Stability of pollen–ovule ratios in pollinator-dependent versus autogamous Clarkia sister taxa: testing evolutionary predictions

Authors


Author for correspondence:
Susan J. Mazer
Tel: +1 805 893 8011
Email: mazer@lifesci.ucsb.edu

Summary

  • • It has been proposed that natural selection should favor distinct temporal patterns of sex allocation in selfing vs pollinator-dependent taxa. In autogamous selfers in which pollen receipt is highly reliable, selection should favor genotypes that maintain low and stable pollen to ovule (P : O) ratios throughout flowering. By contrast, in outcrossers the optimum P : O ratio of an individual's flowers will depend on pollinator abundances and mating opportunities, both of which may vary over time. In this case, selection may favor temporal variation among flowers in the P : O ratio. An opposing prediction is that selfing taxa will be developmentally more unstable than outcrossers because of lower homeostasis caused by high homozygosity.
  • • We compared temporal changes in the P : O ratio in two pairs of sister taxa in the genus Clarkia. We examined hundreds of glasshouse-raised maternal families representing three wild populations each of the outcrossing, insect-pollinated Clarkia unguiculata, the facultatively autogamous Clarkia exilis and the outcrossing and selfing subspecies of Clarkia xantiana: ssp. xantiana and parviflora, respectively.
  • • Temporal change in the P : O ratio was significantly greater in both outcrossers than in their selfing sister taxa, although the proportional changes in the P : O ratio (relative to the first bud produced) did not differ significantly between sister taxa (0.07 < P < 0.10).
  • • Our results provide partial support for the hypothesis that the P : O ratio is more stable in selfing than in outcrossing taxa and reject the hypothesis that selfers are less stable.

Introduction

Sex allocation theory predicts that, all else being equal, autogamously self-fertilizing plant populations will evolve to allocate a smaller proportion of reproductive resources to traits that contribute to male function relative to female function than closely related outcrossing taxa (Lloyd, 1987). This prediction has been largely corroborated in angiosperms with cosexual flowers by comparing pollen to ovule (P : O) ratios per flower in selfing vs outcrossing taxa. The evolution of lower P : O ratios in selfers that are derived from outcrossers has been interpreted as the adaptive outcome of selection on the primary sex ratio in response to local mate competition (Charnov, 1982; Lloyd & Bawa, 1984; Miller & Diggle, 2003) and/or to selection favoring the elimination of surplus pollen production (Cruden, 1977, 2000).

While the adaptive significance of variation in the mean P : O ratio among taxa is relatively well-understood (Cruden, 1977; Mione & Anderson, 1992; Chouteau et al., 2008), the significance of variation in the P : O ratio within individuals is less so. Studies of temporal variation among flowers in the P : O ratio have detected high variation within individuals of many species as a result of genetically determined developmental patterns and/or plastic responses to resource availability (Lloyd, 1980; Lloyd, et al., 1980; Stanton & Galloway, 1990; Wolfe, 1992; Diggle, 1993; Vogler et al., 1999; Mazer & Dawson, 2001; Barrett, 2002; Ishii & Sakai, 2002; Garcia, 2003; Miller & Diggle, 2003; Ehlers & Thompson, 2004; Hiraga & Sakai 2007; Zhao et al., 2008; but see Cao et al., 2007). The effects of resource availability and inter-ovary competition have been demonstrated by flower removal experiments that reduce or eliminate the temporal decline in female allocation that is otherwise observed (Garcia, 2003; Kliber & Eckert, 2004). In addition, plant size and flower position can interact to affect gender expression, which is sensitive to the resource status of individual flowers (Ashman et al., 2001). It has not been demonstrated, however, that temporal variation in primary sexual investment is necessarily the adaptive outcome of natural selection.

If the magnitude of developmental flexibility (or stability) in sex allocation among sequentially produced flowers is heritable, then populations may evolve in response to temporal variation in environmental factors that differentially affect male vs female fitness per unit investment in sex-specific functions. In outcrossing populations, for example, natural selection may favor changes in the P : O ratios of flowers over time, depending on the abundance and behavior of pollinators, on the effects of plant size on visitation and pollen export rates, on the number of mates available and on resource-mediated temporal changes in male vs female fitness returns (Brunet & Charlesworth, 1995).

