Plantago coronopus is an herbaceous, short-lived perennial rosette plant with indeterminate growth that mainly occurs in salt marshes along the Dutch coast. The species has perfect flowers on top of long flowering stalks (spikes), which consist of a stem and an ear part. Each leaf axil has an axillary meristem that can produce either a spike or a lateral side rosette, or stay dormant. It is a wind pollinated, gynodioecious and predominantly outcrossing species (Koelewijn 1998). Male steriles (MS) are frequently observed in most populations (Koelewijn & Van Damme 1996), though frequencies never exceed those of hermaphrodites (H). The species flowers from the beginning of May through to September and overwinters as a rosette. The flowers are protogynous, with an overlap in sexual phase. Flowering and subsequent maturation occur from the base of the ear upwards and each flower produces five ovules.
Mean seed weight in P. coronopus is on average 18% higher in male steriles than in hermaphrodites and self-fertilization of hermaphrodites produces slightly smaller seeds compared with cross-fertilization (average 8%). Although small, and variable, among experiments and treatments (summarized in Table 1), both results were significant when calculated over all experiments using either a combined probability test (Sokal & Rohlf 1980) or a meta-analysis (Hedges’d, the standard meta-analytical metric; Hedges & Olkin 1985).
Table 1. Summary of experiments with Plantago coronopus where seed weight (mean ± SE, number of plants) was determined as a function of either sex morph (MS or H) or inbreeding level (outcross (H), one (S1) or two (S2) generations of selfing). Paired t-tests were used when crosses were made on the same plant or when the relatedness among the maternal parents was known (sibling comparison, i.e. male sterile and hermaphrodite parent originate from the same maternal plant). Combined analysis over all studies was done in two ways: (i) a combined probability test (Sokal & Rohlf 1980); (ii) calculate Hedges’d from a random-effects model meta-analysis (Hedges & Olkin 1985). When d is positive male steriles have a larger effect size than hermaphrodites or outcrossing has a larger effect size than selfing. If 95% confidence intervals (CIs) do not include 0 the effect is significant over all studies
|Type of experiment||MS||H||S1||S2||MS vs. H||H vs. Self|
|(A) Random selection of plants|
|Growth (glasshouse, water culture, free access to nutrients; µg C)a||129.1 ± 9.3 (n = 11)||122.8 ± 4.9 (n = 29)|| || ||Two sample t-test t38 = 0.64 P = 0.527|| |
|Growth (glasshouse, water culture, limiting access to nutrients; µg C)a||111.0 ± 8.2 (n = 13)|| 89.2 ± 2.7 (n = 30)|| || ||Two sample t-test t41 = 2.79 P = 0.008|| |
|Hand pollination (field; mg)b||0.130 ± 0.008 (n = 9)||0.094 ± 0.006 (n = 20)|| || ||Two sample t-test t27 = 4.59 P < 0.001|| |
|Growth analysis (glasshouse, water culture; mg)b|| ||0.175 ± 0.009 (n = 12)||0.154 ± 0.01 (n = 12)||0.134 ± 0.008 (n = 11)|| ||One-way anova F2,32 = 4.78 P = 0.015|
|Crossing studies (glasshouse; mg)c||0.179 ± 0.012 (n = 14)||0.185 ± 0.01 (n = 18)|| 0.16 ± 0.01 (n = 18)|| ||Two sample t-test t30 = 0.36 P = 0.721||Paired t-test t17 = 4.54 P < 0.001|
|Crossing studies (glasshouse; mg)c||0.160 ± 0.009 (n = 19)||0.146 ± 0.005 (n = 12)|| || ||Two sample t-test t29 = 1.58 P = 0.126|| |
|(B) Sibling comparison|
|Hand pollination (garden; mg; n = 10)d||0.202 ± 0.013||0.168 ± 0.011||0.152 ± 0.01|| ||Paired t-test t9 = 2.85 P = 0.019||Paired t-test t9 = 1.52 P = 0.168|
|Transplant experiment (field; mg; n = 29)d||0.118 ± 0.005||0.101 ± 0.004|| || ||Paired t-test t28 = 4.14 P < 0.001|| |
|Crossing studies (glasshouse; mg; n = 11)b||0.167 ± 0.014||0.128 ± 0.006||0.131 ± 0.009|| ||Paired t-test t10 = 2.60 P = 0.026||Paired t-test t10 = 0.22 P = 0.833|
|(1) Combined probability test (−2 Σ ln P) || || || || ||χ216 = 55.7 P < 0.001||χ28 = 28.5 P < 0.001|
|(2) Hedges’d|| || || || ||0.65||0.45|
|95% CI|| || || || ||0.31–0.98||0.07–0.84|
origin of experimental plants
