SEARCH

SEARCH BY CITATION

Keywords:

  • constraints;
  • hierarchical selection analysis;
  • intrafloral integration;
  • natural selection;
  • phenotypic integration;
  • variance–covariance structure

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • • 
    Floral integration has been deemed an adaptation to increase the benefits of animal pollination, yet no attempts have been made to estimate its adaptive value under natural conditions.
  • • 
    Here, the variation in the magnitude and pattern of phenotypic floral integration and the variance–covariance structure of floral traits in four species of Rosaceae were examined. The intensity of natural selection acting on floral phenotypic integration was also estimated and the available evidence regarding the magnitude of floral integration reviewed.
  • • 
    The species studied had similar degrees of floral integration, although significant differences were observed in their variance–covariance structure. Selection acted on subsets of floral traits (i.e. selection on intrafloral integration) rather than on the integration of the whole flower. Average integration was 20% and similar to the estimated mean value of flowering plants.
  • • 
    The review indicated that flowering plants present lower integration than expected by chance. Numerical simulations suggest that this pattern may result from selection favouring intrafloral integration. Phenotypic integration at the flower level seems to have a low adaptive value among the species surveyed. Moreover, it is proposed that pollinator-mediated selection promotes the evolution of intrafloral integration.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

One of the major evolutionary transitions, the origin of multicellularity, also marked the beginning of the process of phenotypic integration (Maynard-Smith & Szathmáry, 1995; Pigliucci & Preston, 2004). The transition from unicellular to multicellular organisms favoured the evolution of cells that differentiate and specialize in reproductive or vegetative survival-enhancing functions (Michod et al., 2006). Eventually, this process resulted in the evolution of phenotypically and functionally integrated complex organs that depend on each other for the successful maintenance and reproduction of multicellular organisms. The magnitude (the degree to which the traits are tied) and the pattern (the arrangement of the relationships among traits) of covariation among sets of functionally related traits have been proposed to be a measurable estimate of phenotypic integration (Arnold, 1992; Pigliucci, 2003; Armbruster et al., 2004). Phenotypic integration results from the simultaneous occurrence of historical, physiological, developmental and adaptive sources of variation (Armbruster et al., 2004). For example, phenotypic integration within populations may arise from genetic correlations or functional linkage, while covariation of traits across populations or species results from selection acting on functionally related traits, when traits respond in a similar manner under a selective pressure that varies geographically or temporally, and/or when traits respond to several selective pressures that themselves covary (Endler, 1995; Armbruster & Schwaegerle, 1996).

In this context, the flower, as a modular structure, represents a clear example of a set of related traits that shows genetic constraints (Armbruster et al., 2004), developmental and functional linkage (Olson & Miller, 1958; Weberling, 1989, Murren, 2002) and adaptive phenotypic integration (Berg, 1960; Bell, 1985; Armbruster, 1991; Anderson & Busch, 2006; Pérez et al., 2007). Flowers are complex organs composed of several interrelated modules (calix, corolla, androecium and gynoecium) that mediate the interaction with pollinators. Accordingly, scenarios of strong selective pressures imposed by pollinators (Stebbins, 1950, 1970; Campbell, 1989; Galen, 1989; Herrera, 1993) should favour the association among floral attributes (adaptive floral integration; for a review, see Ashman & Majetic, 2006) that better matches the pollinator's morphology and behaviour (Nilsson, 1988; Alexandersson & Johnson, 2002). This scenario was first envisioned by Darwin (1862), who proposed that insects with long proboscises should pollinate flowers with deep tubes. Almost a century later, Berg (1960) formalized these ideas and proposed that floral integration should be higher in plants with specialized rather than generalized pollination systems. Since then, the idea that flowers are highly integrated organs promoting an efficient pollen transfer through their interaction with pollinators has become a paradigm among floral biologists (Stebbins, 1950, 1970; Faegri & van der Pijl, 1966).

The initial approach to the study of floral integration focused on disentangling the relative importance of constraints and selection on the patterns of variation in the magnitude of phenotypic integration (Armbruster, 1991; Armbruster & Schwaegerle, 1996). These pioneer studies used the comparative method and were based on the idea that differences in the magnitude of integration among groups at the same taxonomic level are indicative of an adaptive component, while similarities would suggest that genetic/developmental constraints might be more important. For instance, Murren et al. (2002) found different degrees of phenotypic integration among closely related species of Brassicaceae, suggesting a possible role of selection in the diversification of floral integration. Similarly, selection has also been invoked as one of the sources of diversification in floral integration among Dalechampia species (Armbruster et al., 2004). To date, the relative importance of these two sources of variation on the patterns of phenotypic integration remains a major question in evolutionary biology (Arnold, 1992; Pigliucci & Preston, 2004; Frey, 2007).

Although the evidence supporting one of the premises for floral integration to evolve, the functional adjustment between floral architecture and pollinator morphology, is abundant (Nilsson, 1988; Muchhala, 2006), there is little evidence showing that natural selection promotes the evolution of combinations of floral traits (floral integration). In a recent review, Kingsolver et al. (2001) found that only three out of eight studies aiming to measure the correlated selection on pairs of floral traits obtained a significant result. Since then, only one additional study reported the presence of correlational selection on floral traits (Benítez-Vieyra et al., 2006). This failure to detect correlational selection may be the result, at least in part, of the large sample sizes necessary to detect a significant effect of selection on combinations of traits (Conner, 2001). Overall, the lack of available evidence is thus insufficient to assess whether natural selection favours the correlation among floral traits, and thus the evolution of floral integration (O’Connell & Johnston, 1998; Caruso, 2000; Gómez, 2000; Maad, 2000; Benítez-Vieyra et al., 2006).

