Is plasticity across seasons adaptive in the annual cleistogamous plant Lamium amplexicaule?

Abstract

Background and aims Many angiosperms exhibit cleistogamy, the production of both cleistogamous flowers (CL), which remain closed and obligately self-pollinated, and chasmogamous flowers (CH), which are potentially open-pollinated. The CH proportion can be plastic. Plasticity is adaptive if environmental changes can be reliably assessed and responded to with an appropriate phenotype and if plastic genotypes have higher fitness in variable environments than non-plastic ones.

Methods We studied the plastic response of four natural populations from northern and southern France of an annual cleistogamous plant, Lamium amplexicaule, to predictable seasonal variation. Plants were grown in a semi-controlled environment in spring and in autumn. We assessed the variation in flower number, phenology and cleistogamy-related traits, which were all plastic with respect to season. The CH proportion was higher in spring than in autumn in all four populations.

Key Results We showed significant stabilizing selection for cleistogamy traits, with higher optimal CH proportions and more pronounced stabilizing selection in spring than in autumn. Observed CH proportions were close to the predicted optimal CH proportions in each season except in autumn for southern populations, which do not experience the autumnal growing season in nature.

Conclusions These results are consistent with adaptive plasticity across seasons of cleistogamy in L. amplexicaule. We propose that adaptive plasticity of cleistogamy could be driven by pollination environment variation, with CL flowers providing reproductive assurance when pollinators are scarce and CH flowers reducing the inbreeding depression in offspring when pollinators are abundant.

INTRODUCTION

Plasticity is the ability of a given genotype to modify its phenotype in response to environmental variation (Bradshaw, 1965). If a plastic genotype has a higher fitness across variable environments than a non-plastic genotype, then plasticity can be adaptive (Sultan, 1995). Plasticity can also be non-adaptive if an organism is unable to respond suitably to environmental changes, e.g. if it is unable to maintain a constant phenotype in a changing environment (canalization; Debat and David, 2001). For plasticity to be adaptive, several conditions need to be fulfilled. First, plastic individuals need to be subject to environment-dependent selection (Via and Lande, 1985); each of the phenotypes expressed by the plastic genotypes should confer a higher fitness in the environment inducing its expression, and a lower fitness in other environments. This could be achieved if stabilizing selection, with different optimal phenotypes, operates in each environment (Lande and Arnold, 1983; Via and Lande, 1985). Second, the selective advantage conferred by an adaptive phenotype must be higher than the costs associated with plasticity (De Witt, 1998). Third, the organism must be able to reliably predict environmental features and produce the appropriate phenotype rapidly (De Witt, 1998). If the time necessary to establish a plastic response is longer than the duration of the environmental change, or if environmental variation is unpredictable, then a fixed phenotype will be appropriate (Roff, 2002). Inversely, if environmental change takes place over a longer period of time and is easily anticipated, then a plastic response will be appropriate. A way to capture environmental variation is to rely on regular periodical changes, such as seasonal variation.

Adaptive plasticity in plants has been well documented for morphological traits such as individual size or leaf form (Sultan, 1995; Donohue et al., 2000). Mating system plasticity has been less studied, though empirical evidence has been reported in the literature. For instance, sex expression has sometimes been found to depend on environmental conditions, with favourable conditions increasing femaleness in some species (Korpelainen, 1998; Dorken and Barrett, 2004) and decreasing it in others (Delph, 2003; Dorken and Mitchard, 2008). In mixed mating systems the selfing rate can also vary in response to environmental change (Kay and Picklum, 2013; Jorgensen and Arathi, 2013; Van Etten and Brunet, 2013). General conditions for adaptive plasticity described above also apply to plastic mating systems; if relative values of different reproductive modes depend on environmental conditions, then a plastic reproductive system should be favoured in variable environments. For instance, selfing reproductive strategies should be favoured in environments where the possibility of an encounter between gametes from different parents is low, while outcrossing strategies should be favoured in environments with high levels of inbreeding depression (Lloyd, 1979). However, compared with plastic morphological traits, the evolution of plastic reproductive traits is additionally limited by frequency-dependent selection (Ernande and Dieckmann, 2004). This could be easily illustrated in dioecious species. Producing a high proportion of females in a given environment will automatically increase the reproductive success of males. In such situations, an unbalanced sex ratio generates negative frequency-dependent selection on sexual phenotypes (Shaw and Mohler, 1953), which should prevent plasticity from evolving (Lloyd and Bawa, 1984). Frequency-dependent selection implies that there are no environment-dependent optimal phenotypes, and thus constrains the evolution of plasticity for reproductive traits.

Frequency dependence also occurs in hermaphroditic mating systems. In hermaphrodites, individuals capable of selfing leave more copies of their genes because they can use their pollen to fertilize their own ovules as well as the ovules of other plants. The magnitude of this advantage, called the automatic advantage of selfing (or the cost of outcrossing; Fisher, 1941; Lloyd, 1979), depends on the opportunity to sire ovules devoted to outcrossing, and thus on the frequency of outcrossers in the population. Contrary to selection on sex ratio in dioecious species, selfing and outcrossing exclude each other in this type of selection, and only complete selfing or complete outcrossing are stable strategies, depending on the magnitude of inbreeding depression (Lloyd, 1979; Lande and Schemske, 1985). Modified forms of selection in hermaphrodites, taking into account e.g. pollen discounting, can, however, stabilize a mix of outcrossing and selfing (i.e. mixed mating, reviewed in Goodwillie et al., 2005).