Under some scenarios, temporal shifts towards increased allocation to male function per flower are expected towards the end of the reproductive season. For example, if pollinator service predictably declines as the season progresses, then seed production may become increasingly pollen-limited. In this case, natural selection may favor a temporal increase in the P : O ratio, either through an increase in pollen production (which may increase the probability of pollinator attraction because of a higher reward) or a temporal reduction in ovule production per flower (which may reduce the number of unfertilized ovules). Protandrous outcrossers are particularly likely to exhibit pollen-limited seed production late in the reproductive season because at this time most flowers will be in the female phase and pollen-producing flowers may be rare (Brunet & Charlesworth, 1995; Brunet, 1996). In this situation, among the flowers that are still in the male phase, selection may favor those with male-biased sex allocation (i.e. with high P : O ratios). Similarly, if nutrients or water availability decline over time, allowing flower development (and pollen production) but not full fruit maturation (Lloyd 1980; Lloyd et al., 1980; Thomson, 1989; Jennersten, 1991), then theory also predicts that selection should favor a shift towards male function. It has also been proposed that increases in male allocation may be favored when the number of open flowers, the number of pollinator visits and the rate of pollen removal increase over time, as observed in Narthecium asiaticum (Ishii & Sakai, 2002).

Alternative scenarios predict a temporal increase in the proportional allocation to female function. For example, if pollinators become predictably scarce among the last-blooming flowers, reproductive assurance may favor the production of autogamously selfing flowers with low P : O ratios. In addition, if as flowering proceeds, nitrogen or phosphorous (which may limit pollen production more than seed production; Ashman & Baker 1992; Poulton et al., 2002) becomes more limiting than carbon and water (which often limit seed production), selection may favor a disproportionate temporal reduction in pollen production per flower.

In all of the scenarios mentioned, animal-pollinated outcrossing taxa may be expected to evolve floral P : O ratios that change over the lifetime of an individual plant. By contrast, mating opportunities among the flowers of highly autogamously selfing individuals remain constant throughout the flowering period (Mazer & Delesalle, 1998; Delesalle et al., 2008). As a result, autogamous taxa should evolve to produce sequential flowers with a highly stable P : O ratio, in which each flower produces the minimum amount of pollen necessary to ensure complete self-fertilization. The absence of other selective forces (e.g. temporal variation in pollen limitation) in selfers should also favor the evolution of highly consistent P : O ratios (Brunet & Charlesworth, 1995).

The degree of temporal variation in the P : O variation should diverge most conspicuously between closely related taxa with highly contrasting mating systems (e.g. obligate outcrossers vs fully autogamous selfers). Nevertheless, wherever taxa differ in mating system sufficiently to result in evolutionary divergence in traits associated with the degree of outcrossing, including flower size, floral longevity, and the mean P : O ratio, we predict that the magnitude of temporal variation in the P : O ratio should also diverge, with the more outcrossing taxa characterized by greater variation.

These predictions are sound if selfers and outcrosser exhibit similar levels of intrinsic developmental instability. An opposing prediction may prevail, however, if habitually selfing genotypes are developmentally less stable owing to reduced homeostasis (Lerner, 1954; Kramer et al., 2002; Hall, 2005; Pedersen et al., 2005; Pertoldi et al., 2006; but see Taylor, 2001). In this case, even if selection favors higher canalization of the P : O in selfers, its expression may be obscured such that selfing taxa regularly exhibit significantly higher variation in the P : O ratio than their selfing counterparts.

In the glasshouse study described here, we compare two pairs of annual herbaceous sister taxa with contrasting mating systems but overlapping geographic distributions in the genus Clarkia (Onagraceae) to examine the association between mating system and variation in the P : O ratio within and among genotypes. The comparison of sister taxa has the benefit of allowing one to control for their many shared traits while detecting the evolutionary divergence of other traits. We address the following sets of questions: (1) Within individual plants, does the P : O ratio differ between the first and last flowers produced? If so, is temporal change in the P : O ratio (ΔP : O) caused by change in pollen production, ovule production or both? (2) Is the P : O ratio more stable within individuals of pollinator-dependent or autogamous taxa? The magnitude of temporal change in the P : O ratio was calculated in four ways: (a) ΔP : O; (b) ΔP : O/Early P : O; (c) | ΔP : O |; and (d) | ΔP : O |/Early P : O. The first two parameters addressed the question: Do both the absolute and the proportional temporal changes in the P : O differ between pollinator-dependent and autogamous taxa? Given that the mean P : O ratio of autogamous Clarkia taxa is lower than that of their outcrossing sister taxa, comparing them with respect to the proportional change in the P : O controls for potential covariation between the mean and the variance of the P : O within individuals. The second two parameters address the question: Does the absolute value of the change in the P : O ratio (| ΔP : O |) differ between sister taxa, independently of the direction of the change? This is a critical comparison because the mean value of ΔP : O could be close to zero in a taxon even if maternal families differ greatly in the magnitude and direction of variation among families. In this case, our prediction would be supported if the mean values of | ΔP : O | and | ΔP : O |/Early P : O are lower in selfers than in outcrossers.