Lande & Arnold (1983) proposed the use of principal components analysis (PCA) as an alternative approach to study selection on highly correlated traits. PCA transforms a number of correlated variables into a smaller number of independent variables (the so-called principal components), each representing a particular linear combination of the original variables (Kleinbaum et al., 1988; Marcus, 1990; Reyment & Jöreskog, 1993). In this sense, principal components can be visualized as the different Pleiades of correlation (sensu Terentjev, 1931) that make up the phenotype of an organism. Analyses of selection applied to the scores of principal components estimate the strength and pattern of selection acting on these Pleiades (i.e. intrafloral integration sensu Herrera et al., 2002). Although the former approach has been used to measure simultaneous selection on several floral traits (Campbell, 1989; Galen, 1989), it has not yet been used to test the adaptive value of phenotypic integration (or intrafloral integration). Accordingly, a thorough test of the adaptive value of floral integration should include the estimation of the magnitude and direction of the selective pressures acting on this complex trait. Furthermore, the comparison of selection acting on individual traits, on intrafloral integration and on the integration of the whole flower would help to assess if natural selection influences the evolution of epistatic/pleiotropic interactions among floral traits. In the present study, we exemplify the use of a selection analysis at different levels of trait organization (hierarchical selection analysis) as a tool to understand the evolution of phenotypic integration.

In this study we describe the pattern and the extent of phenotypic integration among floral traits of four species of Rosaceae, and estimate the strength of natural selection acting on individual floral traits, scores from principal components (intrafloral integration) and the whole-flower integration. We also performed a review of the available literature to describe the distribution of floral integration values among flowering plants. Finally, we complemented our study with a numerical simulation to examine the relationship between intrafloral and whole-flower integration.

Materials and Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study site

Fieldwork was carried out at the ‘Parque Natural de Las Sierras de Cazorla, Segura y Las Villas’, Jaén Province, in the southeast of Spain, within the Guadahornillos Valley. This 21 400 ha reserve ranges in elevation between 700 and 2000 m asl. A detailed description of the vegetation of the Guadahornillos Valley can be found in Herrera (1984).

Study species

We used four species of Rosaceae (Prunus mahaleb, P. prostrata, Crataegus monogyna and Sorbus torminalis), all belonging to the ‘Maloidea’sensu lato clade proposed by Potter et al. (2002). Within Maloidea, the genus Prunus is an independent ancient clade (hereafter ‘Prunoid’ clade) with respect to C. monogyna and S. torminalis, which belong to a sister clade (‘Maloid’ clade). The four species are deciduous and produce hermaphroditic flowers (Herrera, 1982). Prunus mahaleb is a small (≤ 6 m high) tree producing corymbs with three to 10 white to light-beige flowers, typically visited by 41 species of bees and flies (Jordano, 1993). In southeastern Spain, populations of P. mahaleb are functionally gynodioecious and no differences have been detected in the morphology of hermaphrodite and male-sterile flowers (Jordano, 1993). Although hermaphrodites are self-compatible, cross-pollination is required for successful reproduction (Jordano, 1993; García et al., 2005). Prunus prostrata is a small (≤ 1 m high) shrub with pink tubular solitary flowers mostly visited by Bombus sp. (C. A. Domínguez, unpublished). Crataegus monogyna is a self-compatible small (≤ 10 m high) thorny tree with corymbs of nine to 18 white flowers; its main pollinators are Apis mellifera (Hymenoptera: Apidae) and several other Apidae, calliphorid and syrphid flies (Garcia & Chacoff, 2007). Previous examinations indicated that cross-pollination significantly increases seed production in this species (Guitián & Flores, 1992). Sorbus torminalis is a self-incompatible tree (≤ 10 m high) producing compound corymbs of 30–50 white flowers. Pollinators are mainly small flies, bees and beetles (Oddou-Muratorio et al., 2006). Unlike P. prostrata, which apparently is only visited by Bombus sp., all other species have several insect visitors, suggesting that these plants present a generalized pollination system.

Data collection

From February to April 1991, before the onset of the flowering season, two populations per species were selected in the Guadahornillos Valley and all plants within each population were marked with permanent tags. At the beginning of the flowering season (May–June), a subsample of reproductive individuals of each species in every population was selected. For S. torminalis, 27 plants were selected at Hoyo de Muñoz, and 30 at Roble Hondo. For the other species, populations were located at the Calarillo and Correhuelas sites (25 and 30 individuals for P. mahaleb; 30 and 25 for P. prostrata; 30 and 30 for C. monogyna, respectively).

From each reproductive plant we randomly collected 10 inflorescences (flowers in the case of P. prostrata), from which two flowers were also randomly selected, resulting in a sample of 20 flowers from each plant (five flowers per plant were sampled for P. prostrata because of the low availability of flowers) that were filmed with a portable video camera (VM-H1000P, Sanyo). A total of 3700 flowers were recorded (1100 for P. mahaleb, 275 for P. prostrata, 1200 for C. monogyna, and 1125 for S. torminalis). Measurements were taken directly from individual frames in a TV monitor using a ruler (mm). For each image we measured corolla diameter (CD), petal length (PL), style length (SL), gynoecium width (GW), hypanthium width (HW), and hypanthium length (HL). Herkogamy (HK) was calculated as the average distance between five anthers and the upper side of the stigmatic surface (Fig. 1). Petal length was the average of the lengths of five petals. Plant size was estimated as the sum of the diameters of all trunks at 20 cm from the ground. It has been shown that this measure is highly correlated with fruit production (Jordano, 1993). From May to September 1991, female fitness (fruit production) was estimated for all the studied plants except for P. prostrata which was excluded owing to its meagre fruit production. Because all studied species have between one and 2.5 seeds per fruit (Herrera, 1982), fruit production is a reliable estimate of maternal fitness.

image

Figure 1. Schematic representation of floral traits measured in the four Rosaceae species (Crataegus monogyna, Sorbus torminalis, Prunus mahaleb and P. prostrata). PL, petal length; CD, corolla diameter; SL, style length; HK, herkogamy; HL, hypanthium length; HW, hypanthium width; GW, gynoecium width.