Maternal plants were germinated from seeds of plants collected at the Westplaat, a salt-meadow in the south-west of the Netherlands (Koelewijn & Van Damme 1995a). One hermaphrodite and one male-sterile individual were selected from the offspring of each of five collected plants. The male-sterile maternal individuals were crossed with mixed pollen from a random selection of hermaphrodites (outcross MS × H, expected inbreeding coefficient F = 0). Hermaphrodites were outcrossed (H × H, F = 0) or self-pollinated (S1, F = 0.5) and five S1 seeds from each mother were subsequently sown and self-pollinated (S2, F = 0.75). After ripening, seeds were taken out of the capsules, pooled per cross type, and divided into two groups. One group was used as an unsorted control group, while the other group was separated into seven seed size classes (class range 0.02 mg, class means from 0.11 to 0.21 mg and > 0.22 mg; Fig. 1) by means of an Astell Hearson SCB016 Ottawa seed blower (Waterreus C.V., The Hague, the Netherlands). Seeds were stored in a cold room at 4 °C until further usage. No difference in seed set was observed among the cross types.
Figure 1. Size frequency distribution of seeds of Plantago coronopus originating from four different cross types: male steriles crossed with hermaphrodites (MS × H); hermaphrodites crossed with hermaphrodites (H × H); self-fertilization of the hermaphrodites (S1) and self-fertilization of offspring from these S1 plants (S2). Also indicated are the three categories of seed size (small, medium and large) used in the experiment.
Download figure to PowerPoint
growth experiment (glasshouse)
The aim of the experiment was to compare the performance of seeds from two size classes (small, 0.12–0.14 mg; large, 0.18–0.22 mg) and three cross types (MS × H, H × H and S1) during the juvenile stage of plant development, i.e. a classical growth analysis (Evans 1972). Seeds were germinated in a climate cabinet on wet (demineralized water) glass beds (day = 16 hours, 20 °C; night = 8 hours, 15 °C). After 14 days 60 seedlings of each combination of size class and cross type were transferred to three separate growth units placed together in a growth room maintained at 14/10 hours day/night, 23° ± 0.5 C day/night, quantum flux density at mean plant height 305 µmol m−2 s−1 (photo synthetically active radiation) and 70% relative humidity. Light was provided by fluorescent lamps and incandescent bulbs. Each growth unit was connected to a 5-L container of a 0.125 strength Hoagland nutrient solution (625 µm KNO3, 625 µm Ca(NO3)2, 250 µm MgSO4, 125 µm KH2PO4, 90 µm Fe-EDTA, 45 µm H3BO3, 9 µm MnCl2, 0.75 µm ZnSO4, 0.5 µm Na2MoO4 and 0.3 µm CuSO4; Hewitt 1966) whose pH was automatically adjusted to 6.0 by a Radiometer pH-stat unit (Radiometer Analytical, Lyon, France) using H2SO4. The nutrient solution was applied as a mist (see Freijsen et al. 1990 for further details). To prevent nutrient depletion, the solution was renewed each week. Seedlings for the growth units were randomly selected and the remainders were pooled in groups of 10 to estimate seedling dry weight. To prevent root or shoot competition the plants were gradually moved apart over the course of the experiment as harvesting reduced the initial number of 120 plants. To minimize the effect of spatial variation in growth conditions the units were regularly rotated within the growth room. At 4, 8, 11, 14, 16 and 18 days after transfer to the growth unit, 27 plants per size class (nine from each cross type), were selected and, when possible, dissected into roots and leafs. The experiment was finished 3 days after initiation of spikes was first observed. Dry weights were determined on oven-dried (48 hours at 70 °C) material.
Because no significant differences in weight or growth rate were observed among the three growth units, the data were analysed as a completely randomized design (procedure GLM, SAS 1988). A full factorial model including time, cross type, seed size and their interactions was fitted. Data were log transformed prior to analysis to improve homogeneity of variance. Difference in relative growth rate (RGR) between the seed size classes and among cross types was tested according to Poorter & Lewis (1986) as treatment by time interaction of the natural logarithm of total dry weight. Second-order polynomials were fitted to test for a trend of RGR with time.
