Mutualists and antagonists drive among-population variation in selection and evolution of floral display in a perennial herb

Jon Ågrena,1,2, Frida Hellströma, Per Toränga, and Johan Ehrlénb,1

aDepartment of Ecology and Evolution, Evolutionary Biology Centre, Uppsala University, SE-752 36 Uppsala, Sweden; and bDepartment of Ecology, Environment, and Plant Sciences, Stockholm University, SE-106 91 Stockholm, Sweden

Edited by Rodolfo Dirzo, Stanford University, Stanford, CA, and approved September 26, 2013 (received for review January 22, 2013) Spatial variation in the direction of selection drives the evolution herbivores have been inferred in many systems (15, 19, 20), but of adaptive differentiation. However, few experimental studies no study has simultaneously manipulated the intensity of both have examined the relative importance of different environmental interactions to determine their relative importance for spatio- factors for variation in selection and evolutionary trajectories in temporal variation in selection on plant traits. There is also natural populations. Here, we combine 8 y of observational data a lack of studies experimentally examining the importance of and field experiments to assess the relative importance of mu- biotic interactions for the evolutionary trajectories of natural tualistic and antagonistic interactions for spatial variation in se- plant populations (28, 29). lection and short-term evolution of a genetically based floral Here, we combine long-term observational data and field display dimorphism in the short-lived perennial herb far- experiments to examine causes and consequences of spatial and inosa. Natural populations of this species include two floral temporal variation in selection on floral display in the rosette- morphs: long-scaped that present their flowers well above forming, short-lived, perennial herb . This spe- the ground and short-scaped plants with flowers positioned close cies offers an ideal system to examine the outcome of conflicting to the ground. The direction and magnitude of selection on scape selection by mutualists and antagonists. It is dimorphic for scape

morph varied among populations, and so did the frequency of the length, with a long-scaped morph displaying the umbellate in- EVOLUTION short morph (median 19%, range 0–100%; n = 69 populations). A florescence well above the soil surface and a short-scaped morph field experiment replicated at four sites demonstrated that varia- with the inflorescence very close to the ground. The segregation tion in the strength of interactions with grazers and pollinators of scape morphs in controlled crosses is consistent with scape were responsible for among-population differences in relative fit- morph being determined by a single biallelic locus with a domi- ness of the two morphs. Selection exerted by grazers favored the nant allele coding for short scape (SI Text, SI Segregation of Scape short-scaped morph, whereas pollinator-mediated selection fa- Morphs in Crosses and Table S1). This difference in floral display vored the long-scaped morph. Moreover, variation in selection affects interactions with both pollinators and antagonists. In pre- among natural populations was associated with differences in vious studies, we have shown that seed production in the long- morph frequency change, and the experimental removal of graz- scaped morph is less likely to be limited by pollen availability (14, ers at nine sites significantly reduced the frequency of the short- 30, 31), whereas the short-scaped morph is less frequently at- scaped morph over 8 y. The results demonstrate that spatial var- tacked by seed predators (14, 18, 32, 33). The inflorescence of iation in intensity of grazing and pollination produces a selection the long-scaped morph should also have a higher probability of mosaic, and that changes in biotic interactions can trigger rapid being damaged by grazers compared with that of the short-scaped genetic changes in natural plant populations. morph. These interactions influence plant fitness largely via fruit production, which is a key fitness component of the study species adaptive evolution | divergent selection | floral trait | herbivory | and straightforward to quantify. In P. farinosa populations in the natural selection Significance patial variation in the intensity of biotic interactions is an Sintegral part of the geographic mosaic model of coevolution A prominent floral display may increase attractiveness to pol- (1, 2), and may result in divergent selection and the maintenance linators but also the risk of damage from herbivores. Here, we of genetic variation in traits influencing the strength and out- show experimentally that differences in the relative strength come of interactions (3, 4). However, few studies have presented of interactions with grazers and pollinators could explain var- quantitative estimates of spatiotemporal variation in selection on iation in selection on floral display among natural populations traits influencing the outcome of biotic interactions across more of an -pollinated primrose. In addition, we demonstrate than a handful of populations. In plants, variation in the com- that differences in selection translate into rather rapid changes position of the mutualist and antagonist assemblages may result in the genetic composition of local plant populations. The in spatially varying selection on morphology, phenology, and life- results indicate that interactions with mutualists and antago- history traits (e.g., 5–12). Of particular interest are traits such as nists can drive adaptive differentiation not only across broad floral display that may be subject to conflicting selection from geographic scales but also among populations across relatively mutualists and antagonists, and where the magnitude and di- short distances. rection of net selection should depend on the relative strength of these interactions (13–20). Author contributions: J.