Stability of Floral Specialization in Trollius

doi:10.1111/j.1420-9101.2009.01731.x

Stability of ﬂoral specialization in Trollius europaeus in contrasting ecological environments

S. IBANEZ, G. DUJARDIN & L. DESPRE´ S Laboratoire d’Ecologie Alpine (CNRS UMR 5553) and Station Alpine Joseph Fourier (UMS UJF-CNRS 2925), Universite´ J. Fourier, Grenoble, France

Keywords: Abstract floral morphology; Specialization of some plants on seed-eating pollinators is intriguing, especially male and female fitness; when co-pollinators exclusively feeding on nectar are also present. We most efficient pollinator; examined the stability of the morphological specialization of Trollius europaeus nursery pollination mutualism; (L.) globeflowers with respect to Chiastocheta (Pokorny) flies by artificially pollinator-mediated selection. opening the flowers. In the montane and subalpine environments studied, other visitors contributed 2% and 28% of all the visits, respectively, and visited open flowers nearly eight times more often than closed flowers, but in both environments their contribution to pollination did not compensate for Chiastocheta aversion against open phenotypes. Net seed set (female success) was slightly higher (+4%) and pollen export (male success) was much higher (+85%) for closed than open flowers. Selection in favour of the closed phenotype was even more intense in patches where open phenotypes were most common, precluding the evolution of open flowers in the study populations.

Pollination typically implies a cost for the plant Introduction through nectar or pollen lost to feeding pollinators. The Over the last decade, our view of plant-pollinator cost is particularly severe in seed-eating pollinator interactions has shifted from the long-standing concept mutualisms where the offspring of pollinators feed on of ‘pollination syndromes’ reflecting specialization as the potential plant offspring. Plant specialization involving common outcome of natural selection in those systems, such partners has long been intriguing, especially when to the widely accepted idea that generalization is wide- co-pollinators feeding only on nectar and not on devel- spread in pollination ecology, and that specialized plant- oping seeds are also present, as is often the case (Holland pollinators interactions might be the exception rather & Fleming, 2002; Thompson & Cunningham, 2002). If than the rule (Waser et al., 1996; Johnson & Steiner, seed-eating pollinators consume more seeds than fertilize 2000; Waser, 2006). If this is true, understanding the ovules, they should be viewed as parasites rather than ecological factors and the evolutionary mechanisms that mutualists (Holland & Fleming, 2002; Thompson & lead to specialization or generalization is of fundamental Cunningham, 2002; Thompson & Fernandez, 2006). importance (e.g. Dilley et al., 2000; Gomez & Zamora, Plants may then be expected to develop traits that allow 1999; Mayfield et al., 2001; Thompson, 2001; reviewed in pollination by generalist co-pollinators and minimize the Fenster et al., 2004 and Gomez & Zamora, 2006), as they costs imposed by specialized seed-eating pollinators. could help to explain the emergence of diverse plant- In the Trollius europaeus (L.)–Chiastocheta spp. (Pok- pollinator communities through adaptive co-diversifica- orny) mutualism, Chiastocheta flies contribute to almost tion (Hodges & Arnold, 1994; Dilley et al., 2000; Aigner, all pollination (Pellmyr, 1989; Jaeger & Despres, 1998), 2005). although other insects, mainly Diptera and Coleoptera, account for about 10% of plant visits (Jaeger & Despres, 1998). Among Trollius species and Chiastocheta flies, Correspondence: Se´bastien Ibanez, Laboratoire d’Ecologie Alpine (CNRS various levels of specialization can be found (Pellmyr, UMR 5553) and Station Alpine Joseph Fourier (UMS UJF-CNRS 2925), Universite´ J. Fourier, BP 53, F-38041 Grenoble, France. 1992). Some Trollius species are not visited at all by Tel.: 33 (0) 4 76 63 56 99; fax: 33 (0) 4 76 51 42 79; Chiastocheta flies, such as Trollius pumilus, Trollius ranun- e-mail: [email protected] culinus and Trollius laxus. All of these present an open flat

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS 1183 1184 S. IBANEZ ET AL.

