Skip to main content
NCBI home page
Annals of Botany logoLink to Annals of Botany
. 2013 Mar 6;111(5):935–944. doi: 10.1093/aob/mct050

Decoupled phenotypic variation between floral and vegetative traits: distinguishing between developmental and environmental correlations

Christophe Pélabon 1,*, Nora C Osler 2,3, Martin Diekmann 3, Bente J Graae 2
PMCID: PMC3631334  PMID: 23471008

Abstract

Background and Aims

In species with specialized pollination, floral traits are expected to be relatively invariant and decoupled from the phenotypic variation affecting vegetative traits. However, inferring the degree of decoupling between morphological characters from patterns of phenotypic correlations is difficult because phenotypic correlations result from the superimposition of several sources of covariance. In this study it is hypothesized that, in some cases, negative environmental correlations generated by non-congruent reaction norms across traits overshadow positive developmental correlations and generate a decoupling of the phenotypic variation between vegetative and floral traits.

Methods

To test this hypothesis, Campanula rotundifolia were grown from two distinct populations under two temperature treatments, and patterns of correlation were analysed between leaf size and flower size within and among treatments.

Key Results

Flower size was less sensitive to temperature variation than leaf size. Furthermore, flower size and leaf size showed temperature-induced reaction norms in opposite directions. Flower size decreased with an increasing temperature, while leaf size increased. Consequently, among treatments, correlations between leaf size and flower size were negative or absent, while, within treatments, these correlations were positive or absent in the cold and warm environments, respectively.

Conclusions

These results confirm that the decoupling of the phenotypic variation between vegetative and floral traits can be dependent on the environment. They also underline the importance of distinguishing sources of phenotypic covariance when testing hypotheses about phenotypic integration.

Keywords: Berg hypothesis, Campanula rotundifolia, correlation pleiades, insect pollination, modularity, phenotypic integration, phenotypic plasticity, temperature

INTRODUCTION

In flowering plants with specialized pollination, a tight fit is expected between floral and pollinator morphologies (Fenster et al., 2004; Armbruster et al., 2009), and variation in flower size is likely to be maladaptive because pollinators remain of more or less fixed size and behaviour (Stebbins, 1974; Armbruster et al., 2004, 2009). Berg (1959, 1960) hypothesized that the maintenance of the fit between the floral traits responsible for pollen transfer and the pollinators in the face of the phenotypic response of plants to environmental variation was allowed by the decoupling of the phenotypic variation between vegetative and floral traits. However, studies analysing phenotypic variation and covariation between these two types of traits yield inconsistent evidence for the Berg hypothesis. While several studies support the prediction that floral traits are less variable than vegetative traits (Fenster, 1991; Cresswell, 1998; Herrera, 2001; Hansen et al., 2007; Pérez-Barrales et al., 2007; Chalcoff et al., 2008; Van Kleunen et al., 2008; Rosas-Guerrero et al., 2010; Pélabon et al., 2011), others show that floral traits still respond to environmental variation, sometimes markedly (Vogler et al., 1999; Carroll et al., 2001; Dorken and Barrett, 2004; Brock et al., 2005; Mal and Lovett-Doust, 2005; Caruso, 2006; Brock and Weining, 2007; for a review, see Herrera 2009). Furthermore, several studies analysing the morphological integration of vegetative and floral traits confirm the decoupling of phenotypic variation between these two types of traits (Conner and Sterling, 1996; Magwene, 2001; Juenger et al., 2005; Chalcoff et al., 2008), while other studies observe positive correlations (Waitt and Levin, 1993; Conner and Sterling, 1996).

Following the influential book from Olson and Miller (1958) on morphological integration, the decoupling of the phenotypic variation between vegetative and floral traits has been generally inferred from patterns of phenotypic correlation within and among series of traits belonging to one or the other category (Armbruster et al., 2004; Pigliucci and Preston, 2004). This approach is not without problems, however. First, comparisons of non-homologous floral and vegetative traits are difficult, because variational properties strongly depend on the trait complexity and dimensionality (Hansen et al., 2007). Hallgrímsson et al. (2009) further argued that integration and modularity (the organization of complex organisms into quasi-independent parts) ‘cannot be reliably studied from phenotypic covariance patterns alone’, because of the superimposition of multiple determinants of phenotypic covariance. Building on this argument, we suggest that part of the inconsistencies observed in the tests of the Berg hypothesis may have resulted from the effect of environmental variation on the relative magnitude of non-congruent sources of covariation that generate phenotypic correlations among traits.

Phenotypic correlations result from the combination of at least two, more or less independent, processes that generate covariation. On the one hand, homologous or structurally adjacent traits that share similar developmental pathways are expected to covary positively because variation in these developmental pathways (i.e. genetic variation, resource allocation or developmental noise) will generate covariation at the phenotypic level (‘direct interaction’ sensu Klingenberg, 2008). Phenotypic correlations resulting from this type of interaction will be referred to as ‘developmental correlations’ hereafter. On the other hand, covariation between traits may result from a similar phenotypic response to environmental variation (i.e. phenotypic plasticity; Bradshaw, 1965), and traits showing similar reaction norms should be more strongly integrated than traits showing different reaction norms (Schlichting, 1986; Schlichting and Pigliucci, 1998). Hence, congruent reaction norms may result from the effects of environmental variation on shared developmental pathways (Donohue and Schmitt, 1999), but may also result from a similar response to environmental variation from traits with more or less independent developmental pathways (‘parallel variation’ sensu Klingenberg, 2008). In the following, correlations generated by phenotypic plasticity and measured among environments are referred to as ‘environmental correlations’. Phenotypic correlations will therefore result from the combination of developmental and environmental correlations. This view is similar to the one developed by Searle (1961) on genetic and environmental correlations. However, because developmental correlations may involve localized epigenetic effects (Zelditch and Swiderski, 2011), genetic and developmental correlations are not necessarily synonymous. Importantly, directions of developmental and environmental correlations are not necessarily congruent. While developmental correlations between homologous or structurally adjacent traits are likely to be mostly positive, norms of reaction in the opposite direction may occur and generate negative environmental correlations among traits (Fig. 1A). Phenotypic correlations will therefore depend on the direction and the relative magnitude of the covariance generated by either developmental or environmental variation (Fig. 1B). Under constant environmental conditions, developmental correlations should prevail and phenotypic integration should mainly reflect these correlational patterns. Under a varying environment, however, phenotypic integration will be strongly affected by phenotypic plasticity, and developmental correlations may be masked by environmental correlations (see Fig. 1B for further explanations). Consequently, the decoupling of the phenotypic variation between floral and vegetative traits inferred from patterns of phenotypic correlations may depend on the degree of environmental variation, or on the type of environmental factor varying, with different factors being expected to generate different reaction norms.