Å. and J.E. designed research; J.Å., F.H., P.T., and J.E. performed Experimental manipulation of environmental conditions is a research; J.Å. and J.E. analyzed data; and J.Å. and J.E. wrote the paper. powerful approach to identify agents of selection and to deter- The authors declare no conflict of interest. mine the evolutionary consequences of changes in the selection This article is a PNAS Direct Submission. regime (21, 22). Experimental manipulation of pollen deposition 1J.Å. and J.E. contributed equally to this work. (6, 23, 24) and interactions with herbivores (25–28) can be used 2To whom correspondence should be addressed. E-mail: [email protected]. to assess the roles of pollinators and herbivores for patterns of This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. selection. Conflicting selection on floral traits by pollinators and 1073/pnas.1301421110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1301421110 PNAS Early Edition | 1of6 Downloaded by guest on October 2, 2021 fi study area, plant mortality is high, overall tness is strongly in- 2 2000 fl uenced by successful seedling recruitment (34), and total seed 1.5 fi production is signi cantly correlated with number of intact ma- 1 = = ture fruits produced (r 0.838, n 442). 0.5 na na na na na na na na na na na na We documented variation in scape morph frequencies among 0 na 69 populations and asked the following questions (1): Does se- -0.5 lection on scape length vary among populations and years? We -1 Relative fitness of short-scaped morph 1 2 3 74 82 12 13 15 16 35 38 40 42 43 45 46 47 54 55 56 58 59 61 62 70 76 78 81 84 86 87 91 94 96 64 69 93 fi 18 41 66 71 72 73 75 77 79 85 quanti ed selection on scape morph in about 40 populations in 104 109 128 fi each of 2 y, and in ve populations across 5 y (2). What are the 3 2001 drivers of variation in selection on scape morph? We docu- 2.5 mented the relationship between grazing intensity and selection 2 on scape morph, and with a field experiment, we tested the hy- 1.5 pothesis that spatial variation in grazing pressure and pollination 1 intensity cause among-population variation in selection on scape 0.5 na na morph (3). Do among-population differences in selection result na na 0 in different evolutionary trajectories? We used observational -0.5 data to examine whether changes in scape morph frequencies Relative fitness of short-scaped morph -1 1 2 3 46 85 13 15 38 40 42 43 47 54 56 58 59 61 62 64 66 69 70 71 72 73 76 77 78 82 84 86 87 93 94 96 12 16 18 35 41 45 were correlated with estimates of selection on scape morph, and 55 74 75 79 81 91 128 104 109 an 8-y field experiment to test whether the exclusion of grazers Population resulted in a reduced frequency of the short-scaped morph. Fig. 2. Relative fitness of the short-scaped morph (ln[mean number of in- Results tact mature fruits produced by short-scaped plants divided by mean number Morph Frequency Variation. The proportion of the short-scaped of intact mature fruits produced by long-scaped plants]) in P. farinosa morph varied from zero to 100% (median 19%) among 69 P. populations on Öland, southeastern Sweden in 2000 (n = 37 populations) farinosa populations within a 4 × 10-km large area on the island and 2001 (n = 46 populations). Populations in which the difference in fruit fi Öland, southeastern Sweden (Fig. 1). production between scape morphs was statistically signi cant (according to contrasts in mixed-model ANOVA) are indicated in red (S > L) and blue (S < L); statistically nonsignificant differences are indicated in green. Spatiotemporal Variation in Selection on Floral Display. The relative fitness of the two scape morphs varied from a twofold advantage of the long-scaped morph to a 13-fold advantage of the short- short-scaped morph in 14 of 37 populations in 2000, and in 18 of scaped morph among the populations for which fecundity data fi 46 populations in 2001; the reverse was true in 2 populations in were available from 2000 and/or 2001 (Fig. 2; signi cant scape 2001 (Fig. 2). The overall higher fecundity of the short-scaped morph × population interaction in mixed-model ANOVA of fi χ2 = morph was mainly the result of morph-speci c differences in variation in number of mature fruits produced, 2000, 23.1, fruit set (number of mature fruits per flower). The relative fitness df = 1, P < 0.0001; 2001, χ2 = 23.1, df = 1, P < 0.0001, Table S2). fi fi of the short-scaped morph was strongly correlated with its rela- Contrasts indicated that tness was signi cantly higher in the tive fruit set (number of mature fruits per flower) in both years (2000, r = 0.92, n = 37; 2001, r = 0.95, n = 46), and weakly to moderately correlated with differences in relative number of flowers (2000, r = 0.14; 2001, r = 0.59; Figs. S1 and S2). Five of the populations were sampled over 5 y. In this set of populations, the relative fitness of the of the two scape morphs varied among years, and this variation was not synchronous across populations (Fig. 3). The effect of the scape morph × population × year interaction on number of intact mature fruits was statistically significant in a mixed-model ANOVA (χ2 = 14.6, df = 1, P < 0.001), whereas effects of two-way interactions or main effects were not significant (P > 0.05). The mean relative fitness of the short-scaped morph over the 5-y period was posi- tive in all five populations (range 0.047–0.485, n = 5) indicating selection favoring the short morph, but the direction of selection on scape morph varied among years in four of the populations (Fig. 3). 