corolla and are likely to be pollinated by a wide range of globeflowers in five populations (montane and subal- insects (Pellmyr, 1992). Molecular phylogenetic analysis pine environments), recorded visits from insects, eval- showed that an open flower is the ancestral state in the uated pollen export, and counted final seed production genus Trollius (Despres et al., 2003) and that all species in order to compare the relative fitness of closed visited by Chiastocheta flies throughout their range form a and open flower phenotypes. More specifically, we derived clade (altaicus, europaeus, ledebouri, asiaticus and addressed the following questions: (i) is floral mor- chinensis). The only morphological character clearly phology a key trait in globeflowers’ specialization, (ii) linked to the presence of Chiastocheta is the bowl shape do flowers always benefit from being specialized in of the corolla, which reaches its extreme in T. europaeus present-day populations, including contrasting ecologi- by forming a totally closed globe. This globular shape of cal conditions, and (iii) is pollinator selection of open T. europaeus is thus likely to be responsible for the vs. closed flower phenotypes frequency-dependent? exclusion of pollinators other than Chiastocheta flies. These small flies can navigate past the sepals to reach Materials and methods the inner part of the globe where larger insects are prevented from entering. We hypothesize that T. euro- Study system paeus and Chiastocheta flies coevolved reciprocal specialization through a number of traits, one being the closed The European globeflower T. europaeus L. (Ranuncula- flower shape coevolving with the flower shape prefer- ceae) is a hermaphroditic, homogamous, arctic-alpine ence of the insect. perennial species that grows in moist meadows Other potential pollinators are present in communi- (Despres et al., 2007). In the Alps, natural populations ties where T. europaeus occurs, so that it is not obvious range from 700 to 2500 m above sea level (a.s.l.). We why the plant would be specialized on Chiastocheta flies. did field experiments around the ‘Station Alpine The shape of T. europaeus corolla in natural populations Joseph Fourier’, col du Lautaret, France, in two is always mostly globular, but in some populations populations at the subalpine level (above tree-line, open flowers can be found at very low frequency Ruillas, 2025 m a.s.l.; Pre´ Gele´, 2374 m a.s.l.) and in (< 10%; S. Ibanez, personal observation), suggesting three populations located in the Chartreuse range at that this trait is potentially variable, and insects other the montane level (below tree-line, Sarcenas, 1101 m than Chiastocheta have been reported inside the globe a.s.l.; Fontanil 1020 m a.s.l.; Cherlieu, 980 m a.s.l.) by various authors (Pellmyr, 1989; Jaeger & Despres, from May to July in 2006 and 2007. Each flower 1998). Previous observations have shown that the contains about 30 multiovulate carpels dehiscing other visitors were more diversified at the montane 3–4 weeks after the end of flowering. Flowering is than at the subalpine level, where mostly Dipterans are synchronized within populations and typically lasts found (Jaeger & Despres, 1998). It is possible that the 2–3 weeks (Jaeger et al., 2000). Apart from T. europaeus presence of pollinators of different types would result which represented the most abundant flowering plant in disruptive selection on floral morphology (e.g. Medel during the study period in all the sites, main co-flow- et al., 2003). Given that T. europaeus populations are ering and insect-pollinated plants included Geranium exposed to different insect communities in their mon- sylvestris at the montane level, and Narcissus poeticus at tane and subalpine environments, they could be the subalpine level. Other species were present at low subject to variable selective pressures from male and ⁄ or densities (Alchemilla vulgaris, Dactylorhiza maculata and female functions, hence creating a geographical mosaic Polygonum bistorta at the montane level; Veratrum with spatial variability in strength and direction of album, Polygonum viviparum, Bartsia alpina and Primula selective pressures on floral morphology (Thompson, farinosa at the subalpine level). In the Alps, T. europaeus 2005). In the Trollius case where the specific pollinator is passively pollinated by six species of Chiastocheta flies is also a seed predator, one might expect that female (Anthomyiidae), but for this work, we will consider and male functions will respond differently to change these at the genus level because identification at the in pollinator communities, the female function favour- species level requires the dissection of genitalia and ing the open morph when alternative pollinators are cannot be performed in the field (Collin, 1954; available, while male function still favours the closed Michelsen, 1985). Chiastocheta larvae only feed on phenotype. Finally, local densities of flower phenotypes T. europaeus seeds; they are obligate associates of could affect selection regimes, because insects often globeflowers. Both male and female flies exclusively prefer common phenotypes to which they are accus- visit the globe-shaped flower (we observed only once a tomed and neglect rare phenotypes (Chittka et al., Chiastocheta fly on Ranunculus acris during the whole 1999). Open flowers might therefore be avoided when study) where they eat nectar and pollen, and where rare but regularly visited when locally abundant. To they mate (Pellmyr, 1989). Both activities contribute to determine the direction of selection in favour of the pollination (Despres, 2003). Females deposit one to specialized closed phenotype, in different community several eggs, on or between the carpels. Seed matura- contexts, we manipulated the floral morphology of tion and larval development last 3–4 weeks (Jaeger

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS Stable ﬂoral specialization in globeﬂowers 1185

et al., 2000). Each larva eats several seeds and falls to mal random effects, and the model used the ‘Poisson’ the soil to over-winter as a pupa. family (link function: log).

Experimental modification of floral shape Measurement of fitness components Globeflowers were artificially opened (‘O’ flowers) sim- Male fitness was measured using the movement of ilarly to the most naturally open flowers by placing a fluorescent dye particles as a surrogate for pollen export plastic ring inside the corolla, the diameter of which was (Price & Waser, 1982). Ten flowers of the same age chosen according to flower size (22 or 25 mm, see were chosen from the centre of a 5-m square patch; five Aigner, 2004; Johnson et al., 1995 for similar experi- of them were opened with a ring (‘O’) and their mental modification of floral morphology). The ring stamens dyed yellow using a paintbrush, five remained touches the sepals (and sometimes the filaments of the morphologically unmodified (‘C1’) and their stamens stamens) but neither the anthers nor the carpels. We did dyed red. A 100 m away, the same experimental not cut any sepals in order to avoid modification of the treatment was applied to a second patch, with the dye volatiles emitted by the flower. Control flowers were colours reversed, in order to check for a dye colour either untouched (‘C1’ flowers) or harboured smaller effect. Then 24 h later, 60–65 previously untouched plastic rings that did not modify the globular morphology target flowers were collected from each patch and (D = 18 mm, ‘C2’ flowers), in order to check for a ring scanned under a UV lamp to reveal fluorescence. effect. All the flowers we manipulated or surveyed were Preliminary experiments showed that during a period selected from vigorous plants. Flowers were of roughly of 24 h, the dye contained in a given flower was never the same age and height, with dehiscing stamens, found farther than 20 m away, precluding massive dye abundant pollen and well-developed calyxes. The exper- transfer between patches. We recorded the presence or iments were conducted when populations were at the absence of each dye colour on the stamens and the peak of flowering in the five populations described stigmas. This procedure was conducted in three mon- above. tane populations and two subalpine populations to test for environmental variability. In each population, we recorded the proportion of target flowers marked with Visits census red and ⁄ or yellow in each patch (2 patches · 2 col- Visits were recorded for observation periods ranging ours = 4 observations), so a total of 20 observations from 30 to 90 min in variable weather conditions were used in the statistical analysis. We compared the between 9 AM and 4 PM, corresponding to the peak of proportion of flowers that received dye from O vs. C1 pollinator activity. During each period, nine randomly flowers using ‘glmmPQL’ (binomial family, link func- chosen flowers were surveyed, three for each treat- tion: logit). The fixed effects included dye colour, floral ment O, C1 and C2. For each insect visit, we recorded if morphology (C1 as reference level, and O) and environ- the insect entered the globe or not, if it belonged to ment (subalpine as reference level, and montane); the genus Chiastocheta or not, and if not, to which interactions between morphology and dye colour, and order it belonged. For each period, we pooled the between morphology and environment were also number of visits received by the three flowers that included. The populations nested in environmental received the same treatment in order to reduce the effects had normal random effects. occurrence of zero values in the data set. The total Net seed production after pollination and predation, number of observations was then 135, divided in 45 a measure of female fitness, was estimated by weighing observation periods nested in five populations (three in the seeds of dehiscent fruits. Preliminary results montane and two in subalpine environments). There showed a high correlation between this measure and were from 7 to 11 observation periods per population. the number of viable seeds (seed number = 2.11 · total We recorded a total of 1177 insect visits during a total seed mass in mg, R2 = 0.96, n = 31). A total of 120 of 151 h of observation. All statistical analyses were flower buds were bagged to prevent pre-experimental done with the software R 2.6.0 (R Development Core pollination. At flower maturity (i.e. anther dehiscence), Team, 2007). We compared the number of visits per the bags were removed and treatments O, C1 and C2 given insect type using Generalized Linear Mixed were each applied to 40 flowers. Flowers remained Models, with the Penalized Quasi-Likelihood method untouched until carpel dehiscence (which was also the (function ‘glmmPQL’ in library MASS, Venables & end of larval development), and were then collected. Ripley, 2002). The fixed effects included the duration After hatching, the empty egg shell remains attached to of observation, floral morphology (C1 as reference level, the carpels, so that we were able to count egg number O and C2) and environment (subalpine as reference and to weigh seed mass. This experimental set-up was level, and montane), and excluded the interactions replicated in the five study sites. A large fraction of the between these factors. The observation periods and the 600 manipulated flowers were either lost or destroyed populations nested in environmental effects had nor- in the field, or were too damaged to be dissected in the