Cleistogamy is a particular case of mixed mating in which closed flowers (cleistogamous, CL), which are obligately self-pollinated, and open flowers (chasmogamous, CH), which are potentially outcrossed, are found within the same individual or among individuals of the same species (Lord, 1981; Culley and Klooster, 2007). Because of their particular morphology, CL flowers are unable to export pollen (complete pollen discounting; Holsinger, 1991). For many cleistogamous species the proportion of CH flowers can vary according to environmental conditions such as photoperiod, temperature, light intensity, nutrient and water availability, competition and herbivory, or a combination of these factors (Oakley et al., 2007). Variation in the CH proportion across environments can be associated with different production costs of the two floral types (Schemske, 1978), differences in fruiting success or seed set, differences in flowering phenology (Redbo-Torstensson and Berg, 1995; Winn and Moriuchi, 2009), different seed dispersal strategies (Berg and Redbo-Torstensson, 1998), adaptation to pollinator abundance (Masuda et al., 2004), or inbreeding depression (Stewart, 1994). In cleistogamous species, plasticity of CH proportion translates to plasticity of the outcrossing rate if CH flowers substantially outcross. Thus, CH adaptive plasticity allows cleistogamous species to potentially maintain stable plastic mixed mating throughout variable environments (Schoen and Lloyd, 1984).

Lamium amplexicaule is an annual cleistogamous species that produces CH flowers in variable proportions. The variation of the CH proportion in this species is influenced by environmental cues such as photoperiod and temperature (Lord, 1982), with long photoperiods and warm temperatures increasing the CH proportion. In temperate climates, it is known either as a winter annual that germinates in autumn and flowers in spring, a spring annual that germinates in winter and flowers in spring or a summer annual that germinates in spring and flowers in autumn (Baskin and Baskin, 1981). Thus, L. amplexicaule can have up to two generations per year (Baskin and Baskin, 1981), provided that environmental conditions are appropriate for its germination and development. Seasonal change is by definition a periodic and regular environmental change, associated with cues that are reliable, such as temperature, photoperiod or light intensity, and therefore easy to predict. Because CH proportion in L. amplexicaule is known to respond to photoperiod and temperature (Lord, 1982), it is plausible that populations experiencing two different seasons have developed adaptive plasticity in the proportion of CH in response to seasonal variation.

In order to test the adaptive plasticity of cleistogamy to seasonal changes in L. amplexicaule, we studied the influence of seasonal variation (spring and autumn) in a common garden for individuals from four contrasted natural populations in France. For each population studied we measured flower number, phenology and different traits related to CH and CL flower production (cleistogamy-related traits). Our goals were to (1) estimate plasticity across seasons for these traits; (2) characterize the form of selection operating on the measured traits within seasons (directional versus stabilizing selection); and (3) test for the adaptive character of plastic cleistogamy across seasons for these populations.

MATERIALS AND METHODS

Model plant

Lamium amplexicaule is a weedy annual of the mint family (Lamiaceae) native to Europe and Asia and introduced in all other continents (USDA-NRCS, 2015). Flowering in this species is ascending within whorls; at the beginning of the flowering season only CL flowers are produced on the lowest whorls (constitutive cleistogamy sensu Lord, 1981), followed by CH flowers in variable proportions in subsequent whorls, i.e. plastic cleistogamy (Lord, 1980). The proportion of CH flowers produced during the flowering season can vary from 0 to 50 % in response to environmental cues such as photoperiod and temperature (Lord, 1982), with long photoperiod and high temperatures increasing the CH proportion and short photoperiod and cold temperatures decreasing the CH proportion. These experimental conditions correspond to natural conditions in spring (long days, warm temperatures) and autumn (short days and cold temperatures).

Studied populations

In spring 2010, four French populations of L. amplexicaule were monitored during the flowering season: two populations in northern France near Dijon, where the species is abundant, and two populations in southern France near Montpellier, where the species is sparser. In each geographical region, we sampled regularly ploughed vineyards (early successional vegetation cover) with low vegetation cover density and non-cultivated or undisturbed sites (such as lawns or fallows), with dense vegetation cover. Early successional populations were dense, with several hundred individuals (hereafter called ‘large’, L) while undisturbed populations were sparse, with about 40 individuals (hereafter called ‘small’, S). The two populations from Dijon (DS, GPS coordinates 47°16′37.14″ N, 5°03′43.37″ E, and DL, 47°15′58.88″ N, 4°59′11.51″ E) and two populations from Montpellier (ML, 43°44′56.28″ N, 3°51′06.49″ E, and MS, 43°46′16.36″ N, 3°47′28.03″ E) were those studied by Stojanova et al. (2014). Plants in large populations are visually bigger than plants in small populations; the number of flowers and the number of secondary axes per plant are indeed higher in the former.

Small and large populations within a region are not differentiated for their flowering phenology within a given geographical region, with a delay of ∼2 weeks in Dijon compared with Montpellier (population flowering peak is 5 April in Montpellier and 18 April in Dijon; B. Stojanova, pers. obs.; for details see Stojanova et al., 2014). The CH proportion observed in the field in spring differs between small and large populations, with 20 % of CH flowers in large and 30 % of CH flowers in small populations. In spite of these differences in the floral display, the average outcrossing rate of CH flowers was estimated at 25 % in all four populations (Stojanova et al., 2014).

A 2-year survey also showed that populations in Dijon can behave as spring annuals (flowering in mid-spring) and if the summer climate is appropriate, as summer annuals too (flowering in early autumn), whereas those in Montpellier behave only as spring annuals because of the dry summer climate, which prevents germination in late spring. Therefore, Dijon populations are exposed to a more variable environment and have lower CH proportions in autumn than in spring (B. Stojanova, pers. obs.). We checked all population sites in late summer or early autumn for recently emerged plants. Population DL was the only one that had grown in the autumn 2010, and the same growth pattern was observed in 2011 (data not shown).

Production of plant material

Seed sampling in natural populations. In each of the four populations monitored, 20 individuals were tagged with coloured tape on the principal axis. Several calices bearing CH flowers of each tagged individual were marked with acrylic paint. After corollas fell out, calices were closed using acrylic paint. At the end of the flowering season, seeds from marked calices were collected and stored in a dry place at room temperature. Seeds were sampled from CH flowers in order to maximize genetic variability in our experimental design.

