Spatial variation in pollinator-mediated selection on phenology, floral display and spur length in the orchid Gymnadenia
conopsea
Elodie Chapurlat, Jon Agren and Nina Sletvold
Plant Ecology and Evolution, Department of Ecology and Genetics, Evolutionary Biology Centre, Uppsala University, Norbyv€agen 18D, SE-752 36 Uppsala, Sweden
Author for correspondence:
Elodie Chapurlat Tel: +46 18 471 2870
Email: elodie.chapurlat@ebc.uu.se Received: 24 April 2015
Accepted: 9 June 2015
New Phytologist (2015)208: 1264–1275 doi: 10.1111/nph.13555
Key words: correlational selection, diurnal vs nocturnal pollination, floral evolution, Gymnadenia conopsea (fragrant orchid), plantanimal interactions.
pollinator-mediated selection, semi- generalized pollination, spatial variation.
Summary
Spatial variation in plant –pollinator interactions may cause variation in pollinator-mediated selection on floral traits, but to establish this link conclusively experimental studies are needed.
We quantified pollinator-mediated selection on flowering phenology and morphology in four populations of the fragrant orchid Gymnadenia conopsea, and compared selection medi- ated by diurnal and nocturnal pollinators in two of the populations.
Variation in pollinator-mediated selection explained most of the among-population varia- tion in the strength of directional and correlational selection. Pollinators mediated correlational selection on pairs of display traits, and on one display trait and spur length, a trait affecting pollination efficiency. Only nocturnal pollinators selected for longer spurs, and mediated stronger selection on the number of flowers compared with diurnal pollinators in one popula- tion. The two types of pollinators caused correlational selection on different pairs of traits and selected for different combinations of spur length and number of flowers.
The results demonstrate that spatial variation in interactions with pollinators may result in differences in directional and correlational selection on floral traits in a plant with a semi-gen- eralized pollination system, and suggest that differences in the relative importance of diurnal and nocturnal pollinators can cause variation in selection.
Introduction
Identifying the causes of geographical variation in natural selec- tion is central for the understanding of adaptive differentiation and speciation (MacColl, 2011). In angiosperms, much of floral evolution is thought to have been driven by pollinator-mediated selection (Darwin, 1862; Faegri & Van der Pijl, 1966; Fenster et al., 2004) and there is increasing evidence from phylogenetic studies in favour of this view (Fenster et al., 2004; Harder &
Johnson, 2009; Van der Niet & Johnson, 2012). However, the connection between macroevolutionary patterns of diversity and microevolutionary processes remains poorly understood. Selec- tion on floral traits can be mediated by pollinators but also by antagonistic biotic agents (Gomez, 2003; Parachnowitsch &
Caruso, 2008; Burkhardt et al., 2012; Agren et al., 2013; Sletvold et al., 2015) and abiotic agents (Petit & Thompson, 1998; Galen, 2000; Totland, 2001). The quantification of pollinator-mediated selection in multiple populations is therefore an important step to understanding the role of pollinators in plant adaptive differ- entiation and patterns at the macroevolutionary level (Wilson &
Thomson, 1996; Herrera et al., 2006; Kay & Sargent, 2009).
In generalized pollination systems, pollinator-mediated selec- tion can be expected to vary spatially for several reasons. First, the
pollinator communities with which plant populations interact may differ in species composition, or in the relative abundance of different species (Gomez et al., 2009, 2010; Zhao & Huang, 2013). Second, the behaviour of a given pollinator can vary with plant community composition (Geber & Moeller, 2006; Hersch
& Roy, 2007), and pollinator-mediated selection can be affected by plant population size (Weber & Kolb, 2013) and vegetation height (Sletvold et al., 2013). Numerous studies have linked floral and pollinator traits across multiple populations (Anderson &
Johnson, 2008; Boberg et al., 2014; Newman et al., 2014), and in the generalist herb Erysimum mediohispanicum, among-popu- lation variation in selection on floral shape has been linked to variation in the composition of the local pollinator communities through path analysis (Gomez et al., 2008, 2009). By contrast, very few studies have quantified current pollinator-mediated selection in multiple populations, and the evidence for the pres- ence of a geographical mosaic of pollinator-mediated selection is predominantly correlative (but see Weber & Kolb, 2013; Agren et al., 2013).
Pollinators may exert selection not only on single floral traits but also on combinations of traits, that is, correlational selection.
In particular, correlational selection is expected when one trait
influences the rate of pollinator visitation and a second trait
influences pollinator effectiveness per visit, because pollination success will depend on the product of these two components (Sletvold & Agren, 2011; Campbell et al., 2014). In the orchid Dactylorhiza lapponica, phenotypic manipulations documented nonadditive effects on reproductive success of floral display and spur length, a trait affecting pollination efficiency (Sletvold &
Agren, 2011). Correlational selection may also be expected for pairs of display traits if they influence pollinator visitation nonad- ditively. For example, in Senecio jacobaea, the manipulation of inflorescence size and corolla size revealed nonadditive effects on fruit set (Andersson, 1996). Among the studies that detected cor- relational selection on floral traits, all found selection on combi- nations of one display trait and one efficiency trait (O’Connell &
Johnston, 1998; Benitez-Vieyra et al., 2006; Cuartas-Dom ınguez
& Medel, 2010; Reynolds et al., 2010; Bartkowska & Johnston, 2012) and some on pairs of display traits (O’Connell &
Johnston, 1998; Reynolds et al., 2010; Bartkowska & Johnston, 2012). These studies show that such combinations of floral traits can indeed be the targets of selection. However, only Bartkowska
& Johnston (2012) experimentally tested whether pollinators contribute to correlational selection.
In this study, we experimentally quantified pollinator- mediated selection on flowering phenology and flower morphol- ogy in four populations of the fragrant orchid Gymnadenia conopsea in southern Sweden. Gymnadenia conopsea has a semi- generalized pollination system: it is mostly visited by lepi- dopteran species (Claessens & Kleynen, 2011), which include both diurnal and nocturnal species belonging to similar func- tional groups (sensu Fenster et al., 2004). Selective pollinator exclusion experiments documented that nocturnal pollinators were more important than diurnal pollinators for seed produc- tion in two German populations (Meyer et al., 2007), whereas the contrary was true in two populations in central Norway (Sletvold et al., 2012b). Studies of two Norwegian G. conopsea populations demonstrated that variation in pollinator-mediated selection explained much of spatiotemporal variation in selection on morphological traits involved in pollinator attraction and pol- lination efficiency (Sletvold & Agren, 2010, 2014), and that dif- ferences in the contributions of diurnal and nocturnal pollinators
could explain part of the variation in pollinator-mediated selec- tion (Sletvold et al., 2012b). Here, we examined spatial variation in pollinator-mediated selection in southern Sweden, where nights are longer and night-flying pollinators can be expected to be more important than in the Norwegian study area. We specifi- cally asked: whether selection on flowering start, three display traits (plant height, number of flowers and corolla size) and one pollination efficiency trait (spur length) varies among populations;
whether such variation can be explained by variation in pollinator- mediated selection; and whether diurnal and nocturnal pollinators exert different patterns of selection, which would imply that varia- tion in the importance of the two pollinator categories may con- tribute to spatial variation in pollinator-mediated selection.