Download figure to PowerPoint

Data analysis

Phenotypic variation in floral traits  To compare the amount of variation in floral traits among clades, species and populations, raw data were averaged for each individual plant. Except for HK (which was transformed to loge(x + 1)), all variables were transformed as loge(x). The variation in floral traits was analysed using a nested MANOVA (type III SS), without including plant size as covariable, because in a previous model this effect was not significant (Wilks’λ = 0.957, F7,207 = 1.341, P = 0.23). These analyses were performed with STATISTICA (StatSoft, Inc., 2004).

Floral integration and variance–covariance structure  Floral integration was measured by calculating the index of phenotypic integration (INT) proposed by Wagner (1984). This index measures the variance among the eigenvalues of the phenotypic or genetic correlation matrix (V(λ); Cheverud et al., 1989; Herrera et al., 2002). Each eigenvalue represents the amount of variation accounted for by a given principal component. A high variance among eigenvalues indicates high degrees of integration because most of the phenotypic or genetic variation is accounted for by the first principal components (i.e. strong association among traits). Given that we measured several flowers from each plant, we were able to calculate individual INT values. The variance in floral integration among clades, species (clades) and populations (species, clades) was then partitioned by means of nested ANOVA. We also calculated the magnitude of floral integration for each taxonomic level (clade, species and populations) (Wagner, 1984; Cheverud et al., 1989; Herrera et al., 2002) and bootstrapping was used to calculate their 95% confidence intervals (S-PLUS, Mathsoft, Inc., 1999). Because the number of plants varied among populations, we applied a correction proposed by Wagner (1984) (also see Cheverud et al., 1989).

To examine the variation in the structure of covariation among floral traits, we calculated variance–covariance phenotypic matrices for each population (Supporting Information, Table S1). Then we examined the variation in the phenotypic variance–covariance matrix structure (hereafter VCS) across clades, species and populations following the Jackknife-MANOVA method proposed by Roff (2002) and implemented in S-PLUS (Mathsoft, Inc., 1999). Results from this analysis were not affected by the inclusion of plant size as a covariate, thus we present the results without this variable.

Hierarchical selection analysis (HSA)   We assessed the relative importance of natural selection acting on: (i) individual traits, (ii) intrafloral integration (scores from principal components), and (iii) the integration of the whole flower (i.e. individual INT values). Selection analyses were carried out for each of the two populations of C. monogyna, S. torminalis and P. mahaleb for which we had fitness estimates. Before the analyses, individual floral traits and INT values were standardized to x̄ = 0 and σ = 1 (Lande & Arnold, 1983). Individual fitness (wi) was relativized to the population mean fitness (w̄) (Lande & Arnold, 1983). Sample sizes ranged from 25 to 30 plants per population per species. Because previous analyses showed that plant size was positively related to fitness (adjusted R2 = 0.07 to 0.47, 0.0001 < P < 0.09), this variable was included as a covariate in all analyses. For each level of analysis, linear (directional) and nonlinear (stabilizing/disruptive) selection gradients were obtained from multiple regressions (JMP version 5.1, SAS Institute, 2005) following the standard procedure proposed by Lande & Arnold (1983).

In this study we used PCA applied to floral morphology to determine whether the flower is composed by one or several correlation Pleiades (i.e. intrafloral integration). Examination of the magnitude and sign of each PC factor loading can be further used to identify the different subsets of correlated variables (Pleiades) and infer its potential functional role. Finally, detecting selection on the scores derived from principal components would be indicative of the adaptive value of intrafloral integration. Selection analyses on intrafloral integration were performed on the scores derived from each of the first three principal components (which jointly explained c. 80% of the variation in the floral phenotype). For these analyses the scores from each principal component were considered as independent variables. Scores for each species were obtained from three independent PCAs, since no differences were detected in the VCS at the population level (see the Results section). Although we acknowledge the low power of the selection analyses (because of the relatively small sample size), the main goal of the HSA is to determine which levels are the target of natural selection for a given sample size. In the present study, the term HSA is not used in the context of kin selection or group selection as in previous studies (Heisler & Damuth, 1987).

Floral integration among angiosperms   We carried out a literature search in the ISI Web of Science using the following keywords: floral, morphology, traits, correlation, phenotypic and integration for all available document types, languages and years (1995–2007). We searched again for new data in the cited references of the resulting list of papers. The final dataset included all the studies providing INT values and/or correlation matrices of floral traits. When only correlation matrices were found, we calculated the INT value (n = 19 papers). Finally, we calculated the percentage of mathematically maximum possible integration (hereafter % maximum possible integration) by scaling INT values on the number of measured traits for each entry from our database (Herrera et al., 2002). In order to evaluate whether the observed distribution of INT is indicative of a consistent trend favouring floral integration, we generated an expected distribution of INT values based on randomly generated correlation matrices. Random matrices indicate all possible associations among traits of hypothetical organisms with no constraints in its VCS. Thus, our null model assumes that the magnitude of INT is not conditioned by either adaptive or developmental genetic or physiological constraints. Integration values higher than those produced by chance would indicate that evolution favoured floral integration or that floral structure is constrained by functional and/or phylogenetic constraints that limit the exploration of a wider phenotypic morphospace. Lower values would indicate that selection has probably acted against floral integration (perhaps favoured intrafloral integration) to provide more plasticity to the phenotypic expression of the flower. Finally, because the observed distribution of INT was composed of an uneven frequency distribution of matrices of different dimensions (three to 15 characters), the expected distribution was built controlling for this source of variation. For each of the observed dimensions of matrices, we generated 1000 matrices whose entries were randomly chosen from a uniform probability distribution of correlations. From each set of random matrices corresponding to a given dimension, we selected the number of matrices corresponding to the relative frequency of each matrix dimension in the observed distribution of INT values. In this way, the expected distribution has the same relative frequencies of matrices of different dimensions as the observed distribution. The observed distribution of INT values was contrasted against the expected distribution using Kolmogorov–Smirnov (KS) and median tests.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Phenotypic variation in floral traits