This study complements a field survey (Delesalle et al., 2008) that reported phenotypic variation in the P : O ratio within and among field populations of the same taxa. The field observations were in partial agreement with the results reported here; in one pair of sister taxa (Clarkia exilis vs Clarkia unguiculata) the highly autogamous C. exilis exhibited lower absolute (but not proportional) temporal change in the P : O than the outcrossing unguiculata, but the other pair (Clarkia xantiana ssp. parviflora vs ssp. xantiana) did not differ significantly in either measure of temporal variation. Variation in the P : O ratio observed in field populations includes the combined effects of environmental variance and phenotypic plasticity. Where phenotypic variance is inflated because of environmental heterogeneity, it may be difficult to detect genetically based differences between taxa. In the study reported here, environmental variance was reduced by raising maternal families in a common environment, elevating the proportion of phenotypic variation that should be attributable to genetic sources. We compared the patterns observed in the field and in the glasshouse to assess their generality.

Materials and Methods

Study system

Clarkia (Onagraceae) comprises around 41 self-compatible, annual herbaceous taxa, including 34 self-compatible diploid species whose center of distribution is California, USA (Hickman, 1993). Systematic relationships in Clarkia have been inferred from phenotypic and molecular variation (Gottlieb, 1984, 1988; Lewis & Lewis, 1955; Sytsma & Gottlieb, 1986; Sytsma, 1990; Sytsma et al. 1990; Gottlieb & Ford, 1996), and autogamy has evolved independently many times. Clarkia has been used extensively to explore the ecological and evolutionary causes and consequences of alternative mating systems (Moore & Lewis, 1965; Vasek & Harding, 1976; Holtsford & Ellstrand, 1992; Runions & Geber, 2000; Fausto et al., 2001; Eckhart et al., 2004; Mazer et al., 2004, 2007; Geber & Eckhart, 2005; Moeller & Geber, 2005; Moeller, 2006). As in many taxa, outcrossing rates in Clarkia are associated with the degree of herkogamy and dichogamy (Lewis, 1953; Vasek 1964a, 1965, 1967; Vasek & Harding, 1976; Holtsford & Ellstrand, 1992).

Here we focus on two pairs of diploid sister taxa, in each of which there is an evolutionary shift in the degree of pollinator-dependence. All four taxa are self-compatible winter annuals, with differences between sister taxa in herkogamy, dichogamy, flower size, floral longevity, P : O ratio and fruit set in the absence of pollinators being consistent with an evolutionary shift between predominantly pollinator-dependent outcrossing and autogamous selfing (Vasek, 1958, 1960, 1965, 1967, 1977; Moore & Lewis, 1965; Vasek & Sauer, 1971; Vasek & Harding, 1976; Vasek & Weng, 1988; Eckhart & Geber, 1999; Dudley et al. 2007). The first pair of sister taxa comprises the predominantly outcrossing C. unguiculata Lindley and the facultatively selfing C. exilis Lewis & Vasek; the second pair includes two subspecies of C. xantiana Gray: the outcrossing ssp. xantiana and the autogamous ssp. parviflora (Lewis & Raven, 1992). Hereafter, we refer to these taxa as unguiculata, exilis, xantiana and parviflora, respectively. In both pairs, the outcrossers are the proposed progenitors of the selfers (Vasek, 1964b, 1977; Gottlieb, 1984).