Common garden generation. To control for maternal effects, a common garden generation (G₁) was produced from the seeds collected in the field. In February 2011, a maximum of 12 CH seeds per family were randomly selected and sown in a sterile Petri dish with two pieces of Whatman paper that was humidified as necessary. Petri dishes were placed in growth chambers with 12 h light exposure at 25 °C and 12 h in the dark at 13 °C. Successfully germinated seeds were planted in 14 ten-well plates until they produced at least four leaves. One plant per family was then transplanted directly into the soil of a ploughed plot in the experimental field of the Plateforme des Terrains d’Experience du LabEx CeMEB, resulting in 18, 13, 16 and 15 plants for populations ML, MS, DL and DS, respectively, that were to be used as the maternal parents of the G₂ generation used in the experiment.

Experimental design for plasticity measurements on the experimental generation. In order to avoid gene flow among populations, only CL flower seeds were collected from the plants of the G₁ generation. In May 2011, the calices of several CL flowers were marked on each G₁ plant. At the end of the flowering season, seeds in marked CL calices were collected and stored at room temperature, resulting in 18, 13, 16 and 13 full-sib families, which constituted the experimental descendants for populations ML, MS, DL and DS, respectively.

For each family, seeds were randomly divided into two lots in order to measure trait values in autumn and spring. Up to eight seeds per family were germinated under the same conditions as described above on 15 August 2011 for the autumn treatment and on 15 February 2012 for the spring treatment. Germination timing was chosen in order to maximize the difference in photoperiod magnitude among seasons during the flowering period, ranging from 14 to 15 h of daylight in spring (ascending) and from 10 to 9 h in autumn (descending). When most of the plants had at least four developed leaves, a maximum of five seedlings per family were transplanted into 450-cm³ pots and placed outdoors in the experimental field. The soil used was a mixture of sand and humus, and the same mixture was used for the autumn and the spring experiment. In order to facilitate manipulations, pots were assembled in blocks of 20. The position of the pots within blocks and the position of the blocks were regularly randomized. Because some of the families did not germinate at all, the number of families available was 13 (15), 10 (11), 13 (13) and 10 (10) for populations ML, MS, DL and DS in autumn (spring), respectively, with a total of 223 plants in autumn and 224 plants in spring. Pots were watered ad libitum.

Data collection

Phenological traits. Plants were checked for flowers every other day in order to record the date of the first CL flower (d_CL) and the date of the first CH flower (d_CH). For the purpose of statistical analysis, the data were encoded as number of days counted from the beginning of the experiment in each season (15 August and 15 February in autumn and spring, respectively).

Flower type counts and cleistogamy. Flower counts were made on the principal axis only. The number of calices and CH proportion on the principal axis were highly correlated with the total number of calices and total CH proportion, respectively (number of flowers, Spearman’s ρ = 0·78, p < 0·0001, and CH proportion, Spearman’s ρ = 0·89, P < 0·0001, in a sample of 98 individuals). For each plant, the number of CH flowers was assessed every 4 d after the emergence of the first CH flower. Since CH flowers wilt in ∼4 d, only recent, fresh flowers were recorded on each monitoring date, thus avoiding recounting the same flower. The sum of all CH flowers counted on the principal axis during the flowering period was termed the total number of CH flowers (N_CH). At the end of the flowering season, we counted the calices on the principal axis in order to estimate the total number of flowers produced (N_total). The flowers produced before the first CH flower were considered as constitutive cleistogamous flowers (N_CLC).

Three different flower ratios related to cleistogamy proportions were calculated. Overall CH proportion (on the principal axis) was estimated as p_CH = N_CH/N_total. The constitutive cleistogamy proportion, which corresponds to the portion of flowers produced at the beginning of the flowering period when only CL flowers are produced, was estimated as p_CLC = N_CLC/N_total. Hence, p_CLC was 1 for the plants that produced no CH flowers (most of which were observed in autumn) and 0 for the few plants that produced no CL flowers (all in spring). The induced CH proportion, corresponding to the ratio of CH to CL flowers produced after constitutive CL flowering, was estimated as p_CHpl = N_CH/(N_total – N_CLC). Induced CH proportion data were not estimated for plants that produced no CH flowers (mostly in autumn).

Fertilization success estimates. Throughout the flowering season up to 15 CH and CL calices were marked on each individual using acrylic paint according to the number of flowers available per plant. This method does not affect seed production as long as the acrylic paint is not in direct contact with the seeds (B. Stojanova, pers. obs.). After corollas fell out, calices were closed with a small drop of acrylic paint. At the end of the flowering season, seeds produced in each marked calyx were counted. Since the maximal number of seeds per flower is invariably four, the proportion of fertilized CH and CL seeds (G_CH and G_CL, respectively) per individual was estimated as the number of seeds counted in all marked CH or CL flowers divided by four times the number of marked CH or CL calices, respectively. The variables G_CH and G_CL are estimators of fertilization success in CH and CL flowers, respectively.

Data analysis

Plasticity of traits across seasons. Plasticity analyses were performed with type III general linear mixed models (PROC GLM procedure, SAS). For each model, family (nested in population) was fitted as random factor, while season, population and their interaction were fitted as fixed factors. Data were normalized by square root transformation of count data (N_total, d_CH, d_CL), and arcsine square root transformation of proportion data (p_CH, p_CLC, p_CHpl, G_CH, G_CL). The significance of fixed and random effects was tested using the appropriate error structure. Population was tested using family (nested in population) as error. Season, family (nested in population) and season × population were tested using model residuals as error. Because the design was not totally balanced, F tests were performed using Satterthwaite approximation (Littell et al., 2006). Visual examination of the residuals showed that some models did not fulfil the heteroscedasticity requirements, mostly because of zero inflated distributions, but their conclusion was consistent with the patterns observed in raw data.