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS 1186 S. IBANEZ ET AL.

lab, so only 291 and 294 were included in the Results statistical analysis of egg number and net seed production per flower respectively (from 28 to 93 per population). We compared seed production and egg Visits from Chiastocheta and other insects deposition in flowers of different morphologies using Unsurprisingly, Chiastocheta and other insects’ visits ‘glmmPQL’ (‘Poisson’ family). The fixed effects number increased with the length of the observation included floral morphology (C1 as reference level, O and C2), environment (subalpine as reference level, Table 1 Generalized Linear Mixed Model of the number of visits to and montane) and carpel number; the interaction flowers (either from Chiastocheta or from other insects) against the between morphology and environment was also duration of observation, floral morphology and environment (fixed included. The populations nested in environmental effects), and observation period (random effects, not shown). effects had normal random effects. Chiastocheta entered Other insects entered

Manipulation of the local frequency of open Parameter Parameter phenotypes Effect d.f. value t-value value t-value We selected eight circular patches of T. europaeus,ina Observation 1;39 0.02 3.49*** 0.01 0.68 duration (min) single large population near the Col du Lautaret. Each O flowers 1;88 )1.46 )9.23**** 2.05 4.69**** patch had a diameter of 5 m, contained approximately C2 flowers 1;88 )0.14 )1.34 0.41 0.76 100–200 flowers, and was separated from others by a Environment 1;3 )1.32 )5.92*** 1.43 3.13* distance of at least 50 m. Fifty target flowers in the centre of each patch were bagged until maturity, and Interactions were not included. For floral morphology, C1 is the reference level and the comparison between O and C is also either treatment O,orC1 were each applied to 25 2 flowers. Additionally, we defined a low-frequency significant (not shown, P < 0.001). For environment, subalpine is treatment in which all the flowers surrounding target the reference level. Both numerator and denominator d.f. are given. The number of observations was 135, the number of populations flowers inside the patch were left untouched, so that was five and the number of groups nested in populations only 25 flowers were locally open; and a high-fre- (observation periods) is 45. *P £ 0.1; ***P £ 0.01; ****P £ 0.001. quency treatment where all the surrounding flowers were experimentally opened (O flowers), so that only 14 25 flowers were locally closed. Both treatments were 13 each applied to four randomly chosen patches. High- frequency patches were visited every 2 days during 12 10 days in order to open recently blooming flowers. As Flowers soon as the sepals of the target flowers had fallen, they Closed 10 were collected, and egg number and pollination rate Open were determined for each flower. Pollination rate was evaluated by randomly choosing five carpels and 8 counting the number of initiated seeds vs. the number of unfertilized ovules in each (15 229 ovules counted in total). We did not wait for carpel dehiscence to weigh 6 seed mass because seed initiation provides more imme- 32 32 diate information about the selective regimes acting on 4 13 Number of insect visits per hour floral morphology. Out of the 400 flowers manipulated, 234 were finally used in the analysis, the remaining 2 being either lost or destroyed in the field, or too 13 32 damaged to be dissected in the lab. We compared egg 32 13 deposition and pollination rate of flowers of different 0 morphologies in low and high-frequency patches using Montane SubalpineMontane Subalpine ‘glmmPQL’ (‘Poisson’ and ‘binomial’ families respec- Chiastocheta Other pollinators tively). The fixed effects included floral morphology (C1 as reference level, and O), frequency of open flowers in Fig. 1 Effect of environment (montane or subalpine) and morphology (closed or open) of observed flowers on the number of visits per the focal patch (low frequency as the reference level, flower by Chiastocheta and other insects per hour of observation. and high frequency) and carpel number; the interaction Error bars indicate standard errors of the residuals of the number of between morphology and the frequency of open flow- visits explained (Generalized Linear Mixed Model) by the duration ers was also included. The patches nested in the of observation (fixed effect) and observation period (random effect). frequency of open flowers treatment had normal The numbers above each bar correspond to the number of random effects. observation.