Fig. 1.

Fig. 1.

Open in a new tab

Schematic representation of how developmental and environmental sources of variation generate covariation among traits and how these combine to generate phenotypic correlation. In (A), we first represent how positive and negative correlations can be generated by congruent (left) or non-congruent (right) response to variation. Environmental correlations will be generated by phenotypic response to environmental variation (phenotypic plasticity), while developmental correlations will be generated by variation affecting the development of the traits (genetic variation, developmental noise). In the central panel, trait 2 is canalized against different sources of variation (genetic or environmental canalization). In this case, we expect no correlation between the two traits. Note also that developmental correlations are expected to be mostly positive, but negative developmental correlation may occur if the developmental pathways of two traits include a trade-off (e.g. Tomkins et al. 2005). In (B), we represent how environmental and developmental correlations will combine to generate phenotypic correlation between the two traits. In cases 1, 5 and 9, both types of correlation are congruent, environmental variation generating the same type of phenotypic variation as genetic variation or developmental noise [direct interaction sensu Klingenberg (2008) in case 1; developmental and environmental trade-off in case 9]. Consequently, phenotypic variation follows the direction of both environmental and developmental correlations. In cases 4 and 6, despite the two traits being developmentally independent, the environment generates some correlation between them, and the overall phenotypic correlation follows the pattern of environmental correlation (parallel variation sensu Klingenberg, 2008). In cases 2 and 8, despite the environmental canalization of trait 2, the developmental correlation between the two traits may generate some phenotypic correlations. We are not aware of any example of this particular scenario. In cases 3 and 7, developmental and environmental correlations are in opposite directions, and the phenotypic outcome is uncertain and depends on the relative amplitude of the two sources of variation. On the right-hand side of the matrix (C), we represent two possible scenarios. In both graphs, circles represent individuals raised in one of two environments (the grey and black circles representing individuals raised in environment 1 or 2). The solid lines represent developmental correlations, while the dotted lines represent environmental correlations. In the upper graph, the environmental variation is strong and the phenotypic correlation is primarily determined by this environmental correlation. In the lower graph, the limited environmental variation generates only a weak negative environmental correlation between the two traits. The two sources of covariance tend to cancel each other out, leading to an apparent absence of phenotypic correlation.

These considerations suggest that distinguishing between environmental and developmental correlations may be important when studying the decoupling of phenotypic variation. Indeed, failure to identify non-congruent correlation patterns may seriously undermine the interpretation of studies testing the Berg hypothesis. To illustrate this idea, we analysed patterns of variation and covariation in leaf size and flower size in two populations of Campanula rotundifolia from contrasting environments that were exposed to different temperature treatments. Temperature is a particularly important factor for plants, affecting growth, morphology and life histories (Larcher, 2003). In the face of the compelling evidence of the ongoing changes in global climate, understanding temperature-induced phenotypic plasticity seems particularly important especially in arctic and alpine environments where temperature changes are the most pronounced (Maxwell, 1992; Stendel et al., 2008; Gottfried et al., 2012). By analysing correlations between flower size and leaf size within and among temperature treatments, we expect to be able to distinguish between developmental and environmental correlations and test whether the decoupling of the phenotypic variation between these two series of traits depends on environmental variation.

MATERIALS AND METHODS

Study species

Campanula rotundifolia L. (Campanulaceae) is a circumpolar perennial herb spanning large altitudinal gradients (Giblin, 2005). The flowers of C. rotundifolia are bell-shaped, protandrous and predominantly outcrossing even though they are not self-incompatible (Nyman, 1991). Pollination biology including the role of secondary pollen presentation has been well documented (Nyman, 1992; Cresswell and Robertson, 1994). The main pollinators are bees and bumble-bees, at lower and higher altitudes, respectively (Bingham and Orthner, 1998). This altitudinal change in pollinator predominance is probably caused by the ability of bumble-bees to tolerate lower average temperatures at higher altitudes (Bishop and Armbruster, 1999; Bingham and Ranker, 2000). Flower size and shape in C. rotundifolia are expected to be important for pollination accuracy. Indeed, in species with secondary pollen presentation (i.e. pollen placement by the anther onto another floral structure which in turn places the pollen on the pollinator; Nyman, 1992), or species where pollinators have to crawl into the inner surface of the corolla to reach the reward, we expect a tight fit between the pollinator size and the corolla size to ensure pollen transfer efficiency (Fenster et al., 2004; Armbruster et al., 2009; but see Kobayashi et al., 1999). Earlier observations on different populations of C. rotundifolia suggested a decoupling of the phenotypic variation in flower size and plant size, because high-altitude populations displayed lower plant height but larger flowers (Turesson, 1925; Shelter, 1982; Maad et al., 2013). Furthermore, the increase in flower size with a decreasing temperature (J. Maad, Uppsala University, Sweden, and W. S. Armbruster, University of Portsmouth, UK, unpubl. res.) suggests that, in this species, part of the decoupling of the phenotypic variation is generated by temperature-induced reaction norms in opposite directions in vegetative and floral traits.

Experimental design

Second greenhouse-generation seeds were grown under two different temperature treatments. At flowering, we recorded traits reflecting the vegetative size of the plants, including leaf area, and traits reflecting the size of the flower that is assumed to affect the fit with the pollinator. We then analysed patterns of variation and covariation between flower size and leaf size, within and among environments.

The seeds used in this experiment were derived from individuals collected at two different sites in the Dovre mountain region (Norway). One population was located at 510 m a.s.l (low-altitude population, 62°32′N, 09°36'E), while the other population was located at 1100 m a.s.l (high-altitude population, 62°31'N, 09°40'E); these populations are referred to as populations G and H, respectively, in Maad et al. (2013). In 2005, individuals collected in the wild and maintained in the greenhouse (NTNU, Trondheim) were crossed in a partial diallel experiment in order to estimate genetic variance of floral traits (J. Maad, Uppsala University, Sweden, and W. S. Armbruster, University of Portsmouth, UK, unpubl. res.). The F1 generation was grown under greenhouse conditions and crossed to produce new seeds to be used in the present experiment. Thirty genetic families per population were chosen randomly at the beginning of the experiment in November 2009. For each of the 30 families in each population, approx. 200 seeds were sown in ten individual pots (20 seeds per pot) in early November 2009 and kept at 19 °C to ensure germination. Following germination, seedlings were thinned in January 2010 to make sure that each pot contained only one plant. In all pots, we kept the largest seedling close to the centre of the pot. We subsequently placed half of the pots from each family in a room at an average temperature of approx. 12 °C, and the other half in a room at an average temperature of approx. 19 °C (Supplementary Data Fig. S1). The position of each pot on one of the four tables in each room was randomly assigned. We applied a light:dark cycle of 20 h:4 h, resembling the average summer conditions in mid-Norway. For practical reasons, the temperature treatment was confounded with the room. However, the two rooms used in this study were contiguous in the greenhouse and strictly similar with respect to the light environment, the supplementary lighting being provided by two high-pressure 400 W sodium lamps over each table in each room. We also took great care in maintaining plants under similar moisture conditions in the two rooms. Consequently, although the treatment truly corresponds to the combined effects of the temperature treatment and the rooms, we refer to it as temperature treatment hereafter. It is also important to note that the difference in temperature between the two treatments may represent an extreme situation compared with the differences normally encountered by natural populations. However, because our primary aim was to compare environment-induced correlations between vegetative and reproductive traits with developmental correlations, we chose treatment levels that were expected to maximize the effect size at the expense of the ecological realism.