Results from nested MANOVA revealed significant differences at all levels (clade, Wilks’λ = 0.023, F7,208 = 1243.6; species, Wilks’λ = 0.071, F14,416 = 81.9; populations, Wilks’λ = 0.386, F28,751.4 = 8.1; P < 0.0001 in all cases). Univariate ANOVAs further indicated that HK and GW differed at the clade level, and that HL differed at the species (within clade) level. By contrast, significant differences were detected in all traits between populations within species (Table 1). Most of the variation at the clade level was related to floral traits related having to do with fruit development, such as GW, HW and SL. By contrast, most of the variation at the population level was related to floral traits possibly involved in pollinator attraction and mating success, such as PL, CD and HK (Table 1).

Table 1.  Results from univariate nested ANOVAs testing for the effect of clade, species (clade), and population (species, clade) on seven floral traits in four Rosaceous species (Crataegus monogyna, Sorbus torminalis, Prunus mahaleb and P. prostrata)
TraitCladeSpeciesPopulationsError
MSFProportion of variance (%)MSFProportion of variance (%)MSFProportion of variance (%)MSProportion of variance (%)
  1. PL, petal length; CD, corolla diameter; SL, style length; HK, herkogamy; HL, hypanthium length; HW, hypanthium width; GW, gynoecium width; MS, mean square. Degrees of freedom for clade, species (clade), populations (species, clade) and error were 1, 2, 4, and 214, respectively. *P < 0.05; **P < 0.01; ***P < 0.001.

PL0.911.5415.90.594.620.40.199.2***8.80.1351.6
CD1.114.1623.80.262.411.20.119.2***9.20.1253.4
SL8.8612.766.10.695.010.40.149.4***4.10.1323.1
HK0.2886.4*12.10.000.00.30.1624.3***27.90.1261.5
HL48.16.472.57.51180***22.60.042.6*0.20.135.0
HW5.76.953.50.826.115.30.1310.8***5.00.1324.6
GW27.3293**84.10.090.70.60.127.4***1.50.1411.2

Floral integration among Rosaceae species

The magnitude of phenotypic integration was significantly different from zero in all populations, species and clades (INT range = 0.86–2.38). Table 2 presents INT mean values and confidence limits for each population, species and clade. Nested ANOVA revealed that the variance in the magnitude of phenotypic integration (INT) was not explained by differences between clades, species (clades) or populations (species, clades) (all F < 1.68, P > 0.28, n = 170). The percentage of variance accounted for by all the three effects was less than 2% of the total variance.

Table 2.  Mean estimates of the integration index (INT) per population, species and clade in four Rosaceous species. Subindices indicate populations within species
 Mean INT values (95% confidence limits)
PopulationsSpeciesClades
  1. 95% confidence limits were obtained after bootstrapping.

C. monogyna11.267 (0.894–1.590)1.320 (0.883–1.787)1.236 (0.981–1.520)
C. monogyna21.279 (0.664–1.869)
S. torminalis12.382 (1.774–2.724)1.844 (1.264–2.222)
S. torminalis20.907 (0.510–1.160)
P. mahaleb10.860 (0.500–0.967)1.004 (0.768–1.116)1.301 (1.052–1.480)
P. mahaleb20.991 (0.700–1.150)
P. prostrata11.572 (0.944–2.000)1.567 (1.022–2.049) 
P. prostrata21.165 (0.648–1.470)

Variation in the variance–covariance structure

Despite the fact that we did not observe differences in the magnitude of floral integration (INT), the VCS showed marked differences between clades (Wilks’λ = 0.593, F28,187 = 4.5, P < 0.0001) and species (Wilks’λ = 0.426, F56,374 = 3.5, P < 0.0001), but not between populations (Wilks’λ = 0.541, F112,745.3 = 1.1, P = 0.21). The VCS at all levels was characterized by higher variances than covariances (see Table S1).

Hierarchical selection analysis

We detected selection acting on individual traits and on intrafloral integration (i.e. using principal components scores), but found no evidence of selection on individual INT values. Multiple regression analyses indicated that natural selection acts directly upon individual flower traits in S. torminalis. Natural selection favoured a decrease in style length (linear selection gradient, β (SE hereafter) = −1.18 (0.435), F1,15 = 7.35, P = 0.014), HK (β = −0.64 (0.307), F1,15 = 4.39, P = 0.051), and GW (β = −0.73 (0.231), F1,15 = 9.97, P = 0.005) in S. torminalis. We found no evidence of selection currently acting on individual floral traits of P. mahaleb.

Selection analyses on intrafloral integration (scores from principal components) detected selection acting on P. mahaleb and S. torminalis. In P. mahaleb, we detected stabilizing selection acting on PC1 in population 1 (nonlinear selection gradient,γ = −0.12 (0.07), F1,17 = 4.69, P = 0.044), and directional selection on PC1 and PC2 in population 2 (PC1, β = 0.11 (0.07), F1,25 = 5.89, P = 0.022; PC2, β = −0.28 (0.13), F1,25 = 6.49, P = 0.017). The PCA for P. mahaleb indicated that the higher factor loadings on PC1 were all positive and related with pollinator attraction (PL, CD and SL). The higher loadings on PC2 corresponded to HK and SL. PC1 and PC2 together accounted for 65% of the variation in flower morphology. The overall selective pattern for P. mahaleb indicated that natural selection favoured a reduction in SL and HK, while traits associated with pollinator interaction were kept at intermediate values.