Estimates of fitness in cleistogamous plants. The fitness of a hermaphrodite is the sum of its maternal and paternal contributions to the offspring: each individual passes on two copies per selfed progeny, one copy per outcrossed progeny and one copy per sired progeny through successful pollen export (Morgan and Schoen, 1997). The contribution of each of these portions to parental fitness is weighted by the relative fitness of the offspring types, which could be affected by various factors, such as inbreeding depression in selfed progeny (Lloyd, 1979) or, in cleistogamous species, flower type from which the progeny is issued (Oakley et al., 2007). We then estimate the individual fitness of a CL plant as twice the contribution of CL progeny and selfed CH progeny, to which is added the contribution of outcrossed ovules and successful pollen export competing with the outcross pollen in the population (Lloyd, 1979):

$$w=N_{\mathrm{CL}}{G}_{\mathrm{CL}}2w_{\mathrm{CL}}+N_{\mathrm{CH}}{G}_{\mathrm{CH}}\left(t{w}_{\mathrm{CHO}}+2\left(1-t\right)w_{\mathrm{CHS}}\right)+\frac{N_{\mathrm{CH}}{G}_{\mathrm{CH}}}{{\displaystyle {\sum }_{}^{\mathrm{pop}}{N}_{\mathrm{CH}}{G}_{\mathrm{CH}}}}{\displaystyle \sum _{}^{\mathrm{pop}}{N}_{\mathrm{CH}}{G}_{\mathrm{CH}}t{w}_{\mathrm{CHO}}}$$ (1)

where N_CL, N_CH, G_CL and G_CH are the number of CL flowers, the number of CH flowers, the fertilization success of CL flowers, and the fertilization success of CH flowers of the focal plant, respectively. We assume that attractivity at the individual level will impact pollen export and seed production of CH flowers in the same way. Consequently, the pollen export of a plant is proportional to N_CHG_CH, given that CL flowers do not export pollen (total pollen discounting). The variables w_CL, w_CHO and w_CHS are the viabilities of CL progeny, outcrossed CH progeny and selfed CH progeny, respectively; the variable t is the outcrossing rate of CH flowers. Thus, the equation simplifies to:

$$w=N_{\mathrm{CL}}{G}_{\mathrm{CL}}2w_{\mathrm{CL}}+N_{\mathrm{CH}}{G}_{\mathrm{CH}}2\left(t{w}_{\mathrm{CHO}}+\left(1-t\right)w_{\mathrm{CHS}}\right)$$ (2)

It is possible that siring success via pollen export is not equal to seed fertilization success, but this appears as a parsimonious hypothesis. Moreover, given that only ∼10 % of total seeds are produced by outcrossing, the deviation from equality is not expected to have a major impact on total fitness.

As a consequence, and contrary to the classic single flower morph mating systems theory, the fitness of a CL individual does not depend on the mating system traits in the population, i.e. frequency dependence is cancelled in CL populations (see also Lloyd, 1979). Thus, individual fitness (through male and female fitness components) can be estimated directly by counting seeds produced by each flower type on an individual plant. This estimate of fitness was used to estimate selection coefficients.

Estimates of selection coefficients within seasons. For plasticity of a trait to be adaptive, stabilizing selection with different optimal phenotypes needs to operate in each of the seasons that an individual encounters. We therefore tested directional and stabilizing selection effects in spring and in autumn on the number of flowers (N_total), phenology (date of first CL and first CH flower, d_CL and d_CH, respectively) and cleistogamy-related traits (overall CH proportion, p_CH, constitutive CL proportion, p_CLC, and induced CH proportion, p_CHpl). Individual fitness was first estimated according to eqn (2) assuming no viability differences between CL and CH progeny, i.e. as the overall seed production (see below for relaxation of this assumption). Since the number of ovules per flower in L. amplexicaule is invariably four, overall seed production for each individual was estimated as: G = G_CL × (N_total – N_CH) × 4 + G_CH × N_CH × 4. In order to cope with zero-inflated distributions of the proportion variables (p_CH, p_CLC and p_CHpl), we calculated means per family for all traits. For each trait a generalized quadratic model (GQM) was fitted using the lm function in software R2.15.1 as follows: G_rel = population × (trait + trait²), where G_rel is the fitness relative to the overall mean fitness (Lande and Arnold, 1983). Significance levels of each variable were assessed using the drop1 function (type II model testing), and whenever possible the model was simplified by omitting the non-significant interaction terms. The regression coefficients for the linear and quadratic terms of the GQM correspond to directional (linear coefficients) and stabilizing (negative quadratic coefficients) selection in each season. The explicative variables p_CH, p_CLC and p_CHpl were arcsin(sqrt) transformed and d_CL, d_CH and N_total were square root transformed as this improves the fit of the models based on AIC (Akaike information criterion) comparisons.

Testing the adaptive character of CH proportion plasticity. In order to test whether the plastic CH proportion is adaptive, we analysed the relationship between relative fitness and CH proportion. If plasticity of p_CH is adaptive, we expect the mean p_CH value expressed in each season to be close to the optimal p_CH value predicted by the selection curve (p_CH*). Under stabilizing selection (quadratic regression) the maximum of the fitness curve is obtained for p_CH* = −β₁/2β₂, with β₁ (respectively β₂) as the linear (respectively quadratic) regression coefficient. The p_CH* values were calculated in spring and in autumn as the maximal values of the GQM regression of CH proportion on parental fitness [eqn (2)] as described in the previous section. Since fitness is a priori sensitive to the viabilities of each progeny type [w_CL, w_CHO and w_CHS; see eqn (2)], which were not estimated in this study, we consider different scenarios for which we assigned contrasted progeny survival values to test the robustness of optimal CH predictions:

• (1) General scenario, with no difference in survival between the three seed types (w_CL = w_CHO = w_CHS = 1). The fitness calculated this way corresponds to overall seed production, G.