Table 2 Generalized Linear Mixed Model of the proportion of target observed, while in montane environment, Hymenoptera flowers marked with dye (either on stigmas or on stamens) against were mostly Apidae, and Coleoptera (including Odomera dye colour, floral morphology (C1 is the reference level) and genus) and Hemiptera were occasionally observed in environment where subalpine is the reference level (fixed effects), addition to the numerous visits by syrphids and other and population (random effects, not shown). Diptera. Non-Chiastocheta visitors showed a clear prefer- Stigmas Stamens ence for open flowers (with nearly eight times more visits in open than in closed flowers). There were no differ- Parameter Parameter ences between C1 and C2 flowers for visits of both Effect d.f. value t-value value t-value Chiastocheta and other insects, indicating that the pres- Dye colour 1, 11 )0.75 )2.32** )1.02 )2.94** ence of a ring inside the flowers had no impact in itself on Environment 1, 3 )1.86 )2.83* )2.39 )4.30** visitation rate. Floral morphology 1, 11 )2.09 )4.18*** )2.22 )4.72**** Dye colour · floral 1, 11 )0.19 )0.22 )0.96 )0.89 morphology Dye export Environment · floral 1, 11 0.48 0.43 0.86 0.75 Although only pollen transfer to receptive stigmas morphology contributes to pollination, we included in our analysis Interactions between morphology and dye colour, and between observations of dye deposited on stamens, because it morphology and environment were included. Both numerator and indicates that flies not only efficiently transfer dye to denominator d.f. are given. There were 10 patches (with two stigma but also come into close contact with stamens. observations per patch, depending on the origin of dye), and the Red dye was less exported than yellow dye (Table 2), but number of populations was five. *P £ 0.1; **P £ 0.05; ***P £ 0.01; this factor did not interact with the floral morphology: ****P £ 0.001. Chiastocheta flies are less attracted by reddish flowers, whether they are open or not. Dye export was more intense in subalpine populations, again without any period (Table 1). Chiastocheta showed a preference for regard to morphology. Open flowers exported closed over open flowers, with an estimated 77% dye toward many fewer flowers than closed ones, e.g. reduction of visits to open flowers. Chiastocheta were for the stigmas the observed decrease reached 85% three times more frequent in the subalpine than in the (Fig. 2). montane environment (Fig. 1). Other insects belonged to Diptera, Coleoptera, Hymenoptera and Hemiptera and Egg load and net seed production were rare compared with Chiastocheta, representing < 15% of the visits. They were more frequent in the Unsurprisingly, both egg load and net seed production montane than in the subalpine environment, represent- were positively correlated with carpel number (Table 3). ing 28.4% and 2.3% of the visits respectively. In both Mean egg load was lower in montane populations environments, Syrphidae represented most of non-Chi- compared with subalpine populations. Net seed produc- astocheta visitors, followed by Hymenoptera. In the tion was also lower in montane than in subalpine subalpine environment, only Diptera (mostly Syrphidae) environment but due to small sample size, the environ- and non-Apidae Hymenoptera (Tenthredinidae) were ment effect was not significant. Open flowers received

8 56 (a)0.25 (b) (c) 100 56 363 58

0.20 6 Flowers 80 Flowers Dye donor 58 Closed Closed Fig. 2 Effects of environment (montane or Closed subalpine) and morphology (closed or open) Open Open Open 0.15 of flowers [dye donor flowers in case of (a)] 60 on (a) the percentage of target flowers 4 40 marked by dye particles deposited on stig- 45 0.10 mas, (b) Chiastocheta egg number per flower 38 40

and (c) net seed production per ﬂower in mg. per flower Egg number In (b) and (c), error bars indicate standard 2 43 errors of the residuals of the response vari- 0.05 232 20 363 (mg) Net seed production per flower able explained (Generalized Linear Mixed Proportion with dye marked of target flowers

Model) by carpel number (ﬁxed effect) and 232 population (random effect). The numbers 0.00 0 0 above each bar correspond to the number of Montane Subalpine Montane Subalpine Montane Subalpine ﬂowers used in the analysis. Environment Environment Environment

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS 1188 S. IBANEZ ET AL.

Table 3 Generalized Linear Mixed Model of the egg number and Table 4 Generalized Linear Mixed Model of the pollination rate net seed production per flower against floral morphology, environ- and egg number per flower against floral morphology (C1 is the ment (subalpine is the reference level), and carpel number (fixed reference level), frequency of open flowers in the focal patch (low effects), and population (random effects, not shown). frequency is the reference level), carpel number (fixed effects), and patches (random effects, not shown). Egg number per flower Net seed production per flower Pollination rate per Egg number per flower flower Parameter Parameter Effect d.f. value t-value d.f. value t-value Parameter Parameter Effect d.f. value t-value value t-value O flowers 1, 281 )0.38 )3.28*** 1, 284 )0.17 )1.88*

C2 flowers 1, 281 )0.10 )0.92 1, 284 0.06 0.74 Floral morphology 1, 223 )0.27 )1.89* )0.17 )1.02 Environment 1, 3 )1.74 )3.22** 1, 3 )0.72 )1.49 Open type frequency 1, 6 0.11 0.75 0.32 1.49 Carpel number 1, 281 0.03 4.39**** 1, 284 0.04 7.64**** Carpel number 1, 223 )0.01 )1.41 0.03 5.99**** O flowers · 1, 281 )0.07 )0.2 1, 284 )0.05 )0.32 Open type frequency · 1, 233 )0.26 )1.31 )0.24 )1.09 environment floral morphology

C2 ﬂowers · 1, 281 0.37 1.31 1, 284 0.16 1 environment The interaction between morphology and frequency of open ﬂowers in the focal patch was included. Both numerator and denominator

