Relative embryo length as an adaptation to habitat and life cycle in Apiaceae

Authors


Author for correspondence:
Filip Vandelook
Tel: +49 6421 2822053
Email: filip.vandelook@biologie.uni-marburg.de

Summary

  • The factors driving the evolution of the relative embryo length in Apiaceae were examined. We tested the hypothesis that seeds with large relative embryo length, because of more rapid germination, are beneficial in dry and open habitats and for short-lived species. We also analyzed to what extent delayed germination as a result of embryo growth can be considered a dormancy mechanism.
  • Hypotheses were tested by correlating the relative embryo length with other plant traits, habitat and climatic variables. The adaptive nature of the relative embryo length was determined by comparing the performance of a pure drift, Brownian motion (BM) model of trait evolution with that of a selection–inertia, Ornstein–Uhlenbeck (OU) model.
  • A positive correlation of the relative embryo length with germination speed and negative correlations with the amount of habitat shade, longevity and precipitation were found. An OU model, in which the evolution of longer embryos corresponded to a transition to habitats of high light, or to a short life cycle, outperformed significantly a BM model.
  • The results indicated that the relative embryo length may have evolved as an adaptation to habitat and life cycle, whereas dormancy was mainly related to temperature at the sampling sites.

Introduction

Angiosperm seeds usually contain not only an embryo, but also nutrient reserves consisting of either endosperm or perisperm. To understand evolutionary changes in the size of the embryo relative to the amount of nutritive tissue, both phylogenetic and ecological factors should be considered (Nikolaeva, 2004). Over 100 yr ago, the ecological significance of relative embryo size was already recognized (Goebel, 1898; Crocker, 1916; Findeis, 1917). The importance of phylogeny in the distribution of relative embryo size among angiosperms became clear from the work of Martin (1946). Species with small embryos embedded in copious endosperm are generally considered to be the plesiomorphic condition in angiosperms, whereas more derived species often have a more developed embryo (Martin, 1946; Stebbins, 1974; Forbis et al., 2002). Small relative embryos sizes are typical of primitive taxa, such as the Ranunculales in the Eudicots and some representatives of the ANITA grade, the most basal angiosperms (Chien et al., 2011; but see Baskin & Baskin, 2007). Verdú (2006) suggested that this evolutionary trend towards increased relative embryo size was not driven by either anagenesis or cladogenesis, but that the evolution of embryo size rather occurred as a passive process away from a minimum size.

The storage of food reserves in an external tissue, rather than in the embryo, has been suggested to be related to germination timing and seedling vigor (Stebbins, 1974). The predicted positive relation between germination speed and relative embryo size was confirmed for Mediterranean plant species (Vivrette, 1995), but later disputed when phylogeny was taken into account (Verdú, 2006). It has been argued that a large relative embryo size is especially beneficial in dry habitats, where rapid germination during short wet periods is advantageous (Hodgson & Mackey, 1986; Vivrette, 1995). Embryo size may also be related to adult longevity, as short-lived species often have long-lived seeds that are incorporated in the soil seed bank (Rees, 1993; Thompson et al., 1998). Seeds in a seed bank usually germinate during short spells of suitable environmental conditions, for example after disturbance of soil or vegetation (Fenner & Thompson, 2005). The production of seeds with a small underdeveloped embryo that requires an extensive period of embryo growth before germination would be disadvantageous under such conditions. Species with small embryos are, however, common in moist habitats such as woodlands or damp grasslands (Baskin & Baskin, 1988, 1998). It is in these environments that seeds are imbibed for an uninterrupted sufficiently long period for extensive embryo growth to be completed (Fenner & Thompson, 2005).

Some angiosperm taxa typically have other mechanisms not related to embryo size that control germination rate and dormancy. These mechanisms may obscure analyses of selective forces that drive changes in relative embryo size across all angiosperms. Therefore, we confined our study to the Apiaceae, which all require post-dispersal embryo growth. This family includes almost 3500 species divided into four subfamilies of which the Saniculoideae and Apioideae are by far the largest (Plunkett & Lowry, 2001; Magee et al., 2010). A southern African origin of these two subfamilies, followed by migrations northwards into Eurasia, has been suggested (Calvino et al., 2006, 2008). Fruits typically consist of two mericarps, each containing a seed with copious nuclear endosperm and a small embryo (Corner, 1976). Despite these general characteristics, considerable interspecific variation in relative embryo size exists within the Apiaceae, ranging from species with tiny rudimentary embryos to species with embryos more than one-half the length of the seed (Martin, 1946). A period of post-dispersal embryo growth, varying from 1–2 wk to several months, usually precedes root protrusion and germination (reviewed in Vandelook, 2009). These seed characteristics, together with the large variation in habitats in which Apiaceae grow, and the enormous amount of effort that has been put into resolving the phylogenetic relationships between Apiaceae during the last 15 yr (see Downie et al., 2010), make the Apiaceae a good choice for studying questions related to factors driving embryo size evolution.

