SEARCH

SEARCH BY CITATION

Keywords:

  • generalist pollination syndrome;
  • hybridization;
  • Piriqueta cistoides ssp. caroliniana;
  • retrospective selection analysis;
  • transgressive floral traits;
  • turneraceae

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Hybridization between closely related lineages is a mechanism that might promote substantive changes in phenotypic traits of descendants, resulting in transgressive evolution. Interbreeding between divergent but morphologically similar lineages can produce exceptional phenotypes, but the potential for transgressive variation to facilitate long-term trait changes in derived hybrid lineages has received little attention. We compare pollinator-mediated selection on transgressive floral traits in both early-generation and derived hybrid lineages of the Piriqueta cistoides ssp. caroliniana complex. The bowl-shaped flowers of morphotypes in this complex have similar gross morphologies and attract a common suite of small insect pollinators. However, they are defined by significant differences in characters that generate pollinator interest and visitation, including floral area and petal separation. In common garden experiments, patterns of pollen deposition in early-generation recombinant hybrids indicate that Piriqueta's pollinators favour flowers with greater area and reduced petal separation. Changes in floral morphology in derived hybrid lineages are consistent with predictions from selection gradients, but the magnitude of change is limited relative to the range of transgressive variation. These results suggest that hybridization provides variation for evolution of divergent floral traits. However, the potential for extreme transgressive variants to contribute to phenotypic shifts may be limited due to reduced heritability, evolutionary constraints or fitness trade-offs.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Limits to expressed genetic variation can constrain potential responses to selection. Although adaptation may occur through selection on standing genetic variation or the appearance of advantageous mutations (Lenski & Travisano, 1994; Barrett & Schluter, 2008; Rutter et al., 2010), hybridization among divergent lineages could introduce more substantial variation that allows evolution of novel phenotypes or colonization of new habitats (Anderson & Stebbins, 1954; Lewontin & Birch, 1966; Cruzan & Arnold, 1993; Arnold & Emms, 1998; Ellstrand & Schierenbeck, 2000; Lexer et al., 2003b; Rosenthal et al., 2008; Shahid et al., 2008; Campbell et al., 2009). Recombinant hybrids from crosses between divergent lineages may express exceptional phenotypes due to transgressive segregation (Rieseberg et al. 1999). In these cases, the range of phenotypic variation is exaggerated such that some recombinant genotypes ‘exceed the limits of variation found in the parental types’ (Stebbins, 1959). Crosses between genetically divergent but morphologically similar lineages are more likely than dissimilar lineages to produce extreme recombinant phenotypes, mostly due to intergenomic epistasis (Rieseberg et al., 1999; Stelkens et al., 2009). Such variation provides the raw materials for diversifying phenotypic evolution (e.g. Tobler & Carson, 2010). Although early-generation recombinant hybrids commonly display levels of phenotypic variation exceeding that of parentals, the potential for this transgressive variation to produce a hybrid lineage that is phenotypically divergent from parental progenitors (i.e. transgressive evolution) in natural systems has been evaluated only rarely (e.g. Lexer et al., 2004; Rieseberg et al., 2007; Grant & Grant, 2009).

Evolution of divergent hybrid lineages might be promoted by transgressive evolution (e.g. Stebbins, 1959; Ranganath & Aruna, 2003; Rieseberg et al., 2007). However, comparisons of selection gradients on transgressive variation in early-generation hybrids to outcomes in derived hybrid phenotypes have seldom been made (e.g. Dudley, 1996). In this study, we test the potential for transgressive variation to lead to evolution of hybrid lineages with floral traits that diverge from parental progenitors. Populations of recombinant hybrids have proven particularly effective for assessing patterns of selection on a wide variety of plant characters (e.g. Jordan, 1991; Lexer et al., 2003a; Erickson et al., 2004; Whitney et al., 2006; Latta et al., 2007; Picotte et al., 2007; Wright & Stanton, 2007), including floral traits in natural environments (Melendez-Ackerman et al., 1997; Schemske & Bradshaw, 1999; Martin et al., 2008). Using recombinant hybrid populations allows for disassociation of phenotypic qualities that co-occur in parental taxa so that the strength and patterns of selection on individual traits can be assessed more effectively (Lexer et al., 2003a; Picotte et al., 2007).

We investigate patterns and consequences of pollinator-mediated selection on floral morphology and evaluate a potential case of transgressive evolution in the Piriqueta cistoides L. ssp. caroliniana Arbo (1985) complex (Turneraceae; P. caroliniana; Wang & Cruzan, 1998; Martin & Cruzan, 1999; Maskas & Cruzan, 2000; Cruzan, 2005; Rhode & Cruzan, 2005; Benz et al., 2007; Picotte et al., 2007) from Southeastern North America (Fig. S1). Flowers of these plants are visited by suites of pollinators with similar ranges of body sizes and foraging behaviours (Anton, 2008), and their shallow, bowl-shaped flowers have consistent differences among morphotypes and derived hybrids. Our preliminary observations suggested that the range of variation in floral traits of early-generation recombinant hybrids can be substantial relative to that observed in parental lineages (Fig. 1). These exceptional phenotypes present an ideal system for testing the potential for transgressive variation to lead to evolution of stable hybrid lineages with divergent trait expression (Ranganath & Aruna, 2003; Stelkens et al., 2009).

image

Figure 1. Examples of the range of variation observed in floral morphology in Piriqueta cistoides ssp. caroliniana. Flowers vary in petal area (a–b) and shape (c–e) due to differences in the degree of separation among petals and the petal width. These pictures include typical flowers found in derived hybrid populations (b) and in the parental morphotype populations (d). Flowers in pictures a, c and e are from recombinant hybrids.