In March 1997 seeds of four size classes (small, 0.12–0.14 mg; medium, 0.14–0.18 mg; large, 0.18–0.22, see Fig. 1, plus unsorted) and four cross types (MS × H, H × H, S1 and S2) were buried in a salt meadow at the site from which the source parents were collected. Five plots of 1.25 by 1.00 m were laid out randomly in an apparently homogeneous area of 25 × 25 m. These plots were considered to be replicates for the different treatments. Plots were subdivided into 20 subplots in a 5 by 4 grid. Each combination of size and cross type was randomly assigned to one subplot in each plot and seeds (n = 25) were buried 3 mm deep at 5-cm spacing in a 5 by 5 grid. Germinates were located by laying a 25 by 25 cm square dish containing 25 holes with a 1-cm diameter over the existing vegetation and scoring the appearance of seedlings in the holes. No seeds were buried in the remaining four subplots, which served as controls for natural germination and to adjust for possible misidentifications in the treated subplots. Plots were checked once a month for germination, survival, growth and flowering (April through September). In April 1997 rabbits destroyed one plot. In August 1997 the number of leaves, length of the longest leaf and number of ripe spikes of germinated seedlings were measured. Monthly checks continued in 1998 but size measurements were now also made monthly and at least one ripe spike was sampled from each flowering plant to estimate seed production. Plant size was estimated from the linear regression of the product of leaf length and number of leaves (total leaf length, TLL) on shoot dry weight: SDW (mg) = −1.509 + 0.191 × TLL (mm), r2 = 0.78, F1,66 = 235.6, P < 0.001. This relationship was based on plants dug up in another experiment in the same area (Koelewijn 1998).
The germination potential of the different size–cross type combinations was assessed at the start of the field experiment, except for the S2 cross type, where there were too few seeds. For each combination 100 seeds were placed in Petri dishes filled with 0.5% aqueous agar, and germinated under natural light conditions in the glasshouse during March 1997.
Analysis of categorical variables was performed by fitting generalized linear models with GLIM (Baker 1987). For data involving proportions (germination, survival and number of plants alive) a log-linear model with a binomial error distribution was used. Probability of flowering was analysed by fitting a multinomial logit model (three flowering states were distinguished: 0 = non-flowering; 1 = flowering in only one season; 2 = flowering in both seasons). In all cases the analysis started with a null model with all main effects and interactions. Subsequently, a χ2 test was used to determine whether dropping a term from the model significantly reduced the explained variance. The difference in unexplained variance between the categorical models (deviance) is approximately χ2 distributed, with the number of degrees of freedom equal to the difference between the model with and without the term to be tested (McCullagh & Nelder 1983). Contrasts were used to test for specific differences within the main factors seed size (small vs. large) and cross type (MS vs. H, and H vs. selfing). Quantitative variables were analysed using the SAS statistical package (procedures GLM and t-test, SAS 1988).
Field experiments are often prone to difficulties in statistical analysis and interpretation. Because of variation in germination and the death of plants during the experiment, the experimental design can become unbalanced or, even worse, incomplete. This is especially true for traits related to reproduction, as only a fraction of all plants produce seeds. Therefore, only three plots were used for the statistical analysis of germination and survival (plots 1–3), and only two for the analysis of flowering data (plots 1 and 3), because the log-linear models based on all four plots did not converge. Quantitative traits of flowering plants (size measurements, number of seeds) could only be analysed by one-way anova, because of small sample sizes.
Epistasis at the population level can be detected by calculating the linear (f) and quadratic (f2) coefficients of the quadratic regression of (log)-fitness on inbreeding level (Crow & Kimura 1970). To examine the effect of inbreeding on fitness, calculated as the number of seeds produced per initial buried seed (standardized seed production), we performed a linear regression with linear and quadratic components as the independent variables. A significant effect of the quadratic component indicates a non-linear relationship between inbreeding level and fitness.
Cumulative fitness for selfed (F > 0) and outcrossed progeny (F = 0) was calculated as the product of germination, probability of flowering and number of seeds per flowering plant at the end of the second year (standardized seed production). Inbreeding depression (δ) per inbreeding level was estimated as:
where ws and wo are, respectively, the mean fitness of selfed and outcrossed progeny.