Although the functional ecology of embryo size has been studied extensively, the environmental conditions driving the evolution of relative embryo size, and correlations with other plant traits, have never been analyzed explicitly. In this study, we first aim to analyze the factors driving the evolution of the relative embryo length in the Apiaceae by applying phylogenetic regressions. We tested the hypotheses that seeds with a large relative embryo length germinate faster and to a higher percentage. As a result of the advantages related to faster germination, we expect species with a large relative embryo length to be frequent in open, dry habitats and among short-lived species. Although previous work has suggested that only weak relationships between plant traits and climatic factors are to be expected, because of often considerable differences in plant traits between coexisting species (Wright et al., 2004; Moles et al., 2005), we expect more species with a large relative embryo length in dry regions.

Finally, we tested to what extent the relative embryo length is an adaptation to habitat and plant longevity by means of Ornstein–Uhlenbeck (OU) models of trait evolution (Hansen, 1997). If the relative embryo length is a strongly adaptive trait, the OU models of trait evolution should outperform a random walk or Brownian motion (BM) model of trait evolution. This method also enabled us to examine whether the evolutionary optimal relative embryo length differs according to habitat conditions and plant longevity. If the relative embryo length is a strongly adaptive trait, we should find the optimal relative embryo length to be lowest in moist and shady habitats, and in perennial species.

Materials and Methods

Data sampling

A dataset of 275 Apiaceae species was compiled, containing data on seed traits, plant traits, niche characteristics and the climate conditions in which the species grow. Seed material of the study species was obtained from seed banks, botanical gardens and by field sampling. A complete species list with detailed information on the source of the data and references is available in Supporting Information Table S1.

The mean relative embryo length (embryo length × seed length−1) of each species was determined by incubating 20 seeds in water for 24 h, cutting them in half and measuring embryo and seed length under a dissecting microscope equipped with an ocular micrometer. Seeds were incubated in water for 24 h before measurement. Data on seed dry mass were either obtained from the seed information database (Royal Botanic Gardens Kew Seed Information Database, 2008) or, if not available, by weighing 100 air-dried seeds and calculating the mean mass per seed. Germination data were based on germination trials performed at the Millennium Seed Bank (Kew), using 20 seeds incubated on moist filter paper in Petri dishes and placed in temperature-controlled incubators at 5 or 15°C for at least 30 d. Germinated seeds were counted and usually discarded weekly, until no further germination occurred for at least 2 wk. The germination rate was expressed as the mean time to germinate = (∑(gd × d))/D, where gd is the number of seeds that germinated on day d from sowing and D is the total number of seeds that germinated in the test. Species with < 10% germination were excluded from the analyses of germination rate. Before testing germination, seeds were stored for variable periods of time at − 18°C after drying to equilibrium at 15% relative humidity (RH). No correlation existed between the storage time (d) and mean time to germinate at 5°C (r2 = 0.04, P = 0.08) or 15°C (r2 < 0.01, P = 0.56), or with the final germination percentage at 5°C (r2 = 0.01, P = 0.36) or 15°C (r2 < 0.01, P = 0.95). Data on the germination rate of seeds incubated at 5 and 15°C were obtained for a subset of 60 and 59 species, respectively. The final germination percentages at 5°C and 15°C were determined for 69 and 122 species, respectively.

Maximum plant height and adult longevity were retrieved from floras and online databases. The species were divided into three adult longevity classes: annuals, biennials and perennials. When plants were listed as both annual and biennial, or as monocarpic perennial, we assigned them to the biennial class. Habitat moisture and habitat shade characteristics were based on statements of habitats in floras and online databases. Species were classified into three classes according to habitat moisture (1, dry; 2, moist; 3, wet) and habitat shade (1, open; 2, semi-shaded; 3, shaded). The geographical coordinates of the sampling sites of most species were given by the institutes that provided seed material. If coordinates were not available, coordinates of species occurrence were retrieved from the Global Biodiversity Information Facility website (GBIF Data Portal, 2011). When more than one set of coordinates was available, we chose coordinates on the basis of two criteria: the most recent observation and the observation closest to the presumed sampling site of the species. Using these coordinates, climatic data associated with the sampling or occurrence sites were estimated using the Worldclim database version 1.3 (Hijmans et al., 2005), accessed through DIVA-GIS version 7.1.7 (Hijmans et al., 2001). Climatic data consisted of the annual mean temperature, mean diurnal temperature range (mean of monthly maximum minus minimum temperature), maximum temperature of the warmest month, minimum temperature of the coldest month and annual precipitation. In addition, the altitude of the sampling sites was incorporated into our dataset. The relative embryo length, seed mass, plant height, annual precipitation and mean time to germinate were log-transformed before statistical analyses to meet the assumption of a normal distribution of residuals.