Out of the 600 pots (2 treatments × 2 populations × 30 families × 5 replicates), germination failed in 48 pots (12 in the low-altitude population and 36 in the high-altitude population). Overall, 294 plants produced flowers, 42 and 98 from the low-altitude population, and 44 and 110 from the high-altitude population, in the cold and warm environment, respectively (Supplementary Data Table S1). For logistic reasons, the experiment was terminated in early June 2010 on day 208 after sowing.

Measurements

From March 2010 onward, plants were monitored every second to third day to record the emergence of the first flower. Floral traits were measured on one inflorescence per plant in early pistillate phase in order to control for possible ontogenetic variation in flower size. Measurements of floral traits included three measurements of flower length [flower length (FL), corolla length (CL) and tube length (TL)] and two measurements of the flower diameter [corolla width (CW) and tube diameter (TD); Fig. 2]. All floral measurements were taken with a digital calliper with 0·01 mm precision by a single person (N.C.O.). To account for a possible trade-off between the number of flowers and flower size, we recorded the number of flowers and buds present at the time of the floral measurements. Because number of flowers is not a morphological trait affecting the fit with pollinators, it was not included among the floral traits analysed to test the Berg hypothesis.

Fig. 2.

Fig. 2.

Open in a new tab

Drawings of the different floral traits measured (solid arrows). FL, flower length; CL, corolla length; CW, corolla width; TL, tube length; TD, tube diameter. Abbreviations: OFL, orthogonal flower length; and CD, corolla diameter, are measurements derived from FL and CW, respectively. OFL and CD were not used in the analysis.

We also measured several traits related to the vegetative size of the plant 3 weeks after the first flower opened. These measurements included rosette diameter (average of two diameters perpendicular to each other) and stalk height (height of the highest flower above the soil surface). The three largest leaves were collected and scanned, and we estimated their area using Image-J 1·43 software (http://rsb.info.nih.gov/ij/). Leaf area was square root transformed in order to provide a measure of leaf size on a scale (mm) similar to the floral measurements. After all measurements were completed, the total above-ground biomass was estimated by drying the plants for 48 h at 40 °C and weighing them to the nearest 0·001 g. This biomass is referred to as dry weight.

Statistical analyses

We first tested whether floral traits were less variable than vegetative traits by comparing the variational properties of the different traits. We estimated the amount of variation due to the temperature treatment and the population of origin by conducting a variance-component analysis on each trait separately. We used mixed-effects models where temperature, population and family were entered as random factors to partition the variance between these different levels (Crawley, 2007, p. 638). We expressed the different components of the variance as mean-square scaled variances in order to compare the variance across traits. Mean-square scaled variances are equivalent to square coefficient of variation but have the advantage of being additive.

Leaf size was estimated as the mean of the three leaf area measurements. To estimate the overall flower size, we ran a principal component analysis (PCA) on the floral traits. For morphological traits, the first axis of the principal component analysis (PC1) essentially captures size variation (Wagner, 1984). We therefore defined flower size as the scores on the first axis of the PCA on floral traits. The first axis of the PCA comprised 69 % of the total variance in floral traits (Supplementary Data Table S2). We then tested the effects of temperature and population on flowering time and both size variables using mixed-effects models with population and temperature as fixed factors and family as a random factor. Model selection was conducted on models fitted with maximum likelihood (ML), while parameter estimates were obtained from the best models fitted with restricted maximum likelihood (REML). Because none of the floral traits was negatively affected by the number of flowers [model on flower size with flower number as predictor variable, Akaike information criterion (AIC) = – 457·5; without flower number, AIC = – 464·2], we did not correct our estimates of flower size for variation in the number of flowers.

It is not possible to disentangle environmental from developmental correlations entirely because developmental correlations will always be present between the two size measurements. By analysing the phenotypic correlations within and among treatments, we intended to obtain estimates of the developmental correlations within treatment, i.e. when the effect of the environmental variation on covariance was minimal. Among-treatment correlations, on the other hand, provided estimates of the sum of the developmental and environmental correlations when environmental variation was expected to represent the primary source of trait covariance.

Patterns of covariation between leaf size and flower size were estimated using Pearson correlation coefficients between the two variables, within and among treatments. However, in order to account fully for the structure of our data (family nested in populations) and obtain unbiased parameter estimates and P-values for the covariance between leaf size and flower size, we used mixed-effects models where flower size was regressed on leaf size with population as a fixed factor and family as a random factor. Models were conducted on the whole data set (i.e. including the among-treatment variation) and within each treatment separately. Graphical inspections of the residuals from all the models revealed that these were normally distributed. All analyses were computed using R version 2.11.0 (R Development Core Team, 2010).

Because our experimental design included several individuals per family, we should have been able to estimate genetic correlations among traits. However, due to germination and flowering failures, the final design was unbalanced and several families were represented by only one individual. Because this strongly jeopardized any analysis of the genetic correlation (Lynch and Walsh, 1997), we decided to focus here on the phenotypic correlations only.

RESULTS

Variational properties of vegetative and floral traits

Variational properties of vegetative and floral traits differed markedly. Vegetative traits, including leaf size, were much more variable than floral traits as demonstrated by their higher square coefficients of variation (Table 1). Additionally, while the temperature treatment generated on average 57 % (range 50–75 %) of the phenotypic variation in vegetative traits, with 61 % for leaf size, it only generated an average of 18 % (range 14–28 %) of the phenotypic variation in floral traits (Table 1). Variation in floral traits was essentially expressed at the among-individual level, the differences between populations and temperature treatments accounting for an equal proportion of the phenotypic variance.

Table 1.