• (2) Flower type scenario, with (a) higher survival rates for CL progeny (w_CL = 1, w_CHO = w_CHS = 0.5) or (b) lower survival rates for CL progeny (w_CL = 0·5, w_CHO = w_CHS = 1).

• (3) Inbreeding depression scenario, in which inbred progeny, regardless of the type of flower they are produced by, have lower survival rates than outbred progeny (w_CL = w_CHS = 0·5, w_CHO = 1).

• (4) Outbreeding depression scenario, with inbred progeny having a higher survival rate than outbred progeny (w_CL = w_CHS = 1, w_CHO = 0·5).

For the estimates of fitness we used the CH outcrossing rate [t in eqn (2)] estimated for these populations in a previous study, which is t = 0·25 (Stojanova et al., 2014), and the values for G_CH, G_CL, N_CH and N_CL as estimated in this study. The quadratic regressions of CH proportion on fitness were made using the average values per family, as described in the previous section.

We also analysed the relationship between annual fitness (the sum of fitness estimates in spring and autumn) and plasticity (the absolute difference of arcsine square root transformed CH proportions in autumn and in spring) in order to test whether the most plastic genotypes have the highest fitness. For this we fitted linear regressions on annual fitness as calculated above using plasticity, population, and their interaction as explanatory variables. A significant positive regression coefficient would indicate that more plastic genotypes have higher annual fitness

Results

Photoperiod conditions during the experiment

Photoperiod values at the flowering peak and at the end of the flowering season for each region during the field survey in 2010 and the experimental study in Montpellier in 2011/2012 are given in Supplementary Data Table S1. Compared with spring flowering in natural populations, peak photoperiod in our experimental design was >1 h longer for the Montpellier populations, but coincided with the natural photoperiod in Dijon. In autumn, the photoperiod at flowering peak was ∼1·5 h shorter than that occurring in population DL in field observations in 2010.

Trait variation across seasons

Total number of flowers. Season had a significant effect on the total production of flowers on the principal axis (Table 1), with a >2-fold N_total increase in spring compared with autumn (mean ± s.e. for populations ML, MS, DL and DS in autumn 19·49 ± 1·41, 6·43 ± 0·60, 21·39 ± 1·08 and 19·07 ± 1·20, respectively, and spring 48·68 ± 1·62, 17·05 ± 0·93, 50·12 ± 1·28 and 52·55 ± 1·71). Population MS produced significantly fewer flowers and was less plastic than the other three populations; excluding it from the analysis rendered the population and the interaction factors non-significant (data not shown).

Table 1.

Results of general linear mixed model analyses of among-season variation in number of flowers (N_total), date of the first CL flower (d_CL), date of the first CH flower (d_CH), generalized linear mixed model (GLMM) CH proportion (p_CH), constitutive CL proportion (p_CLC), induced CH proportion (p_CHpl), CL fertilization success (G_CL) and CH fertilization success (G_CH)

Source	Ntotal			dCL			dCH			pCH			pCLC			pCHpl			GCL			GCH
	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F	d.f.	F	Pr > F
Season	1	808	<0·001	1	35·15	<0·001	1	15·56	<0·001	1	272·5	<0·001	1	191·7	<0·001	1	24·84	<0·001	1	0·59	0·444	1	6·76	0·01
Population	3	66·95	<0·001	3	26·54	<0·001	3	38·39	<0·001	3	2·79	0·051	3	2·2	0·099	3	2·78	0·046	3	14·24	<0·001	3	5·28	0·002
Familypopulation	46	2·81	<0·001	46	3·78	<0·001	45	1·19	0·205	46	2·06	<0·001	46	1·54	0·018	45	1·03	0·433	46	1·4	0·051	45	1·77	0·004
Season × population	3	10·94	<0·001	3	2·13	0·097	3	1·46	0·226	3	2·61	0·052	3	2·09	0·101	3	0·98	0·405	3	10·19	<0·001	3	16·88	<0·001

Season, population, and their interaction are fixed explanatory variables, and family nested in population is declared as random.

Dependent variables were transformed (square root transformation for count data, arcsin square root transformation for proportion data). Pr > F is the associated P-value with the observed F-statistic.

Flowering dates. The CL flowering of the plants began earlier in spring than in autumn (Fig. 1A, Table 1). Population MS began flowering with a delay of almost 1 month in both seasons compared with the other three populations, which were rather synchronized (Fig. 1A). The interaction term was not significant. The inverse seasonal pattern was observed for CH flowering: all populations produced CH flowers significantly later in spring than in autumn (Fig. 1B, Table 1). Population MS has a significant delay of 10–15 d compared with the other three populations.

Fig. 1.

Phenological traits across seasons. (A) Mean date of the first CL flower. (B) Mean date of the first CH flower. Black lines, Montpellier (southern) populations; grey lines, Dijon (northern) populations; dashed lines, small populations (from unfavourable habitats); solid lines, large populations (from favourable habitats). Vertical bars present the standard error.

Comparing d_CL and d_CH values for each population showed that in autumn CH flowering was earlier than CL flowering for population MS and almost at the same time for population ML. Though this may be true on the population level, individual plants always produced CL flowers first (constitutive cleistogamy) followed by the production of CH and CL flowers in variable proportions. The results observed here could be explained by the delayed flowering onset of plants that produced only CL flowers, which could lead to a higher estimate of d_CL in autumn. Excluding plants that produced only CL flowers from the mean estimates did not noticeably affect d_CH in either season or d_CL in spring. However, d_CL values in autumn were considerably lower for plants that produced both CL and CH flowers (mean ± s.d., 53·33 ± 4·90, 73·89 ± 6·15, 59·59 ± 3·73 and 62·20 ± 4·26 for populations ML, MS, DL and DS, respectively; compare with Fig. 1A).