Construction and dating of phylogeny

To construct a phylogenetic tree, 323 sequences in total of the internal transcribed spacer (ITS) of the order Apiales were obtained from GenBank (Table S2) using the Geneious software package v 4.7.5. (Biomatters Ltd., Auckland, New Zealand). Of these, one species belongs to the Griseliniaceae, one to the Pennantiaceae, 10 to the Pittosporaceae, 48 to the Araliaceae and 263 to the Apiaceae. Brighamia insignis, Asyneura campanuloides and Campanula acarnanica (Campanulaceae, Asterales), and Haplocarpha scaposa (Asteraceae, Asterales), were used as outgroup taxa. The usefulness of ITS sequence data to resolve the intergeneric phylogenetic relationships within the species-rich Apiaceae family has been shown by Downie et al. (2010). Initial sequence alignment was performed with MUSCLE using default parameters (Edgar, 2004) and subsequently fine-tuned by hand in MacClade 4.05 (Maddison & Maddison, 2002). Phylogenetic analyses of the nuclear ribosomal ITS datasets were carried out using the probabilistic maximum likelihood (ML) method. ML analyses were performed using the RaxML search algorithm (Stamatakis et al., 2005) under the GTRGAMMA approximation of rate heterogeneity for each gene (Stamatakis, 2006). Five hundred bootstrap trees were inferred using the RaxML Rapid bootstrap algorithm (ML-BS) to provide support values for the best-scoring ML tree. The ML topology with the highest likelihood obtained with RaxML was used for further dating analysis. A strict molecular clock for the combined dataset had to be rejected using a χ2 likelihood ratio test (P < 0.001). Because of rejection of the null hypothesis of constant rate, age estimates for the Apiales and Apiaceae were therefore inferred with a relaxed clock model (penalized likelihood algorithm; Sanderson, 2002) implemented in the r8s software package (Sanderson, 2004). The most favorable rate-smoothing penalty parameter was calculated in r8s with a statistical cross-validation method. As a result of a rather precarious fossil record for the Apiacaeae (Martinez-Millán, 2010), we used a fossil-based age estimate of the order Apiales to infer dating estimates within the Apiaceae. This calibration point was obtained from the study of Janssens et al. (2009), who analyzed the whole asterid clade using nine different reference fossils to calibrate their phylogeny. The sophisticated method of Janssens et al. (2009) corroborated the results of Bremer et al. (2004), estimating the stem group age at 87 and 84 million yr (ma), respectively.

Phylogenetic regressions were performed using subtrees derived from this original phylogenetic tree. These subtrees were constructed using Phylocom 4.1 software (Webb et al., 2009). Most taxa analyzed (> 95%) were already included in the original tree. For the remaining < 5% of the taxa, no ITS sequence was available; instead, the ITS sequence of a congeneric taxon was used.

Analyses

Phylogenetic regression  We explored our ecological and evolutionary hypotheses by examining correlations between species traits or between a trait and habitat or climate characteristic, whilst correcting for the nonindependence of species as a result of phylogenetic relatedness (Freckleton & Pagel, 2006). Recent simulations have shown that the decision on whether to use a phylogenetic regression cannot be based solely on measures of the phylogenetic signal calculated on individual variables in the analysis (Hansen & Orzack, 2005; Labra et al., 2009; Revell, 2010). Instead, we applied an alternative approach, consisting of an ML procedure, whereby the phylogenetic signal and regression model were estimated simultaneously. As a measure of the phylogenetic signal, we applied Pagel’s λ (Pagel, 1999). The phylogenetic regression was performed with a phylogenetic tree whose internal branches were all multiplied by λ, leaving tip branches as their original length. The λ statistic proposed by Pagel (1997, 1999) measures how well a BM model fits the data, that is, measures the phylogenetic signal (Lynch, 1991; Freckleton et al., 2002; Housworth et al., 2004). When λ equals zero, related taxa are not more similar than expected by chance and the trait is evolving as in star-like phylogeny (Pagel, 1999). In such a scenario, phylogenetic correction becomes redundant. Significant phylogenetic signal or clumping of trait states on the phylogenetic tree occurs when λ > 0, meaning taxa are more similar than expected by chance. When λ = 1, the trait is evolving following a constant variance random walk or BM model. If 1 > λ > 0, traits are less similar among species than expected from their phylogenetic relationships, but more similar than expected by chance.