20 Median = 19% Agents of Selection. The short-scaped morph was less damaged by 15 Range 0-100% grazers than was the long-scaped morph, and the relative fitness n = 69 10 of the short morph was positively related to grazing intensity. The proportion of plants that had their inflorescence grazed was

populations 5 5.4 times higher in the long-scaped morph in 2000 (20.9% vs. # 3.9%, paired test, t = 6.3, P < 0.0001, n = 37 populations), and 0 4.6 times higher in 2001 (18.2% vs. 3.9%, t = 4.4, P < 0.0001, n = 0102030 40 50 60 70 9080100 46 populations). The relative fitness of the short-scaped morph was positively related to the grazing intensity (quantified as the Proportion short (%) proportion of long-scaped plants grazed) in 2000 (linear re- fi = = 2 = = Fig. 1. (Lower) Frequency distribution of the proportion of the short-sca- gression coef cient, b 1.193, P 0.007, R 0.17, n 37 2 ped morph in 69 populations of P. farinosa on Öland, southeastern Sweden populations), in 2001 (b = 2.164, P < 0.0001, R = 0.59, n = 46 in 2001. (Upper) The photos show the long scaped morph (Left) and the populations), and in the five populations sampled over 5 y (b = short-scaped morph (Right). Photo: J.Å. 1.767, P = 0.010, R2 = 0.22, n = 25 population × year combinations).

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1301421110 Ågren et al. Downloaded by guest on October 2, 2021 3.0 whereas in adjacent control plots, it remained essentially the same (0.50 vs. 0.48, t = 0.1, df = 8, P = 0.891; Fig. 5). 2.5 Pop 13 Pop 46 Discussion 2.0 Pop 59 fi Pop 81 This study identi ed interactions with pollinators and grazers as 1.5 Pop 84 major drivers of spatial variation in the direction of selection on floral display in P. farinosa, and showed that differences in se- 1.0 lection regimes translated into differences in the evolutionary trajectories of local populations. The experimental results sug- 0.5 gest that grazing avoidance is a major factor favoring the short- 0.0 scaped morph whereas pollinator attraction favors the long-

Relative fitness short scaped morph, and that shifts in the relative strength of these -0.5 interactions change the direction of selection and trigger rapid

-1.0 changes in the genetic composition of local populations. 1999 2000 2001 2002 2003 2004 2005 Spatial variation in the direction of selection should contribute to the maintenance of adaptive genetic variation among pop- Year ulations, and may in combination with gene flow also promote Fig. 3. Spatiotemporal variation in relative fitness of the short-scaped the maintenance of genetic variation within populations (cf. 35). morph in five P. farinosa populations sampled over 5 y. In P. farinosa, selection favored the short-scaped morph over the long-scaped morph in many populations, but the reverse was true in some populations. Gene flow among P. farinosa populations A field experiment replicated in four other populations dem- subject to divergent selection may thus contribute to the main- onstrated that interactions with mammalian grazers and insect tenance of the dimorphism in floral display within populations. pollinators explained almost all variation in selection on scape In addition, previous experiments have demonstrated negative fi morph. For unmanipulated control plants, the relative tness of frequency-dependent selection on scape morph mediated by pol- the short-scaped morph varied from -0.580 to 1.679 among pop- linators within large patchy populations (18). Both variable se- EVOLUTION ulations (i.e., from a 1.7-fold advantage of the long-scaped morph lection among populations and frequency-dependent selection to a 5.3-fold advantage of the short-scaped morph; Fig. 4A). within populations may thus promote the persistence of the scape- Outside exclosures, the proportion of plants whose inflorescence length polymorphism in this system. was removed by grazers varied among populations (range 0.12– 0.63; F3,7 = 40.6, P < 0.0001), was between 1.5 and 4.0 times higher in the long- than in the short-scaped morph (median 2.7 times higher; effect of morph, F1,7 = 44.1, P = 0.0003; population × morph interaction, F3,7 = 9.5, P = 0.0074), but was not affected by supplemental hand-pollination (P > 0.27). Exclusion of mam- malian grazers reduced the relative fitness of the short-scaped morph (ANOVA, F1,9 = 9.5, P = 0.01), whereas hand-pollination increased the relative fitness of the short-scaped morph (F1,9 = 5.7, P = 0.04; the grazing × pollination treatment interaction was not statistically significant, F1,9 = 0.6, P = 0.46; Fig. 4 B and C). Among plants protected from grazers and provided with a sur- plus of pollen, there was no significant selection on scape morph (Fig. 4D). The proportion of fruits consumed by seed predators was generally low (median 6%), varied among populations (F3,25 = 7.6, P = 0.0009) and scape morphs (F1,25 = 9.5, P = 0.0049), but was not affected by experimental treatments (P > 0.18).