For morphology, C1 is the reference level and the comparisons d.f. are given. The number of observations was 234, the number of between O and C2 are significant (not shown, P = 0.021 for egg groups (patches) was eight. *P £ 0.1; ****P £ 0.001. number and P = 0.008 for seed production). The interaction between morphology and environment was included. Both numerator and denominator d.f. are given. The number of observations was 291 (egg number) and 294 (seed production), the number of Egg load, pollination rate and the frequency of open groups (populations) was five. *P £ 0.1; **P £ 0.05; ***P £ 0.01; phenotypes ****P £ 0.001. There was a positive correlation between carpel number and egg load, and a negative nonsignificant correlation (a) (b) 0.7 between carpel number and pollination rate (Table 4). 62 Flowers 8 Flowers Mean egg load and pollination rate were lower in open Closed Closed 0.6 Open Open than in closed flowers (Fig. 3), although these differences 62 were not significant (Table 4). When open flowers were 59 59 locally more frequent, there was a slight nonsignificant 0.5 6 63 50 50 increase in mean egg load and pollination rate. The 63 difference between egg load and pollination rate in closed 0.4 and open flowers was larger when the open phenotype was frequent than when it was rare (Fig. 3), but the 4 interactions between the floral morphology and the local 0.3 frequency of open phenotypes for both egg load and pollination rate were nonsignificant (Table 4). Pollination rate per flower 0.2 Egg number per flower 2 Discussion 0.1 Specialization and selection pressures on male and female functions 0.0 0 Low High Low High In all populations studied, the experimental opening of Open flowers local density Open flowers local density T. europaeus flower resulted in a decrease in both male and female fitness. We cannot make a formal comparison Fig. 3 Effect of local frequency of open floral types (low or high) and morphology of flowers (closed or open) on (a) pollination rate between male and female fitness because we used different and (b) Chiastocheta egg number per flower. In (b), error bars indicate proxies to estimate them (the proportion of target flowers SE of the residuals of the number of eggs explained (Generalized marked with dye for male fitness, and seed weight for Linear Mixed Model) by carpel number (fixed effect) and patch female fitness). The pattern of dye and pollen movement (random effect). The numbers above each bar correspond to the may vary between systems. For example, the correlation number of flowers used in the analysis. between the total amount of dye and pollen found on individual stigmas lies between 0.4 and 0.8 in systems significantly fewer eggs ()30%) and produced fewer using hummingbirds and bumblebees (Waser & Price, seeds ()4%) than closed flowers (Fig. 3). The differences 1982; Thomson et al., 1986; Waser, 1988; Fenster et al., between C1 and C2 flowers were not significant. 1996; Rademaker et al., 1997; Adler & Irwin, 2006). Dye

cannot be used to estimate absolute male fitness because open flowers: the pollinator community context dramat- dye particles are smaller than pollen grains and more ically changed with the environment. Geographically particles are carried by insects than grains (Thomson et al., variable pollinator communities are a key feature of 1986); and because post-pollination mechanisms such as plant-pollinator interactions (Waser & Ollerton, 2006) incompatibility can lead to a discrepancy between dye and lead to selection mosaics and local co-evolution movement and gene flow estimated via paternity tests (Thompson, 2005). Despite a highly variable community (Campbell, 1991). However, the use of dye is appropriate context in the montane and subalpine populations, we for comparisons on the relative dispersal of pollen from found no interaction between environment and floral plants receiving different treatments because treatments morphology on either female or male fitness, which affect dye and pollen carryover in similar ways (Thomson suggests that alternative pollinators play a minor role in et al., 1986; Aigner, 2004). on going coevolution of the specialization of the T. euro- Differences in seed production between floral pheno- paeus–Chiastocheta spp. interaction at the present time. types were small (4%) compared with the radical change Nevertheless, alternative pollinators may influence the in pollen export (85%). Despite the use of different proxies morphological evolution of other Trollius species. For to estimate male and female fitness, the 20-fold difference example, the Asiatic species T. chinensis that is visited both is large enough to suggest that the main selective pressure by Chiastocheta and alternative pollinators displays an on floral morphology is male function. Several other intermediate floral morphology between a closed globe studies have shown that floral attractive traits evolve and a widely open corolla, while an open flat corolla is mainly through selection on male fitness (male-function found in T. laxus, T. pumilus and T. ranunculinus, three hypothesis, reviewed in Burd & Callahan 2000). The species outside the range of Chiastocheta. Furthermore, the stability of highly specialized plant-pollinators mutualisms evolution of floral traits also involves species other than is often questioned in a different way: do insects provide a potential pollinators, such as pollen feeders, florivores and sufficient pollination service to account for the persistence nectar robbers (Herrera et al., 2002). The closed morphol- of their host-plant populations? (Anstett et al., 1997; ogy of T. europaeus might contribute to excluding these, Ferriere et al., 2002; Morgan et al., 2005; Bloch et al., leading to an additional selection pressure favouring 2006). Our results show that in the Trollius–Chiastocheta closed phenotypes. Although some harmful insects might interaction, seed production is not limited by pollen indeed be excluded thanks to the closed morphology, we availability, so that competition between male gametes did occasionally observe Oedemera (Coleoptera), small for seed siring is likely to be high, especially because unidentified Coleoptera and Thysanoptera consuming globeflowers produce large amounts of pollen. More pollen, and Lepidoptera larvae consuming developing specifically, we show that the main effect of floral anthers and carpels inside the globes after having pierced a specialization on Chiastocheta flies is to increase pollen hole through the sepals. These insects were not more export rather than seed production: specialization of frequently observed in artificially opened flowers, and are T. europaeus on Chiastocheta flies results more from the unlikely to exert a strong selective pressure on Trollius competition between male gametes, rather than from floral morphology. increasing overall effective pollination and seed set at the population level. Cheptou & Schoen, 2007 emphasized the Negative frequency-dependent selection against importance of considering both female and male fitness in open flowers the study of the evolution of self-fertilization; here, we show that considering both components is also necessary When open flowers were most common, the egg load to understand the evolution of floral traits in response to was high in closed flowers and the pollination rate was pollinator selection pressures. Female and male fitness can low in open flowers (Fig. 3), although this was not respond differently to phenotypic changes (Aigner, 2004), significant (Table 4). The lack of statistical significance sometimes in opposite ways (Lankinen et al., 2006). In the was due to a small number of replicates (eight patches) globeflower populations documented here, both male and given the low magnitude of differences between treat- female fitness decrease when the flower shape is open, ments. The number of patches was small because it was ranging from a slight to a very sharp decrease. impossible to design more than eight distant patches in the population we studied. However, although nonsignificant, egg loads and pollination rates tended to be Geographical variation of selection pressures higher in closed flowers surrounded by open flowers Chiastocheta visits were more frequent and dye export more than in closed flowers surrounded by closed flowers, intense in subalpine populations than in montane popu- suggesting that when closed flowers were rare, Chiast- lations. Indeed, egg number per flower, which partly ocheta flies tended to congregate in them, thereby reflects fly population density (Johannesen & Loeschke, advantaging even more the closed phenotype (negative 1996), was higher in subalpine populations. In contrast, frequency-dependent selection). Our initial hypothesis other potential pollinators were much more abundant in was that locally abundant open flowers would help montane populations, especially Syrphid flies able to enter flies learn to deal with the open phenotype (positive