Simple and multiple generalized least-squares regressions were performed, with the relative embryo length as the dependent variable and all other variables as independent variables. The three discrete variables in our dataset, adult longevity, habitat moisture and light, were implemented as continuous variables in this analysis. Using a backward selection approach, we successively excluded the least significant variables from the model until the Akaike information criterion (AIC; Akaike, 1973) was minimal. Correlations between other traits and habitat variables are included as supplementary data (Table S3).

The relationship between germination rate and final percentage germination, on the one hand, and embryo length and all other variables measured, on the other, was tested by means of multiple ordinary least-squares regression (OLS) with germination rate and the final germination percentage (at 5°C and 15°C) as dependent variables. In all these analyses, λ = 0 (i.e. no phylogenetic signal) resulted in the lowest AIC score, indicating that the phylogenetic regressions collapsed to OLS regressions. Again, we used a backward selection approach to exclude the least significant variables from the model until AIC was minimal. Analyses were performed using the APE (Paradis, 2006) and nlme (Pinheiro et al., 2009) packages in R version 2.12.0 (R Development Core Team, 2009).

Models of trait evolution  We tested the adaptive nature of the relative embryo length using two models reflecting hypothetical selective regimes on the relative embryo length. The simplest model is a BM mode of trait evolution which assumes that embryo length evolved following a pure drift process. The second model, an OU model, is a simple linear model that allows the quantification of the effects of both natural selection and inertia (Hansen, 1997; Butler & King, 2004; Hansen et al., 2008). In its simplest form, the model assumes that the relative embryo length evolves towards a single hypothetical optimum θs. The model also includes a parameter α measuring the rate of adaptation towards the optimum and a stochasticity component σ which is a measure of the intensity of the random fluctuations in the evolutionary process. If α is large, species will adapt very rapidly to new conditions, whereas a low α makes ancient adaptations relatively more important (Hansen, 1997). We calculated a more intuitive measure of the phylogenetic signal in this OU model: phylogenetic half-life, t1/2 = loge(2)/α. This half-life indicates how long it takes before adaptation to a new selective regime is expected to be more influential than the constraints from the ancestral state (Hansen, 1997).

As the evolutionary optimum is expected to differ according to habitat conditions and plant longevity, we further refined this model so that it included different evolutionary optima based on different hypotheses regarding evolutionary adaptation to different levels of habitat shade (θ1, open; θ2, semi-shaded; θ3, shaded), habitat moisture (θ1, dry; θ2, moist; θ3, wet) and evolution of adult longevity (θ1, annual; θ2, biennial; θ3, perennial). Models of evolutionary adaptation to habitat conditions and evolution of longevity were based on a parsimony analysis and on an ML analysis of the three trait states of each of these variables performed in Mesquite, version 2.72 (Maddison & Maddison, 2009). Parsimony analysis was unable to provide unequivocal results for all nodes, especially basal nodes for habitat variables. Equivocal nodes were assigned a value of four, thus allowing evolution towards a fourth optimum (θ4). In the ML analysis, the nodes were assigned the most likely trait state.

The performance of the models was tested by means of AIC. Percentile confidence intervals for the model parameters were computed using parametric bootstrap with 10 000 replicates. Analyses were performed in the Geiger (Harmon et al., 2009) and OUCH (Butler & King, 2004) packages for R.

Results

Phylogenetic reconstruction and dating analysis

The aligned nuclear ribosomal ITS data matrix consisted of 1274 characters, 385 of which were variable. ML analysis of the ITS sequences yielded a highly resolved phylogeny that was fairly congruent with the results of Downie et al. (2010) (Supporting Information Fig. S1). A small number of discrepancies between the two phylogenies were probably the result of differences in sampling size and alignment. However, these small disparities had only a minor influence on the ecological analyses carried out in the present study, as the main branching events leading to different subfamilies in Apiaceae were the same for both studies. Interestingly, our results illustrate that a clade consisting of the genus Hydrocotyle and the family Araliaceae are a sister group to the Apiaceae. Applying the divergence time of the previously estimated age of the Apiales crown group by Janssens et al. (2009), we dated the split between Apiaceae and Araliaceae–Hydrocotyle at 73.5 ma, whereas the crown group of Apiaceae was estimated at 64.3 ma. Age estimations of the main Apiaceae clades are listed in Table 1.