Download figure to PowerPoint

Estimating selection gradients requires measuring both the traits of interest and plant fitness. Several studies assessing pollinator-mediated selection on floral morphology have used pollinator visitation rates as a surrogate for male and female reproductive success (i.e. for rates of pollen removal and deposition on stigmas: Martin et al., 2008; Schemske & Bradshaw, 1999; Melendez-Ackerman et al., 1997; Kudo & Harder, 2005; Mitchell et al., 2004). As pollinators often differ in their effectiveness, individual visits do not necessarily result in equivalent amounts of pollen removal and deposition (Castellanos et al., 2003; Larsson, 2005; Sahli & Conner, 2006, 2007; Theiss et al., 2007; Young et al., 2007). This may be especially true for plants with generalist pollinator syndromes, whose pollinators often represent a wide range of taxa and body sizes (e.g. Kandori, 2002; Anton, 2008; Gomez et al., 2009). In this study, we measure pollen deposition on stigmas, which is generally considered to be correlated with pollinator visitation rate (Kearns & Inouye, 1993; Engel & Irwin, 2003) and male reproductive success (Harder & Thomson, 1989). Furthermore, in P. c. caroliniana, the amount of pollen deposited on stigmas is a good predictor of seed set under greenhouse conditions (Wang & Cruzan, 1998). Under field conditions, seed production is often an inaccurate indicator of fitness for pollinator-mediated selection studies because of fruit depredation by larvae of the gulf fritillary butterfly (Agraulis vanillae; J.R. Ward, personal observation). Quantifying pollen loads requires excision of the style, which may have negative effects on ovule fertilization. Since pollen load size effectively integrates both visitation frequency and pollinator effectiveness, and is correlated with seed set, it provides an adequate assessment of male and female reproductive success and serves as a reliable surrogate for fitness.

The Piriqueta cistoides ssp. caroliniana complex, found in Florida and southern Georgia, includes two parental morphotypes (caroliniana, C; and viridis, V) with distinct vegetative morphologies and habitat associations (Martin & Cruzan, 1999; Maskas & Cruzan, 2000; Picotte et al., 2007). An historical hybridization event between these two morphotypes resulted in a broad hybrid zone that extends over much of the Florida peninsula. The hybrid zone's centre, which extends across nearly 100 km of the Lake Wales Ridge, is occupied by a stable hybrid derivative (the H morphotype: Benz et al., 2007; Picotte et al., 2007; Rhode & Cruzan, 2005). The H morphotype originated from interbreeding between C and V morphotypes as much as 5000 years ago has been in central Florida for at least 20 generations (Cruzan, 2005), and is isolated from parental populations by at least 60 km. These plants display intermediate vegetative traits that are relatively consistent within and among populations (Martin & Cruzan, 1999; Maskas & Cruzan, 2000). The hybrid zone is bounded by a region of sharp clines and high levels of gametic disequilibria (indicating recent admixture) for populations to the south, and a region of higher morphological and marker frequency variation to the north (Martin & Cruzan, 1999; Cruzan, 2005). The variance in vegetative traits among individuals within populations and interpopulation genetic differentiation among populations of the H morphotype are similar to those found for allopatric parental populations.

Flowers of P. c. caroliniana morphotypes typically have three styles, five anthers and five yellow petals arranged into a shallow, bowl-shaped corolla (diameter = 0.5–5 cm) with nectaries at the base of each petal (Arbo, 1995). Individual plants typically produce one or two flowers that last one day and have between 5 and 48 ovules. Under field conditions, flowers open from approximately 0800–1300, at which point petals close around their stamens and carpels. Plants of all three morphotypes and hybrids flower throughout the summer (field conditions) or year-round (greenhouse conditions). The wide array of solitary bees, butterflies and bee-flies observed on Piriqueta flowers in Florida, and the behaviour of these insects, indicate that most visitors forage for nectar (Anton, 2008). The sexual system of P. c. caroliniana is distylous, a form typical for this genus (Ornduff & Perry, 1964; Shore & Barrett, 1985; Arbo, 1995), with long styles above anthers (L morph) or short styles below anthers (S morph). Flowers are self- and intra-morph incompatible, such that pollen transfer from the same flower or the same style morph will typically not result in seed set (Ornduff & Perry, 1964; Wang & Cruzan, 1998).

Here, we tested the hypothesis that the derived hybrid (H) morphotype expresses transgressive phenotypic shifts relative to its parental progenitors. Specifically, we investigated the hypothesis that the derived hybrid lineage has undergone transgressive evolution for floral morphology by asking whether (i) floral traits in early-generation hybrids are transgressive relative to the parental morphotypes and display a range of variation encompassing floral trait expression found in derived hybrids; (ii) there are significant selection gradients for floral traits; (iii) the direction of change in floral traits in the observed in the derived hybrid is consistent with predictions from selection gradients; and, (iv) pollinator behaviours may be affected by aspects of the floral display that are correlated with stigma morphology (i.e. larger flowers may have larger stigmas), and may influence how much pollen is deposited. We compared the floral morphology of the derived hybrid (H) to parental morphotypes (C and V) to assess whether there is evidence for transgressive evolution of floral traits in the derived hybrid lineage.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Generation of recombinant hybrids

Fruits were collected from at least ten plants in at least ten populations of the C and V parental morphotype and derived hybrids (H; Rhode & Cruzan, 2005). Greenhouse plants from field-collected seeds were used to create first-generation hybrids (F1) by crossing all combinations of parental morphotypes (C, V) and the derived hybrid (H; treated as a morphotype for these crosses), including reciprocal crosses (CH, CV, HC, HV, VC and VH, where the first morphotype is the maternal parent), for a total of 50 crosses. Backcross lines (first-generation recombinant lines: BC1) were created by crossing F1 genotypes back to parental genotypes from at least five different populations of each morphotype, for a total of 170 crosses. Crosses among a total of 50 BC1 genotypes were used to produce the BC1F1 generation. Although all recombinant hybrids were used for estimating selection gradients in the field, only hybrids from crosses between parental morphotypes (C,V) were used in tests of transgressive trait expression for comparison to early- and advanced-generation (H) hybrids.