Evolution of Morph Frequencies. Both observational and experi- mental data indicated that selection on scape morph result in rather rapid changes in morph frequencies. Scape morph fre- quencies were monitored in 24 of the 69 populations from 2006 to 2012, and over this 6-y period the short-scaped morph tended to increase in 19 populations and decrease in five (significant population × year interaction in ANCOVA; F23,120 = 2.7, P = 0.0002). Across all 24 populations, the short-scaped morph in- creased by on average 1% per year. In five populations observed over 8 y, the mean relative fitness of the short-scaped morph 2000–2004 was positively correlated with the change in the proportion of the short-scaped morph over the period 2000–2007 (Spearman rank correlation rs = 0.90, P = 0.037, n = 5; change in Fig. 4. Effects of grazer exclusion and supplemental hand-pollination on morph frequency quantified as the regression of proportion relative fitness of the short-scaped morph of P. farinosa in a field experiment short-scaped plants on year). Removal of mammalian grazers conducted in four populations: (A)control,(B) grazers excluded, (C) plants receiving supplemental hand-pollination, and (D) grazers excluded and plants from exclosures established in nine populations caused the mean receiving supplemental hand-pollination. Population × treatment combina- proportion of the short morph to decrease from 0.48 to 0.36 tions for which the 95% confidence interval of relative fitness (estimated between 2004 and 2012 (paired t test, t = 3.2, df = 8, P = 0.013), through bootstrapping 1000 times) did not overlap zero are indicated.

Ågren et al. PNAS Early Edition | 3of6 Downloaded by guest on October 2, 2021 We cannot completely rule out that the scape morphs differ also in other components of fitness. However, available evidence sug- gests that selection on scape morph mainly acts through differential reproductive success. Consistent with common-garden experiments (39), no significant difference between scape morphs in mortality of flowering plants (Table S3), or in age at first reproduction [mean = 1.2 y after germination for both long-scaped (n = 195), and short- scaped plants (n = 55)] was recorded in a demographic study conducted in three populations in the study area (34). A key finding of the present study was that differences in se- lection regime were associated with rapid changes in the genetic composition of natural populations and in field experiments. Changes in morph frequencies in natural populations across 8 y were correlated with the mean relative fitness of the short-scaped morph over a 5-y period. Moreover, exclusion of grazers in- creased the relative fitness of the long-scaped morph and had after 8 y resulted in a 10% decrease in the proportion of the short-scaped morph. High mortality and fast development from Fig. 5. Change in the proportion of the short-scaped morph from 2004 to seedling to flowering individual in P. farinosa populations in the 2012 in control plots and exclosures established in nine populations of P. farinosa. study area should have promoted the rapid changes in morph Difference in change between the two treatments tested with paired t test. frequencies. In three populations monitored for 7 y (34), mor- tality among vegetative and reproductive plants combined ranged from 9.9% to 100% (median 61.9%; n = 18 annual transitions). fi fi The eld experiment identi ed pollinators and grazers as major Of seedlings surviving to the reproductive stage, 85% had begun fl agents of selection on oral display in P. farinosa, and showed flowering the year after germination (n = 250), and all had done that variation in net selection on scape length was largely the so by the third year after germination (data extracted from the result of differences in the relative strength of selection mediated study reported in ref. 34). Geographic variation in biotic inter- by these mutualists and antagonists. This study assessed the causes actions has been hypothesized to drive population differentiation of selection on floral display by experimentally manipulating the in plant traits influencing interaction strength and outcome (2, interaction intensities with both mutualists and antagonists. 12, 40). The present study provides experimental evidence for Differential pollination success favored the long-scaped morph, a causal link between selection mediated by biotic agents and whereas differential grazing damage favored the short-scaped microevolutionary change in natural plant populations. morph. Pollinator-driven selection for tall plants has been dem- The field experiment identified grazers and pollinators as onstrated experimentally in previous studies of P. farinosa (14, major agents of selection, suggesting that factors governing the 30, 31) and other -pollinated plants (e.g., 36, 37), and is relative abundances of mutualists and antagonists are likely to likely to be related to the capacity of pollinators to detect in- influence the evolutionary trajectories of P. farinosa populations. florescences of different height and their willingness to forage The results are consistent with a large literature indicating the close to the ground. The advantage of a long scape during pol- importance of interactions with pollinators (41, 42) and herbi- lination was counteracted by higher probability of grazing dam- vores (5, 43) for the evolution of plant traits, but they also provide age. Because grazers in this area remove the vegetation only a striking example of how land use can influence the evolutionary above ca. 1–5 cm, the inflorescence of the short-scaped morph dynamics of plant populations. Deer and moose may graze runs a lower risk of being eaten than does that of the long-scaped P. farinosa, but cows, horses, and sheep are currently the main morph. Grazer-mediated selection for short scape is consistent grazers in the study area. The grazing pressure in the alvar with positive correlations between plant stature and risk of grazing grasslands is thus largely determined by management decisions. damage in other systems (19), and the prevalence of prostrate The observation of selection favoring the short-scaped morph in growth forms in areas subject to high grazing pressure (38). In