Under glasshouse and field conditions, xantiana and unguiculata flowers develop and senesce more slowly and are more herkogamous and dichogamous than those of their selfing sister taxa (Vasek, 1958; Eckhart et al., 2004; Dudley et al., 2007). This corresponds to differences between sister taxa in their ability to set seeds autogamously. Under insect-free conditions, mean seed production per fruit in exilis was three times higher than in unguiculata (60.9 vs 20.6 seeds, respectively; Vasek, 1958) relative to a maximum of about 100 seeds per fruit. In common garden experiments conducted in the field, fruit set in xantiana was as low as 0.3% where its pollinators are absent, while parviflora exhibited 75% fruit set in the same location (Geber & Eckhart, 2005).

Genetically based estimates of outcrossing rates in unguiculata do not differ from 100% (Vasek, 1965), while exilis facultatively self-fertilizes but may outcross up to 45% of its seeds (Vasek, 1964a, 1967; Vasek & Harding, 1976). Estimates of outcrossing rates in field populations of C. xantiana are unavailable, but rigorous ecological studies and surveys of allozymes demonstrate that ssp. xantiana relies on insect-pollination to achieve high levels of fruit set and comprises genetically polymorphic populations, while ssp. parviflora autogamously self-fertilizes and comprises genetically monomorphic populations (Gottlieb 1984; Geber & Eckhart, 2005; Moeller & Geber, 2005; Moeller, 2006).

Unguiculata occupies a broad geographic range among woodland slopes, grasslands, and road cuts in the Coast Ranges, the Sierra Nevada western and southern foothills, and the Tehachapi, Western Transverse, Peninsular and South Coast Ranges; exilis is much less common and restricted to similar vegetation types in and near the Kern River Valley (Kern and Tulare Counties, CA, USA). Where both species coexist, their habitat preferences differ; exilis typically occupies the bases of large boulders and the banks or channels of streambeds, which may retain more moisture than the more exposed slopes often occupied by unguiculata. The outcrossing and relatively widespread C. xantiana ssp. xantiana occupies hillsides widely distributed in the southern Sierra Nevada, the Tehachapi Mountains, and the Western Transverse Ranges; ssp. parviflora is typically restricted to boulder fields, streambeds and rocky slopes in the eastern portion of the species’ range.

Seed collection  In May and June of 2001, mature seeds were collected from three wild populations per taxon in Kern and Tulare Counties, California (Table 1). Seeds collected from individual plants (maternal families) were stored in paper coin envelopes (one family/envelope) at room temperature until prepared for germination.

Table 1.  Geographic locations of populations from which seeds were collected
PopulationLongitudeLatitude
Clarkia exilis
Cow Flat Road (CF)35°31.79′N118°39.19′W
Dougherty Creek (DC)35°28.36′N118°42.82′W
Stark Creek (SC)35°28.52′N118°43.47′W
Clarkia unguiculata
Bodfish-Caliente (BC)35°20.33′N118°34.76′W
Live Oak (LO)35°28.82′N118°44.89′W
Posey (P)35°48.43′N118°43.90′W
Clarkia xantiana ssp. parviflora
Chimney Peak (CP)35°46.33′N118°05.17′W
Long Valley (LV)35°48.82′N118°05.95′W
Sawmill Road (SM)35°40.69′N118°28.89′W
Clarkia xantiana ssp. xantiana
Camp 3 (C3)35°48.68′N118°27.26′W
Cow Flat Road (CFR)35°31.79′N118°39.19′W
Sawmill Road (SM)35°50.69′N118°28.89′W

Cultivation  In November 2003 for exilis/unguiculata and March 2005 for parviflora/xantiana, 20–50 seeds per maternal family (n = 43–63 families/population for exilis/unguiculata; 51–60 families/population for parviflora/xantiana) were transferred to agar-filled (8 g l−1) Petri dishes (5 cm), wrapped in aluminum foil, and refrigerated for about 1 wk at 10°C. To promote germination, following vernalization, seeds of parviflora were scarified by nicking the seed coat with a scalpel. Petri dishes were then transferred to ambient glasshouse conditions under natural light until seeds germinated (1–5 d). Five to eight days after germination, two to five seedlings from each maternal family were transplanted into each of three plastic tubes (3.8 cm diameter, 21 cm deep, with drainage holes) filled with UC mix (½ part sand and ½ equal parts by volume of vermiculite, Perlite, oak leaf humus and peat moss). Each tube also contained four pellets of slow-release fertilizer (sieved to 2–4 mm diameter; 14 : 14 : 14 nitrogen–phosphorus–potassium (NPK) obtained from Osmocote Scotts-Sierra Horticultural Products Co. (Marysville, OH, USA). For exilis and unguiculata, the pellets were added to the surface 3 wk following seedling transfer; for xantiana and parviflora, the pellets were placed approx. 5 cm below the soil surface just before seedling transfer. All tubes were spatially randomized in the glasshouse and kept moist with daily or biweekly mist irrigation, as needed.