Common gardens planting and experiment design

Multiple vegetative cuttings from early-generation hybrids (F1, BC1, BC1F1), derived hybrids (H) and parental morphotypes (C, V) were made to by rooting short pieces of stem with one leaf each in Jiffy™ peat pellets (Jiffy Products of America Inc., Lorain, OH, USA). In July 2002 and July 2004, established cuttings were transplanted into common gardens at Archbold Biological Station (ABS; Venus, FL, USA), located towards the south end of the hybrid zone, between the distribution of H and V morphotypes. Experiments were conducted in two field common gardens containing around 350 plants each and that had been established several years prior. Gardens were isolated from natural Piriqueta populations by at least 2 km. Genotypes included in these experiments included 354 BC1, 196 BC1F1 and 82 F1 hybrids; 39 derived H hybrids; and 21 and 15 plants of the C and V morphotypes. Genotypes were separated by 15 cm so their foliage and inflorescences did not intermingle, and they were randomized by position within blocks to create locally random mixtures of the hybrid generations and morphotypes. During July 2004 and July 2006, photos of flowers, collection of stigmas and pollinator observations were made in common gardens. Each day, between 0800 and 1100 h, open flowers (one per plant) were placed against a black felt background with ruler for calibration, then photographed with a Nikon Coolpix 5700 digital camera (Nikon Inc., Melville, NY, USA).

Floral morphological traits

Morphological characters of each flower were measured from digital images using Image-Pro software 4.5.1 (Media Cybernetics, Inc., Rockville, MD, USA). Each picture was calibrated by manually resetting the measurement scale on the computer to the ruler displayed in the picture. A grid line framework was generated for each picture so that each landmark on the flower had x and y coordinates, which were used to obtain a two-dimensional quantification of floral characters. An encompassing circle was drawn to touch the outer periphery of the longest petals and surround the entire flower (Fig. 2). Measures taken from the encompassing circle included its area, perimeter and centre point coordinates. The outside edges of petals were manually traced to obtain measures of the total floral perimeter and area. We used an area-weighted centre point function in Image-Pro to obtain the centre of the flower, then compared it with centre of the encompassing circle to assess radial symmetry.

image

Figure 2. Depictions of the measurements taken to quantify variation in floral morphology in parental and hybrid genotypes of the Piriqueta cistoides ssp. caroliniana complex. See text for further details.

Download figure to PowerPoint

The flower's shape, or degree of departure from a circle, was measured in several ways to capture different aspects of petal separation (Fig. 2). First, the corolla perimeter and area were subtracted from the perimeter and area of the encompassing circle (perimeter difference and area difference respectively). For these measures, a greater difference between the encompassing circle and the corolla area or perimeter indicated larger clefts between petals. Area and perimeter differences were divided by total floral area and perimeter of the entire flower to obtain measures that represented a proportion of floral area (proportional floral perimeter difference and proportional area difference respectively). Second, the depth of clefts between petals was measured as the distance from the flower centre to the outermost point of petal overlap (average distance to cleft), which was divided by average petal length to obtain a proportional measure of cleft size (proportional distance to cleft). The average width of petals was measured as the distance between the coordinates of petal overlap points (average petal width) and divided by the radius of the encompassing circle to obtain a proportional measure of petal width (proportional petal width).

Stigma pollen loads

At approximately 1900 h each day, the three styles of each flower were excised at ovary's top with cuticle scissors and collected using forceps. Styles were placed into 1.5 mL microcentrifuge tubes containing 70% ethanol and 3–5 drops of 1% malachite green stain. Styles were mounted onto slides using a drop of corn syrup overlaid by a cover slip. Images of each stigma were captured using a Retiga 1300 Q Imaging Camera® (QImaging, Surrey, BC, Canada) attached to a Leica MZ 16 Stereo Microscope at 50× magnification. Pollen grain counts were made from images by manually marking each pollen grain in Image-Pro. Although most flowers had three styles per flower <1% had four styles, and all grains on all styles were included in total pollen counts. Individual grains were easily distinguished as cases where the entire stigmatic surface was crowded with pollen were rare. Pollen deposition was defined as the total number of pollen grains on all styles, which was log transformed to approximate a normal distribution. Stigmas were generally not saturated with pollen, but in cases of grain overlap, focussing and contrast adjustments were made to discern individual pollen grains. Piriqueta cistoides ssp. caroliniana pollen grains are elliptical in shape with a smooth surface, so pollen grains that were obviously not from P. c. caroliniana could often be distinguished based on their distinct shape and texture; foreign grains were excluded from pollen counts. The amount of pollen deposited on stigmas was analysed during peak flowering season (July) over two summers (2004 and 2006; 9 and 15 days of stigma sampling, respectively, over 2–3 weeks per year), for a total of approximately 700 open-pollinated flowers that included derived hybrids, parental morphotypes and early-generation hybrids from the ABS common gardens. Although only a portion of the pollen deposited on stigmas was compatible, the correlation between pollen load size and compatible pollen in Piriqueta (Feldman, 2008), let us use total load to estimate reproductive success.

Single pollinator visits

Pollen deposition after single pollinator visits was measured to assess the possibility that floral petal characters were correlated with aspects of stigma morphology (e.g. stigma surface area or differences in adherence potential) that influence pollen load size. Single visit observations (ABS, 2006) were used to test the hypothesis that pollen loads after one visit would be correlated with floral traits. Each morning a subset of flowers were bagged before opening. After flowers opened, between one and three trained observers uncovered one to three individual flowers each and observed them for pollinator visits. The type of pollinator and time spent on each flower was recorded. Immediately following a visit, flowers were re-bagged and stigmas collected at the end of the day as above. Stigma loads were quantified and flowers photographed as above.