Components of the phenotypic variance in vegetative and floral traits

Trait Mean ± s.e. Temperature Population Family Residual CV2
Vegetative traits
Stalk height (mm) 225·00 ± 54·2 9·89 0·63 0·61 5·96
57·87 % 3·67 % 3·59 % 34·87 % 17·09
Rosette diameter (mm) 85·66 ± 30·51 19·94 0·00 0·57 6·27
74·46 % 0·00 % 2·11 % 23·43 % 26·78
Dry weight (mg) 615·10 ± 387·6 33·85 1·63 0·93 18·91
61·19 % 2·95 % 1·68 % 34·18 % 55·32
Leaf size (mm) 3·57 ± 0·81 8·23 0·23 0·00 5·03
60·97 % 1·74 % 0·00 % 37·29 % 13·49
Floral traits
Flower length (mm) 24·08 ± 0·91 0·27 0·43 0·15 1·01
14·69 % 22·82 % 8·18 % 54·31 % 1·86
Corolla length (mm) 21·44 ± 0·97 0·41 0·57 0·15 1·04
18·90 % 26·08 % 6·86 % 48·16 % 2·17
Tube length (mm) 15·26 ± 0·88 0·47 1·44 0·15 1·32
13·79 % 42·68 % 4·36 % 39·16 % 3·38
Corolla width (mm) 20·44 ± 1·10 0·39 0·15 0·14 1·68
16·42 % 6·31 % 6·09 % 71·19 % 2·36
Tube diameter (mm) 14·65 ± 0·79 0·64 0·00 0·32 1·37
27·46 % 0·14 % 13·78 % 58·62 % 2·33
Ln (no·of flowers) 1·25 ± 0·39 14·47 1·56 0·00 13·21
49·48 % 5·33 % 0·00 % 45·19 % 29·24
Open in a new tab

For each trait, the different variance components are reported as mean-square scaled variances (Inline graphic) on the top line, and percentage of the total variance on the bottom line. The reported means correspond to the estimates of the fixed effect in the mixed-effects models. These were used to scale the variance components. The residual variance component corresponds to the variance among full-sibs (n = 294). Total square coefficients of variation are presented in the last column.

Effects of temperature and population of origin on flowering time, flower size and leaf size

Plants flowered on average 28·3 ± 3·6 d earlier in the warm environment, and the plants from the high-altitude population flowered on average 8·0 ± 3·4 d earlier than those from the low-altitude population (Table 2). Similarly, the temperature treatment and the population of origin had an additive effect on leaf size, leaves being larger in the warm environment and in the plants from the low-altitude population (Table 2; Fig. 3). In contrast, flowers from the high-altitude population were larger than those from the low-altitude population, and an increasing temperature negatively affected flower size, this latter effect being more marked in the high-altitude population (Table 2; Fig. 3).

Table 2.

Effects of treatment and population on flowering time, flower size and leaf size

Fixed effects AIC AIC weights Low-altitude and cold treatment Temperature effect: warm treatment Population effect: high altitude Interaction effect: warm treatment × high altitude
Flowering date ∼Pop × Temp 3076·04 0·257 188·53 ± 4·23 –26·68 ± 5·17 –5·99 ± 5·85 –3·16 ± 7·23
Flowering date ∼Pop + Temp 3074·23 0·635 189·61 ± 3·42 –28·30 ± 3·60 –8·06 ± 3·42
Flowering date ∼Pop 3122·27 2·3 × 10−11
Flowering date ∼Temp 3077·76 0·109
Flowering date ∼1 3123·09 1·6 × 10−11
Leaf size ∼Pop × Temp 793·7 0·242
Leaf size ∼Pop + Temp 791·7 0·659 2·948 ± 0·122 1·580 ± 0·110 –0·304 ± 0·126
Leaf size ∼Pop 947·3 9·71 × 10−35
Leaf size ∼Temp 795·5 0·098
Leaf size ∼1 950·5 2·17 × 10−35
Flower size ∼Pop × Temp 1042·64 0·776 –0·060 ± 0·280 –1·022 ± 0·284 2·114 ± 0·381 –0·812 ± 0·384
Flower size ∼Pop + Temp 1045·13 0·224
Flower size ∼Pop 1094·95 3·4 × 10−12
Flower size ∼Temp 1070·82 5·9 × 10−7
Flower size ∼1 1121·22 6·7 × 10−18
Open in a new tab

Estimates (± s.e.) obtained from the models fitted with REML are presented for the best model or the two best models when ΔAIC between competing models was <2 AIC units. All models were mixed-effects linear models where family was included as a random effect. Fixed factors were population of origin (Pop) and temperature treatment (Temp). Estimates are given for a trait measured in the low-altitude population in the cold environment. The temperature effect shows how the trait changes when plants from the low-altitude population are grown in the warm environment. The population effect shows how plants from the high-altitude population perform under cold treatment compared with the plants from the low-altitude population. The interaction effect shows how the trait in the plants from the high-altitude population changes from a cold to warm environment compared with the change observed in plants from the low-altitude population. The most parsimonious models (lower AIC and lowest number of parameters) are presented in bold. Models are presented following the R syntax: response variable ∼ predictor variables.

Fig. 3.

Fig. 3.

Open in a new tab

Effects of the temperature treatment on leaf size (Inline graphic in mm) and flower size (PC1 scores) in both populations (low- and high-altitude, as indicated in the key). Means (± s.e.) are estimates from mixed-effects models.

Correlation between leaf size and flower size within and among environments

Due to reaction norms in opposite directions, correlations between flower size and leaf size among treatments tended to be negative or absent (Fig. 4). Within treatment, however, leaf size and flower size were positively correlated in the cold environment, but independent from each other in the warm environment (Fig. 4). These patterns were confirmed statistically when we regressed flower size on leaf size, accounting for the structure of the data using mixed-effects models (Table 3).

Fig. 4.

Fig. 4.

Open in a new tab

Within- and among-treatment phenotypic correlations between leaf size (Inline graphic in mm) and flower size (PC1 scores) in each population (cold and warm environments, as indicated in the key). Dotted lines represent the regression across environment while the solid lines represent the regressions within environment. Among-treatment correlation coefficients (95 % confidence interval) are r = – 0·21 (– 0·36 to – 0·05) and r = – 0·09 (– 0·26 to 0·08), for the high- and low-altitude population, respectively. Within each treatment, the correlation coefficients are: in the cold environment, r = 0·34 (0·05 to 0·58) and r = 0·26 (– 0·08 to 0·54); and in the warm environment, r = 0·12 (– 0·08 to 0·30) and r = 0·03 (– 0·17 to 0·23), for the high- and low-altitude population, respectively.

Table 3.