Materials and Methods Study species and populations
Gymnadenia conopsea (L.) s.l. is a terrestrial orchid distributed across Eurasia (Hulten & Fries, 1986). It occurs on calcareous soils in grazed meadows and at the margins of marshes and fens.
The species is a tuberous, nonclonal and long-lived perennial (Øien & Moen, 2002). The fragrant flowers vary in colour from pale to bright pink, and rarely white. The plants produce a single inflorescence of c. 10–100 flowers. Flowers open sequentially from the bottom to the top of the inflorescence. Individual flow- ers remain open for up to a week while individual plants may flower for a month. A long, narrow spur contains nectar that is produced throughout anthesis (Stpiczynska & Matusiewicz, 2001). Each flower contains two pollinaria which are situated above the spur entrance. Plants are self-compatible, but depend on pollinators for successful fruit set (Sletvold et al., 2012a).
The four study populations, Mel€osa, Langl€ot, Graborg and Kvinneby, are located on the island of € Oland in southern Sweden and separated by 13–40 km. The populations are found in dry/
moist grasslands grazed by cattle and horses and the number of flowering G. conopsea individuals varies tenfold among popula- tions (Table 1). The populations differ in flowering phenology, with three early- and one late-flowering population (peak in
Table 1 Population characteristics, pollinators observed on Gymnadenia conopsea, and the number of nights on which observations of nocturnal pollina- tors were conducted
Population Size Habitat Flowering phenology Diurnal visitors Nocturnal visitors
Kvinneby (56°330N, 16°370E) 400 Grassland grazed by cattle, dry, short grass, many bushes
Early (early June) Aglais urticae Pieris brassicae
Deiliphila porcellus (4 nights)
Mel€osa (56°510N, 16°500E) 2500 Grassland grazed by cattle, dry, sparse and short grass
Early (mid-June) Aglais urticae Zygaena minos
Deiliphila porcellus Autographa gamma Agrotis exclamationis (2 nights)
Langl€ot (56°450N, 16°450E) 1000 Grassland grazed by cattle, dense and tall grass, many bushes
Early (mid-June) Aglais urticae Empis sp.
Siona lineata
Deiliphila porcellus Autographa gamma Cucullia umbratica Hyles gallii (2 nights) Graborg (56°400N, 16°370E) 300 Grassland grazed by cattle
and horses, surrounded by woodland
Late (early July) Zygaena filipendula Hyles gallii
No data
mid-June and early July, respectively). In all populations, the main pollinators were lepidopteran species, including butterflies, moths and hawkmoths, but a dipteran (Empis sp.) was also observed pollinating G. conopsea in the L angl€ot population (Table 1). Some pollinators were observed in all three early-flow- ering populations (Aglais urticae during the day, and Deiliphila porcellus during the night), while observations of additional visi- tors were restricted to single populations. Diurnal pollinators observed in the late-flowering population differed from those observed in early-flowering populations (Table 1). Day-time observations were made throughout the flowering season while recording plant traits and conducting experimental manipula- tions, whereas night-time observations were available only for the three early-flowering populations and were based on video recordings (two video cameras) throughout two to four nights during the peak of flowering as well as one evening of direct observation between 21:00 h and 24:00 h in each of the three populations (the number of observation nights is indicated in Table 1). Pollinator observations in the two following years (2013 and 2014) indicate that the main pollinators and their sea- sonal timing in the different populations are consistent across years (E. Chapurlat, pers. obs.).
Field experiments
To quantify pollinator-mediated selection on flowering phenol- ogy, floral display and spur length, we conducted experimental hand-pollination in all four study populations during summer 2012. Before flowering, 240 plants with flower buds were ran- domly chosen and individually tagged in each population, except at Kvinneby, where 300 plants were tagged because of a higher risk of grazing damage. Half of the plants were randomly submit- ted to each of two treatments: natural pollination (control (C)) or supplemental hand-pollination (HP). At Kvinneby, the extra 60 plants were assigned to the C treatment. We visited the popu- lations at least twice a week and supplemental hand-pollinations were conducted as flowers opened. Hand-pollinated plants were exposed to pollinators throughout the flowering period, and thus received a mixture of pollen transferred by insects and pollen added by hand. We collected pollinia with cocktail sticks from other plants in the HP treatment or from untagged individuals located more than 5 m away from the recipient plant. Each flower on an inflorescence was pollinated at least twice by rub- bing one or two pollinia across each stigma, saturating the surface with pollen.
To separate the effects of diurnal and nocturnal pollinators, we added two pollination treatments in two of the study popula- tions. In the L angl€ot and Mel€osa populations, 240 additional plants were tagged. We randomly assigned 120 plants to each of two treatments: diurnal pollination (D) or nocturnal pollination (N). The peak of diurnal visits occurred between 10:00 h and 14:00 h, whereas the peak of nocturnal activity occurred between 22:00 h and midnight (E. Chapurlat, pers. obs.). Plants in the D treatment were caged during the night (18:00–06:00 h), receiving only diurnal visits, and plants in the N treatment were caged dur- ing the day (06:00–18:00 h), receiving only nocturnal visits.
Caging continued until all flowers had wilted. The cages were made of a white mosquito net wrapped around a metallic wire cylinder of c. 10 cm diameter. A previous experiment in a Norwe- gian population of G. conopsea showed that caging per se does not affect female reproductive success (Sletvold et al., 2012a).
Measured traits
We visited the populations every 2–3 d and flowering start was recorded for each individual as the estimated day on which the first flower opened. At the onset of flowering, we recorded the height of all plants included in the experiment (distance from ground to topmost flower; Fig. 1a). On one of the three lowermost flowers on each individual, we measured spur length (distance from the corolla to the spur tip; Fig. 1b) and maximum corolla width and height to the nearest 0.1 mm with digital callipers (Fig. 1c).
Corolla size was quantified as the product of corolla width and height. The number of flowers was counted after fruiting.
To quantify female reproductive success, we recorded the number of fruits (Fig. 1d) and, when possible, harvested three mature nondehisced capsules spread across the inflorescence to determine mean fruit mass for each plant. Fruits were dried at room temperature for at least a month, and their dry mass was determined to the nearest 0.01 mg. Fruit mass is positively related to number of seeds with embryos in G. conopsea (Sletvold
PH
CH CW SL
(a) (b)
(c)
(d)
Fig. 1 Illustration of the phenotypic traits measured in this study on Gymnadenia conopsea inflorescences (a) and flowers (b, c) and during fruiting (d). PH, plant height; CH, corolla height; CW, corolla width; SL, spur length. Corolla size was estimated as the product of CH and CW.
Bars, 1 cm.
& Agren, 2010). The capsules of some plants had dehisced before fruit collection (67, 81 and 74 plants at Kvinneby, Mel€osa and L angl€ot, respectively). For these individuals, we collected open capsules and measured fruit width and length to the nearest 0.01 mm using digital callipers. Fruit volume was approximated by the volume of a cylinder. To relate fruit volume to fruit mass, we measured the volume of nondehisced fruits from 50 HP plants and 50 C plants from each of the Mel€osa and Langl€ot pop- ulations. The relationship between fruit mass and fruit volume did not differ among populations or treatments (two- and three- way interactions; P > 0.21). We pooled data from all 200 plants to determine the relation between fruit mass and fruit volume:
fruit mass (mg) = 0.136 9 fruit volume (mm
3) + 1.65; R
2= 0.87.