many populations and the increase in the proportion of short- line with previous studies (18, 32), the long-scaped morph lost scaped plants in most of the 24 populations monitored over 6 y is likely the result of the recent increase in grazing pressure in the a higher proportion of fruits to seed predators compared with area. Grazing intensity has increased considerably during the last the short-scaped morph in the field experiment. However, in the 15 y. The current high grazing pressure is thus a relatively recent experiment differential seed predation was not large enough to feature of the area, which may explain why the long-scaped outweigh the advantage of the long-scaped morph when grazers morph still dominates in most populations. The experimental were excluded (Fig. 4B), or to result in a clear advantage of the results imply that land use policies may strongly influence the short-scaped morph when plants were pollinated by hand and selection regime and evolutionary trajectories in seminatural grazers excluded (Fig. 4D). Taken together, the results suggest grasslands, and illustrate the need to consider the effects of that among-population variation in direction and strength of management not only for species conservation but also for the selection on scape morph to a large extent can be attributed to maintenance of adaptive genetic variation. spatial variation in the balance between selection exerted by A comprehensive understanding of selection requires that pollinators and grazers. both the targets and agents of selection are identified, and the fi It is likely that the tness advantage of the short-scaped morph genetic basis of adaptive variation documented (4, 22). Here we observed in many populations would be less strong if also male have experimentally both identified the drivers of spatial varia- reproductive success had been considered. production tion in selection among natural populations and demonstrated did not differ between scape morphs, except in one population how this variation translates into changes in the genetic com- (population 81 in 2001; Fig. S1). Because seed predation and position of local populations. Divergent selection on floral dis- most grazing damage occurs only after flowering, these inter- play in the primrose studied was largely explained by spatial actions should influence relative male reproductive success very variation in the relative strength of interactions with mutualist little, and interactions with pollinators are, if anything, likely to pollinators and antagonist grazers, and manipulation of the se- favor the long morph also through male function. lection regime resulted in rather rapid changes in scape morph

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1301421110 Ågren et al. Downloaded by guest on October 2, 2021 frequency. The results indicate that interactions with mutualists that also included population and the population × year interaction as in- and antagonists can drive adaptive differentiation not only across dependent variables. In the study populations, scape morph varied in- broad geographic scales (2, 11, 12), but also among plant pop- dependently of style morph. We had information about style morph for 71% of the plants scored in 2000 (n = 2,576), and for 42% of the plants ulations across relatively short distances. This kind of information = fi is fundamental to link environmental heterogeneity to mosaic scored in 2001 (n 2,833). Preliminary analysis indicated no signi cant effect of style morph or its interactions with scape morph and population on fruit selection and adaptive evolution. production (Table S4), and style morph was therefore not included in the final analysis. For each population and year, we quantified the relative fit- Materials and Methods ness of the short-scaped morph as ln(mean number of intact mature fruits Study System. P. farinosa L. () is a hermaphroditic, self-incompatible, produced by short-scaped plants divided by mean number of intact mature distylous perennial herb found primarily in moist meadows on calcareous fruits produced by long-scaped plants). Positive values indicate a selective ground (44). It is distributed in Europe from central Sweden and Scotland to advantage of the short-scaped morph, whereas negative values indicate the central Spain and Bulgaria (45). On the island Öland, off the southeastern reverse. To quantify relative number of flowers per plant and relative fruit Swedish coast, it occurs as two distinct scape morphs (46). The long-scaped set (proportion of flowers developing into intact mature fruits) of the two morph produces a 3- to 30-cm scape with a smooth surface, whereas the scape morphs in each population, we similarly calculated the natural log of short-scaped morph produces a 0- to 3-cm short, thick and striated scape. In the ratio of the means for the short- and long-scaped morphs. the population survey conducted in 2001, we determined the scape length of 2,442 plants at fruit maturation, and the overlap in scape length was small Field Experiment. To assess experimentally the importance of interactions between plants classified as short- and long-scaped, respectively. Ninety-nine with grazers and pollinators for among-population variation in selection on percent of all long-scaped plants had a scape length of 30 mm or longer scape morph and in scape morph frequencies, we established a field ex- (mean ± SD, 101.4 ± 36.9 mm, n = 1318), whereas 99% of all short-scaped periment in 2004. In each of nine P. farinosa populations located in the plants were shorter than 30 mm (9.7 ± 6.7 mm, n = 1124; Fig. S3). are middle and southern parts of the Great Alvar on Öland, we established two arranged in an umbel at the top of the scape, and flowering takes place in plots, to which one of two experimental treatments was randomly assigned May. In the study area, butterflies (especially Pyrgus malvae) and solitary (exclosure vs. control). The size of the experimental plots varied, depending bees (especially Osmia bicolor) are the main pollinators. The tortricid , 2 2 fi on the density and shape of the local population, from 70 m to 960 m ru ciliana, is the main seed predator, and wild (deer and 2 moose) and domesticated (cattle, sheep, and horses) grazers may