Data analyses

Floral trait variables (floral area, perimeter, asymmetry, proportional perimeter difference, proportional area difference, average proportional distance from petal overlap to centre of the flower and average width of petals) were analysed using anova (SAS Proc Mixed, version 9.1; SAS Institute Inc., Cary, NC, USA). Residuals from these analyses approximated normal distributions as untransformed values, except in the case of floral area, where values were square-root transformed. Residuals grouped by generation and year displayed similar variances among groups. The effects of style morph (long and short styles), generation (parental morphotypes, C and V; derived hybrids, H; and early hybrid generations, F1, BC1, and BC1F1) and the interaction between style morph and generation were tested as fixed effects. Year was treated as a random effect to reflect the fact that it represented a sample of possible years that could have been included. manova was used to simultaneously evaluate all dependent floral trait variables.

We used a priori contrasts to compare parental morphotypes (C and V) with derived (H) and recombinant hybrids (BC1 and BC1F1, from C × V crosses only). To compare hybrids (early-generation and derived hybrids) with their parental ancestors, contrasts were made for the C and V morphotypes (combined) against derived hybrids, and for the C and V morphotypes (combined) against recombinant hybrid generations (i.e. the C × V BC1 and BC1F1 generations). We used a separate analysis with the same model described for floral traits above that included only recombinant hybrids to test for floral trait differences between hybrids derived from C × V crosses and those derived from crosses with the H morphotype, and to test for the effect of reciprocal crosses (maternal cytoplasmic effects, designated as C or V). Correlations among floral traits were estimated using SAS Proc Corr (version 9.1). Variance components for floral traits within and among recombinant hybrid genotypes and across seasons were estimated using restricted maximum likelihood with SAS Proc Varcomp (version 9.1).

Brown and Forsyth tests (Brown & Forsythe, 1979) were used to test the hypothesis that early-generation hybrids are more variable in floral characters than established parental morphotypes. This test of variance equality uses deviations from the median rather than the mean (i.e. as in the Levene test) and is considered more robust to departures from normality. Tests were implemented for each floral character by calculating absolute values of the deviation of each observation from the median for that parental morphotype or hybrid generation, then testing for differences among average deviations using SAS Proc Mixed (version 9.1). We used a priori contrasts as described above to test differences in the average variances of floral characters for the parental and hybrid morphotypes, and hybrids from crosses between the C and V parental morphotypes.

To assess the strength and pattern of selection for recombinant hybrid (all BC1 and BC1F1 genotypes including those from crosses with the H morphotype) floral traits, the effects of style morph, generation and year were removed in a mixed-model anova using Proc Mixed, and residuals were exported for submission to regression analyses. Analysis of residuals allowed for direct tests of linear and quadratic components of the selection gradient. We tested for selection on traits with multiple regression analyses (Lande & Arnold, 1983; Schluter, 1996) with pollen load size as the dependent variable. Selection coefficients were estimated as standardized regression coefficients, and quadratic regression coefficients were doubled to convert them to quadratic selection coefficients (Stinchcombe et al., 2008). Standardized regression coefficients were estimated in multiple regression models that included the effects of linear and quadratic terms for floral area, asymmetry, proportional perimeter difference, proportional area difference, average proportional distance from petal overlap to centre of the flower and average proportional width of petals between petal overlaps on the residual stigma pollen loads using SAS Proc Reg (version 9.1). The regression model for selection analysis included both linear and quadratic terms for all floral traits to control for the effects of colinearity. Linear regression coefficients were interpreted as directional selection, and negative quadratic coefficients as evidence for stabilizing selection (Lande & Arnold, 1983; Rausher, 1992; Schluter, 1996). We further assessed the shape of selection gradients for local optima using spline analyses (Schluter, 1988) and compared trait expression in derived hybrid populations (H) to expectations based on selection estimated from early-generation recombinant populations. Variation in pollen deposition was also analysed using anova (SAS Proc Mixed, version 9.1). The effects of style morph and generation, and their interaction were tested as fixed effects. Year and date within year were entered as random effects.

Pollen loads after single visits were analysed using anova models that included floral trait covariates along with pollinator type (three different classes of bees, as well as butterflies, and flies: Anton, 2008), pollinator body size (large or small), style morph (long or short) and generation entered as fixed effects using the SAS Proc Mixed (version 9.1). Parental and hybrid generations were effectively randomized across dates by haphazard selection so each would be pollinated by similar sets of pollinators. Since residuals pollen load size analyses were skewed, values were natural log transformed to produce approximately normal distributions.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Morphological variation in common gardens

Floral area and shape varied substantially within and among Piriqueta parental morphotypes and C × V hybrid generations. Flowers in early-generation recombinant hybrids ranged from 0.5 to 5.0 cm in diameter, and floral shapes ranged from nearly round to deeply clefted (Fig. 1, Table S1). Flowers with extreme petal separation are best described by a large difference in floral perimeter or area between the flower and the encompassing circle. There were differences in the way these two measures reflected differences in floral shape, with large area but small perimeter differences associated with flowers having shallow and wide gaps between petals, and flowers with deep clefts having larger differences in perimeter than area. Variation in cleft shape probably explained the moderate correlation between proportional area and perimeter differences (= 0.647, < 0.0001). Cleft shape differences were also captured by measures of average distance from the centre of the flower to the point of petal overlap and by proportional petal width (Fig. 2), both of which displayed moderate associations with area and perimeter differences (= −0.4 to −0.5). This combination of measures appears to comprise an accurate description of shape variation among Piriqueta's bowl-shaped flowers.

There were significant differences in floral area among parental morphotypes, derived hybrids and early-generation hybrids. Flowers of P. c. caroliniana parental morphotypes (C: N = 21; V: N = 15), derived hybrids (H: N = 39) and greenhouse-bred early hybrid generations from C × V crosses (F1: N = 14; BC1: N = 52; and BC1F1: N = 32) differed for several size and shape variables when grown in ABS common gardens (Fig. 3, Table 1). Contrasts confirmed that H flowers were significantly larger than those of C and V morphotypes (Table 1), and indicated that early-generation hybrid flowers were similar in size to those of C and V flowers (Table 1).