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS 1190 S. IBANEZ ET AL.

frequency-dependent selection). Patches where open evolved fully closed phenotypes, i.e. they did not flowers were locally abundant could appear by chance, specialize on Chiastocheta (Pellmyr, 1992). One may think and then would have been able to invade thanks to that the open flower phenotypes occasionally observed in positive frequency-dependent selection (similar mecha- natural populations of T. europaeus did not emerge nisms might be at play in the evolution of altruism, e.g. Le through mutation but only through environmental Galliard et al., 2003). We found the opposite: the chances effects. While we argue that environmental effects of locally abundant open flowers to invade are even cannot entirely account for flower polymorphism in further reduced as compared with rare open flowers. T. europaeus, our findings demonstrate that if open flower mutants happened to emerge in natural populations, they would be counter-selected under a wide range of Stability of the closed phenotype in present ecological conditions. populations In this work, we focused on the stability of the reciprocal Previous theoretical predictions suggested that the globe specialization on the globeflowers’ side. Specialization of a shape could enhance egg survival, increase larval com- pollinator to a particular flower may be unstable, depend- petition and, eventually, reduce seed consumption ing on the community context, but specific seed-eating because of contest competition between larvae, and pollinators that rely entirely on their host-plant for larval would have evolved as a partner sanction to prevent development require complex adaptations (e.g. Pellmyr & overexploitation (Ferdy et al., 2002). This original expla- Krenn, 2002). Host shifts have been documented in nation for the evolution of closed corolla is not supported specific seed-eating pollination mutualisms inside a given by field observations: the survival of eggs harboured by genus or between related genera (Molbo et al., 2003; artificially open vs. globular flowers does not differ Kawakita et al., 2004; Michaloud et al., 2005), but T. euro- (L. Despre´s, unpublished data). The empirical results paeus is the only member of its genus in Europe, so that presented here suggest instead that the evolution of a Chiastocheta flies do not have another potential partner closed flower shape is involved in the pollination readily available, and they cannot get out from the specialization of globeflowers on Chiastocheta. specialized interaction. Finally, the stability of the mor- The mechanisms that ensure the current stability of a phological specialization of T. europaeus on Chiastocheta trait and those responsible for the evolution of this trait flies does not imply evolutionary stability for the whole may well be different. In the case of coevolving mutu- interaction. Although the system studied here is consid- alisms, one main reason is that both partners evolve (de ered to be mutualistic, agonistic traits can coexist with Mazancourt et al., 2005): ancestral Chiastocheta flies were antagonistic traits. For example, Chiastocheta larvae induce no doubt adapted to the open floral morphology they a plant chemical defence (Gallet et al., 2007), which acts as dealt with. The ecological conditions necessary for the a potentially destabilizing factor in the specialized inter- persistence of Trollius specialization documented here action. (high Chiastocheta pollination efficiency, Chiastocheta preference for closed corolla) are not necessarily the same as Acknowledgments those required for specialization to evolve in the first place in a population of open flowers in a mixed, This work was supported by a grant from the French fluctuating, pollinator environment (Thompson & Pell- Ministe`re de l’Education Nationale, de l’Enseignement myr, 1992; Althoff et al., 2005). Nevertheless, the selec- Supe´rieur et de la Recherche. The authors thank Fanny tive pressures against open flowers discussed above can Dommanget and Thomas Martin for their help during explain the absence of high levels of variation in floral field work, and the authors are grateful to Laura closure in present populations of globeflowers. Closure of Galloway, Se´bastien Lavergne and Ire`ne Till-Bottraud the flower in T. europaeus is an effective way of filtering for their helpful comments. floral visitors and enhancing the efficiency of the most effective pollinator, as shown in other specialized systems References (Schemske & Bradshaw, 1999; Galen & Cuba, 2001; Aigner, 2005), and specialization is not merely the reflect Adler, L.S. & Irwin, R.E. 2006. Comparison of pollen transfer of depauperate pollinator communities (Johnson & dynamics by multiple floral visitors: experiments with pollen Steiner, 2000). Evolution of traits through natural and fluorescent dye. Ann. Bot. 97: 141–150. selection requires variable, heritable traits that are Aigner, P.A. 2004. Floral specialization without trade-offs: associated with differences in fitness. Corolla closure optimal corolla flare in contrasting pollination environments. Ecology 85: 2560–2569. varies considerably among species, demonstrating Trollius Aigner, P.A. 2005. Variation in pollination performance gradi- that this trait has a genetic basis, and fully open ents in a Dudleya species complex: can generalization promote phenotypes are found only in Trollius species out of the floral divergence? Funct. Ecol. 19: 681–689. range of Chiastocheta. Other species such as T. ledebouri Althoff, D.M., Segraves, K.A. & Pellmyr, O. 2005. Community and T. riederianus are visited by both Chiastocheta and context of an obligate mutualism: pollinator and florivore other pollinators, and, unlike T. europaeus, they have not effects on Yucca filamentosa. Ecology 86: 905–913.