consume (median 158 m ). Exclosures were fenced to exclude large mammalian × the entire inflorescence. grazers (120-cm tall fence with mesh size of 15 15 cm), and no damage from grazers was recorded inside exclosures during the experiment. Scape morph frequencies in control plots and exclosures were documented during Morph Frequency Variation. In 2001, we documented the proportion of long- fl fl EVOLUTION and short-scaped plants in 69 P. farinosa populations in a 4 × 10 km large owering in 2004 and in 2012 by determining the scape morph of all ower- area in the northern part of the Great Alvar on Öland, southeastern Sweden. producing plants. In 2006, we added a hand-pollination treatment to the A population was operationally defined as a group of plants that was iso- experimental design in four populations. In these populations, we marked lated from its closest conspecific by at least 50 m. In populations that in- up to 60 long- and 60 short-scaped plants in each experimental plot; half of cluded fewer than 200 flowering plants, the scape morph was recorded for the plants of each scape morph were randomly assigned to supplemental all plants. In larger populations, we recorded the number of long- and short- hand-pollination, while the remaining plants served as open-pollinated scaped plants in transects established across the population. Sample sizes controls. Because pollinating short-styled (thrum) plants by hand without fl fi ranged from 4 to 408 (median = 207). damaging the ower is dif cult and very time consuming, only long-styled (pin) plants were included in the hand-pollination experiment. Populations fl fl Selection on Scape Morph and Changes in Scape Morph Frequencies in Natural were visited regularly during the owering period to ensure that all owers Populations. In populations with at least 10 flowering plants of each morph, on plants in the pollination treatment received supplemental hand-polli- we determined the relative fitness of the two morphs. We marked up to 30 nation with compatible pollen during the period of stigma receptivity. At fl short-scaped and 50 long-scaped plants during flowering in each of 37 fruit maturation, we scored the number of owers, the number of fruits populations in 2000 (2,576 plants in total) and 46 populations in 2001 (2,833 damaged by grazers and seed predators, and the number of intact mature plants in total). At fruit maturation, we recorded the number of flowers fruits produced by each plant and calculated the relative fitness of the short- produced and the number of intact mature fruits (not consumed by grazers scaped morph in the four treatment combinations of each population as or the seed predator) of all marked plants. In addition to intact fruits and described above. We used ANOVA to examine the effects of population, fruits damaged by the seed predator, the pedicels and the dry remains of grazing treatment (exclosure vs. control), pollination treatment (supple- flowers that did not initiate fruit development are still present on the plants mental hand-pollination vs. open pollination), and the grazing × pollination at the time of fruit maturation, making it possible determine total flower treatment interaction on relative fitness of the short-scaped morph (n = 16), production. In five populations, flower and fruit production were estimated and to determine the effects of population, scape morph, pollination treat- in the same manner over a period of 5 y (2000−2004) by marking a different ment, and the population × scape morph interaction on the proportion of set of flower-producing plants each year. We used mixed-model ANOVA to plants outside exclosures whose inflorescence was removed by grazers (n = examine the effects of scape morph (fixed factor), population (random 16), and the effects of population, scape morph, grazing treatment, and pol- factor), and year (random factor; multiyear study only) on the number of lination treatment on proportion of fruits consumed by the seed predator intact mature fruits produced. The statistical significance of random factors (analysis conducted on means calculated for each population × scape morph was determined by using as test statistic the difference between the 2 log × grazing treatment × pollination treatment combination; n = 32). Less than likelihood of the full model and that of a model from which the random half of the plants were subject to supplemental hand-pollination in the factor of interest had been removed (47). To determine whether estimates experiment conducted in 2006, and to minimize the effects of the experi- of selection on scape morph were correlated with changes in morph fre- mental hand-pollination on the relative fitness of the two scape morphs in quencies over time, the frequency of the short-scaped morph in the five the experiment, fruits produced by hand-pollinated plants were removed populations with estimates of fruit production from 5 y was monitored each from the plots. The proportion of hand-pollinated plants varied from 17% to year from 2000 until 2007. To determine whether any consistent change in 31% (median 27%; n = 4) in the control, and from 7% to 38% (median 30%) scape morph frequencies could be observed across a wider sample of pop- in the exclosures in the 2006 experiment. The hand-pollination treatment ulations, we annually recorded scape morph frequencies in large permanent was conducted only in a single year, and should therefore have had a mini- plots in 24 populations from 2006 to 2012 (sample sizes for individual mal impact on the evolution of scape morph frequencies in the experiment. population × year combinations ranged from 2 to 3,057, median 290). The 2 2 plots ranged in size from about 100 m to 2000 m and included all or the ACKNOWLEDGMENTS. We thank Camille Madec, Didrik Vanhoenacker, Tove majority of plants in each population. Trends in scape-morph frequency von Euler, and a large number of dedicated field assistants for help with data change were identified by regressing the proportion of the short-scaped collection, and Martin Breed for R-script. This study was financially supported morph (weighted by the square-root of the sample size) on year in a model by grants from Formas and the Swedish Research Council (to J.Å. and J.E.).