Table 1. Results from anova and a priori contrasts (last three rows) for differences in the means for morphological traits in Piriqueta cistoides ssp. caroliniana parental morphotypes and hybrids growing in common gardens. Floral traits were measured as described in Fig. 2 and in the text. Significant sources are in bold
Source of variationd.f.Floral areaAsymmetryPerimeter differenceArea differenceDistance to overlapWidth of petals
Model12F = 2.68F = 0.60F = 2.53F = 2.20F = 3.56F = 3.28
P  = 0.0028 P = 0.8364 P  = 0.0047 P  = 0.0147 P  < 0.0001 P  = 0.0003
Year1F = 4.29F = 0.00F = 6.15F = 0.44F = 0.01F = 0.03
P  = 0.0402 P = 0.9507 P  = 0.0143 P = 0.5097P = 0.9098P = 0.8607
Style morph1F = 0.28F = 0.07F = 0.34F = 0.11F = 0.12F = 0.50
P = 0.5968P = 0.7986P = 0.5611P = 0.7450P = 0.7308P = 0.4789
Generation5F = 4.95F = 0.96F = 3.37F = 3.61F = 1.90F = 1.81
P  = 0.0003 P = 0.4442 P  = 0.0066 P  = 0.0042 P = 0.0983P = 0.1145
Style morph*generation5F = 1.96F = 0.50F = 1.63F = 1.05F = 5.36F = 4.75
P = 0.0887P = 0.7737P = 0.1560P = 0.3905 P  = 0.0002 P  = 0.0005
H vs. C and V1F = 7.76F = 0.16F = 3.01F = 4.31F = 4.35F = 4.07
P  = 0.0061 P = 0.6930P = 0.0850 P  = 0.0398 P  = 0.0389 P  = 0.0455
H vs. recombinant hybrids1F = 19.94F = 0.06F = 7.03F = 15.61F = 0.02F = 0.30
P  < 0.0001 P = 0.8014 P  = 0.0089 P  < 0.0001 P = 0.8808P = 0.5828
C and V vs. recombinant hybrids1F = 3.57F = 0.02F = 1.09F = 4.22F = 3.59F = 2.00
P = 0.0608P = 0.8984P = 0.2988 P  = 0.0417 P = 0.0601P = 0.1595
image

Figure 3. Mean (±1 SE) proportional perimeter difference (compared to the encompassing circle), petal width, distance from centre of the flower to petal overlap, floral area and floral area difference (compared with the encompassing circle) across parental morphotypes (C and V), advanced-generation hybrids (H) and early-generation hybrids of the Piriqueta cistoides ssp. caroliniana complex in the ABS common gardens. Flower area is rescaled by 1/10. See text for further details on these measurements.

Download figure to PowerPoint

Flowers of the derived hybrid (H) had significantly rounder shapes (smaller differences between their areas compared to the encompassing circle), shallower petal clefts, and wider petals than parental morphotypes (Table 1, Fig. 3). Flowers of the advanced-generation H morphotype were significantly larger and had less petal separation (i.e. they had smaller differences in perimeter and area compared with the encompassing circle) than flowers of the recombinant hybrids from crosses between parental morphotypes (Table 1, Fig. 3). Maternal cytoplasmic origin (C or V) did not have significant effects on any measured floral trait (P > 0.05 for all). The manova Wilks' Lambda test was significant among generations, indicating that morphotypes and hybrids differed for the combination of all measured floral traits (F30,546 = 2.02, P = 0.0012).

Between years there were differences in floral measures for area (flowers tended to be larger in 2006) and shape (greater petal separation in 2006, as measured by the area difference; Table 1). No floral morphology differences were observed between style morphs; however, the interaction of style morph and generation did have an effect on two shape variables (Table 1). Examination of interaction plots revealed that, for H and V morphotypes, short-styled flowers had deeper clefts and narrower petals than long-styled flowers (P < 0.05). There were no significant differences in floral symmetry among morphotypes and hybrids in either year.

There were significant differences among individual recombinant genotypes for all measured floral traits except symmetry (P < 0.05 for all traits). Variance among recombinant genotypes ranged broadly from 23% for floral area to 58% for average petal width, and variance among seasons ranged little, from 0% for average petal width to 8% for proportional area difference. For variation among sampling periods within a single season, variation ranged from 3% for floral area to 22% for proportional perimeter difference. The strong differences among recombinant genotypes suggest that a substantial proportion of floral character variation was genetically determined.

Variation around the means of floral characters differed among generations (Table 2). Brown and Forsyth tests indicated differences in variance for all floral morphological variables with the exception of symmetry (Table 2, Fig. 4). The floral traits of parental morphotypes were less variable (more morphologically consistent) than recombinant hybrids for floral area, perimeter difference and area difference, but not for petal width or distance to petal overlap (Table 2, Fig. 4). Derived hybrids displayed similar levels of variability for these floral traits as C and V morphotypes (Table 2), but had marginally lower variation for floral area and higher variation for petal width compared to recombinant hybrids (Table 2, Fig. 4). The manova Wilks' Lambda test was significant, indicating that the generations differed in variability of combined floral measures (F30,550 = 1.70, P = 0.0128).