We used this equation to estimate fruit mass for the individuals for which we had sampled dehisced capsules. For each plant, we estimated female fitness as the product of number of fruits and mean fruit mass. For each population, we quantified pollen limi- tation as 1 (mean female fitness of C plants divided by mean female fitness of HP plants). We bootstrapped a 95% confidence interval by randomly drawing plants with replacement in the C and HP treatments, respectively, calculating pollen limitation for each permuted sample (n = 2000 replicates), and extracting the 2.5th and 97.5th percentiles from the obtained distribution of pollen limitation estimates.
In the Kvinneby population, 135 plants were lost as a result of grazing by cattle and 26 plants were not retrieved (the probability of grazing or retrieval did not differ between the two pollination treatments; v
2= 0.67, df = 1, P = 0.41 and v
2= 1.23, df = 1, P = 0.27, respectively). Grazing probability was not related to any of the measured floral traits (analysed with a generalized lin- ear model including the floral traits as independent variables and with a binomial error distribution; all P > 0.14).
Statistical analyses
The effects of supplemental hand-pollination (control vs supple- mental hand-pollination) and population on plant traits and reproductive performance were examined with a two-way ANOVA, including the C and HP treatments in all four popula- tions. In a second two-way ANOVA, we examined the effect of diurnal vs nocturnal pollination and population on plant traits and reproductive performance, including the D and N treatments in the Mel€osa and Langl€ot populations. The numbers of flowers and fruits were analysed with generalized linear models with a quasi-Poisson error distribution, because of overdispersion.
When necessary, data were transformed before the analyses to improve normality of residuals (indicated in Supporting Infor- mation Tables S1, S2).
Selection was estimated following Lande & Arnold (1983), using multiple regression analyses with relative fitness (individual fitness divided by mean fitness) as the response variable and stan- dardized trait values (with a mean of 0 and a variance of 1) as explanatory variables. Relative fitness and standardized trait val- ues were calculated separately for each treatment and population.
We estimated directional selection gradients b
ifrom multiple regression models including only linear terms, and separately for
each treatment and population. For all populations except Kvinneby, we quantified nonlinear (c
ii) and correlational (c
ij) selection gradients from the quadratic and cross-product terms of the full regression models. The reported c
iiare obtained by dou- bling the coefficients extracted from the regression model to rep- resent quadratic selection gradients (Stinchcombe et al., 2008).
Too few plants were sampled for the Kvinneby population to allow the analysis of full models. To quantify multicollinearity in the regression models, we computed variance inflation factors (VIFs) for the linear terms. All VIFs were < 2.2, indicating no problem of multicollinearity (Quinn & Keough, 2002).
To determine whether net directional selection varied among populations, we analysed data from the control treatment with ANCOVA. The model included relative fitness as the response variable and the five standardized traits (flowering start, plant height, number of flowers, corolla size and spur length), popula- tion and the trait 9 population interactions as explanatory vari- ables. In a second ANCOVA model including both the C and HP treatments in the four populations, we determined whether pollinator-mediated directional selection varied among popula- tions. The model included relative fitness as the response variable and the five standardized traits (flowering start, plant height, num- ber of flowers, corolla size and spur length), pollination treatment (C vs HP), population, and trait 9 pollination treatment, trait 9 population, and trait 9 pollination treatment 9 popula- tion interactions as explanatory variables. Because some three-way interactions were close to significant, we further tested the effect of pollination treatment (C vs HP) on linear selection gradients separately for each population to determine whether there was sig- nificant pollinator-mediated selection. To test for the effect of pol- lination treatment (C vs HP) on quadratic and correlational selection gradients, we used an ANCOVA on the full models sep- arately for each population, because our sample was too limited to include population as a factor in the analysis. To quantify pollina- tor-mediated selection, we subtracted for each trait the estimated selection gradient for hand-pollinated plants (b
HPor c
HP) from the estimated gradient for open-pollinated control plants (b
Cor c
C), such that Db
poll= b
Cb
HPand Dc
poll= c
Cc
HP(see Sletvold & Agren, 2010; Bartkowska & Johnston, 2012). This approach includes effects of trait variation on both the quantity and the quality of pollen deposited, acknowledging that pollina- tor-mediated selection is the integrated result of effects on both components of pollination success. To estimate the extent to which variation in pollinator-mediated selection explained varia- tion in net selection, we regressed net selection on pollinator-me- diated selection separately for each trait.
To determine whether linear selection gradients differed
between the D and N treatments and between populations
(L angl€ot and Mel€osa), we used a similar ANCOVA model as
described in the previous paragraph with pollination treatment
now being either D or N. Because a significant three-way interac-
tion was detected, we further tested the effect of pollination treat-
ment (D vs N) on linear selection gradients separately for each
population. To test the effect of pollination treatment on
quadratic and correlational selection gradients, we used an
ANCOVA on the full models separately for each population.
All analyses were conducted with R 3.1.3 (R Core Team, 2015). Type III sum of squares tests were used for all analyses of linear models (ANOVA function of the
CARpackage (Fox &
Weisberg, 2011)).
Results
Floral traits and phenotypic correlations
Flowering onset, floral display and spur length varied among the four populations (Tables 2, S1; Fig. S1). The Gr aborg population started to flower markedly later, and consisted of plants that pro- duced more flowers with shorter spurs than did plants in other populations. In the Mel€osa population, plants were shorter with
larger flowers compared with other populations. Otherwise, the variation in trait means was limited (Table 2; Fig. S1). In all pop- ulations, the range in trait expression was similar (Table 2), and floral traits were moderately positively correlated, except flower- ing start, which tended to be negatively correlated to the other traits (Table S3).