Among- and within-treatment relationship between flower size and leaf size

Fixed effects AIC AIC weights Intercept Leaf size Population effect: high altitude Interaction effect: leaf size × altitude
Among environments
Flower size ∼leaf size × Pop 1092·76 0·434 0·036 ± 0·527 –0·205 ± 0·116 2·248 ± 0·720 –0·205 ± 0·169
Flower size ∼leaf size + Pop 1092·24 0·562 0·435 ± 0·399 –0·302 ± 0·084 1·436 ± 0·269
Flower size ∼leaf size 1113·76 1·19 × 10−5
Flower size ∼ Pop 1102·23 0·004
Flowering date ∼1 1128·46 7·67 × 10−9
Cold environment
Flower size ∼leaf size × Pop 314·67 0·343 –1·977 ± 1·336 0·636 ± 0·444 0·925 ± 1·801 0·536 ± 0·627
Flower size ∼leaf size + Pop 313·46 0·627 –2·716 ± 0·957 0·894 ± 0·310 2·395 ± 0·469
Flower size ∼leaf size 331·96 6·0 × 10−5
Flower size ∼ Pop 319·55 0·030
Flowering date ∼1 334·59 1·6 × 10−5
Warm environment
Flower size ∼leaf size × Pop 742·60 0·132
Flower size ∼leaf size + Pop 740·95 0·302 –1·492 ± 0·526 0·091 ± 0·108 1·324 ± 0·265
Flower size ∼leaf size 760·44 1·8 × 10−5
Flower size ∼ Pop 739·69 0·566 –1·074 ± 0·189 1·293 ± 0·263
Flowering date ∼1 758·49 4·7 × 10−5
Open in a new tab

These are the results of mixed-effect models where flower size is regressed on leaf size with population as fixed factor and family as a random factor. Estimates (± s.e.) were obtained from models fitted with REML. We present here estimates for the best or the two best models when ΔAIC between competing models was <2 AIC units. The most parsimonious models (lower AIC and lowest number of parameters) are presented in bold.

DISCUSSION

We hypothesized that the superimposition of the different sources of covariation that generate phenotypic correlations may hamper our capacity to assess the degree to which variation in vegetative and floral traits is decoupled in flowering plants with specialized pollination. We further suggested that manipulating the environment and analysing phenotypic correlations within and among environmental treatments should enable us to distinguish between environmental and developmental correlations. This approach should provide a better insight into the mechanisms producing phenotypic correlations and should help understanding if and how phenotypic variation is decoupled between the two types of traits. We illustrated this idea by analysing correlations between leaf size and flower size in C. rotundifolia, a bee-pollinated perennial herb, grown at two different temperatures.

Flower size was less variable than leaf size and less sensitive to the temperature treatment. This result supports previous observations showing that floral traits are generally more canalized than vegetative traits against environmental variation (Fenster, 1991; Diggle, 1992; Waitt and Levin, 1993, 1998; Cresswell, 1998; Armbruster et al., 1999; Wolfe and Mazer, 2005; Hansen et al., 2007; Pélabon et al., 2011), possibly due to the stabilizing selection generated by pollen transfer efficiency (Conner and Sterling, 1995; Cresswell, 1998, 2000; Hodgins and Barrett, 2008; Rosas-Guerrero et al., 2010). Nevertheless, flower size responded to the temperature treatment with a reaction norm opposite to the one observed in leaf size. Flower size decreased with an increasing temperature, while leaf size increased. Interestingly, these patterns of phenotypic plasticity were mimicking patterns of variation generated by differences in altitude among populations. Plants from the low-altitude population that experienced a warmer environment grew larger leaves than the plants from the high-altitude population, but developed smaller flowers in both experimental treatments. Because these plants were second greenhouse generation, we interpreted these differences as resulting from genetic differences among populations.

The decrease in plant size with altitude (or latitude) has been generally interpreted as resulting from the negative effect of low temperature on plant growth (Larcher, 2003). Our results show that such an effect of altitude in C. rotundifolia is generated by the combination of phenotypic plasticity and genetic differences among populations, a pattern also observed by Olsson and Ågren (2002) in Lythrum salicaria, Byars et al. (2007) in Poa hiemata, and Gonzalo-Turpin and Hazard (2009) in Festuca eskia. Similarly, while the increase in flower size with altitude (Tureson, 1925; Shetler, 1982; Galen et al., 1987; Scobel and Scott, 2002; Fabbro and Körner, 2004; Maad et al., 2013; but see Totland, 2001, 2004) has been generally interpreted as an adaptive response to different selection pressures (i.e. local adaptation), our results show that flower size variation also results from phenotypic plasticity in response to temperature variation. Similar variations in flower size have been observed for other abiotic factors such as water availability (Galen, 1999; Carroll et al., 2001; Herrera, 2005; Caruso, 2006), soil nutrients (Delesalle and Mazer, 1996; Frazee and Marquis, 2004) or temperature (Vogler et al., 1999; Catley et al., 2002; (J. Maad, Uppsala University, Sweden, and W. S. Armbruster, University of Portsmouth, UK, unpubl. res.). Note also that our results do not support the hypothesis of a trade-off between flower size and flower number mediated by altitude, a result also observed in the congeneric C. rapunculoides (Volger et al., 1999).

Reaction norms in response to the temperature treatment for flower size and leaf size were in opposite directions, inducing a negative environmental correlation between the two series of traits in one population, and the decoupling of the variation in the other population. When plants were grown under similar conditions, however, positive correlations between the two size measurements were observed for both populations in the cold environment, but not in the warm environment. These results support the idea that, in C. rotundifolia, the decoupling of the phenotypic variation between vegetative and floral traits depends on the relative magnitude of the different sources of variation generating phenotypic correlations. Under limited variation in temperature, developmental correlations (correlations measured within treatment in our study) prevail, and positive correlations are observed between vegetative and floral traits, suggesting an incomplete decoupling of the phenotypic variation between the two types of trait. When temperature variation increases, developmental correlations are gradually overshadowed by the effect of phenotypic plasticity that generates negative environmental correlations. Consequently, conclusions about the decoupling of the phenotypic variation between flower size and leaf size depend on the relative contribution of environmental correlations to the overall phenotypic correlation, this contribution depending on the level of environmental variation. Such a scenario provides a possible explanation for the environmental dependency of phenotypic correlations often observed in plants (Schlichting, 1989; Donohue and Schmitt, 1999; Volger et al., 1999; Brock and Weinig, 2007) and for the inconsistent results of the studies testing the Berg hypothesis.