Table 2. Results from anova and a priori contrasts (last four rows) for differences in the variance in morphological traits (absolute value of the deviation of each observation from the median) among Piriqueta cistoides ssp. caroliniana parental morphotypes and hybrids growing in common gardens. Floral traits were measured as described in Fig. 2 and in the text
Source of variationd.f.Floral areaAsymmetryProportional perimeter differenceProportional area differenceProportional distance to overlapProportional width of petals
Model11F = 1.80F = 0.94F = 1.59F = 1.86F = 1.72F = 2.49
P = 0.0590P = 0.5006P = 0.1085 P  = 0.0494 P = 0.0739 P  = 0.0069
Style morph1F = 1.30F = 0.07F = 0.10F = 1.25F = 0.05F = 0.01
P = 0.2567P = 0.7871P = 0.7471P = 0.2648P = 0.8228P = 0.9406
Generation5F = 3.43F = 0.54F = 2.96F = 3.10F = 2.96F = 4.22
P  = 0.0059 P = 0.7472 P  = 0.0141 P  = 0.0109 P  = 0.0142 P  = 0.0013
Style morph*generation5F = 0.33F = 1.37F = 0.73F = 0.91F = 1.29F = 1.74
P = 0.8964P = 0.2406P = 0.6027P = 0.4743P = 0.2701P = 0.1285
H vs. C and V1F = 2.68F = 0.30F = 1.86F = 2.57F = 0.06F = 0.09
P = 0.1037P = 0.5850P = 0.1752P = 0.1113P = 0.8051P = 0.7670
H vs. recombinant hybrids1F = 4.25F = 0.00F = 1.66F = 2.79F = 2.17F = 3.94
P  = 0.0411 P = 0.9786P = 0.1996P = 0.0972P = 0.1430 P  = 0.0490
C and V vs. recombinant hybrids1F = 15.63F = 0.36F = 8.17F = 12.37F = 3.13F = 2.77
P  < 0.0001 P = 0.5504 P  = 0.0049 P  = 0.0006 P = 0.0789P = 0.0980
image

Figure 4. Absolute deviations from the means (±1 SE) as a measure of variation in proportional perimeter difference (compared with the encompassing circle), petal width, distance from centre of the flower to petal overlap, floral area and floral area difference (compared with the encompassing circle) across parental morphotypes (C and V), advanced-generation hybrids (H) and early-generation hybrids of the Piriqueta cistoides ssp. caroliniana complex in the ABS common gardens.

Download figure to PowerPoint

Pollen deposition after single visits

No floral trait measured had a significant effect on pollen load size after single pollinator visits (P > 0.08 for all traits; = 119 visits) confirming that most of the variation in pollen load size observed on open-pollinated flowers was due to multiple visits by pollinators. The amount of pollen deposited by a single visit was, however, affected by pollinator type (F5,103 = 2.81, P = 0.0203). Floral generation had a marginal effect on visitation (F3,103 = 2.89, P = 0.0604; F1 hybrids had less pollen after one visit), but this was not associated with any specific trait difference.

Daily pollen deposition

Variation in floral characters had significant effects on pollen deposition for the open-pollinated parental and hybrid genotypes in the ABS common garden (F36,520 = 4.71, P < 0.0001, R2 = 0.246 for the full model in Table 3; means for parental and hybrid generations appear in Fig. S2). Foreign pollen grains, observed on approximately 4% of the slides, were not included in total grain counts. Pollen load sizes averaged 105 grains per flower (standard error = 4.15; range = 0–780). There was a significant difference in pollen load size among generations, with parental morphotypes (mean = 85.6, error = 10.8, = 47) and F1 (mean = 82.4, error = 11.3, = 86) hybrids receiving slightly less pollen than recombinant (mean = 111.5, error = 4.8, = 531) and derived hybrids (mean = 108.2, error = 20.8, = 15). Over all generations, long-styled morphs had greater pollen deposition (mean = 126.1 grains, error = 6.45, = 365) than flowers with short styles (mean = 81.6, error = 4.55, = 307; Table 3).

Table 3. Sources of variation in pollen deposition for Piriqueta cistoides ssp. caroliniana growing in common gardens at Archbold Biological Station. The mixed-model anova included fixed effects (year, style morph, and generation), a random effect (year and date within year) and all of the floral characters measuredThumbnail image of

When the strength and pattern of selection on floral traits were assessed for recombinant hybrids (BC1 and BC1F1 genotypes), we found significant selection gradients for both floral area and shape. There was a significant positive relationship between floral area and pollen load size in early-generation hybrids, and spline analysis revealed a gradual, but significant, selection gradient for increased floral area (Fig. 5). Multiple regression analysis indicated that only the linear component of floral area was significant (linear selection coefficient = 0.12, P = 0.0052), suggesting linear selection for larger flowers in recombinant hybrids. In contrast, there appeared to be stabilizing selection for floral shape since there was a significant linear coefficient for perimeter difference (linear selection coefficient = −0.10, P = 0.0121; negative slope indicates selection for less perimeter difference) and a significantly negative quadratic term for area difference (quadratic selection coefficient = −0.30, P = 0.0006). When the selection gradient for proportional area difference was visualized with spline analysis, the selection on floral shape appeared to be primarily linear (Fig. 6); flowers with less petal separation had higher fitness across most of the phenotypic range, with a decrease in fitness for extreme phenotypes. Selection coefficients estimated using only C × V recombinant hybrids were qualitatively similar for floral area (linear coefficient = 0.15) and proportional area difference (−0.27 and −1.94 for linear and quadratic components), but only the linear component of floral area difference was significant (P = 0.0109). C × V recombinant hybrids and C × V × H hybrids spanned a similar range for flower size (95% confidence interval: 1.16–11.28 cm2 vs. 1.88–11.15 cm2), but the range of variation for C × V hybrids was greater for proportional area difference (95% confidence intervals: 0.71–5.61 and 0.67–3.58 cm2 for C × V and C × V × H hybrids respectively). Other floral characters measured, including depth of the petal cleft, petal width, petal overlap and floral symmetry, did not have significant effects on levels of pollen deposition (Table 3). Analyses using untransformed pollen load size data (Stanton & Thiede, 2005) provided results qualitatively similar to those in Table 3.

image

Figure 5. Selection gradient (±95% confidence intervals) for floral area estimated using recombinant hybrids of the Piriqueta cistoides ssp. caroliniana complex in the ABS common gardens. Arrows show the position of morphotype (C and V) and derived hybrid (H) means. Thicker horizontal bars are (±1 SE) and lighter bars are (±95% confidence intervals). The mean for floral area in the derived hybrid is significantly greater than the two parental morphotypes combined (Table 1).