Table 1.   Estimated age (million years, ma) for crown and stem groups of the main Apiaceae lineages
Apiaceae lineagePenalized likelihood age estimate
StemCrown
Saniculoideae64.360.7
Heteromorpheae63.460.7
Bupleureae59.337.4
Pleurospermeae52.5
Oenantheae46.233.4
Erigenieae48.1
Smyrnieae36.121.3
Scandiceae41.337.4
Aciphylleae38.920.7
Apioid superclade39.136.5

Phylogenetic regression

Almost all the variables tested, including the relative embryo length, showed a significant phylogenetic signal, with λ differing significantly from zero (Table 2). Altitude (λ = 0, CI 0, 0.28), mean time to germinate (λ = 0.33, CI 0, 0.95) and final germination percentage (λ = 0, CI 0, 0.91) at 5°C were randomly distributed across the phylogenetic tree. Phylogenetic signal was also absent in mean time to germinate at 15°C (λ = 0, CI 0, 0.82), but not in the final germination percentage for seeds tested at this temperature (λ = 0.546, CI 0.22, 0.80). λ differed significantly from unity for all variables tested, implying that traits are less similar between species than expected from their phylogenetic relationships, and that evolution did not result from a pure drift process.

Table 2.   Simple phylogenetic regressions fitted by maximum likelihood between the relative embryo length and other plant traits, habitat or climatic conditions
 Pagel’s λPhylogenetic signal
coefficientSEt valueλλ
  1. Phylogenetic generalized least-squares regressions were based on a model in which λ was estimated simultaneously. The last column provides an estimate of the phylogenetic signal λ in the single variables tested. Numbers in parentheses denote the confidence intervals. ns, P > 0.05; *, P < 0.05; ***, P < 0.001; n = 275.

Log[relative embryo length]0.78 (0.59, 0.89)
Log[seed mass] (mg)−0.180.03−6.13***0.720.81 (0.66, 0.90)
Adult longevity−0.070.02−3.69***0.740.90 (0.78, 0.96)
Log[plant height] (cm)−0.060.04−1.50ns0.770.76 (0.54, 0.97)
Habitat moisture−0.010.02−0.63ns0.770.64 (0.43, 0.81)
Habitat shade−0.080.02−3.89***0.670.75 (0.53, 0.88)
Altitude (m)−0.010.01−0.95ns0.770 (0, 0.28)
Mean temperature (°C)0.010.010.15ns0.770.50 (0.19, 0.74)
Max. temperature (°C)0.010.011.30ns0.770.58 (0.30, 0.76)
Min. temperature (°C)−0.010.01−0.81ns0.790.40 (0.10, 0.70)
Temperature fluctuation (°C)0.010.010.19ns0.780.76 (0.58, 0.87)
Log[annual precipitation] (mm)−0.090.05−1.97*0.760.62 (0.28, 0.83)

Significant negative correlations (P < 0.05) between the relative embryo length and seed mass, adult longevity, habitat shade and annual precipitation resulted from both simple and multiple linear regressions (Tables 2, 3). No significant correlations (P > 0.05) with any of the other variables tested were found. The estimate of λ for the relative embryo length (λ = 0.78, CI 0.59, 0.89) was very similar to the value of λ estimated in the simple regressions between the relative embryo length and all other variables measured (λ ranging from 0.67 to 0.79; Table 2). This contrasts with the considerably smaller estimate of λ in the multivariate model (λ = 0.53).

Table 3.   Multiple phylogenetic regression fitted by maximum likelihood between the relative embryo length and other plant traits, habitat and climatic conditions
Log[relative embryo length]
 CoefficientSEt valueP value
  1. The regression was based on a model in which λ was estimated simultaneously. Nonsignificant terms were removed successively until only significant terms remained.

  2. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Log[seed mass] (mg)−0.170.03−5.75***
Adult longevity−0.040.02−2.23*
Habitat shade−0.070.02−3.50***
Log[annual precipitation] (mm)−0.110.04−2.49**

The multiple regression models with mean time to germinate at 5°C and 15°C as dependent variables were highly significant and a large amount of variation was explained (F4, 54 = 18.2, r2 = 0.59 and F5, 53 = 12.1, r2 = 0.48, respectively). For seeds germinated at 5°C, the relative embryo length and mean temperature at collection sites were highly significant predictors (P < 0.001) of germination rate, and both were negatively related to mean time to germinate (Table 4). The mean time to germinate at 5°C was also significantly negatively correlated with seed mass (P < 0.05), meaning that large seeds germinate more rapidly. The picture was different for seeds germinated at 15°C (Table 4). For these seeds, highly significant predictors (P < 0.001) were seed mass and mean diurnal temperature range, which were both positively related to mean time to germinate. As at 5°C, the relative embryo length was significantly negatively correlated with mean time to germinate at 15°C. Germination rate did not correlate significantly with adult longevity in the multiple regression models at either 5°C or 15°C.