In S. torminalis (Pop 1) we found a significant negative linear selection gradient acting on PC3 (β = −0.80 (0.33), F1,22 = 7.10, P = 0.014). Also, stabilizing selection was detected on PC1 (γ = −0.10 (0.05), F1,18 = 4.68, P = 0.043) and disruptive selection on PC3 (γ = 1.29 (0.35), F1,17 = 10.76, P = 0.003). Although we found a disruptive selection gradient on PC3, its significance was explained by a slight curvature of the fitness function at the end of the positive range of PC3 score values. Since we have no a priori reason to eliminate these values, and the number of points involved was < 10% of the sample, the positive value of γ we found can be interpreted as consistent with the significant negative directional selection detected on PC3. Except for HK, all factor loadings of PC1 were high and positive, indicating that this component is related to flower size. Gynoecium width had a strong positive loading on PC3. PC1 and PC3 accounted for 68% of the variation in flower morphology. Accordingly, these results indicate that selection acting on S. torminalis favours flowers of intermediate size and a reduction in GW. We found no evidence of selection acting at any level on C. monogyna.

Floral integration among flowering plants

We found 36 species from which 80 INT values were obtained. This information derived from 55 studies spanning 12 orders, 16 families and 26 genera. The range of INT values per species was 1–3, except for Dalechampia scandens (n = 11), Helleborus foetidus (n = 9) and Lythrum salicaria (n = 6). With the exception of Rhyzophora mangle (Domínguez et al., 1998) and the Rosaceae species included in this study, all previously published data correspond to herbaceous plants. Overall, the studies included in this review estimated the INT value with three to 15 floral characters (a complete list of the species and INT values is available in Table S2). Although the low number of studies on floral integration prevented a thorough account of the possible phylogenetic effects, there were no differences among the frequency distributions of INT values constructed using population, species, genera or family values. This suggests that the observed distribution of INT values is unlikely to be biased by phylogenetic effects (all P > 0.1 after KS tests).

The observed distribution based on population values is left-skewed, indicating that most species had low integration (Fig. 2). The lowest value (INT = 0.69%) corresponds to Fragaria virginiana (Rosaceae) while the highest (INT = 73.23%) to Trillium grandiflorum (Liliaceae). Most values (n = 33) fall between 10 and 20% of the maximum possible integration. The four Rosaceae species examined in this study had an average integration (22.03% ± 5.11) around the mean of all angiosperms examined (21.5%). We found a significant and marked difference between the observed and expected distributions of INT values (KS test, D= 0.65, P < 0.001; Fig. 2), and significant differences between the medians of both distributions (inline image = 45.3, P < 0.0001). Average amounts of integration were 21.5% (SD = 15.4) and 32.7% (SD = 8.5) for the observed and expected datasets, respectively. Based on the randomly generated distribution (Fig. 2), few species have higher than expected degrees of integration (more than 2 SD above the mean percentage of maximum possible integration). These species were Stylidium brunonianum (Stylidaceae), Anacamptis pyramidalis (Orchidaceae), Collinsia sparsiflora (Scrophulariaceae) and Trillium grandiflorum (Liliaceae). Within each of these families, other members have lower or similar amounts of integration than those expected by chance.

image

Figure 2. Probability distribution of observed (grey bars) and randomly generated (open bars) integration levels expressed as a percentage of the maximum possible integration.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Taken together, the results of this study revealed that despite the commonly held assumption that flowers are highly integrated organs, flowering plants have lower floral integration (21.5%) than expected by a randomly generated distribution. This finding was further reinforced by HSAs showing that both individual floral traits and intrafloral integration, but not floral integration, were the targets of selection. Numerical simulations indicated that the low amounts of floral integration observed among the angiosperms could result from selection acting on intrafloral integration (Fig. S1, Text S1). Finally, the variance in the magnitude of integration was not explained by differences among clades, species or populations, suggesting that this attribute has low adaptive value or is maintained by selection favouring evolutionary lability.

Parallel to the absence of differences in the magnitude of integration at all levels of analyses, significant differences were detected in the VCS. This pattern may indicate that during cladogenesis, selection produced important changes in the covariances among floral traits, highlighting the possible adaptive value of the VCS during speciation. By contrast, the presence of few changes in individual traits during cladogenesis suggests their low value during speciation. Besides HK, only two traits related to the development of the kernel and pulp of the fruits (HL, HW) showed significant differences between clades and species (Weberling, 1989). It is possible that the differences between clades in both the VCS and these traits reflect past selection on flower attributes related to fruit morphology rather than on flower functioning during the diversification of clades and species. It is well known that several flower attributes can influence the phenotypic expression of the fruits because of developmental linkage (Weberling, 1989). This evidence, plus the fact that, unlike the flower, the fruit has a recognized taxonomic value within the Rosaceae (Judd et al., 2002), suggests that the floral phenotype we observed in the studied species could have been the result of selection acting on the fruit. Nonetheless, the pattern of differentiation in the VCS between clades, species and populations may still reflect the presence of an important phylogenetic component (Armbruster et al., 2004, Dalechampia; Widén et al., 2002, Brassica cretica; Waldmann & Andersson, 2000, Scabiosa; Schlichting, 1989, Phlox; but see Pérez et al., 2007). Overall, the low degrees of differentiation in individual traits at high taxonomic levels suggest that the flower is a highly conserved ‘labile’ structure within the Rosaceae.

The similar amounts of integration found between clades, species and populations suggest that this complex trait is shared by the whole family, and therefore was weakly related to the pattern of diversification within the Rosaceae. The contrast between the lack of variation of phenotypic integration and the substantial variability found in the VCS might result from three nonmutually exclusive explanations: (i) the extent of integration did not function as a constraint during the radiation of the Rosaceae; (ii) different VCS converged to similar amounts of integration; or (iii) the degree of integration of the flower per se has little adaptive value and constitutes a by-product of the genetic architecture or selection favouring intrafloral integration. If high integration is in fact a constraint, natural selection could even have favoured a reduction in this trait value. It is important to note that the average amount of integration among the study species did not differ from the average of all the angiosperms found in our survey (see later). A profitable approach to understand the adaptive value of floral integration (or its role as a constraint) during evolution would be to determine if plant lineages with high rates of diversification have lower amounts of floral integration (Eble, 2004).