Pollen limitation and reproductive success
All populations were pollen limited, and pollen limitation of female fitness and its components was similar in the four study populations and among diurnally and nocturnally pollinated plants. Pollen limitation of female fitness ranged from 0.22 to 0.30 (reported in Fig. 2; Table S4), and did not vary among
Table 2 Plant traits and reproductive performance (mean SD) for plants receiving supplemental hand-pollination (HP), open-pollinated control plants (C), plants with diurnal pollination (D) and plants with nocturnal pollination (N) in the Gymnadenia conopsea populations at Mel€osa, Langl€ot, Graborg and Kvinneby (sample sizes n are given in that order) in 2012
Trait, by population
HP C D N
n= 54/104/95/109 n= 83/116/105/114 n= /112/112/ n= /117/112/
Flowering start (day of the year)
Kvinneby 167 3.4 167 3.3 NA NA
Mel€osa 169 4.1 169 4.6 171 2.8 171 2.6
Langl€ot 172 4.3 172 4.7 173 3.7 173 3.8
Graborg 188 2.6 188 2.3 NA NA
Plant height (cm)
Kvinneby 24.0 5.5 24.4 5.1 NA NA
Mel€osa 20.8 5.2 20.8 5.1 19.8 4.7 20.7 4.9
Langl€ot 26.6 6.1 26.7 6.2 26.2 5.0 26.9 5.9
Graborg 28.6 7.4 28.2 7.3 NA NA
Number of flowers
Kvinneby 30.5 10.1 32.6 10.2 NA NA
Mel€osa 34.9 10.9 34.3 12.2 33.5 11.2 34.0 12.1
Langl€ot 32.7 10.0 31.2 10.6 31.3 10.6 32.2 13.6
Graborg 43.6 12.9 41.2 12.8 NA NA
Corolla size (mm2)
Kvinneby 104.0 26.5 101.3 28.2 NA NA
Mel€osa 109.6 25.7 113.2 26.2 103.4 26.4 105.3 23.9
Langl€ot 111.2 27.0 100.8 26.2 99.4 23.4 98.2 23.2
Graborg 104.3 20.3 103.5 19.0 NA NA
Spur length (mm)
Kvinneby 14.8 1.8 15.1 1.8 NA NA
Mel€osa 15.2 1.8 14.8 1.8 15.0 1.7 15.0 1.6
Langl€ot 15.3 2.1 14.7 1.7 14.6 1.9 14.9 2.0
Graborg 13.7 1.3 14.1 1.3 NA NA
Number of fruits
Kvinneby 30.1 10.2 25.5 10.2 NA NA
Mel€osa 33.2 11.0 23.5 13.5 22.5 11.5 25.1 12.6
Langl€ot 31.0 9.9 24.1 12.5 24.6 10.4 25.3 12.6
Graborg 42.0 12.7 33.6 13.5 NA NA
Fruit mass (mg)
Kvinneby 10.9 3.5 9.8 3.6 NA NA
Mel€osa 12.4 3.9 11.4 4.1 10.3 3.3 10.6 3.5
Langl€ot 11.3 3.9 10.2 3.7 9.6 3.1 10.6 3.4
Graborg 7.3 2.2 7.1 2.1 NA NA
Female fitness
Kvinneby 338 188 264 174 NA NA
Mel€osa 420 210 293 263 248 186 283 206
Langl€ot 360 193 262 179 242 142 282 201
Graborg 320 165 252 146 NA NA
NA, not applicable.
populations, as indicated by the absence of a significant popula- tion 9 pollination treatment interaction for female fitness (Table S1). Fruit production was also pollen limited to a similar extent in all populations, whereas fruit mass was only marginally increased by the supplemental hand-pollination (Tables 2, S1).
In both the Mel€osa and Langl€ot populations, number of fruits and female fitness did not differ between the D and N treat- ments, while fruit mass was significantly higher among plants exposed to nocturnal pollinators than among plants exposed to diurnal pollinators (Tables 2, S2). Plants exposed to only noctur- nal or diurnal pollinators had similar female fitness to open-polli- nated plants (Table 2).
Among-population variation in phenotypic selection Among open-pollinated plants, flowering start, floral display (plant height, number of flowers and corolla size), and spur length were subject to selection in one or more populations, and selection on flower production and corolla size varied among populations (Fig. 2a; Table S4). The production of more flowers was favoured in all populations, but the strength of selection var- ied (number of flowers 9 population interaction, P = 0.0038),
with a stronger gradient in the Mel€osa population (b
C= 0.56) compared with the three other populations (b
C= 0.30–0.39).
There was selection for larger corollas in the Mel€osa and Langl€ot populations, whereas selection on corolla size was not statistically significant in the other two populations (corolla size 9 popula- tion interaction, P = 0.026). Earlier flowering was favoured in the Kvinneby population, whereas no significant selection on flower- ing start was recorded in the other populations (flowering start 9 population interaction, P = 0.088). Selection for taller plants was similar in all populations, although selection gradients only approached statistical significance at Mel€osa and Kvinneby (plant height 9 population interaction, P = 0.67). Finally, there was similar and significant selection for longer spurs in the Mel€osa and Kvinneby populations (b
C= 0.14 and b
C= 0.13, respectively), whereas no statistically significant selection on spur length was detected in the two other populations (spur length 9 population interaction, P = 0.25).
Significant correlational selection was detected among open- pollinated plants (Table S5), but no significant quadratic gradi- ents (Table S4). At Mel€osa, tall plants with many flowers were disproportionally favoured, as well as plants with a combination of long spurs and many flowers (Fig. 3a,b; Table S5). At L angl€ot, there was positive correlational selection on flowering start and corolla size and negative correlational selection on flowering start and spur length (Fig. 3c,d; Table S5).
Contribution of pollinator-mediated selection to variation in phenotypic selection
Pollinators contributed significantly to directional selection on number of flowers, corolla size and spur length (Fig. 2b), and variation in pollinator-mediated selection could explain much of among-population variation in net phenotypic selection on these three traits (Figs 2, 4; Table S4). At Mel€osa, the significant pollinator-mediated selection for more flowers (Db
poll= 0.23) accounted for 41% of the selection observed among open-polli- nated plants (Fig. 2; Table S4). At Kvinneby, pollinators selected for smaller corollas (Db
poll= 0.20), accounting for all of the observed net selection. Although not statistically significant, pollinator-mediated selection explained 65% of net selection for early start of flowering at Kvinneby (Db
poll= 0.098; Fig. 2;
Table S4). Pollinator-mediated selection accounted for all net selection on spur length (Mel€osa, Db
poll= 0.18; Kvin- neby, Db
poll= 0.15; Fig. 2; Table S4). Variation in pollinator- mediated selection explained 99%, 94%, 94% and 64% of the among-population variation in net selection on spur length, number of flowers, corolla size and flowering start, respectively, whereas it explained only 17% of the variation in net selection on plant height (Fig. 4). Pollinator-mediated selection tended to vary among populations (trait 9 pollination treatment 9 popula- tion interactions in the ANCOVA including C and HP plants) for number of flowers (P = 0.093), corolla size (P = 0.073) and spur length (P = 0.14).
Pollinators contributed significantly to both quadratic and cor- relational selection, but the favoured trait combinations varied among populations (Fig. 3; Tables S4, S5). Although no
–0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Flowering date
Plant height Number of flowers
Corolla size Spur length Net selecon gradient (βC) ± SE
Kvinneby (PL = 0.22 [0.045 ; 0.36]) Melösa (PL = 0.30 [0.17 ; 0.42]) Långlöt (PL = 0.28 [0.15 ; 0.39]) Gråborg (PL = 0.21 [0.090 ; 0.31])
*
* *
*
* *
*
* * * *
(*)
* ns
*
(*)
(*)
(*) ns
–0.3 –0.2 –0.1 0 0.1 0.2 0.3
Flowering date
Plant height Number of flowers
Corolla size Spur length Pollinator-mediated selecon (Δβpoll)
ns
ns ns (*)
(*)
*
*
(*)
*
(a)
(b)
Fig. 2 Phenotypic linear selection gradients among open-pollinated plants (a) and pollinator-mediated selection gradients (b) on flowering start, plant height, number of flowers, corolla size and spur length in the four Gymnadenia conopsea populations in 2012. The different populations are represented by different bar shading, and PL is the pollen limitation of female fitness estimated as 1 (mean female fitness of open-pollinated control plants/mean female fitness of hand-pollinated plants), given with its 95% bootstrapping confidence interval between square brackets.