Cleistogamy-related traits. Overall CH rate (p_CH) was significantly higher in spring than in autumn for all four populations (Fig. 2A, Table 1). Neither population nor season × population had a significant effect on p_CH. Family was highly significant, because several families produced no CH flowers in autumn. Constitutive cleistogamy (p_CLC) was significantly lower in spring than in autumn (Fig. 2B, Table 1). The population and interaction terms had no significant effect on constitutive cleistogamy proportion, but family was significant, again because of the high variation of p_CLC in certain families in autumn (data not shown). The induced CH proportion (p_CHpl) could be calculated only for individuals that produced at least one CH flower. Therefore, there are fewer data for the analysis of this trait, especially in autumn, compared with other traits presented in this study, as shown by the standard error estimates (mean ± s.e. for populations ML, MS, DL and DS in autumn 0·249 ± 0·05, 0·259 ± 0·04, 0·312 ± 0·04 and 0·366 ± 0·05, respectively, and in spring 0·409 ± 0·01, 0·303 ± 0·02, 0·401 ± 0·02 and 0·395 ± 0·01). It was observed that p_CHpl was less variable across seasons than the other two cleistogamy-related traits, but a significant increase in p_CHpl in spring compared with autumn was observed (Table 1). Population was marginally significant, but the family and interaction terms were not significant.

Fig. 2.

Cleistogamy-related traits across seasons. (A) Mean CH proportion. (B) Mean constitutive cleistogamy proportion. (C) Mean fertilization success of CH flowers. (D) Mean fertilization success of CL flowers. Black lines, Montpellier (southern) populations; grey lines, Dijon (northern) populations; dashed lines, small populations (from unfavourable habitats); full lines, large populations (from favourable habitats). Vertical bars present standard error estimates.

Seed production. Significant population × season interaction was detected for the probability of a CL ovule developing into a seed (CL fertilization success, Fig. 2C, Table 1). This was due to the fact that the variation pattern of G_CL in population MS differed from the other three populations. Mean CL fertilization success was higher in autumn than in spring for population MS, while it varied little across seasons for the other three populations (Fig. 2C). Population had a significant effect on both CL and CH seed production, with plants from small populations producing less seed than plants from large populations. All explanatory variables were significant for CH fertilization success estimates (Table 1, Fig. 2D). The interaction season × population was highly significant, probably because of population MS, which is the only one that produced fewer seeds in spring than in autumn.

The overall seed production, G, was 2- to 5-fold higher in spring than in autumn (mean ± s.d. for populations ML, MS, DL and DS in autumn 61·41 ± 28·07, 18·86 ± 17·56, 61·45 ± 25·83 and 31·65 ± 18·63, respectively, and in spring 169·40 ± 59·54, 30·87 ± 20·34, 159·89 ± 49·11 and 152·75 ± 65·34).

Selection on plasticity

Positive directional selection acted on total number of flowers in both seasons and in all populations, as shown by the significant linear selection coefficients in Table 2. This result is not surprising, since increasing the total number of flowers logically increases the number of seeds produced in a plant.

Table 2.

Selection coefficients obtained for the best minimum model for the number of flowers (N_total), date of the first CL flower (d_CL), date of the first CH flower (d_CH), CH proportion (p_CH), constitutive CL proportion (p_CLC) and induced CH proportion (p_CHpl) for populations ML, MS, DL and DS

	Autumn								Spring
	ML		MS		DL		DS		ML		MS		DL		DS
	β1	β2	β1	β2	β1	β2	β1	β2	β1	β2	β1	β2	β1	β2	β1	β2
Ntotal	0·440	n.s.	0·440	n.s.	0·440	n.s.	0·440	n.s.	0·450	n.s.	0·450	n.s.	0·450	n.s.	0·450	n.s.
dCL	−2·245	0·105	−2·245	0·105	−2·245	0·105	−2·245	0·105	−0·675	n.s.	−0·675	n.s.	−0·675	n.s.	−0·675	n.s.
dCH	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.
pCH	2·636	−3·113	2·636	−3·113	2·636	−3·113	2·636	−3·113	103·810	−79·740	3·788	−3·300	62·063	−51·110	81·308	−66·599
pCLC	2·999	−1·923	2·999	−1·923	2·999	−1·923	2·999	−1·923	3·582	−3·773	3·582	−3·773	3·582	−3·773	3·582	−3·773
pCHpl	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	129·317	−88·409	4·153	−3·618	−20·070	14·901	93·3	−70·473

β₁, linear coefficient (directional selection); β₂, quadratic coefficient (stabilizing selection if negative). Only significant selection coefficients (P < 0·05) are reported except when β₂ is significant, when both coefficients are reported.

n.s., not significant

Linear selection coefficients for CL flowering date were negative and significant in both seasons, meaning that early flowering increased seed production. We also observed significant positive quadratic selection coefficients for this trait in autumn that could indicate disruptive selection on the flowering date (plants that flowered early or late were favoured, but not plants with intermediate flowering date). This result was due to two observations, in one family in population ML and one in population MS, with relatively high seed production in spite of late flowering. Excluding these two points from our analysis rendered the quadratic term of the regression non-significant (data not shown). According to our analysis, CH flowering date was not under selection in either season. Considering the plasticity of cleistogamy-related traits, patterns of stabilizing selection (significant negative quadratic regression coefficients) were observed for p_CH and p_CLC in both seasons, as well as for p_CHpl in spring. Selection on p_CH was stronger in spring than in autumn (higher |β₂| values) except for population MS (Table 2, Fig. 3). Additionally, pCH associated with maximum fitness (p_CH*) was higher in spring than in autumn in all populations. Similar selection patterns were observed for p_CLC: selection for p_CLC was stronger in spring than in autumn, and p_CLC* was higher in autumn than in spring. Selection coefficients were not significant for induced CH proportion (p_CHpl) in autumn, probably because of the smaller size of the analysed sample. Selection on p_CHpl in spring was stabilizing for three out of the four populations and divergent for population DL (positive β₂).