As mentioned above, we found evidence of selection acting on individual floral traits and on intrafloral integration, but not on the integration index. This result suggests that natural selection is not favouring the integration of the whole flower of the three studied species of Rosaceae. Specifically, because of its effect on size- and pollination-related floral attributes, natural selection is likely to have favoured the tightening of covariation among sets of attributes and thus intrafloral integration. The only other study attempting to measure the selective value of floral integration also found a nonsignificant result in Lavandula latifolia (Herrera, 2001). In accordance with Berg's prediction that higher floral integration should be expected in specialized rather than generalized pollination systems, our results are consistent with the observation that the rosaceous species examined are visited by several insect species (Guitián & Flores, 1992; Jordano, 1993; Garcia & Chacoff, 2007). Unfortunately, because nobody has estimated the intensity of natural selection acting on floral integration on a highly specialized system, we were unable to contrast our results with other studies. While trying to reduce the number of traits to perform selection analyses in Polemonium viscosum and Ipomopsis aggregata, previous studies also found evidence of selection acting on intrafloral integration (selection on scores from principal components; Galen, 1989; Campbell, 1989). This evidence is consistent with the proposal that analyses of the patterns of covariation among subsets of traits would help to understand the possible adaptive value of floral integration (Armbruster et al., 1999). Hence, the results of the present study and previous evidence suggest that the average magnitude of floral integration observed among flowering plants could be a by-product of selection favouring specific associations of traits within the flower.

Our numerical simulations further demonstrated that the degree of phenotypic integration of the flower could be a by-product of the strength of intrafloral integration (Fig. S1, Text S1). According to these analyses, selection on intrafloral integration can produce values of floral integration similar to those commonly observed among wild species. Thus, our simulations suggest that selection acting on intrafloral integration could be responsible for the observed amounts of floral integration found among flowering plants. In addition, in order to avoid misleading conclusions about the adaptive value of the magnitude of floral integration, HSA can be used to test whether selection is acting on intrafloral integration or whole-flower integration. From an organismic point of view, our results also highlight the potential of the HSA to better understand the patterns of selection in multivariate phenotypes.

Floral integration or intrafloral integration

The ground plan of all bisexual flowers normally consists of an outer sterile perianth, male organs in the middle and female organs in the centre (Smyth, 2005). Within this basic plan, most of the diversity that we can now recognize results from variation in the shape and size of floral organs rather than from evolutionary innovations related to the genetic and developmental control of the morphogenesis of the flower (Weberling, 1989; Smyth, 2005). Besides the structural and developmental organization of the flower, the observed distribution of INT values (expressed as a percentage of maximum possible integration) indicates that flowers have rather low integration. Also, some species display higher than expected degrees of phenotypic integration, suggesting that there is wide variation available for further evolutionary changes. Being randomly generated, our expected distribution of integration values was not constrained by development or physiology. Thus, the observed low mean value obtained here suggests that this may have been favoured during evolution. In other words, natural selection may have promoted low amounts of floral integration or intrafloral integration.

Recently, detailed analyses in the genus Schizanthus (Solanaceae) showed that species with different correlation structures among floral traits have similar amounts of integration (Pérez et al., 2007). This evidence supports the hypothesis of a low adaptive value of the magnitude of phenotypic integration. Moreover, the different patterns of trait covariation were associated with different groups of pollinators (Pérez et al., 2007; Pérez-Barrales et al., 2007). Overall, the available evidence supports the hypothesis that pollinators may select for specific patterns of intrafloral integration, promoting differentiation in the VCS, rather than on the integration level of the whole flower (Armbruster et al., 1999). Because the flower is the target of complex selection regimes derived from conflicts among sexual functions, pollinator attraction, antagonist avoidance, and efficient pollen donation/reception, it is unlikely that all flower traits would be optimized for all these functions. In other words, if different selection pressures are acting upon different suits of floral traits, natural selection should favour intrafloral over whole-flower integration.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