1. Thompson JN (1994) The Coevolutionary Process (Univ Chicago Press, Chicago). 3. Schluter D (2000) The Ecology of Adaptive Radiation (Oxford Univ Press, Oxford, UK). 2. Thompson JN (2005) The Geographic Mosaic of Coevolution (Univ Chicago Press, 4. Mitchell-Olds T, Willis JH, Goldstein DB (2007) Which evolutionary processes influence Chicago). natural genetic variation for phenotypic traits? Nat Rev Genet 8(11):845–856.

Ågren et al. PNAS Early Edition | 5of6 Downloaded by guest on October 2, 2021 5. Marquis RJ (1992) The selective impact of herbivores. Plant Resistance to Herbivores 26. Stinchcombe JR, Rausher MD (2001) Diffuse selection on resistance to deer herbivory and Pathogens, eds Fritz RS, Simms EL (Univ Chicago Press, Chicago), pp 301–325. in the ivyleaf morning glory, Ipomoea hederacea. Am Nat 158(4):376–388. 6. Galen C (1996) Rates of floral evolution: Adaptation to bumblebee pollination in an 27. Gómez JM (2003) Herbivory reduces the strength of pollinator-mediated selection in alpine wildflower, Polemonium viscosum. Evolution 50(1):120–125. the Mediterranean herb Erysimum mediohispanicum: Consequences for plant spe- 7. Nuismer SL, Thompson JN, Gomulkiewicz R (2000) Coevolutionary clines across se- cialization. Am Nat 162(2):242–256. lection mosaics. Evolution 54(4):1102–1115. 28. Agrawal AA, Hastings AP, Johnson MTJ, Maron JL, Salminen J-P (2012) Insect herbi- 8. Herrera CM, Castellanos MC, Medrano M (2006) Geographical context of floral evo- vores drive real-time ecological and evolutionary change in plant populations. Science lution: Towards an improved research programme in floral diversification. Ecology 338(6103):113–116. and Evolution of Flowers, eds Harder LD, Barrett SCH (Oxford Univ Press, New York), 29. Turley NE, et al. (2013) Contemporary evolution of plant growth rate following ex- – pp 278–294. perimental removal of herbivores. Am Nat 181(Suppl 1):S21 S34. fl 9. Elzinga JA, et al. (2007) Time after time: Flowering phenology and biotic interactions. 30. Ågren J, Fortunel C, Ehrlén J (2006) Selection on oral display in insect-pollinated Trends Ecol Evol 22(8):432–439. Primula farinosa: Effects of vegetation height and litter accumulation. Oecologia – 10. Siepielski AM, Benkman CW (2010) Conflicting selection from an antagonist and 150(2):225 232. a mutualist enhances phenotypic variation in a plant. Evolution 64(4):1120–1128. 31. Vanhoenacker D, Ågren J, Ehrlén J (2006) Spatio-temporal variation in pollen limi- 11. Anderson B, Alexandersson R, Johnson SD (2010) Evolution and coexistence of pol- tation and reproductive success of two scape morphs in Primula farinosa. New Phytol – lination ecotypes in an African Gladiolus (Iridaceae). Evolution 64(4):960–972. 169(3):615 621. 12. Benkman CW, Smith JW, Maier M, Hansen L, Talluto MV (2013) Consistency and 32. Vanhoenacker D, Ågren J, Ehrlén J (2009) Spatial variability in seed predation in variation in phenotypic selection exerted by a community of seed predators. Evolu- Primula farinosa: Local population legacy versus patch selection. Oecologia 160(1): 77–86. tion 67(1):157–169. 33. Vanhoenacker D, Ågren J, Ehrlén J (2013) Non-linear relationship between intensity 13. Ehrlén J (1997) Risk of grazing and flower number in a perennial plant. Oikos 80(3): of plant-animal interactions and selection strength. Ecol Lett 16(2):198–205. 428–434. 34. Toräng P, Ehrlén J, Ågren J (2010) Linking environmental and demographic data to 14. Ehrlén J, Käck S, Ågren J (2002) Pollen limitation, seed predation and scape length in predict future population viability of a perennial herb. Oecologia 163(1):99–109. Primula farinosa. Oikos 97(1):45–51. 