Symbols above individual bars indicate the level of significance of the gradient. Symbols above the lines spanning several gradients show whether ANCOVA indicated significant variation among populations in selection gradients (significant trait9 population interaction): *, P < 0.05;
(*), P < 0.10; ns, P > 0.10.
quadratic gradient was significant among open-pollinated plants, there was a significant difference between pollination treatments for quadratic selection on corolla size in the L angl€ot population, indicating that pollinators contributed to this gradient (Dc
poll= 0.24; Table S4). At Mel€osa, pollinator-mediated selec- tion explained all of the correlational selection on number of flowers and spur length (Dc
poll= 0.22) and 82% of the correla- tional selection on number of flowers and plant height (Dc
poll= 0.14), even though the latter contribution was not sta- tistically significant (Fig. 3a,b; Table S5). At L angl€ot, estimates of correlational selection differed significantly between the control and the hand-pollination treatment for combinations between flowering start and three other traits (number of flowers, corolla size, and spur length; Table S5), indicating that pollinators influ- enced patterns of correlational selection (Fig. 3c,d).
Comparison of selection generated by diurnal and nocturnal pollinators
Selection for more flowers mediated by nocturnal pollinators was twice as strong as that mediated by diurnal pollinators at L angl€ot,
but directional selection on floral traits did not otherwise differ between plants subject to diurnal and nocturnal pollinators, respectively (Fig. 5; Table S6). In both treatments and at both sites, there was selection for more flowers and larger corollas. At Mel€osa, there was in addition selection for taller plants in both treatments, and significant selection for longer spurs mediated by nocturnal pollinators. At L angl€ot, we documented selection for taller plants mediated by diurnal pollinators.
Correlational selection on number of flowers and spur length differed between day- and night-pollinated plants in the L angl€ot population, but in no other case was a statistically significant dif- ference in quadratic or correlational selection gradients detected between the two treatments (Fig. S2; Tables S6, S7). At L angl€ot, although not statistically significant in either pollination treat- ment, correlational selection on number of flowers and spur length was negative among plants with diurnal pollination and positive among plants with nocturnal pollination (Fig. S2d; Table S7). At Mel€osa, there was significant positive correlational selection on plant height and number of flowers among plants subject to diur- nal pollination and on number of flowers and spur length among plants subject to nocturnal pollination (Fig. S2a,b; Table S7).
(a) Number of flowers × Plant height
(b) Number of flowers × Spur length
(c) Flowering start × Corolla size
(d) Flowering start × Spur length
Melösa Långlöt
Number of flowers
Plant height Plant height
Number of flowers
Number of flowers
Number of flowers
Corolla size
Flowering start Corolla
size
Spur length
Spur length
Numberof flowers Numberof flowers
Numberof flowers Numberof flowers
Plant height Plant height
Spur length
Corolla size Corolla size
Spurlength Spurlength
Spur length Flowering start
C* HP
C* HP
C* HP
C* HP
Flowering start Flowering start Flowering start
Spur length Spur length Flowering start Flowering start
Flowering start
Fig. 3 Added-variable plots for the trait combinations where significant correlational selection was detected among open-pollinated control plants in the Mel€osa (a, b) and Langl€ot (c, d) Gymnadenia conopsea populations in 2012. C, open-pollinated control plants; HP, supplementally hand-pollinated plants.
*, significant correlational selection gradient. To produce these plots, the residuals from a linear regression of relative fitness on all floral traits except the two focal traits are plotted against the two sets of residuals obtained when regressing each of the two focal traits on the remaining floral traits (Cook &
Weisberg, 1989). In each panel, the added-variable plots are represented as estimated 3D surfaces in the upper graphs, and as contour plots with the actual data points indicated in the lower graphs. To facilitate comparison between treatments, each pair of 3D graphs in the panels have the same z-axis scale. The surfaces were estimated and plotted using the gam and vis.gam functions of the mgcv R package (Wood, 2006).
Discussion
The present study documents among-population variation in the strength of directional selection on two floral display traits (corolla size and number of flowers) and in the targets of correla- tional selection in G. conopsea. Spatial variation in both linear and correlational net selection could largely be attributed to polli- nator-mediated selection. Additionally, selection mediated by diurnal and nocturnal pollinators differed, indicating that varia- tion in the relative importance of the two pollinator categories may contribute to spatial variation in selection.
Pollinator-mediated selection largely explains variation in net phenotypic selection
Pollinator communities often vary across the distributional range of plant species with generalized pollination systems, potentially causing variation in selection on floral traits associated with attraction or mechanical fit (Gomez et al., 2008, 2009; Sletvold
& Agren, 2010, 2014). In E. mediohispanicum, variation in the direction of selection on corolla shape and tube width has been associated with geographical variation in the composition of the pollinator community (Gomez et al., 2009), and the strength of selection on floral traits has been shown to vary spatially in several species with semi-generalized pollination systems (Caruso et al., 2003; Weber & Kolb, 2011). In the present study, we were able to directly link variation in selection to interactions with pollina- tors. Pollinators contributed substantially to the strong selection for more flowers at Mel€osa, consistent with the well-known role of display size in pollinator attraction (Grindeland et al., 2005;
Makino et al., 2007 and references therein), and pollinators mediated selection for smaller corollas at Kvinneby. In G. conopsea, smaller flowers have narrower spurs (correlation between corolla size and spur diameter, r = 0.28, P < 0.0001, n = 197; N. Sletvold, unpublished data), which may increase
pollination efficiency by facilitating pollinium transfer to the pro- boscis of the pollinator. Lepidopteran pollinators select for nar- rower floral tubes in other systems (Campbell et al., 1997;
Kulbaba & Worley, 2012). Our results suggest that the advantage associated with producing small flowers is not outweighed by any negative effects of a reduced floral display. Only the Kvinneby population experienced significant selection for earlier flowering, and this was mainly mediated by pollinators. In this population, both diurnal and nocturnal visitors were more abundant early in the flowering season, whereas visitors present later in the season in other populations (Zygaena sp. and Autographa gamma) were not observed. This contrasts with a Norwegian G. conopsea popu- lation, where pollinator abundance increased during the flower- ing season, resulting in pollinator-mediated selection for later flowering (Sletvold et al., 2015), and suggests that variation in seasonal intensity of interactions with pollinators may underlie among-population variation in selection on phenology across var- ious spatial scales. Finally, all observed selection for longer spurs was pollinator-mediated (Mel€osa and Kvinneby), paralleling findings from Norwegian G. conopsea populations (Sletvold &
Agren, 2014). Taken together, the results indicate that pollina- tors are the main selective agent acting on pollination efficiency traits, whereas their contribution to selection on display traits
–0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
–0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
Net selection gradient (βC)
Pollinator-mediated selection (Δβpoll)
Flowering start (R2 = 0.64, P = 0.20) Plant height (R2 = 0.17, P = 0.59) Number of flowers (R2 = 0.94, P = 0.033*) Corolla size (R2 = 0.94, P = 0.033*) Spur length (R2 = 0.99, P = 0.0027**)
Fig. 4 Correspondence between pollinator-mediated selection and net selection for each floral trait in the four Gymnadenia conopsea study populations. Each symbol is the estimate of selection in one of the populations. Solid lines represent significant linear regressions (R2and P of each regression given within brackets). The dashed line represents a 1 : 1 relationship betweenbCandDbpoll.*, P < 0.05; **, P < 0.01.