M. O. was supported by a postdoctoral fellowship from the Universidad Nacional Autónoma de México at the Instituto de Ecología, UNAM. Rubén Pérez helped us during the course of the study. The authors want to thank Derek Roff, Ignacio Méndez, Patricia Romero, Roberto Munguía and Santiago Benítez-Vieyra for statistical advice. Scott Armbruster, Courtney Murren, Constantino Macias, Frank M. Frey, Carlos Herrera and one anonymous reviewer made valuable comments that improved the final version of the manuscript. C. D. was supported by a postdoctoral fellowship from the European Economic Community at the Estación Biológica de Doñana, Spain. We are also grateful to Carlos Herrera for his encouragement to develop this study.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Alexandersson R, Johnson SD. 2002. Pollinator-mediated selection on flower-tube length in a hawkmoth-pollinated Gladiolus (Iridaceae). Proccedings of the Royal Society of London B 269: 631636.
  • Anderson IA, Busch JW. 2006. Relaxed pollinator-mediated selection weakens floral integration in self-compatible taxa of Leavenworthia (Brassicaceae). American Journal of Botany 93: 860867.
  • Armbruster WS. 1991. Multilevel analysis of morphometric data from natural plant populations: insights into ontogenetic, genetic, and selective correlations in Dalechampia scandens. Evolution 45: 12291244.
  • Armbruster WS, Di Stilio VS, Tuxill JD, Flores TC, Velásquez-Runk JL. 1999. Covariance and decoupling of floral and vegetative traits in nine neotropical plants: a re-evaluation of Berg's correlation pleiades concept. American Journal of Botany 86: 3955.
  • Armbruster WS, Pélabon C, Hansen TF, Mulder CPH. 2004. Floral integration, modularity, and accuracy. Distinguishing complex adaptations from genetic constraints. In: PigliucciM, PrestonK, eds. Phenotypic integration. Studying the ecology and evolution of complex phenotypes. New York, NY, USA: Oxford University Press, 2349.
  • Armbruster WS, Schwaegerle KE. 1996. Causes of covariation of phenotypic traits among populations. Journal of Evolutionary Biology 9: 261276.
  • Arnold SJ. 1992. Constraints on phenotypic evolution. American Naturalist 140: S85S107.
  • Ashman T-L, Majetic CJ. 2006. Genetic constraints on floral evolution: a review and evaluation of patterns. Heredity 96: 343352.
  • Bell GB. 1985. On the function of flowers. Proceedings of the Royal Society of London B 224: 223265.
  • Benítez-Vieyra S, Medina AM, Glinos E, Cocucci AA. 2006. Pollinator-mediated selection on floral traits and size of floral display in Cyclopogon elatus, a sweat bee-pollinated orchid. Functional Ecology 20: 948957.
  • Berg RL. 1960. The ecological significance of correlation Pleiades. Evolution 14: 171180.
  • Campbell DR. 1989. Measurement of selection in a hermaphroditic plant: variation in male and female pollination success. Evolution 43: 318334.
  • Caruso CM. 2000. Competition for pollination influences selection on floral traits of Ipomopsis aggregata. Evolution 54: 15461557.
  • Cheverud JM, Wagner GP, Dow MM. 1989. Methods for the comparative analysis of variation patterns. Systematic Zoology 38: 201213.
  • Conner JK. 2001. How strong is natural selection? Trends in Ecology and Evolution 16: 215217.
  • Darwin CR. 1862. On the various contrivances by which British and foreign orchids are fertilised by insects, and on the good effects of intercrossing. London, UK: Murray.
  • Domínguez CA, Eguiarte LE, Núñez-Farfán J, Dirzo R. 1998. Flower morphometry of Rhizophora mangle (Rhizophoraceae): geographical variation in Mexican populations. American Journal of Botany 85: 637643.
  • Eble GJ. 2004. The macroevolution of phenotypic integration. In: PigliucciM, PrestonK, eds. Phenotypic integration. Studying the ecology and evolution of complex phenotypes. New York, NY, USA: Oxford University Press, 253273.
  • Endler JA. 1995. Multiple-traits coevolution and environmental gradients in guppies. Trends in Ecology and Evolution 10: 2229.
  • Faegri K, Van Der Pijl L. 1966. The principles of pollination ecology. Oxford, UK: Pergman Press.
  • Frey FM. 2007. Phenotypic integration and the potential for independent color evolution in a polymorphic spring ephemeral. American Journal of Botany 94: 437447.
  • Galen C. 1989. Measuring pollinator-mediated selection on morphometric floral traits: bumblebees and the Alpine Sky Pilot Polemonium viscosum. Evolution 43: 882890.
  • García C, Arroyo JM, Godoy JA, Jordano P. 2005. Mating patterns, pollen dispersal, and the ecological maternal neighbourhood in a Prunus mahaleb L. population. Molecular Ecology 14: 18211830.
  • Garcia D, Chacoff NP. 2007. Scale-dependent effects of habitat fragmentation on Hawthorn pollination, frugivory, and seed predation. Conservation Biology 21: 400411.
  • Gómez JM. 2000. Phenotypic selection and response to selection in Lobularia maritima: importance of direct and correlational components of natural selection. Journal of Evolutionary Biology 13: 689699.
  • Guitián J, Flores M. 1992. Reproductive biology of Crataegus monogyna in northwestern Spain. Acta Oecologica 13: 311.
  • Heisler IL, Damuth J. 1987. A method for analyzing selection in hierarchically structured populations. The American Naturalist 130: 582602.
  • Herrera CM. 1982. Breeding systems and dispersal-related maternal reproductive effort of southern Spanish bird-dispersed plants. Evolution 36: 12991314.
  • Herrera CM. 1993. Selection on floral morphology and environmental determinants of fecundity in a hawk moth-pollinated violet. Ecological Monographs 63: 251275.
  • Herrera CM. 2001. Deconstructing a floral phenotype: do pollinators select for corolla integration in Lavandula latifolia? Journal of Evolutionary Biology 14: 574584.
  • Herrera CM, Cerdá X, García MB, Guitián J, Medrano M, Rey PJ, Sánchez-Lafuente AM. 2002. Floral integration, phenotypic covariance structure and pollinator variation in bumblebee-pollinated Helleborus foetidus. Journal of Evolutionary Biology 15: 108121.
  • Herrera J. 1984. Vegetation of the valley of Guadahornillos (Sierra de Cazorla, Jaen). Studia Oecologica 3: 7796.