When seedlings reached a height of 8–16 cm each tube was thinned to leave one healthy seedling per tube and the tubes were suspended in racks at densities of five to eight plants per square foot. Additional details regarding fertilization, watering and supplemental lighting are provided in Dudley et al. (2007). Upon senescence, water was withheld and plants dried in the glasshouse for at least 2 wk. When plants had become fully air-dried under ambient glasshouse conditions, aboveground stems were pruned of leaves and reproductive structures, separated from the root, placed into paper bags, and dried for 3 d at 80°C. Oven-dried shoots equilibrated to ambient indoor humidity at room temperature for at least 24 h before being weighed to obtain total shoot mass (recorded to 0.001 g).

Sampling of buds  The nodal positions of all buds sampled on the primary stem were recorded in sequence from the base of the plant. Typically, each node above the first flower bud produced a fully developed flower. When recording the node position of the buds sampled later in ontogeny, however, the occasional nodes that produced either an aborted bud or no bud were ignored.

Bud sampling in exilis and unguiculata As plants matured in February 2004, the glasshouse population included 546 C. exilis individuals representing 183 maternal families and 473 C. unguiculata individuals representing 160 families (almost always three siblings per family). From each plant, one ‘Early’ and one ‘Late’ bud were sampled to measure pollen and ovule production; all buds were sampled from the primary stem. The ‘Early’ bud was removed from the node just above the first open flower; the ‘Late’ bud was typically the final, fully developed bud produced by the primary meristem (rarely, one or two additional buds were produced several days after we had removed what we considered to be the final bud, but we did not resample these plants).

Bud sampling in parviflora and xantiana As plants matured in May 2005, the glasshouse population comprised 463 individuals of parviflora representing 164 maternal families and 508 individuals of xantiana representing 178 maternal families. From each plant, three buds were sampled: an ‘Early’ bud (from the second reproductive node on the primary meristem), a ‘Middle’ bud (before removal, its position was estimated from the total number of buds present on the primary stem at the time of its removal; it was usually the fourth flower bud produced) and a ‘Late’ bud (which appeared to be the penultimate bud produced by the primary meristem). In 33% and 79% of the maternal families of parviflora and xantiana, respectively, one or more of the siblings exhibited a short additional flush of flowering after this ‘Late’ bud was collected, but to maintain a balanced design we did not resample these plants.

Pollen production per bud  To ensure that total pollen production per bud was measured, buds were collected before they opened. Buds were refrigerated for up to 48 h at 5°C before separating the ovary from the rest of the flower. All anthers in each bud were then manually opened by applying gentle pressure along the line of dehiscence to promote pollen release. Once opened, the anthers were placed in an uncapped 1.5 ml microcentrifuge tube (protected from dust with a sheet of plastic) at room temperature for 7 d during which the anthers dehisced. Tubes were then closed and stored at room temperature for up to 3 months before counting pollen grains (one drop of ethanol was added to the tubes with parviflora or xantiana pollen before closing them).

To prepare pollen for counting, 0.5 ml of solution (9 ml ethanol, 20 ml glycerol, 71 ml deionized (DI) water) was added to each microcentrifuge tube, which was then agitated and vortexed to release and to suspend the pollen. The contents were then transferred to 50 ml glass vials into which a known volume (c. 30 ml) of a saline glycerol solution (100 g NaCl, 7.5 l DI water, 2.5 l glycerol) was added to maintain pollen in suspension. An Elzone 180PC Particle Counter was used to estimate the total number of the pollen grains in each vial (Micromeritics Instrument Corporation, Norcross, GA, USA; Mazer et al., 2007). Five 0.5-ml aliquots from each vial were analysed in succession and the pollen counts of these samples were used to estimate total pollen production per bud based on the known volume of solution contained in each vial. To ensure that pollen remained suspended in solution, the sample vial was gently inverted several times before recording the number of pollen grains. Among the five counts, the highest and the lowest were discarded. The mean value of the remaining counts were used to estimate total pollen production per bud based on the volume of solution held in each vial.

