have been found so far) so that the cost of pollination could be very low. In such a system, the age of flower when oviposition takes place will determine a pollinating or a nonpollinating behaviour with no obvious advantage to one or the other strategy. Since species coexist in the same fruits and competition among larvae has been shown (Jaeger et al., unpublished data), delay in oviposition may be costly because larvae will have late access to seeds and thus will suffer higher interspecific competition costs. However, there may be an advantage to lay eggs only on well-developed fruits and thus decrease the risk of laying eggs on unpollinated or poorly pollinated flowers. Such balance between the advantage of choosing already well-pollinated fruits to lay several eggs and the cost of increased larval competition may be a major stabilizing selective factor explaining coexistence of flower- and fruit-oviposition strategies.

If specialization on different flower parts or stages (receptacle use by larva, fruit- vs. flower-oviposition) can underlie sympatric speciation in Chiastocheta, it is unlikely to be involved in speciation of the four intermediate species for which oviposition dates largely overlap. The several Pleistocene ice ages have produced a series of changes in climate which have repeatedly altered the habitat and range of living organisms ( Hewitt, 1993, 1996). The four intermediate strategies may have evolved under such a situation, in geographically isolated T. europaeus populations, their actual presence in the same populations in the Alps being the result of recent secondary northern recolonization following the last ice age. This hypothesis is supported by two facts. First, although all the individuals from C. inermella sequenced were collected in our alpine populations, this species appears clearly paraphyletic, which may reflect a previous fragmentation of its geographical distribution. Second, the four intermediate species are not found in all parts of T. europaeus distribution: only two were found in Scandinavia ( Pellmyr, 1989; Johannesen & Loeschcke, 1996a), and three in Great Britain ( Collin, 1954). Unfortunately, no reports of Chiastocheta community in other parts of the globeflower range are available. Sampling in the Pyrenees, Italian Appenines and Balkan mountains would be particularly interesting as these three mountain ranges are at the southern limit of T. europaeus distribution and are well known post ice age refuges for several alpine organisms ( Hewitt, 1993, 1996; Konnert & Bergmann, 1995; Taberlet et al., 1998 ). The four intermediate species appear to be very similar in terms of molecular divergence, and only slightly differ in the age of the flowers visited. However, they are well differentiated by the morphology of the male genitalia, a character known to evolve rapidly in Dipteras ( True et al., 1997 ), presumably under sexual selection. Strong male genitalia differences observed between the otherwise morphologically very similar four intermediate species could be the result of reinforcement of reproductive isolation through sexual selection after secondary contact of these very closely related species.

In conclusion, the highly diverse Chiastocheta community found on T. europaeus is likely to be the result of a recent radiation during the Pleistocene climatic events. The specialization of the plant on seed-parasites as pollinators (and thus establishment of an obligate mutualism) was favoured by the mutualistic behaviour of the most ancestral species. Nonpollinating behaviour evolved from the pollinator. Unlike in other plant–insect systems where diversification is usually thought to be driven by cospeciation or host shifts, it is more likely that Chiastocheta speciation took place within the host plant, reproductive isolation between species being reinforced by niche partitioning for oviposition and/or sexual selection.

Acknowledgments

We thank J.-F. Desmet, A. Gaginskaïa and A. Hemborg for fly collection, and V. Plaisance, C. Kerdelhué, C. Dubois-Paganon and L. Gielly for help with the molecular work. V. Manceau, F. Pompanon, P. Taberlet and I. Till-Bottraud provided helpful comments on the manuscript. This work was partly supported by the French Ministry for Research (grant ACC-SV7 and PhD fellowship to N. J.).

Ancillary