35. Lenormand T (2002) Gene flow and the limits to natural selection. Trends Ecol Evol 15. Strauss SY, Irwin RE (2004) Ecological and evolutionary consequences of multispecies 17(4):183–189. plant-animal interactions. Annu Rev Ecol Evol Syst 35:435–466. 36. O’Connell LM, Johnston MO (1998) Male and female pollination success in a deceptive 16. Sandring S, Riihimäki M-A, Savolainen O, Ågren J (2007) Selection on flowering time orchid, a selection study. Ecology 79(4):1246–1260. and floral display in an alpine and a lowland population of Arabidopsis lyrata. J Evol 37. Sletvold N, Ågren J (2010) Pollinator-mediated selection on floral display and spur Biol 20(2):558–567. length in the orchid Gymnadenia conopsea. Int J Plant Sci 171(9):999–1009. 17. Parachnowitsch AL, Caruso CM (2008) Predispersal seed herbivores, not pollinators, 38. Diaz S, Noy-Meir I, Cabido M (2001) Can grazing response of herbaceous plants be fl fi – exert selection on oral traits via female tness. Ecology 89(7):1802 1810. predicted from simple vegetative traits? J Appl Ecol 38(3):497–508. 18. Toräng P, Ehrlén J, Ågren J (2008) Mutualists and antagonists mediate frequency- 39. Toräng P, Ehrlén J, Ågren J (2010) Habitat quality and among-population differen- fl – dependent selection on oral display. Ecology 89(6):1564 1572. tiation in reproductive effort and flowering phenology in the perennial herb Primula 19. Gómez JM, Perfectti F, Bosch J, Camacho JPM (2009) A geographic selection mosaic in farinosa. Evol Ecol 24(4):715–729. – a generalized plant-pollinator-herbivore system. Ecol Monogr 79(2):245 263. 40. Zangerl AR, Berenbaum MR (2005) Increase in toxicity of an invasive weed after re- 20. Armbruster WS, Lee J, Baldwin BG (2009) Macroevolutionary patterns of defense and association with its coevolved herbivore. Proc Natl Acad Sci USA 102(43):15529–15532. pollination in Dalechampia vines: Adaptation, exaptation, and evolutionary novelty. 41. Fægri K, van der Pijl L (1979) The Principles of Pollination Ecology (Pergamon, Oxford, – Proc Natl Acad Sci USA 106(43):18085 18090. UK). 21. Reznick DA, Bryga H, Endler JA (1990) Experimentally induced life history evolution in 42. Fenster CB, Armbruster WS, Wilson P, Dudash MR, Thomson JD (2004) Pollination a natural population. Nature 346(6282):357–359. syndromes and floral specialization. Annu Rev Ecol Evol Syst 35:375–403. 22. MacColl ADC (2011) The ecological causes of evolution. Trends Ecol Evol 26(10): 43. Núñez-Farfán J, Fornoni J, Valverde PL (2007) The evolution of resistance and toler- 514–522. ance to herbivores. Annu Rev Ecol Evol Syst 38:541–566. 23. Fishman L, Willis JH (2008) Pollen limitation and natural selection on floral characters 44. Hambler DJ, Dixon JM (2003) Primula farinosa L. J Ecol 91(4):694–705. in the yellow monkeyflower, Mimulus guttatus. New Phytol 177(3):802–810. 45. Tutin TG, Heywood VH, Burges NA, Valentine DH, eds (1972) Flora Europaea (Cam- 24. Sandring S, Ågren J (2009) Pollinator-mediated selection on floral display and flow- bridge Univ Press, Cambridge, UK), Vol 3. ering time in the perennial herb Arabidopsis lyrata. Evolution 63(5):1292–1300. 46. Lagerberg T (1948) Vilda växter i Norden (Natur och Kultur, Stockholm, Sweden), 2nd 25. Mauricio R, Rausher MD (1997) Experimental manipulation of putative selective Ed. agents provides evidence for the role of natural enemies in the evolution of plant 47. Littell RC, Milliken GA, Stroup WW, Wolfinger RD (1996) SAS System for Mixed defense. Evolution 51(5):1435–1444. Models (SAS Inst, Cary, NC).

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1301421110 Ågren et al. Downloaded by guest on October 2, 2021