The Berg hypothesis has often been interpreted in terms of phenotypic integration and modularity (Armbruster et al., 2004). These two variational properties of complex organisms are considered to result from genetic integration (via pleiotropy) and parcellation, respectively (Cheverud, 1996; Wagner, 1996; Wagner and Altenberg, 1996). Although the role of parcellation of genetic effects in the modularity and the decoupling of the phenotypic variation between vegetative and floral traits has been demonstrated in Arabidopsis thaliana (Juenger et al., 2005), our results also show that both environmental canalization and trait- or module-specific phenotypic plasticity can contribute to the decoupling of phenotypic variation. Environmental canalization of flower size (i.e. buffering of the floral traits against environmental variation) has long been recognized and confirmed by numerous studies, including this one, showing the relative consistency of floral traits compared with vegetative traits in the face of environmental variation. The second mechanism, consisting of reaction norms in opposite directions in leaf size and flower size, has not been recognized as a mechanism generating decoupling of the phenotypic variation. It remains unknown whether reaction norms in opposite directions evolved as the result of selection for decoupled variation or simply because of independent evolution of adaptive plasticity in flower size and plant size. In our case, the second scenario seems more likely due to the apparent disconnection between the mechanisms involved in the phenotypic response to temperature variation of the two types of traits (but see Chapin et al., 1993). Indeed, while the phenotypic response of flower size to temperature variation mimics the expected adaptive response to size variation or rarefication of pollinators with altitude, size variation of the vegetative parts essentially reflects the effect of temperature on growth (note that the variation in plant size due to local adaptation is limited compared with the temperature-induced variation).

Overall, these results show that phenotypic correlations among traits may depend on the environment, but also on the degree of environmental variation. While providing a possible explanation for the inconsistencies encountered among studies testing the Berg hypothesis, these results also underline the importance of distinguishing the different sources of covariation if we are to understand the evolution of morphological integration and modularity in flowering plants.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following. Figure S1: variation in the temperature of the two rooms during the experiment. Table S1: descriptive statistics of the recorded traits in each population and each environment. Table S2: results of the principal component analysis on the floral traits.

Supplementary Data
supp_111_5_935__index.html (1.2KB, html)

ACKNOWLEDGEMENTS

The authors would like to thank Jan Gundermann for essential help during data collection. Liv Antonsen helped with the plant maintenance. Johanne Maad provided the seeds used in this experiment. We thank Jenny Hagenblad, Mihaela Pavlicev, W. Scott Armbruster and two anonymous reviewers for commenting on earlier versions of the manuscript.