Fig. 3.

Selection on CH proportion in (A) autumn and (B) spring. Black lines, Montpellier (southern) populations; grey lines, Dijon (northern) populations; dashed lines, small populations (from unfavourable habitats); full lines, large populations (from favourable habitats). Selection curves were drawn using the linear and quadratic selection coefficients estimated by the general quadratic regressions of CH proportion on relative fitness. Symbols indicate predicted values. Arrows indicate the average CH proportion observed in each population (see Table 3). asin√pCH is the arcsinus-square-root transformed proportion of CH flowers.

Adaptive character of CH proportion plasticity

Overall, the sensitivity of optimal CH proportion (p_CH*) with regard to the viability of each progeny type (w_CL, w_CHO and w_CHS) was weak. The highest variation was observed for the flower type scenarios (CH versus CL) in autumn. It is worth noting that the viability of selfed versus outcrossed seeds had less effect on p_CH* than the viability of CH versus CL seeds in spite of similar range values. For each population, p_CH* values were higher in spring than in autumn in all progeny fitness scenarios except for population MS in the flower type 2 scenario (Table 3, Fig. 3). Average p_CH values observed in each population were close to the estimated optimal values in spring. Observed average values in autumn were lower than the optimal estimates, and this difference was more pronounced for the Montpellier populations (Table 3).

Table 3.

Optimal CH proportion (p_CH*) in autumn and spring for different fitness values of CL (w_CL), selfed CH (w_CHS) and outcrossed CH progeny (w_CHO) for populations ML, MS, DL and DS. p_CH* was estimated using the regression coefficients of a quadratic regression of p_CH on relative fitness according to different scenarios [see eqn (2)]

		Autumn				Spring
Scenario	(wCL, wCHS, wCHO)	ML	MS	DL	DS	ML	MS	DL	DS
General	(1, 1, 1)	0·423	0·423	0·423	0·423	0·651	0·574	0·607	0·610
Flower type 1	(0·5, 1, 1)	0·374	0·374	0·374	0·374	0·647	0·570	0·601	0·607
Flower type 2	(1, 0·5, 0·5)	0·588	0·588	0·588	0·588	0·656	0·579	0·618	0·615
Inbreeding depression	(0·5, 0·5, 1)	0·470	0·470	0·470	0·470	0·653	0·576	0·610	0·612
Outbreeding depression	(1, 1, 0·5)	0·419	0·419	0·419	0·419	0·650	0·573	0·606	0·610
	Observed meana	0·171	0·250	0·347	0·391	0·615	0·554	0·627	0·625

^aMean values for arcsin [√(p_CH)] measured in natural populations in autumn and spring.

Regarding the regression of plasticity of the CH proportion on annual fitness, we found that CH proportion and population had a significant effect in all tested scenarios, but not the interaction term (Table 4). However, the effect of CH proportion was rather small, as shown by the regression slopes (Table 3). Similar to the previous analysis, the largest difference in the slope was observed when comparing the two flower fitness scenarios.

Table 4.

Results of GLM analyses of the effect of plasticity of CH proportion, population, and their interaction on annual fitness for the five fitness scenarios

	General			Flower type 1			Flower type 2			Inbreeding depression			Outbreeding depression
	d.f.	F	P-value	d.f.	F	P-value	d.f.	F	P-value	d.f.	F	P-value	d.f.	F	P-value
Plasticity	1	15·87	<0·001	1	12·67	<0·001	1	17·31	<0·001	1	15·25	<0·001	1	16·47	<0·001
Population	3	49·52	<0·001	3	56·48	<0·001	3	43·11	<0·001	3	53·28	<0·001	3	49·32	<0·001
Plasticity × population	3	1·32	0·282	3	1·08	0·367	3	1·44	0·247	3	1·27	0·297	3	1·37	0·296

DISCUSSION

In this paper we study the plasticity of and selection on number of flowers, phenology and cleistogamy-related traits in four natural populations obtained from contrasted geographical regions and habitats in a common garden in spring and in autumn. We show that season has a significant effect on all traits, except fertilization success of CL flowers, which does not vary across seasons for populations ML, DL and MS. However, in our common garden experiment the significant among-population differences for number of flowers, flowering dates, plastic CH proportion and fertilization success is mainly due to the small population in Montpellier. Most of the traits studied here are under selection: positive directional selection for number of flowers, negative directional selection for the CL flowering date and stabilizing selection for overall CH proportion and constitutive cleistogamy proportion. The optimal CH proportion values are lower in autumn than in spring, and the observed average CH proportion per population shows the same trend. The observed CH proportions are close to the predicted optimal values; they are, however, smaller than the predicted optimal values in autumn for the Montpellier populations.

Plastic variation of traits across seasons and population differentiation

In addition to the higher CH proportion in spring, plants produce more flowers and have higher overall seed production than plants in autumn for all four populations. Flowering date after germination is significantly delayed in autumn compared with spring. At the population level, CH flowering date is later than CL flowering in spring, and the reverse pattern is observed in autumn. This is due to the fact that, in autumn, plants producing CH flowers start flowering earlier than the average flowering date in the population, whereas a remarkable proportion of the plants are late flowering and do not produce CH flowers at all. This result fits the general pattern seen in other cleistogamous species in which favourable environments enhance the production of CH flowers (Schemske, 1978). The lack of CH flowers in autumn in late-flowering plants could be due either to the fact that these plants are very small and thus unable to invest resources in showy, CH flowers, or to the environmental conditions becoming inappropriate for the production of CH flowers in late autumn (e.g. lack of pollinators). Indeed, light and temperature conditions like those found in autumn have been shown to influence CH proportion: CH proportion in L. amplexicaule could be <1 % if photoperiod is <10 h and day–night temperatures are 21–10 °C (Lord, 1982). All these observations (smaller size and lower seed production, delayed CL flowering, lower mean CH proportion and absence of CH flowers for some individuals in autumn) show that environmental conditions during spring are more favourable to the development of L. amplexicaule and increase the production of CH flowers in this season, regardless of the population’s geographical origin and size, mean plant size and CH proportion.