Table 4.   Multiple ordinary least-squares regression with mean time to germinate and final germination percentage (at 5 and 15°C) as dependent variables and plant traits, habitat and climatic conditions as independent variables
 CoefficientsSEt valueP value
  1. Nonsignificant terms were removed successively until only significant terms remained.

  2. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Log[mean time to germinate] at 5°C (d); n = 59
 Log[relative embryo length]−0.630.12−5.24***
 Log[seed mass] (mg)−0.120.05−2.28*
 Habitat shade−0.220.08−2.79**
 Habitat moisture0.180.082.37*
 Mean temperature (°C)−14.073.59−3.92***
Log[mean time to germinate] at 15°C (d); n = 60
 Log[relative embryo length]−0.350.16−2.12*
 Log[seed mass] (mg)0.280.073.94***
 Altitude−0.010.01−2.27*
 Temperature fluctuation (°C)0.110.042.87***
 Maximum temperature (°C)−0.070.02−3.57**
Seed germination at 5°C (%); n = 69
 Mean temperature (°C)−0.040.01−3.16***
Seed germination at 15°C (%); n = 122
 Log[relative embryo length]0.310.083.81***
 Log[seed mass] (mg)−0.110.04−2.73**
 Altitude (m)0.010.013.72***
 Temperature fluctuation (°C)−0.060.03−2.39*
 Minimum temperature (°C)0.040.015.28***
 Maximum temperature (°C)0.030.012.33*

The final germination percentage at 5°C was only significantly (negative) related to the mean annual temperature (t1,67 = − 3.17, r2 = 0.13, P = 0.002) and minimum temperature during the coldest month (t1,67 = − 2.62, r2 = 0.09, P = 0.01). Only the mean annual temperature was significant in a multiple regression model. The minimum temperature during the coldest month and altitude were two of the most important predictors for germination percentage at 15°C (F6, 115 = 18.7, r2 = 0.47; Table 4). Interestingly, a highly significant positive correlation (P < 0.001) existed between the final germination percentage and both minimum temperature and altitude. The germination percentage at 15°C was also significantly positively correlated with the relative embryo length (P < 0.001) and negatively correlated with the seed mass (P < 0.01).

Models of embryo size evolution

The performance and parameters of all models tested are summarized in Table 5. Based on AIC, the best-fitting model is an OU model with multiple evolutionary optima as a function of habitat shade. Only minor differences in fits were found between models with multiple evolutionary optima based on habitat shade (AIC: − 87.1 to − 89.3), habitat moisture (AIC: − 74.9 to − 75.3) and adult longevity (AIC: − 81.2 to − 82.1).

Table 5.   Performance and parameters of eight models of the relative embryo length evolution in Apiaceae
 BMOU.sOU habitat moistureOU habitat shadeOU adult longevity
 ParsimonyML ParsimonyML ParsimonyML
  1. For each model, the likelihood values (− 2 × log L) and the Akaike information criterion (AIC) are given. Ornstein–Uhlenbeck (OU) models with multiple optima were based on trait state reconstruction using parsimony or maximum likelihood (ML) analyses. BM, Brownian motion model; σ, magnitude of stochasticity component; λ, phylogenetic signal; α, rate of adaptation; θs, optimum estimated for evolution towards a single optimum; θ1→4, optima (back-transformed) estimated for evolution towards multiple optima. Numbers in parentheses denote 95% confidence intervals.

− 2 × log L−28.5−81.1 −86.9−85.3 −99.1−99.3 −93.2−92.1
df13 65 65 65
AIC−24.5−75.1 −74.9−75.3 −87.1−89.3 −81.2−82.1
σ0.47 (0.43, 0.51)0.59 (0.70, 0.89) 0.73 (0.62, 0.88)0.72 (0.61, 0.86) 0.73 (0.61, 0.88)0.73 (0.62, 0.88) 0.70 (0.59, 0.83)0.69 (0.59, 0.82)
λ    
α4.63 (3.11, 6.83) 5.58 (3.69, 8.34)5.24 (3.49, 7.86) 5.81 (3.86, 8.70)5.84 (3.91, 8.75) 5.05 (3.32, 7.46)4.90 (3.30, 7.24)
θs0.19 (0.21, 0.24)    
θ1Dry0.23 (0.19, 0.27)0.24 (0.20, 0.28)Open0.24 (0.21, 0.27)0.24 (0.21, 0.27)Annual0.28 (0.22, 0.37)0.30 (0.23, 0.38)
θ2Moist0.19 (0.17, 0.22)0.20 (0.17, 0.23)Semi-shade0.19 (0.15, 0.23)0.19 (0.16, 0.23)Biennial0.27 (0.18, 0.39)0.29 (0.20, 0.38)
θ3Wet0.26 (0.19, 0.38)0.26 (0.19, 0.34)Shade0.12 (0.08, 0.16)0.12 (0.08, 0.16)Perennial0.20 (0.17, 0.22)0.20 (0.17, 0.22)
θ4Equivocal0.36 (0.17, 0.71)Equivocal0.25 (0.10, 0.50)Equivocal0.48 (0.17, 1.00)