  • Jordano P. 1993. Pollination biology of Prunus mahaleb L.: deferred consequences of gender variation for fecundity and seed size. Biological Journal of the Linnean Society 50: 6584.
  • Judd WS, Campbell CS, Kellogg EA, Stevens PF, Donoghue MJ. 2002. Plant systematics. A phylogenetic approach, 2nd edn . Sunderland, MA, USA: Sinauer Associates Inc.
  • Kingsolver JG, Hoekstra HE, Koekstra JM, Berrigan D, Vignieri SN, Hill CE, Hoang A, Gilbert P, Beerli P. 2001. The strength of phenotypic selection in natural populations. American Naturalist 157: 245261.
  • Kleinbaum DG, Kupper LL, Muller KE. 1988. Applied regression analysis and other multivariate methods. Boston, MA, USA: PSW-Kent.
  • Lande R, Arnold SJ. 1983. The measurement of selection on correlated characters. Evolution 37: 12101226.
  • Maad J. 2000. Phenotypic selection in hawkmoth-pollinated Platanthera bifolia: targets and fitness surfaces. Evolution 54: 112123.
  • Marcus L. 1990. Traditional morphometrics. In: RohlfFJ, BooksteinFL, eds. Proceedings of the Michigan Morphometrics Workshop. Special Publication No 2. Ann Arbor, MI, USA: Museum of Zoology, University of Michigan, 77122.
  • Mathsoft, Inc. 1999. S-Plus 2000 guide to statistics. Seattle, WA, USA: Mathsoft.
  • Maynard-Smith J, Szathmáry E. 1995. The major transitions in evolution. Oxford, UK: WH Freeman/Spektrum.
  • Michod R, Viossat Y, Solari CA, Hurand M, Nedelcu AN. 2006. Life-history evolution and the origin of multicellularity. Journal of Theoretical Biology 239: 257272.
  • Muchhala N. 2006. Nectar bat stows huge tongue in its rib cage. Nature 444: 701702.
  • Murren C. 2002. Phenotypic integration in plants. Plant Species Biology 17: 8999.
  • Murren C, Pendleton N, Pigliucci M. 2002. Evolution of phenotypic integration in Brassica (Brassicaceae). American Journal of Botany 89: 655663.
  • Nilsson LA. 1988. The evolution of flowers with deep corolla tubes. Nature 334: 147149.
  • O’Connell LM, Johnston MO. 1998. Male and female pollination success in a deceptive orchid, a selection study. Ecology 79: 12461260.
  • Oddou-Muratorio S, Klein EK, Demesure-Musch B, Austerlitz F. 2006. Real-time patterns of pollen flow in the wild-service tree, Sorbus torminalis (Rosaceae). III. Mating patterns and the ecological maternal neighborhood. American Journal of Botany 93: 16501659.
  • Olson EC, Miller RL. 1958. Morphological integration. Chicago, IL, USA: University of Chicago Press.
  • Pérez F, Arroyo MTK, Medel R. 2007. Phylogenetic analysis of floral integration in Schizanthus (Solanaceae): does pollination truly integrate corolla traits? Journal of Evolutionary Biology 20: 17301738.
  • Pérez-Barrales R, Arroyo J, Armbruster WS. 2007. Differences in pollinator faunas may generate geographic differences in floral morphology and integration in Narcissus papyraceus (Amaryllidaceae). Oikos 116: 19041918.
  • Pigliucci M. 2003. Phenotypic integration: studying the ecology and evolution of complex phenotypes. Ecology Letters 6: 265272.
  • Pigliucci M, Preston K. 2004. Phenotypic integration. Studying the ecology and evolution of complex phenotypes. New York, NY, USA: Oxford University Press.
  • Potter D, Gao F, Bortiri PE, Oh S-H, Baggett S. 2002. Phylogenetic relationships in Rosaceae inferred from chloroplast matK and trnL-trnF nucleotide sequence data. Plant Systematics and Evolution 231: 7789.
  • Reyment R, Jöreskog KG. 1993. Applied factor analysis in the natural sciences. Cambridge, NY, USA: Cambridge University Press.
  • Roff D. 2002. Comparing G matrices: a MANOVA approach. Evolution 56: 12861291.
  • SAS Institute Inc. 2005. JMP Version 5. Cary, NC, USA: SAS.
  • Schlichting C. 1989. Phenotypic integration and environmental change. BioScience 39: 460464.
  • Smyth DR. 2005. Morphogenesis of flowers – our evolving view. The Plant Cell 17: 330341.
  • StatSoft, Inc. 2004. STATISTICA (data analysis software system), version 7. Tulsa, OK, USA: StatSoft, Inc.
  • Stebbins GL. 1950. Variation and evolution in plants. New York, NY, USA: Columbia University Press.
  • Stebbins GL. 1970. Adaptive radiation of reproductive characteristics in angiosperms, I: pollination mechanisms. Annual Review of Ecology and Systematics 1: 307326.
  • Terentjev PV. 1931. Biometrische Untersuchungen über die morphologischen Merkmale von Rana ridibunda Pall. (Amphibia, Salientia). Biometrika 23: 2351.
  • Wagner GP. 1984. On the eigenvalue distribution of genetic and phenotypic dispersion matrices: evidence for a nonrandom organization of quantitative character variation. Journal of Mathematical Biology 21: 7795.
  • Waldmann P, Andersson S. 2000. Comparison of genetic (co)variance matrices within and between Scabiosa canescens and S. columbaria. Journal of Evolutionary Biology 13: 826835.
  • Weberling F. 1989. Morphology of flowers and inflorescences. Cambridge, UK: Cambridge University Press.
  • Widén B, Andersson S, Rao G-Y, Widen M. 2002. Population divergence of genetic (co)variance matrices in a subdivided plant species, Brassica cretica. Journal of Evolutionary Biology 15: 961970.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Fig. S1 Relationship between intrafloral integration (r) and the integration level expressed as the percentage of maximum integration.

Table S1 Phenotypic matrices of variance–covariance of each of two populations for each of the four Rosaceae species

Table S2 Data on integration index (INT) after search in 55 studies indicating order, family, species, group name (referring to the name given in the source), level of analysis, number of measured traits (N traits), integration index values (raw, INT; scaled to number of measured traits, % max INT), and source.

Text S1 Using a series of numerical simulations, the relationship was examined between intrafloral integration and the percentage of maximum possible integration (INT) of hypothetical matrices.

Please note: Blackwell Publishing are not responsible for the content or functionality of any Supporting information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

FilenameFormatSizeDescription
NPH_2523_sm_Fig. S1.doc92KSupporting info item
NPH_2523_sm_Table S1.doc303KSupporting info item
NPH_2523_sm_Table S2.doc272KSupporting info item
NPH_2523_sm_Text S1.doc26KSupporting info item