–0.2 0 0.2 0.4 0.6
Flowering date
Plant height Number of flowers
Corolla size Spur length
Selection gradient ± SE
D N
–0.2 0 0.2 0.4 0.6
Flowering date
Plant height Number of flowers
Corolla size Spur length
Selection gradient ± SE
D N
Melösa
Långlöt
* *
* *
* * *
*
*
*
* *
* (a)
(b)
Fig. 5 Phenotypic linear selection gradients on flowering start, plant height, number of flowers, corolla size and spur length among Gymnadenia conopsea plants with diurnal pollination (D; white bars) or nocturnal pollination (N; grey bars) in the (a) Mel€osa and (b) Langl€ot populations. Symbols above individual bars indicate the level of significance of the gradient. The symbol above the line spanning several gradients indicates that there is a significant difference in selection gradients between treatments.*, P < 0.05.
and phenology is variable. Selection on traits that are closely cor- related to size and resource acquisition may largely be mediated by nonpollinator agents (reviewed in Strauss & Whittall, 2006).
Among-population differences in the direction of selection drive the evolution of adaptive population differentiation and contribute to its maintenance. In the present study, the direction of selection on flowering time and corolla size varied among pop- ulations, but only in the Kvinneby population, which was the earliest flowering population and one of the three small-flowered populations, was there significant selection for earlier flowering and selection for smaller flowers (Fig. S1). A weak relationship between population phenotype and current selection may be attributable to temporal variation in the direction and strength of selection, recent shifts in population optima, or conflicting selec- tion through other components of fitness. Finally, if population means correspond to local adaptive peaks, then no directional selection is expected even if the optimal phenotype varies among populations.
Pollinators mediated most of the correlational selection we detected on pairs of display traits and on display and efficiency traits, as well as on flowering phenology and morphology traits.
At Mel€osa, the positive correlational selection on number of flow- ers and plant height suggests that these two display traits act syn- ergistically on pollinator attraction or, alternatively, that one pollinator responds to display size and another to height, leading to a nonadditive effect on overall visitation rate (Campbell et al., 2014). The positive gradient for number of flowers and spur length rather suggests a multiplicative effect of pollinator attrac- tion and pollination efficiency on fitness (Sletvold & Agren, 2011). Correlational selection on pairs of display and efficiency traits may be common, as it has been reported in all previous studies documenting correlational selection on floral traits (O’Connell & Johnston, 1998; Benitez-Vieyra et al., 2006; Cuar- tas-Domınguez & Medel, 2010; Reynolds et al., 2010;
Bartkowska & Johnston, 2012). At L angl€ot, no linear estimate of pollinator-mediated selection was significant, but pollinators mediated selection for earlier flowering in combination with longer spurs, and against later flowering combined with smaller corollas. This suggests that weak effects of single traits on pollina- tion success can become detectable in combination because of nonadditive effects. In Lobelia cardinalis, pollinators also medi- ated selection on trait combinations in a population with no sig- nificant linear pollinator-mediated selection (Bartkowska &
Johnston, 2012), indicating that it may be necessary to consider trait combinations to detect pollinator-mediated selection.
The strength of pollinator-mediated selection is expected to be positively and nonlinearly related to the magnitude of pollen lim- itation (Benkman, 2013; Vanhoenacker et al., 2013). In the pre- sent study, pollinator-mediated selection varied considerably despite similar pollen limitation in all four populations, indicat- ing the importance of the functional relationship between polli- nators and plant traits in driving variation in pollinator-mediated selection (Sletvold & Agren, 2014).
In hermaphroditic plants, patterns of selection may differ between sex functions (Campbell, 1989; Conner et al., 1996;
Maad, 2000; Benitez-Vieyra et al., 2006; Hodgins & Barrett,
2008), and the selection patterns described here may not hold for male fitness. In orchids, pollen removal is often straightforward to score and has been used to estimate male reproductive success (O’Connell & Johnston, 1998; Maad, 2000). In one Norwegian population of G. conopsea, pollen removal varied among plants but was strongly correlated with female reproductive success, sug- gesting that selection estimates based on pollen removal would parallel those obtained for female function (Sletvold & Agren, 2010). However, pollen removal may often be a poor proxy of pollen export (Johnson et al., 2005; Ellis & Johnson, 2010) and of siring success (Snow & Lewis, 1993), and direct estimates of male fitness through paternity analyses would be needed to reli- ably determine whether patterns of selection via male and female function differ.
Diurnal and nocturnal pollinators exert different selection patterns
In species with both diurnal and nocturnal pollinators, selective pollinator exclusion may improve the understanding of among- population variation in selection. At L angl€ot, the stronger selec- tion for more flowers among nocturnally pollinated plants sug- gests that, in this population, nocturnal pollinators responded strongly to either a visual or an olfactory signal of floral display.
At Mel€osa, only nocturnal pollinators mediated selection on spur length, which may reflect a difference in proboscis length between the two pollinator categories. The diurnal Zygaena minos has a shorter proboscis than the other visitors we observed in this population. Similarly, no selection on spur length was mediated by diurnal short-tongued Empis flies in a Norwegian G. conopsea population (Sletvold et al., 2012b). In the Mel€osa population, the net correlational selection on number of flowers and spur length could also be attributed to nocturnal pollinators, consis- tent with the finding that nocturnal pollinators caused directional selection on spur length. By contrast, the net correlational selec- tion on the pair of display traits, number of flowers and plant height, was mediated by diurnal pollinators. The direction of cor- relational selection on the combination of number of flowers and spur length differed between the two pollinator categories at L angl€ot, suggesting a trade-off in adaptation to the two types of pollinators (Aigner, 2001) because of conflicting correlational selection rather than conflicting selection on single traits. Finally, although patterns of net selection could not always be linked to either type of pollinator, the fact that selection exerted by the two categories differed suggests that differences in their relative abun- dance may cause variation in selection among populations.
Latitudinal gradient in the relative importance of nocturnal pollinators
Plants pollinated exclusively by diurnal or nocturnal pollinators
had similar fitness to open-pollinated plants, showing that each
pollinator category in isolation can efficiently pollinate
G. conopsea. Previous selective pollinator exclusion experiments
demonstrated that nocturnal pollinators were more important
than diurnal pollinators for seed set in two German populations
at 51°N (Meyer et al., 2007), whereas the contrary was true in two populations in central Norway at 62–63°N (Sletvold et al., 2012b). Together with the present study, which was conducted at 56°N, these results suggest a latitudinal gradient in the relative importance of diurnal vs nocturnal pollinators for G. conopsea.
With increasing latitude, day length increases and night tempera- ture decreases, which may be less favourable for nocturnal polli- nators.
Conclusion
Determining the causes of spatial variation in selection among nat- ural populations is crucial to understand the process of adaptive differentiation as well as the maintenance of generalization, but empirical data remain limited (MacColl, 2011; Siepielski et al., 2013). We show that variation in pollinator-mediated selection caused most of the observed variation in net selection among pop- ulations of G. conopsea. Of particular interest is the finding that spatial variation in net selection was partly attributable to variation in pollinator-mediated correlational selection. Functional integra- tion of flowers is assumed to be a product of correlational selection (Armbruster & Schwaegerle, 1996), but despite considerable theo- retical interest, few studies have manipulated the environment or trait combinations to verify pollinator-mediated correlational selection. By manipulating the pollination environment in multi- ple populations, we show that it is possible to test rigorously for the presence of a geographical mosaic of pollinator-mediated selec- tion. Studies of trait effects on pollinator visitation and pollination efficiency are needed to reveal the underlying mechanisms generat- ing directional and correlational selection.