The stochasticity component σ did not differ significantly between the models tested (Table 5). A λ value significantly different from unity and an α value significantly different from zero both indicate that the evolution of the relative embryo length in Apiaceae was not a pure drift process. The rate of adaptation (α) ranged between 4.63 in the OU.s model with a single optimum and 5.84 in the OU model as a function of habitat shade reconstructed using ML. No significant differences in α occurred between the OU models tested. The phylogenetic half-life varied from 150 000 yr in the OU.s model to 120 000 yr in the OU with optima as a function of habitat shade, indicating a high rate of adaptation in the relative embryo length.

The optimum relative embryo lengths towards which species evolved were considerably higher in annual (θ = 0.28–0.30) and biennial (θ = 0.27–0.29) species than in perennial species (θ = 0.20), with only a small overlap in the 95% confidence intervals (Table 5). Similarly, the optimum relative embryo length was significantly higher for species growing in open habitats (θ = 0.24) than for species growing in forests (θ = 0.12), with optima for species growing in semi-shaded habitats (θ = 0.19) situated in between. Species from moist habitats had a lower optimum relative embryo length (θ = 0.19–0.20) than species from dry (θ = 0.23–0.24) and wet (θ = 0.26) habitats, but the overlap in confidence intervals indicated that these differences were not significant. The equivocal node values in the parsimony analyses resulted in an optimum (θ4) higher than that of the known node values for all three traits tested. These optima also had very large confidence intervals. The model parameters of the multiple optima OU models were very similar for traits reconstructed by ML and parsimony.

Discussion

Our main conclusion is that the relative embryo length in Apiaceae has not evolved by a pure drift process. The evolution of the relative embryo length is tightly related to that other very important functional seed trait, seed mass. We also provide evidence that evolution towards a larger embryo length is associated with an increased germination rate. Evolutionary changes in embryo length can be considered as adaptations to particular habitat conditions (light vs shade) and adult longevity.

The hypothesis of an increased germination rate and decreased dormancy with increasing embryo length was largely confirmed. The positive correlation between the germination rate and relative embryo length at the two temperature conditions tested indicates that species requiring more embryo growth before germination also take longer to germinate. The absence of such a relationship across all angiosperms may have two causes (cf. Verdú, 2006). The evolution towards seeds that almost completely lack endosperm has opened the door to the evolution of alternative germination regulation mechanisms, such as physical dormancy (Baskin & Baskin, 1998). Second, many species disperse seeds with copious endosperm, but require no embryo growth before germination. These species are widespread in large families such as the Poaceae and Cyperaceae (Martin, 1946). Seed dormancy in Apiaceae, here expressed as the total percentage germination, decreases with increasing relative embryo length for seeds incubated at 15°C. No relationship between seed dormancy and embryo length exists for seeds incubated at 5°C. We should remark here that the germination of several species that had already germinated to a high percentage at higher temperatures (15 or 20°C) was not tested at 5°C. This may have resulted in a bias towards species germinating only at lower temperatures (< 10°C) in the germination test at 5°C.

The phylogenetic regressions revealed significant negative correlations between the relative embryo length and seed mass, adult longevity, habitat shade and annual precipitation. These results support earlier hypotheses that embryo length is tightly related to the ecology and habitat of a species (Fenner & Thompson, 2005). A well-developed embryo, and thus fast germination, is especially advantageous for species that experience only short spells of suitable climatic or biotic conditions for embryo growth and germination, as is the case for species growing in open and dry habitats (Vivrette, 1995). Seeds with a large relative embryo length also appear to be advantageous for short-lived species. We argue that this advantage is related to the formation of a soil seed bank. Short-lived species often have long-lived seeds to reduce the impact of environmental variability (Venable & Brown, 1988), and small seeds that become easily buried into the soil (Thompson et al., 1998). The latter correlation was also found for the Apiaceae studied here (see Table S3). As seeds incorporated in a soil seed bank only have a small window of opportunity to germinate after a disturbance event, it would be particularly advantageous to germinate rapidly. In line with this, Grime et al. (1981) also found higher germination rates in annual forbs and grasses than in perennial forbs in an analysis of 403 British species. By contrast, species growing in predictable habitats, such as temperate forests, may be able to ‘afford’ small embryos, because conditions for embryo development are suitable over a prolonged period of time.