Anstett, M.C., Hossaert-McKey, M. & McKey, D. 1997. Model- Herrera, C.M., Medrano, M., Rey, P.J., Sanchez-Lafuente, A.M., ling the persistence of small populations of strongly inter- Garcia, M.B., Guitian, J. & Manzaneda, A.J. 2002. Interaction dependent species: the case of figs and figs wasps. Conserv. Biol. of pollinators and herbivores on plant fitness suggests a 11: 204–213. pathway for correlated evolution of mutualism- and antago- Bloch, D., Werdenberg, N. & Erhardt, A. 2006. Pollination crisis nism-related traits. Proc. Natl. Acad. Sci. USA 99: 16823–16828. in the butterfly-pollinated wild carnation Dianthus carthusi- Hodges, S.A. & Arnold, M.L. 1994. Columbines: a geographically anorum? New Phytol. 169: 699–706. widespread species flock. Proc. Natl. Acad. Sci. USA 91: 5129– Burd, M. & Callahan, H.S. 2000. What does the male function 5132. hypothesis claim? J. Evol. Biol. 13: 735–742. Holland, J.N. & Fleming, T.H. 2002. Co-pollinators and special- Campbell, D.R. 1991. Comparing pollen dispersal and gene flow ization in the pollinating seed-consumer mutualism between in a natural population. Evolution 45: 1965–1968. Senita cacti and Senita moths. Oecologia 133: 534–540. Cheptou, P.O. & Schoen, D.J. 2007. Combining population Jaeger, N. & Despres, L. 1998. Obligate mutualism between genetics and demographical approaches in evolutionary stud- Trollius europaeus and its seed-parasite pollinators Chiastocheta ies of plant mating systems. Oikos 116: 271–279. flies in the Alps. C. R. Acad. Sci. Paris 321: 789–796. Chittka, L., Thompson, J.D. & Waser, N.M. 1999. Flower Jaeger, N., Till-Bottraud, I. & Despres, L. 2000. Evolutionary constancy, insect psychology, and plant evolution. Naturwis- conflict between Trollius europaeus and its seed-parasite poll- senschaften 86: 361–377. inators Chiastocheta flies. Evol. Ecol. Res. 2: 885–896. Collin, J.E. 1954. The genus Chiastocheta Pokorny (Diptera: Johannesen, J. & Loeschke, V. 1996. Distribution, abundance Anthomyiidae). Proc. R. Entomol. Soc. Lond. B 23: 95–102. and oviposition patterns of four coexisting Chiastocheta species Despres, L. 2003. Sex and pollen: the role of males in stabilising a (Diptera: Anthomyiidae). J. Anim. Ecol. 65: 567–576. plant-seed eater pollinating mutualism. Oecologia 135: 60– Johnson, S.D. & Steiner, K.E. 2000. Generalization versus 66. specialization in plant pollination systems. Trends Ecol. Evol. Despres, L., Gielly, L., Redoutet, B. & Taberlet, P. 2003. Using 15: 140–143. AFLP to resolve phylogenetic relationships in a morphologi- Johnson, S.G., Delph, L.F. & Elderkin, C.N. 1995. The effect of cally diversified plant species complex when nuclear and petal-size manipulation on pollen removal, seed set, and insect- chloroplast sequences fail to reveal variability. Mol. Phylogenet. visitorbehavior inCampanulaamericana. Oecologia 102: 174–179. Evol. 27: 185–196. Kawakita, A., Takimura, A., Terachi, T., Sota, T. & Kato, M. Despres, L., Ibanez, S., Hemborg, A.M. & Godelle, B. 2007. 2004. Cospeciation analysis of an obligate pollination mutu- Geographic and within-population variation in the globe- alism: have Glochidion trees (Euphorbiaceae) and pollinating flower-globeflower fly interaction: the costs and benefits of Epicephala moths (Gracillariidae) diversified in parallel? Evo- rearing pollinators’ larvae. Oecologia 151: 240–250. lution 58: 2201–2214. Dilley, J.D., Wilson, P. & Mesler, M.R. 2000. The radiation of Lankinen, A., Hellriegel, B. & Bernasconi, G. 2006. Sexual Calochortus: generalist flowers moving through a mosaic of conflict over floral receptivity. Evolution 60: 2454–2465. potential pollinators. Oikos 89: 209–222. Le Galliard, J.F., Ferriere, R. & Dieckmann, U. 2003. The Fenster, C.B., Dudash, M.R. & Hassler, C.L. 1996. Fluorescent adaptive dynamics of altruism in spatially heterogeneous dye particles are good pollen analogs for hummingbird- populations. Evolution 57: 1–17. pollinated Silene virginica (Caryophyllaceae). Can. J. Bot. 74: Mayfield, M.M., Waser, N.M. & Price, M.V. 2001. Exploring the 189–193. ‘most effective pollinator principle’ with complex flowers: Fenster, C.B., Armabruster, W.S., Wilson, P., Dudash, M.R. & bumblebees and Ipomopsis aggregata. Ann. Bot. 88: 591–596. Thompson, J.D. 2004. Pollination syndromes and floral de Mazancourt, C., Loreau, M. & Dieckmann, U. 2005. Under- specialization. Annu. Rev. Ecol. Evol. Syst. 35: 375–403. standing mutualism when there is adaptation to the partner. Ferdy, J.B., Despres, L. & Godelle, B. 2002. Evolution of J. Ecol. 93: 305–314. mutualism between globeflowers and their pollinating flies. Medel, R., Botto-Mahan, C. & Kalin-Arroyo, M. 2003. Pollinator- J. Theor. Biol. 217: 219–234. mediated selection on the nectar guide phenotype in the Ferriere, R., Bronstein, J.L., Rinaldi, S., Law, R. & Gauduchon, andean monkey flower, Mimulus luteus. Ecology 84: 1721–1732. M. 2002. Cheating and the evolutionary stability of mutual- Michaloud, G., Bossu-Dupriez, N., Chevolot, M. & Lasbleiz, C. isms. Proc. Biol. Sci. 269: 773–780. 2005. Pollen waste and unrelated traits in a fig-fig wasp Galen, C. & Cuba, J. 2001. Down the tube: pollinators, predators, symbiosis: a new behaviour suggesting a host shift. C. R. Biol. and the evolution of flower shape in the alpine skypilot, 328: 81–87. Polemonium viscosum. Evolution 55: 1963–1971. Michelsen, V. 1985. A revision of the Anthomyiidae (Diptera) Gallet, C., Ibanez, S., Zinger, L., Taravel, F.R., Trierweiler, M., described by J.W. Zetterstedt. Steenstrupia 11: 37–65. Jeacomine, I. & Despres, L. 2007. Plant chemical defense Molbo, D., Machado, C.A., Sevenster, J.G., Keller, L. & Herre, induced by a seed-eating pollinator mutualist. J. Chem. Ecol. E.A. 2003. Cryptic species of fig-pollinating wasps: implica- 33: 2078–2089. tions for the evolution of the fig-wasp mutualism, sex Gomez, J.M. & Zamora, R. 1999. Generalization vs. specializa- allocation, and precision of adaptation. Proc. Natl. Acad. Sci. tion in the pollination system of Hormathophylla spinosa USA 100: 5867–5872. (Cruciferae). Ecology 80: 796–805. Morgan, M.T., Wilson, W.G. & Knight, T.M. 2005. Plant Gomez, J.M. & Zamora, R. 2006. Ecological factors that promote population dynamics, pollinator foraging, and the selection the evolution of generalization in pollination systems. In: of self-fertilization. Am. Nat. 166: 169–183. Plant-Pollinator Interactions. From Specialization to Generalization Pellmyr, O. 1989. The cost of mutualisms: interaction between (N.M. Waser & J. Ollerton, eds), pp. 145–166. Chicago Trollius europaeus and its pollinating parasites. Oecologia 78: 53– University Press, Chicago. 59.