Acknowledgements
We thank Karl Fritzson, Andreas Johansson and Mattias Vass for field assistance, Eje Rosen for information about G. conopsea pop- ulations on € Oland, and Dave Karlsson for help with identifying pollinators. Diane Campbell, Lynda Delph, Lawrence Harder and two anonymous reviewers provided valuable comments on a previous version of this manuscript. The study was financially supported by grants from the Extensus Stiftelse to E.C., from the Swedish Research Council to J. A., and from the Swedish Research Council Formas to N.S.
References
Agren J, Hellstr€om F, Tor€ang P, Ehrlen J. 2013. Mutualists and antagonists drive among-population variation in selection and evolution of floral display in a perennial herb. Proceedings of the National Academy of Sciences, USA 110:
18202–18207.
Aigner PA. 2001. Optimality modeling and fitness trade-offs: when should plants become pollinator specialists? Oikos 95: 177–184.
Anderson B, Johnson SD. 2008. The geographical mosaic of coevolution in a plant–pollinator mutualism. Evolution 62: 220–225.
Andersson S. 1996. Floral display and pollination success in Senecio jacobaea (Asteraceae): interactive effects of head and corymb size. American Journal of Botany 83: 71–75.
Armbruster WS, Schwaegerle KE. 1996. Causes of covariation of phenotypic traits among populations. Journal of Evolutionary Biology 9: 261–276.
Bartkowska MP, Johnston MO. 2012. Pollinators cause stronger selection than herbivores on floral traits in Lobelia cardinalis (Lobeliaceae). New Phytologist 193: 1039–1048.
Benitez-Vieyra S, Medina AM, Glinos E, Cocucci AA. 2006. Pollinator- mediated selection on floral traits and size of floral display in Cyclopogon elatus, a sweat bee-pollinated orchid. Functional Ecology 20: 948–957.
Benkman CW. 2013. Biotic interaction strength and the intensity of selection.
Ecology Letters 16: 1054–1060.
Boberg E, Alexandersson R, Jonsson M, Maad J, Agren J, Nilsson LA. 2014.
Pollinator shifts and the evolution of spur length in the moth-pollinated orchid Platanthera bifolia. Annals of Botany 113: 267–275.
Burkhardt A, Ridenhour BJ, Delph LF, Bernasconi G. 2012. The contribution of a pollinating seed predator to selection on Silene latifolia females: seed predator selection on flower size and number. Journal of Evolutionary Biology 25: 461–472.
Campbell DR. 1989. Measurements of selection in a hermaphroditic plant:
variation in male and female pollination success. Evolution 43: 318.
Campbell DR, Forster M, Bischoff M. 2014. Selection of trait combinations through bee and fly visitation to flowers of Polemonium foliosissimum. Journal of Evolutionary Biology 27: 325–336.
Campbell DR, Waser NM, Melendez-Ackerman EJ. 1997. Analyzing pollinator- mediated selection in a plant hybrid zone: hummingbird visitation patterns on three spatial scales. The American Naturalist 149: 295–315.
Caruso CM, Peterson SB, Ridley CE. 2003. Natural selection on floral traits of Lobelia (Lobeliaceae): spatial and temporal variation. American Journal of Botany 90: 1333–1340.
Claessens J, Kleynen J. 2011. The flower of the European orchid: form and function. Schrijen-Lipperz, Stein, the Netherlands: Jean Claessens & Jacques Kleynen.
Conner JK, Rush S, Kercher S, Jennetten P. 1996. Measurements of natural selection on floral traits in wild radish (Raphanus raphanistrum). II. Selection through lifetime male and total fitness. Evolution 50: 1137.
Cook RD, Weisberg S. 1989. Regression diagnostics with dynamic graphics.
Technometrics 31: 277–291.
Cuartas-Domınguez M, Medel R. 2010. Pollinator-mediated selection and experimental manipulation of the flower phenotype in Chloraea bletioides.
Functional Ecology 24: 1219–1227.
Darwin C. 1862. On the various contrivances by which British and foreign orchids are fertilised by insects and on the good effects of intercrossing. London, UK: John Murray.
Ellis AG, Johnson SD. 2010. Gender differences in the effects of floral spur length manipulation on fitness in a hermaphrodite orchid. International Journal of Plant Sciences 171: 1010–1019.
Faegri K, Van der Pijl L. 1966. The principles of pollination ecology. Oxford, UK:
Pergamon Press.
Fenster CB, Armbruster WS, Wilson P, Dudash MR, Thomson JD. 2004.
Pollination syndromes and floral specialization. Annual Review of Ecology, Evolution, and Systematics 35: 375–403.
Fox J, Weisberg S. 2011. An R companion to applied regression. Thousand Oaks, CA, USA: Sage Publications.
Galen C. 2000. High and dry: drought stress, sex-allocation trade-offs, and selection on flower size in the alpine wildflower Polemonium viscosum (Polemoniaceae). The American Naturalist 156: 72–83.
Geber MA, Moeller DA. 2006. Pollinator responses to plant communities and implications for reproductive character evolution. In: Harder LD, Barrett SCH, eds. Ecology and evolution of flowers. Oxford, UK: Oxford University Press, 102–119.
Gomez JM. 2003. Herbivory reduces the strength of pollinator-mediated selection in the mediterranean herb Erysimum mediohispanicum: consequences for plant specialization. The American Naturalist 162: 242–256.
Gomez JM, Abdelaziz M, Lorite J, Jesus Mu~noz-Pajares A, Perfectti F. 2010.
Changes in pollinator fauna cause spatial variation in pollen limitation:
pollinator assemblage and pollen limitation. Journal of Ecology 98: 1243–1252.
Gomez JM, Bosch J, Perfectti F, Fernandez JD, Abdelaziz M, Camacho JPM.
2008. Spatial variation in selection on corolla shape in a generalist plant is promoted by the preference patterns of its local pollinators. Proceedings.
Biological Sciences/The Royal Society 275: 2241–2249.
Gomez JM, Perfectti F, Bosch J, Camacho JPM. 2009. A geographic selection mosaic in a generalized plant–pollinator–herbivore system. Ecological Monographs 79: 245–263.
Grindeland JM, Sletvold N, Ims RA. 2005. Effects of floral display size and plant density on pollinator visitation rate in a natural population of Digitalis purpurea. Functional Ecology 19: 383–390.
Harder LD, Johnson SD. 2009. Darwin’s beautiful contrivances: evolutionary and functional evidence for floral adaptation. New Phytologist 183: 530–545.
Herrera CM, Castellanos MC, Medrano M. 2006. Geographical context of floral evolution: towards an improved research programme in floral diversification.
In: Harder LD, Barrett SCH, eds. Ecology and evolution of flowers. Oxford, UK:
Oxford University Press, 278–294.