Download figure to PowerPoint

image

Figure 6. Selection gradient (±95% confidence intervals) for petal separation (proportional floral area difference) estimated using recombinant hybrids of the Piriqueta cistoides ssp. caroliniana complex in the ABS common gardens. Arrows show the position of morphotype (C and V) and derived hybrid (H) means. Thicker horizontal bars are (±1 SE) and lighter bars are (±95% confidence intervals). The mean for petal separation in the derived hybrid is significantly lower (i.e. flowers are rounder) than the two parental morphotypes combined (Table 1).

Download figure to PowerPoint

Heterogeneity of slope tests using interactions between the covariate and categorical independent variable to compare interannual variation in effects of floral area and shape on pollen loads were not significant (P > 0.10), indicating that selection was consistent across the two years of this study. Patterns of selection on floral area and shape were statistically consistent among hybrid generations and parental morphotypes (C, V, and H), as heterogeneity of slopes tests for the interaction between these covariates and generation were not significant (P > 0.25 in all cases). Selection gradients obtained for the C, V and H morphotypes combined (linear s = 0.24 for floral area, and s = −0.03 and −0.41 for linear and quadratic selection coefficients; P > 0.05), and for each morphotype individually for floral area (s = 0.20 and 0.64 for C and H, respectively; P > 0.05), were similar to those estimated for the recombinant hybrids. In contrast, individual morphotype estimates for petal separation (proportional area difference) ranged from 0.07 to −0.33 for linear, and from −0.57 to 0.29 for quadratic coefficients for the C and H morphotypes respectively (P > 0.05); sample sizes were inadequate to provide estimates for the V morphotype due to high field mortality rates. It is notable that the smaller sample sizes and reduced range of variation for individual morphotypes compromised the reliability of these estimates. As all parental and hybrid plants were randomly positioned across gardens and we did not detect any behavioural differences in pollinator responses to flowers produced by different morphotypes and hybrids (Anton, 2008), it is reasonable to assume that selection pressures were largely similar. In general, Piriqueta flowers that were larger and had less difference between the floral perimeter or area and the encompassing circle received more pollen.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The derived hybrid lineage of Piriqueta from Central Florida produced flowers that were both larger and had less petal separation (area difference relative to a circle) than parental progenitors. Recombinant C × V hybrids were transgressive compared with parental morphotypes; they displayed high levels of variation for floral area and petal separation, with some recombinant genotypes expressing trait values well outside the range seen in the parental morphotypes. The range of variation in floral morphology found in recombinant hybrids broadly overlapped the variation in the hybrid lineage, suggesting that transgressive segregation in early-generation hybrids provided the variation necessary to allow significant trait transitions in derived hybrids. Plants of the hybrid (H) lineage had increased reproductive success compared with plants from its parental morphotypes, and phenotypic shifts in floral morphology were consistent with selection gradients estimated in the derived hybrid's native habitat. The combined evidence for the origins and isolation of this hybrid lineage (Martin & Cruzan, 1999; Maskas & Cruzan, 2000; Cruzan, 2005), significant differences in its floral traits compared with parental morphotypes, and its higher reproductive success constitute a strong case for a transgressive evolutionary shift as a consequence of hybridization.

On average, flowers of early-generation recombinant hybrids had broader ranges of variation in floral area and shape compared with the parental morphotypes. When we examined the effects of individual floral characters on pollen deposition, we detected significant selection gradients that were consistent across two field seasons, and were mostly consistent across generations. In particular, flowers that were larger with less petal separation had larger pollen loads, and selection gradients for both of these traits displayed incremental fitness increases (based on pollen load size) across the majority of the phenotypic ranges examined. Interestingly, flowers of the early-generation recombinant hybrids had greater average petal separation than parental genotypes (i.e. greater difference from an encompassing circle), indicating that they were transgressive for floral shape in the opposite direction from the selection gradient. The flowers of the derived H hybrids had less petal separation (area and perimeter differences compared with an encompassing circle) than both parental morphotypes, so apparently selection on the wide range of variation in early-generation hybrid flower shape led to significantly less petal separation in the derived hybrids.

Comparing selection gradients estimated in early-generation recombinant hybrids to phenotypic trait differences observed in derived hybrids provides a strong case for the adaptive evolution of floral morphology; the increase in floral area and decrease in petal separation predicted from our selection experiments are consistent with outcomes in the H morphotype. However, there are alternative explanations for these patterns. First, the selection gradients detected may be considered strong by some assessments (e.g. Conner, 2001), but are relatively weak when we consider that populations sizes are relatively small (usually 100 or fewer individuals; personal observation), so selection pressure could be swamped by genetic drift. It is important to note that even though stigmas were not typically saturated, pollinator activity in both years of this study was relatively high; episodes of pollinator limitation could impose stronger selection that would overcome the effects of drift, and would have disproportionate effects on the lifetime fitness of this perennial species. Second, we estimated selection gradients using recombinant genotypes that included crosses to the derived hybrid. However, the range of variation in floral traits and the shape of selection gradients from crosses among all three morphotypes were broadly similar when only the C × V recombinant hybrids were considered. Including all recombinant hybrids provides more robust estimates for the shape of selection gradients. Third, we assume that selection in contemporary populations accurately reflects historical patterns of selection on floral morphology. This does not assume that the specific composition of the pollinator fauna was similar, but rather that there might have been substantial overlap in distributions of their body sizes and behaviorial types. Such an assumption is reasonable because, although species composition of pollinators differed between years, pollinators' body sizes and behavioural types were similar (Anton, 2008); this may explain the similarity in patterns of selection. Piriqueta's generalist pollination syndrome might grant it more resilience to fluctuating pollinator fauna.