LITERATURE CITED

  1. Armbruster WS, Di Stilio VS, Tuxill JD, Flores TC, Velásquez-Runke JL. 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. 1999;86:39–55. [PubMed] [Google Scholar]
  2. Armbruster WS, Pélabon C, Hansen TF, Mulder C. Floral integration and modularity: distinguishing complex adaptation from genetic constraints. In: Pigliucci M, Preston KA, editors. Phenotypic integration studying the ecology and evolution of complex phenotypes. Oxford: Oxford University Press; 2004. pp. 23–42. [Google Scholar]
  3. Armbruster WS, Hansen TF, Pélabon C, Pérez-Barrales R, Maad J. The adaptive accuracy of flowers: measurement and microevolutionary patterns. Annals of Botany. 2009;103:1529–1545. doi: 10.1093/aob/mcp095. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Berg RL. A general evolutionary principle underlying the origin of developmental homeostasis. American Naturalist. 1959;93:103–105. [Google Scholar]
  5. Berg RL. The ecological significance of correlation pleiades. Evolution. 1960;14:171–180. [Google Scholar]
  6. Bingham RA, Orthner AR. Efficient pollination of alpine plants. Nature. 1998;391:238–239. [Google Scholar]
  7. Bingham RA, Ranker TA. Genetic diversity in alpine and foothill populations of Campanula rotundifolia (Campanulaceae) International Journal of Plant Sciences. 2000;161:403–411. doi: 10.1086/314272. [DOI] [PubMed] [Google Scholar]
  8. Bishop JA, Armbruster WS. Thermoregulatory abilities of Alaskan bees: effects of size, phylogeny and ecology. Functional Ecology. 1999;13:711–724. [Google Scholar]
  9. Bradshaw AD. Evolutionary significance of phenotypic plasticity. Advances in Genetics. 1965;13:115–155. [Google Scholar]
  10. Brock MT, Weining C. Plasticity and environment-specific covariances: an investigation of floral–vegetative and within flower correlations. Evolution. 2007;61:2913–2924. doi: 10.1111/j.1558-5646.2007.00240.x. [DOI] [PubMed] [Google Scholar]
  11. Brock MT, Weinig C, Galen C. A comparison of phenotypic plasticity in the native dandelion Taraxacum ceratophorum and its invasive congener T. officinale. New Phytologist. 2005;166:173–183. doi: 10.1111/j.1469-8137.2004.01300.x. [DOI] [PubMed] [Google Scholar]
  12. Byars SG, Papst W, Hoffmann AA. Local adaptation and cogradient selection in the alpine plant, Poa hiemata, along a narrow altitudinal gradient. Evolution. 2007;61:2925–2941. doi: 10.1111/j.1558-5646.2007.00248.x. [DOI] [PubMed] [Google Scholar]
  13. Carroll AB, Palladry SG, Galen C. Drought stress, plant water status, and floral trait expression in fireweed, Epilobium angustifolium (Onagraceae) American Journal of Botany. 2001;88:438–446. [PubMed] [Google Scholar]
  14. Caruso MC. Plasticity of inflorescence traits in Lobelia siphilitica (Lobeliaceae) in response to soil water availability. American Journal of Botany. 2006;93:531–538. doi: 10.3732/ajb.93.4.531. [DOI] [PubMed] [Google Scholar]
  15. Catley JL, Brooking IR, Davies LJ, Halligan EA. Temperature and irradiance effects on Sandersonia aurantiaca flower shape and pedicel length. Scientia Horticulturae. 2002;93:157–166. [Google Scholar]
  16. Chapin FS, III, Autumn K, Pugnaire F. Evolution of suites of traits in response to environmental stress. American Naturalist. 1993;142:S78–S92. [Google Scholar]
  17. Chalcoff VR, Ezcurra C, Aizen MA. Uncoupled geographical variation between leaves and flowers in a South-Andean Proteaceae. Annals of Botany. 2008;102:79–91. doi: 10.1093/aob/mcn057. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Cheverud JM. Developmental integration and the evolution of pleiotropy. American Zoologist. 1996;36:44–50. [Google Scholar]
  19. Conner JK, Sterling A. Testing hypotheses of functional relationships: a comparative survey of correlation patterns among floral traits in 5 insect-pollinated plants. American Journal of Botany. 1995;82:1399–1406. [Google Scholar]
  20. Conner JK, Sterling A. Selection for independence of floral and vegetative traits: evidence from correlation patterns in five species. Canadian Journal of Botany. 1996;74:642–644. [Google Scholar]
  21. Crawley MJ. The R book. Chichester: John Wiley & Sons Ltd; 2007. [Google Scholar]
  22. Cresswell JE. Stabilizing selection and the structural variability of flowers within species. Annals of Botany. 1998;81:463–473. [Google Scholar]
  23. Cresswell JE. Manipulation of female architecture in flowers reveals a narrow optimum for pollen deposition. Ecology. 2000;81:3244–3249. [Google Scholar]
  24. Cresswell JE, Robertson AW. Discrimination by pollen-collecting bumblebees among differentially rewarding flowers of an alpine wildflower, Campanula rotundifolia (Campanulaceae) Oikos. 1994;69:304–308. [Google Scholar]
  25. Delesalle VA, Mazer SJ. Nutrient levels and salinity affect gender and floral traits in the autogamous Spergularia marina. International Journal of Plant Sciences. 1996;157:621–631. [Google Scholar]
  26. Diggle PK. Development and the evolution of plant reproductive characters. In: Wyatt R, editor. Ecology and evolution of plant reproduction: new approaches. New York: Chapman and Hall; 1992. pp. 326–355. [Google Scholar]
  27. Donohue K, Schmitt J. The genetic architecture of plasticity to density in Impatiens capensis. Evolution. 1999;53:1377–1386. doi: 10.1111/j.1558-5646.1999.tb05402.x. [DOI] [PubMed] [Google Scholar]
  28. Dorken ME, Barrett SCH. Phenotypic plasticity of vegetative and reproductive traits monoecious and dioecious populations of Sagittaria latifolia (Alismataceae): a clonal aquatic plant. Journal of Ecology. 2004;92:32–44. [Google Scholar]
  29. Fabbro T, Körner CH. Altitudinal differences in flower traits and reproductive allocation. Flora. 2004;199:70–81. [Google Scholar]
  30. Fenster CB. Selection on floral morphology by hummingbirds. Biotropica. 1991;23:98–101. [Google Scholar]
  31. Fenster CB, Armbruster WS, Wilson P, Dudash MR, Thomson JD. Pollination syndromes and floral specialization. Annual Review of Ecology Evolution and Systematics. 2004;35:375–403. [Google Scholar]
  32. Frazee JE, Marquis RJ. Environmental contribution to floral trait variation in Chamaecrista fasciculata (Fabaceae caesalpinoideae) American Journal of Botany. 1994;81:206–215. [Google Scholar]
  33. Galen C. Why do flowers vary? The functional ecology of variation in flower size and form within natural plant populations. Bioscience. 1999;49:631–640. [Google Scholar]
  34. Galen C, Zimmer KA, Newport ME. Pollination in floral scent morphs of Polemonium viscosum: a mechanism for disruptive selection on flower size. Evolution. 1987;41:599–606. doi: 10.1111/j.1558-5646.1987.tb05830.x. [DOI] [PubMed] [Google Scholar]
  35. Giblin DE. Variation in floral longevity between populations of Campanula rotundifolia (Campanulaceae) in response to fitness accrual rate manipulation. American Journal of Botany. 2005;92:1714–1722. doi: 10.3732/ajb.92.10.1714. [DOI] [PubMed] [Google Scholar]
  36. Gottfried M, Pauli H, Futschik A, et al. Continent-wide response of mountain vegetation to climate change. Nature Climate Change. 2012;2:111–115. [Google Scholar]
  37. Gonzalo-Turpin H, Hazard L. Local adaptation occurs along altitudinal gradient despite the existence of gene flow in the alpine plant species Festuca eskia. Journal of Ecology. 2009;97:742–751. [Google Scholar]
  38. Hallgrímsson B, Jamniczky H, Young NM, Rolian C, Parsons TE, Boughner JC, Marcucio RS. Deciphering the Palimpsest: studying the relationship between morphological integration and phenotypic covariation. Evolutionary Biology. 2009;36:355–376. doi: 10.1007/s11692-009-9076-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Hansen TF, Pélabon C, Armbruster WS. Comparing variational properties of homologous floral and vegetative characters in Dalechampia scandens: testing the Berg hypothesis. Evolutionary Biology. 2007;34:86–98. [Google Scholar]
  40. Herrera CM. Multiplicity in unity. Chicago, IL: Chicago University Press; 2009. [Google Scholar]