Hersch EI, Roy BA. 2007. Context-dependent pollinator behavior: an explanation for patterns of hybridization among three species of Indian paintbrush. Evolution 61: 111–124.
Hodgins KA, Barrett SCH. 2008. Natural selection on floral traits through male and female function in wild populations of the heterostylous daffodil Narcissus triandrus. Evolution 62: 1751–1763.
Hulten E, Fries M. 1986. Atlas of North European vascular plants (North of the Tropic of Cancer), Vols I–III. K€onigstein, Germany: Koeltz Scientific Books.
Johnson SD, Neal PR, Harder LD. 2005. Pollen fates and the limits on male reproductive success in an orchid population. Biological Journal of the Linnean Society 86: 175–190.
Kay KM, Sargent RD. 2009. The role of animal pollination in plant speciation:
integrating ecology, geography, and genetics. Annual Review of Ecology, Evolution, and Systematics 40: 637–656.
Kulbaba MW, Worley AC. 2012. Selection on floral design in Polemonium brandegeei (Polemoniaceae): female and male fitness under hawkmoth pollination. Evolution 66: 1344–1359.
Lande R, Arnold SJ. 1983. The measurement of selection on correlated characters. Evolution 37: 1210–1226.
Maad J. 2000. Phenotypic selection in hawkmoth-pollinated Platanthera bifolia:
targets and fitness surfaces. Evolution 54: 112–123.
MacColl ADC. 2011. The ecological causes of evolution. Trends in Ecology and Evolution 26: 514–522.
Makino TT, Ohashi K, Sakai S. 2007. How do floral display size and the density of surrounding flowers influence the likelihood of bumble bee revisitation to a plant? Functional Ecology 21: 87–95.
Meyer B, Kr€oger J, Steffan-Dewenter I. 2007. Contribution of diurnal and nocturnal pollinators to the reproductive success of the orchid species Gymnadenia conopsea. Entomologia Generalis 30: 299–300.
Newman E, Manning J, Anderson B. 2014. Matching floral and pollinator traits through guild convergence and pollinator ecotype formation. Annals of Botany 113: 373–384.
O’Connell LM, Johnston MO. 1998. Male and female pollination success in a deceptive orchid, a selection study. Ecology 79: 1246–1260.
Øien DI, Moen A. 2002. Flowering and survival of Dactylorhiza lapponica and Gymnadenia conopsea in the Sølendet Nature Reserve, Central Norway. In:
Kindlmann P, Willems JH, Whigham DF, eds. Trends and fluctuations and underlying mechanisms in terrestrial orchid populations. Leiden, the Netherlands:
Backhuys Publishers, 3–22.
Parachnowitsch AL, Caruso CM. 2008. Predispersal seed herbivores, not pollinators, exert selection on floral traits via female fitness. Ecology 89: 1802–
1810.
Petit C, Thompson JD. 1998. Phenotypic selection and population differentiation in relation to habitat heterogeneity in Arrhenatherum elatius (Poaceae). Journal of Ecology 86: 826–840.
Quinn GP, Keough MJ. 2002. Experimental design and data analysis for biologists.
New York, NY, USA: Cambridge University Press.
R Core Team. 2015. R: a language and environment for statistical computing.
Vienna, Austria: R foundation for Statistical Computing.
Reynolds RJ, Dudash MR, Fenster CB. 2010. Multiyear study of multivariate linear and nonlinear phenotypic selection on floral traits of hummingbird- pollinated Silene virginiva. Evolution 64: 358–369.
Siepielski AM, Gotanda KM, Morrissey MB, Diamond SE, DiBattista JD, Carlson SM. 2013. The spatial patterns of directional phenotypic selection.
Ecology Letters 16: 1382–1392.
Sletvold N, Agren J. 2010. Pollinator-mediated selection on floral display and spur length in the orchid Gymnadenia conopsea. International Journal of Plant Sciences 171: 999–1009.
Sletvold N, Agren J. 2011. Nonadditive effects of floral display and spur length on reproductive success in a deceptive orchid. Ecology 92: 2167–2174.
Sletvold N, Agren J. 2014. There is more to pollinator-mediated selection than pollen limitation: interaction intensity and selection strength. Evolution 68:
1907–1918.
Sletvold N, Grindeland JM, Agren J. 2013. Vegetation context influences the strength and targets of pollinator-mediated selection in a deceptive orchid.
Ecology 94: 1236–1242.
Sletvold N, Grindeland JM, Zu P, Agren J. 2012a. Strong inbreeding depression and local outbreeding depression in the rewarding orchid Gymnadenia conopsea.
Conservation Genetics 13: 1305–1315.
Sletvold N, Moritz KK, Agren J. 2015. Additive effects of pollinators and herbivores result in both conflicting and reinforcing selection on floral traits.
Ecology 96: 214–221.
Sletvold N, Trunschke J, Wimmergren C, Agren J. 2012b. Separating selection by diurnal and nocturnal pollinators on floral display and spur length in Gymnadenia conopsea. Ecology 93: 1880–1891.
Snow AA, Lewis PO. 1993. Reproductive traits and male fertility in plants:
empirical approaches.Annual Review of Ecology and Systematics 24: 331– 351.
Stinchcombe JR, Agrawal AF, Hohenlohe PA, Arnold SJ, Blows MW. 2008.
Estimating nonlinear selection gradients using quadratic regression coefficients:
double or nothing? Evolution 62: 2435–2440.
Stpiczynska M, Matusiewicz J. 2001. Anatomy and ultrastructure of spur nectary of Gymnadenia conopsea (L.) Orchidaceae. Acta Societatis Botanicorum Poloniae 70: 267–272.
Strauss SY, Whittall JB. 2006. Non-pollinator agents of selection on floral traits.
In: Harder LD, Barrett SCH, eds. Ecology and evolution of flowers. Oxford, UK:
Oxford University Press, 120–138.
Totland Ø. 2001. Environment-dependent pollen limitation and selection on floral traits in an alpine species. Ecology 82: 2233–2244.
Van der Niet T, Johnson SD. 2012. Phylogenetic evidence for pollinator-driven diversification of angiosperms. Trends in Ecology and Evolution 27: 353–361.
Vanhoenacker D, Agren J, Ehrlen J. 2013. Non-linear relationship between intensity of plant–animal interactions and selection strength. Ecology Letters 16:
198–205.
Weber A, Kolb A. 2011. Evolutionary consequences of habitat fragmentation:
population size and density affect selection on inflorescence size in a perennial herb. Evolutionary Ecology 25: 417–428.
Weber A, Kolb A. 2013. Population size, pollination and phenotypic trait selection in Phyteuma spicatum. Acta Oecologica 47: 46–51.
Wilson P, Thomson JD. 1996. How do flowers diverge? In: Lloyd DG, Barrett SCH, eds. Floral biology: studies on floral evolution in animal-pollinated plants.
New York, NY, USA: Chapman and Hall, 88–111.
Wood S. 2006. Generalized additive models: an introduction with R. Boca Raton, FL, USA: Chapman and Hall/CRC.
Zhao Z-G, Huang S-Q. 2013. Differentiation of floral traits associated with pollinator preference in a generalist-pollinated herb, Trollius ranunculoides (Ranunculaceae). International Journal of Plant Sciences 174: 637–646.