Floral biology

Our analysis of the effects of floral traits on pollen loads assumes that variation in the load sizes is influenced by the number of visits and the effectiveness of different pollinator types. An alternative hypothesis would be that developmental constraints tend to produce correlations between the floral traits we measured and aspects of stigma morphology that affect its load capacity (e.g. stigma surface area or ‘stickiness’). To test this hypothesis, we assessed stigma loads following single visits by pollinators in the field and statistically controlled for pollinator type, plant generation and style morph. The amount of pollen deposited during a visit is known to vary (Herrera, 1993; Wilson, 1995) but depends on visitation time and behaviours that increase opportunities for contact with the stigma (Thomson & Plowright, 1980; Thomson, 1986). Likewise, the potential for stigma contact with the pollinator's body should depend on the stigma surface area, so we expected that stigmas with greater capacity should be able to capture more grains from pollinators during a visit. However, we did not detect any significant effects of floral traits on pollen loads after single visits, suggesting that the effects of floral traits on stigma loads is not related to stigma morphology.

Even though open-pollinated flowers received multiple visits during the course of these experiments, the large majority of stigmas did not have adequate numbers of pollen grains for maximum seed production. Under greenhouse conditions, these plants require at least 25 grains per ovule to achieve full seed set (Wang & Cruzan, 1998). Since plants typically have between 15 and 30 ovules, the average number of pollen grains received was much less than the number needed for full seed set, and more than 95% of flowers had fewer than 375 grains, the number needed for full seed set for a flower with 15 ovules. Even though pollinator activity was relatively high not all visitors had extensive contact with the anthers and stigmas while foraging, and smaller bees tended to have few hairs that could facilitate pollen transfer (personal observation). The pollen limitation observed during our sampling period also validates our usage of pollen loads as a surrogate for fitness; the relationship between the number of grains and seed set is linear over most of the load size range seen in this study (Wang & Cruzan, 1998).

Transgressive evolution

Our results are consistent with the hypothesis that transgressive variation in early-generation recombinant hybrids has led to evolution of a phenotypically divergent hybrid lineage in Central Florida. Cases of transgressive variation may involve more equal intergenomic mixing in the hybrid lineage (as in the case of homoploid hybrid speciation in sunflowers; Rieseberg et al., 1996) compared with a smaller degrees of genome mixing in cases of adaptive introgression as evidenced by selective sweeps (e.g. Jiggins, 2003; Evans et al., 2006; Gompert et al., 2008; Morgan et al., 2010; Dopman, 2011) or divergent clines between neutral and selected markers (Gompert & Buerkle, 2009; Nolte et al., 2009). The consequences of hybridization may differ for these cases of introgression if there were limited phenotypic variation or lack of reproductive isolation in early-generation hybrids. Although our understanding of the genetic basis of transgressive variation is rudimentary, there appears to be a window of intermediate genomic divergence that may be more likely to generate the transgressive variation required for more pronounced evolutionary transitions.

The transgressive expression of floral area and shape observed in the derived hybrid morphotype is consistent with the hypothesis that hybridization may serve as a stimulus for the evolution of divergent traits. This hybrid derivative Piriqueta morphotype appears to be a solid example of transgressive evolution since (i) the origin of the H morphotype is well supported by genetic marker analyses (Martin & Cruzan, 1999; Maskas & Cruzan, 2000; Cruzan, 2005), (ii) it displays superior reproductive success in its native habitat as measured by pollen load size and (iii) geographical distance from parental populations should provide adequate isolation to allow evolutionary independence. This argument is bolstered by observed increases in floral area and roundness (less petal separation) in the derived hybrid compared with the parental morphotypes, results consistent with predictions from selection gradients estimated from early-generation recombinant hybrids. A number of questions remain concerning the potential for transgressive segregation to promote substantive evolutionary transitions. However, since transgressive expression due to standing genetic variation is thought to be largely a product of intergenomic epistasis (Rieseberg et al., 1999; Stelkens et al., 2009), it is likely that a limited proportion of the variation observed in early-generation hybrids is heritable (Turelli & Barton, 2006; van Heerwaarden et al., 2008). It is also possible that mutations or epigenetic modifications could contribute to the production of novel phenotypes (Sturtevant, 1939; Rieseberg et al., 1999; Michalak, 2009). Comprehensive trait assessments coupled with genomic analyses may provide additional insights into the genetic architecture of transgressive variation and its potential to precipitate substantive trait changes and the evolution of novel phenotypes.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

We thank B. Benz, A. Henderson and J. Picotte for technical assistance in the field and the lab. We thank Archbold Research Station, especially E. Menges and the Plant Ecology Lab for use of equipment and expertise, and M. Deyrup for pollinator identification. This manuscript benefited from the comments of T. Cheeke, S. Eppley, M. Fishbein, R. Mitchell, A. Ramakrishnan and P. Sochacki. We appreciate funding provided by National Science Foundation NSF grant DEB-0080437 to MBC and Portland State University Forbes-Lea Grant to KAA.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
jeb12083-sup-0002-TableS1.docxWord document15KTable S1 Means and standard errors (below means) for flower morphological traits of parental (C and V) and derived hybrid (H) morphotypes, and for the F1, first generation backcross (BC1), and first generation crosses among BC1 plants (BC1F1) in Piriqueta cistoides ssp. caroliniana growing in common gardens at Archbold Biological Station.
jeb12083-sup-0001-FigS1-S2.pdfapplication/PDF91K

Figure S1 Geographical distribution of Piriqueta cistoides ssp. caroliniana parental caroliniana (C) and viridis (V) morphotypes, and their derived hybrids (H) in southeastern North America.

Figure S2 Mean pollen deposition in common gardens over two seasons for Piriqueta cistoides ssp. caroliniana parental morphotypes (C and V), derived hybrids (H) and three generations of hybrids.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.