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS 1192 S. IBANEZ ET AL.

Pellmyr, O. 1992. The phylogeny of a mutualism: evolution and Thompson, J.N. & Pellmyr, O. 1992. Mutualism with pollinating coadaptation between Trollius and its seed parasitic pollinators. seed parasites amid co-pollinators – constraints on specializa- Biol. J. Linn. Soc. 47: 337–365. tion. Ecology 73: 1780–1791. Pellmyr, O. & Krenn, H.W. 2002. Origin of a complex key Thomson, J.D., Price, M.V., Waser, N.M. & Stratton, D.A. 1986. innovation in an obligate insect-plant mutualism. Proc. Natl. Comparative studies of pollen and fluorescent dye transport Acad. Sci. USA 99: 5498–5502. by bumble bees visiting Erythronium grandiflorum. Oecologia 69: Price, M. & Waser, N.M. 1982. A comparison of pollen and 561–566. fluorescent dye carry-over by natural pollinators of Ipomopsis Venables, W.N. & Ripley, B.D. 2002. Modern Applied Statistics with aggregata (Polemoniaceae). Ecology 63: 1168–1172. S, 4th edn. Springer, New York. R Development Core Team 2007. R: A Language and Environment Waser, N.M. 1988. Comparative pollen and dye for Statistical Computing. R Foundation for Statistical Comput- transfer by pollinators of Delphinium nelsonii. Funct. Ecol. 2: ing, Vienna. 41–48. Rademaker, M.C., deJong, T.J. & Klinkhamer, P.G. 1997. Pollen Waser, N.M. 2006. Specialization and generalization in plant- dynamics of bumblu-bee visitation on Echium vulgare. Funct. pollinator interactions: a historical perspective. In: Plant- Ecol. 11: 554–563. Pollinator Interactions. From Specialization to Generalization Schemske, D.W. & Bradshaw, H.D. Jr 1999. Pollinator prefer- (N.M. Waser & J. Ollerton, eds), pp. 3–17. Chicago University ence and the evolution of floral traits in monkeyflowers Press, Chicago. (Mimulus). Proc. Natl. Acad. Sci. USA 96: 11910–11915. Waser, N.M. & Ollerton, J. 2006. Plant-Pollinator Interactions. Thompson, J.D. 2001. How do visitation patterns vary among From Specialization to Generalization. Chicago University Press, pollinators in relation to floral display and floral design in a Chicago. generalist pollination system? Oecologia 126: 386–394. Waser, N.M. & Price, M.V. 1982. A comparison of pollen Thompson, J.N. 2005. The Geographic Mosaic of Coevolution. and fluorescent dye carry-over by natural pollinators Chicago University Press, Chicago. of Ipomopsis aggregata (Polemoniaceae). Ecology 63: 1168– Thompson, J.N. & Cunningham, B.M. 2002. Geographic struc- 1172. ture and dynamics of coevolutionary selection. Nature 417: Waser, N.M., Chittka, L., Price, M.V., Williams, N.M. & Ollerton, 735–738. J. 1996. Generalization in pollination systems, and why it Thompson, J.N. & Fernandez, C.C. 2006. Temporal dynamics of matters. Ecology 77: 1043–1060. antagonism and mutualism in a geographically variable plant- insect interaction. Ecology 87: 103–112. Received 10 September 2008; accepted 18 February 2009

JOURNAL COMPILATION ª 2009 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY J. EVOL. BIOL. 22 (2009) 1183–1192 NO CLAIM TO ORIGINAL FRENCH GOVERNMENT WORKS