Seed dormancy has been considered to be a bet-hedging mechanism for which trade-offs with other bet-hedging traits, such as seed size and plant longevity, have been observed (Venable & Brown, 1988; Rees, 1996). Although species with a small relative embryo length clearly take longer to germinate, as is indicated by an increased mean time to germinate and a higher dormancy, the relative embryo length should not be considered as a bet-hedging mechanism. Otherwise, we would expect it to be positively related to seed size and plant longevity, which is not the case. This contrast with earlier studies (Rees, 1993, 1996) can be traced back to differences in the way in which dormancy is defined. A dormant seed has been defined by physiologists and ecologists as a seed that ‘does not have the capacity to germinate in a specified period of time under any combination of normal physical environmental factors that are otherwise favorable for its germination’ (Baskin & Baskin, 2004). In studies on trade-offs between life history traits, dormancy has usually been defined as seed persistence, a seed characteristic that is poorly related to seed dormancy (Thompson et al., 2003). The way in which we measured dormancy also does not agree with the definition proposed by Baskin & Baskin (2004), because we did not restrict germination to a specified period of time. Some of the species that did not germinate when placed at 15°C did germinate when placed at 5°C, and vice versa. The final germination percentage in the Apiaceae studied was mainly related to temperature conditions and altitude at the collection sites. More specifically, the positive correlation between environmental temperature and germination percentage at 15°C and the negative correlation between environmental temperature and germination at 5°C indicate that evolution towards germination at lower temperatures has occurred in species growing in colder regions. Germination at low temperatures (< 10°C) prevents seeds in regions with a clear winter season from germinating in summer or autumn, which is a mechanism to decrease the risk of frost damage during the vulnerable seedling stage in winter (Baskin & Baskin, 1998; Nikolaeva, 2004).

Like many other morphological traits, the relative embryo length in Apiaceae showed a strong phylogenetic component (λ > 0), suggesting that a phylogenetic correction in comparative analyses is necessary. λ < 1 and α > 0 also show that evolution has not been a mere random walk process. The deviation from a pure drift process of evolution can be explained by the adaptive nature of the relative embryo length, as indicated by the fact that the OU models significantly outperformed the BM model. It is interesting that evolution towards different selective optima as a function of habitat shade and adult longevity explained the data significantly better than evolution towards a single optimum. The evolutionary optima towards which the relative embryo length evolves were higher for more open habitats and for shorter lived species. Thus, the concerted evolution of embryo length with changes in habitat shade or life cycle is very likely the result of adaptive processes as a function of habitat and ecological strategy. Moreover, the low phylogenetic half-life calculated tells us that evolution towards habitat- and life cycle-specific optima occurs at a rapid pace.

The adaptive nature of the relative embryo length in Apiaceae contrasts with the results obtained when embryo size evolution was analyzed across the whole angiosperm phylogeny, where the evolution of embryo size occurred as a random walk process (Verdú, 2006). One cause of this discrepancy is that there are major differences in the measure of relative embryo size. Although we analyzed (log) relative embryo length, Verdú (2006), based on the data of Forbis et al. (2002), used the surface area of the whole embryo relative to the surface area of the endosperm. Second, differences in the taxonomic scale at which the studies were performed will affect the results, because different evolutionary pathways and selection pressures are evident in different groups of plants. Similar evolutionary correlates as those observed in Apiaceae are plausible for other plant families characterized by seeds that require post-dispersal embryo growth, such as Ranunculaceae, Adoxaceae and several liliaceous families.

For the first time, we have analyzed the ecological function of the embryo length and possible evolutionary forces acting on it by means of the comparative method. It is likely that habitat conditions and plant longevity have been important forces driving the evolution of embryo size in Apiaceae. We also emphasize that different mechanisms may have driven embryo size evolution in different families or at different taxonomic levels. The enormous amount of data available on seed characteristics (e.g. Baskin & Baskin, 1998) and the continuous development of new analytical techniques (Freckleton & Pagel, 2006) and molecular phylogenies provide huge opportunities for answering the multitude of questions remaining on the evolution of seeds (Silvertown, 1999).

Acknowledgements

We thank all institutions and individuals that provided seed material for this research. We also thank Miguel Verdú, Diethart Matthies and two anonymous referees for reading the manuscript and for providing useful suggestions for its improvement.

Ancillary