  41. Herrera J. The variability of organs differentially involved in pollination, and correlations of traits in Genisteae (Leguminosae: Papilionoideae) Annals of Botany. 2001;88:1027–1037. [Google Scholar]
  42. Herrera J. Flower size variation in Rosmarinus officinalis: individuals, population and habitats. Annals of Botany. 2005;95:431–437. doi: 10.1093/aob/mci041. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Hodgins KA, Barrett SCH. Natural selection on floral traits through male and female function in wild populations of the heterostylous daffodil Narcissus triandrus. Evolution. 2008;62:1751–1763. doi: 10.1111/j.1558-5646.2008.00404.x. [DOI] [PubMed] [Google Scholar]
  44. Juenger T, Perez-Perez JM, Bernal S, Micol JL. Quantitative trait loci mapping of floral and leaf morphology traits in Arabidopsis thaliana: evidence for modular genetic architecture. Evolution and Development. 2005;7:259–271. doi: 10.1111/j.1525-142X.2005.05028.x. [DOI] [PubMed] [Google Scholar]
  45. Klingenberg CP. Morphological integration and developmental modularity. Annual Review of Ecology, Evolution and Systematics. 2008;39:115–132. [Google Scholar]
  46. Kobayashi S, Inoue K, Kato M. Mechanism of selection favoring a wide tubular corolla in Campanula punctata. Evolution. 1999;53:752–757. doi: 10.1111/j.1558-5646.1999.tb05369.x. [DOI] [PubMed] [Google Scholar]
  47. Larcher W. Physiological plant ecology: ecophysiology and stress physiology of functional groups. Berlin: Springer-Verlag; 2003. [Google Scholar]
  48. Lynch M, Walsh B. Genetics and analysis of quantitative traits. Sunderland, MA: Sinauer; 1997. [Google Scholar]
  49. Maad J, Armbruster WS, Fenster CB. Floral size variation in Campanula rotundifolia (Campanulaceae) along altitudinal gradients: patterns and possible selective mechanisms. Nordic Journal of Botany. 2013 in press. [Google Scholar]
  50. Magwene PM. New tools for studying integration and modularity. Evolution. 2001;55:1734–1745. doi: 10.1111/j.0014-3820.2001.tb00823.x. [DOI] [PubMed] [Google Scholar]
  51. Mal TK, Lovett-Doust J. Phenotypic plasticity in vegetative and reproductive traits in an invasive weed, Lythrum salicaria (Lythraceae), in response to soil moisture. American Journal of Botany. 2005;92:119–125. doi: 10.3732/ajb.92.5.819. [DOI] [PubMed] [Google Scholar]
  52. Maxwell B. Arctic climate: potential for change under global warming. In: Chapin FS III, Jeffries RL, Reynold JF, Shaver GR, Svoboda J, editors. Arctic ecosystem in a changing climate. San Diego: Academic Press; 1992. pp. 11–34. [Google Scholar]
  53. Nyman Y. Crossing experiments within the Campanula dichotoma group (Campanulaceae) Plant Systematics and Evolution. 1991;177:185–192. [Google Scholar]
  54. Nyman Y. Pollination mechanisms in six Campanula species (Campanulaceae) Plant Systematics and Evolution. 1992;181:97–108. [Google Scholar]
  55. Olson EC, Miller RL. Morphological integration. Chicago, IL: Chicago University Press; 1958. [Google Scholar]
  56. Olsson K, Ågren J. Latitudinal population differentiation in phenology, life history and flower morphology in the perennial herb Lythrum salicaria. Journal of Evolutionary Biology. 2002;15:983–996. [Google Scholar]
  57. Pélabon C, Armbruster WS, Hansen TF. Experimental evidence for the Berg hypothesis: vegetative traits are more sensitive than pollination traits to environmental variation. Functional Ecology. 2011;25:247–257. [Google Scholar]
  58. Pérez-Barrales R, Arroyo J, Armbruster WS. Differences in pollinator faunas may generate geographic differences in floral morphology and integration in Narcissus papyraceus (Amaryllidaceae) Oikos. 2007;116:1904–1918. [Google Scholar]
  59. Pigliucci M, Preston KA. Phenotypic integration studying the ecology and evolution of complex phenotypes. Oxford: Oxford University Press; 2004. [Google Scholar]
  60. R Development Core Team. The R project for statistical computing. Vienna: 2010. http://www.r-project.org/ [Google Scholar]
  61. Rosas-Guerrero V, Quesada M, Armbruster WS, Pérez-Barrales R, DeWitt Smith S. Influence of pollination specialization and breeding system on floral integration and phenotypic variation in Ipomoea. Evolution. 2010;65:350–364. doi: 10.1111/j.1558-5646.2010.01140.x. [DOI] [PubMed] [Google Scholar]
  62. Schlichting CD. The evolution of phenotypic plasticity in plants. Annual Review of Ecology and Systematics. 1986;17:667–693. [Google Scholar]
  63. Schlichting CD. Phenotypic plasticity in Phlox II. Plasticity of character correlations Oecologia. 1989;78:496–501. doi: 10.1007/BF00378740. [DOI] [PubMed] [Google Scholar]
  64. Schlichting CD, Pigliucci M. Phenotypic evolution: a reaction norm perspective. Sunderland, MA: Sinauer; 1998. [Google Scholar]
  65. Scobell SA, Scott PE. Visitors and floral traits of a hummingbird-adapted cactus (Echinocereus coccineus) show only minor variation along an elevational gradient. American Midland Naturalist. 2002;147:1–15. [Google Scholar]
  66. Searle SR. Phenotypic, genetic and environmental correlations. Biometrics. 1961;17:474–480. [Google Scholar]
  67. Shetler SG. Variation and evolution of the nearctic harebells (Campanula subsect. Heterophylla) Phanerogam Monography. 1982;11:1–516. [Google Scholar]
  68. Stebbins GL. Flowering plant: evolution above the species level. Cambridge, MA: Belknap Press; 1974. [Google Scholar]
  69. Stendel M, Christensen JH, Petersen D. Artic climate and climte change with a focus on Greenland. Advances in Ecological Research. 2008;40:13–43. [Google Scholar]
  70. Totland O. Environment-dependent pollen limitation and selection on floral traits in an alpine species. Ecology. 2001;82:2233–2244. [Google Scholar]
  71. Totland O. No evidence for a role of pollinator discrimination in causing selection on flower size through female reproduction. Oikos. 2004;106:558–564. [Google Scholar]
  72. Turesson G. The plant species in relation to habitat and climate. Hereditas. 1925;4:147–236. [Google Scholar]
  73. Van Kleunen M, Meier A, Saxenhofer M, Fischer M. Support for the predictions of the pollinator-mediated stabilizing selection hypothesis. Journal of Plant Ecology. 2008;1:173–178. [Google Scholar]
  74. Vogler DW, Peretz S, Stephenson AG. Floral plasticity in an iteroparous plant: the interactive effects of genotype, environment, and ontogeny in Campanula rapunculoides (Campnaulaceae) American Journal of Botany. 1999;86:482–494. [PubMed] [Google Scholar]
  75. Wagner GP. On the eigenvalue distribution of genetic and phenotypic dispersion matrices: evidence for a nonrandom organization of quantitative character variation. Journal of Mathematical Biology. 1984;21:77–95. [Google Scholar]
  76. Wagner GP. Homologues, natural kinds and the evolution of modularity. American Zoologist. 1996;36:36–43. [Google Scholar]
  77. Wagner GP, Altenberg L. Complex adaptations and evolution of evolvability. Evolution. 1996;50:967–976. doi: 10.1111/j.1558-5646.1996.tb02339.x. [DOI] [PubMed] [Google Scholar]
  78. Waitt DE, Levin DA. Phenotypic integration and plastic correlations in Phlox drummondii (Polemoniaceae) American Journal of Botany. 1993;80:1224–1233. [Google Scholar]
  79. Waitt DE, Levin DA. Genetic and phenotypic correlation in plants: a botanical test of Cheverud's conjecture. Heredity. 1998;80:310–319. [Google Scholar]
  80. Wolfe LM, Mazer SJ. Patterns of phenotypic plasticity and their fitness consequences in wild radish (Raphanus sativus, Brassicaceae) International Journal of Plant Sciences. 2005;166:631–640. [Google Scholar]
  81. Zelditch ML, Swiderski DL. Epigenetic interactions: the developmental route to functional integration. In: Hallgrímsson B, Hall BK, editors. Epigenetics: linking genotype and phenotype in development and evolution. Berkeley& Los Angeles: University of California Press; 2011. pp. 290–316. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Data
supp_111_5_935__index.html (1.2KB, html)
supp_mct050_mct050supp2.pdf (192.7KB, pdf)
supp_mct050_mct050supp1.docx (143.8KB, docx)

Articles from Annals of Botany are provided here courtesy of Oxford University Press

RESOURCES