Ovule production per flower  The ovary of each bud was placed in a microcentrifuge tube and stored frozen before dissection. All four locules were opened and the total number of ovules counted.

P : O ratio  For each bud, the P : O ratio was estimated as the number of pollen grains divided by the number of ovules in the ovary.

Composite traits  For each bud position (early, middle and late), maternal family means of pollen number, ovule number and P : O ratio were calculated based on the three experimental siblings and used as independent data points in all analyses.

In all four taxa, we used maternal family means of Early and Late buds to estimate temporal change in pollen and ovule production in two ways: (1) the absolute change in pollen and ovule production (Δpollen and Δovule, respectively) were calculated for each maternal family by subtracting mean pollen (or ovule) production of the Early bud from that of the Late bud and (2) the proportional change in pollen or ovule production over time (Δpollen/Early bud pollen production and Δovule/Early bud ovule production). The temporal change in the P : O ratio (ΔP : O) was calculated for each maternal family by subtracting the mean P : O of the Early bud from that of the Late bud. The proportional change in the P : O ratio was estimated as ΔP : O/Early P : O. The absolute values of these parameters (| ΔP : O | and | ΔP : O |/Early P : O) were also calculated in order to estimate and to compare between sister taxa the magnitude of temporal change in the P : O ratio independent of the direction of the change.

The proportional changes in ovule production, pollen production, and the P : O ratio were used to compare sister taxa to control for the fact that the mean values of these traits differ between sister taxa. When this is the case, even if taxa differ with respect to the absolute magnitude of the temporal change in these traits (e.g. with outcrossers exhibiting a higher mean ΔP : O), they will not necessarily differ with respect to the proportional change.

Statistical analysis

Differences between sister taxa in plant size  We were concerned that plant size or flower production might differ within or between sister taxa and influence the magnitude of Δpollen, Δovule or ΔP : O, potentially confounding or obscuring differences between sister taxa owing to mating system alone. Mixed model nested anovas detected significant differences between sister taxa and populations (nested within taxon) in stem biomass, although the distributions overlap (Fig. 1a,b). In both pairs of sister taxa, stem biomass was significantly greater in the outcrossing taxon (Table 2) and was positively correlated with the number of floral nodes produced between the first and last buds sampled (exilis/unguiculata, r = 0.59, P < 0.0001; xantiana/parviflora, r = 0.39, P < 0.0001).

Figure 1.

Bivariate relationships between the magnitude of temporal change in the pollen to ovule (P : O) ratio and two measures of plant size: stem biomass (a,b) and the number of floral nodes between sampled buds (c,d) for the two sister pairs: Clarkia exilis (selfing)–unguiculata (outcrossing) (a,c) and parviflora (selfing)–xantiana (outcrossing) (b,d).

Table 2.  Mixed model nested anovas to detect significant differences between sister taxa and populations with respect to stem biomass between sampled buds; population is treated as a random effect
(a) exilis vs unguiculata, Stem biomass (g)
SourceNDF/DDFF-ratio or 95% CI of the variance componentP > F
Species1/3.8024.910.0086
Population (Species) (−0.0011 to 0.0037) 
Species and populationLeast square mean (g)SE
exilis
Cow Flat (CF)0.65c0.018
Dougherty Creek (DC)0.67c0.018
Stark Creek (SC)0.69c0.018
unguiculata
Bodfish-Caliente (BC)0.84ab0.018
Live Oak (LO)0.87a0.018
Posey (P)0.80b0.021
(b) parviflora vs xantiana, stem biomass (g)
SourceNDF/DDFF ratio or 95% CI of the variance componentP > F
Species1/4.0211.860.0260
Population (Species) (−0.0004 to 0.0023) 
Species and populationLeast square meanSE
  1. F-ratios and the 95% Confidence Interval (CI) of the variance component (for random effects) and P-values are shown. Where the 95% CI of the variance component overlaps zero, the random effect is not significant. REML was used to estimate variance components and the Satterthwaite method to estimate degrees of freedom (df). NDF/DDF = df in the numerator/df in the denominator of the F-statistic. Population means are shown below each anova table. Pairwise Student's t-tests were conducted to detect significant differences among populations. Distinct superscripts between population means indicate significant differences (α = 0.05) between population means based on Student's t.