Interestingly, the number of flowers, flowering dates and fertilization success of population MS differ from the other three populations. This pattern could be explained by a particular phenology encountered by this population in the field. Observations made on the site of population MS in November 2012 revealed plants in a vegetative stage, which would flower in spring if they survived the winter (i.e. a winter annual cycle). The delay of CL and CH flowering in population MS could therefore be a constraint of our experimental design in which seeds from population MS have germinated in late winter instead of late autumn. As a consequence of this phenology delay, flowering occurred during unfavourable environmental conditions (higher temperatures) in late May compared with natural conditions, which could have affected later vegetative growth, floral development and the pollination efficiency of CL and CH flowers in spring, which is lower compared with the other three populations. However, the particular architecture of plants in population MS (more lateral axes, fewer whorls per axis with smaller internode lengths) and the reduced size of the plants could reflect genetic differentiation for these traits. In line with this, microsatellite estimates of the genetic structure of these four populations showed that the number of private alleles was the highest in population MS (Stojanova et al., 2013), supporting the possibility of population differentiation.

In spite of the morphological and phenological differences between the populations studied, or whether they experienced one or two seasons in their natural habitat, the plasticity pattern of cleistogamy-related traits across seasons does not differ between populations: overall CH proportion is ∼30 % in spring and ∼10 % in autumn. An increase in CH production in conditions that favour plant growth is commonly observed for annual cleistogamous species (reviewed in Oakley et al., 2007). Moreover, in several other studies of plastic cleistogamy with patterns similar to ours it was observed that CH proportion variation depended exclusively on the environmental conditions, with no population differentiation for this trait, though morphological or phenological differences could be observed across populations. For instance, CH proportion did not change between three different populations of Impatiens noli-tangere grown under two different light intensity conditions, though population differentiation was observed for flowering time (Masuda et al., 2004). In reciprocal transplant experiments using individuals from one Calathea micans population, CH production was influenced by light intensity and nutrient availability, but not by plant size (Le Corff, 1993). Light intensity is a reliable environmental cue that indicates habitat quality; in general; low light intensity indicates high vegetation density, which could in turn reflect a set of ecological conditions such as increased interspecific competition levels (Donohue et al., 2000). In our study, environmental variation corresponds to seasonal changes, which are generally easy to predict because they are associated with a set of cues that always change in the same manner, such as photoperiod, which is increasing in spring and decreasing in autumn (Lord, 1980). The fine-tuning of cleistogamy-related traits in response to predictable environmental cues suggests adaptation to environmental variation, provided that each phenotype is advantaged in the environment that induces its production (Via and Lande, 1985).

Selection on plasticity

According to our estimates of selection coefficients, larger plants (with more flowers) and early CL flowering are favoured in both seasons. This result is not surprising, since maximizing seed production and early flowering are advantageous for annual species (Munguía-Rosas et al., 2011). Interestingly, cleistogamy-related traits (CH proportion and constitutive cleistogamy proportion) were under stabilizing selection in all populations in both seasons. The optimal values (i.e. phenotypes that maximize individual fitness) differed among seasons, with the optimal CH proportion being lower in autumn than in spring (17 and 33 %, respectively, after back-transformation of the arcsin square root data in Table 3). Furthermore, our analysis of the regression of plasticity of the CH proportion on annual fitness shows that plasticity slightly but significantly improves annual fitness in all of the scenarios tested. These observations indicate that increasing the CH proportion in spring is likely to be adaptive, as suggested by Lord (1982). The sensitivity of the optimal CH proportion to the viability of each progeny type (w_CL, w_CHO and w_CHS) was not high; the highest sensitivity was observed when floral type fitness variation was assumed (and this in autumn only). Importantly, stabilizing selection on CH proportion is always found, and a higher CH optimal value is always predicted in spring regardless of the scenario considered. Moreover, selection is stronger in spring, as revealed by the narrower shape of the selection curves.

Observed values are close to the optimal predicted values of CH proportion in both seasons except for the two southern populations (Montpellier) in autumn, for which observed values are lower than optimal CH proportion. This result is consistent with the fact that populations from southern France experience only the spring season. Though they have maintained plasticity for cleistogamy, their response to autumn conditions is not adaptive. Population DS, which also flowers only in spring, may have maintained an appropriate average CH proportion in autumn potentially because of low population differentiation at the regional scale.

Conclusion

Overall, we show experimentally that plasticity of cleistogamy is consistent with adaptation to seasonal variation in L. amplexicaule, since selection favours different CL phenotypes in each season and plastic genotypes have higher annual fitness than non-plastic ones. Our experimental results also show that CL fertilization success does not vary across seasons, whereas CH flowers have a higher pollination success in spring for three out of the four populations. Thus, CL flowers could provide reproductive assurance when pollinators are scarce (i.e. in autumn), while CH flowers potentially allow the avoidance of inbreeding depression when pollinators are more abundant. However, our results do not predict that 100 % CH flowers in spring and 100 % CL in autumn are optimal strategies. This could be due to differential costs of the two flower types.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of Table S1: Photoperiod at flowering peak and at the end of the flowering season for Montpellier and Dijon.