parviflora
Chimney Peak (CP)0.15cd0.0079
Long Valley (LV)0.17c0.0074
Sawmill Road (SM)0.15d0.0079
xantiana
Camp 3 (C3)0.22b0.0076
Cow Flat Road (CFR)0.29a0.0074
Sawmill Road (SM)0.22b0.0074

To determine whether these size differences between sister taxa might confound our ability to detect differences between them in ΔP : O, we examined correlations among maternal family means between ΔP : O and (a) stem biomass and (b) the number of flowering nodes between buds sampled. When maternal families of exilis and unguiculata were pooled, plant size was significantly negatively correlated with ΔP : O (Fig. 1a); maternal families that grew into larger plants exhibited greater reductions in the P : O ratio than smaller plants. When maternal families of xantiana and parviflora were pooled, ΔP : O was independent of each whole-plant trait (Fig. 1b,d).

Given the effects of stem biomass and node number on ΔP : O among maternal families in the exilis-unguiculata sister pair, we retained both variables as independent covariates in the analyses described later. Although the bivariate analyses detected no significant effect of plant size on ΔP : O in the parviflora–xantiana pair (Fig. 1), when we included these covariates in the analyses of covariance, one or both whole-plant traits accounted for significant variation in each independent variable (see the Results section), justifying their inclusion in the models.

Differences between sister taxa in pollen and ovule production and the P : O ratio

To detect significant differences between sister taxa in mean pollen production, ovule production, and the P : O ratio per bud (within and across Early, Middle, and Late buds), maternal family means were used in mixed-model ancovas (REML method, JMP v. 6.0; SAS Institute, 2005) in which taxon (species or subspecies) was treated as a fixed effect and population as a random effect nested within taxon. The number of nodes produced on the primary meristem between the Early and Late bud and stem biomass were included as covariates.

Differences between sister taxa in the magnitude of temporal change in the P : O ratio

The mixed-model ancova was also used to detect differences between sister taxa with respect to our four estimates of temporal change in the P : O ratio: the absolute and proportional change in the P : O ratio between Early and Late buds (ΔP : O and ΔP : O/Early P : O and in absolute values of these measures (| ΔP : O | and | ΔP : O |/Early P : O).

Differences between sister taxa in temporal change in pollen and ovule production and the P : O ratio (Δovule, Δpollen, ΔP : O)

Differences between sister taxa in the magnitude and direction of temporal change in gametophyte production per flower and in the P : O ratio were evaluated using repeated measures ancovas. In these ancovas, we included Taxon (species or subspecies) as a fixed effect, Time (Early vs Late buds in exilis/unguiculata; Early, Middle and Late buds in parviflora/xantiana) as the repeated effect and node number and stem biomass as covariates, including interactions between and within subjects. Significant time × Taxon interactions indicate significant ontogenetic differences between sister taxa in the focal traits. The effect of Population (nested within Taxon) was excluded from the model because the mixed model ancovas detected no significant differences among populations with respect to any of our observed traits (Table 3).

Table 3.  Mixed-model ancovas to detect effects of Taxon and Population (nested within taxon) on components of sex allocation
(a) exilis vs unguiculataCovariate: Stem biomassCovariate: number of nodes between Early and Late budsPopulation (species) 95% CI of variance componentSpecies
NDF/DDFF-ratioP > F
Mean pollen/bud (mean of early and late)< 0.00010.0014(−10.33 x 10^4 to 58.11 x 10^4)1/4.04319.650.0111
Mean ovules/bud (mean of early and late)0.00560.5820(−3.50 to 14.28)1/4.057436.21< 0.0001
Mean P : O (mean of early and late)0.00360.0017(−12.71 to 68.31)1/4.06699.660.0005
Early bud pollen production< 0.00010.0518(−17.41 x 10^4 to 95.44 x 10^4)1/4.06035.150.0039
Late bud pollen production< 0.0001< 0.0003(−59.75 x 10^3 to 31.15 x 10^4)1/4.0732.840.1659
Early bud ovule production0.18700.2546(−4.80 to 16.31)1/4.320320.78< 0.0001
Late bud ovule production0.00190.0259(−5.28 to 23.01)1/4.130299.95< 0.0001
Early bud P : O0.00580.0178(−17.25 to 89.54)1/4.084107.810.0004
Late bud P : O0.04570.0047(−10.35 to 49.36)1/4.12786.810.0006
ΔP : O0.17920.7201(−4.27 to 4.90)1/5.762111.560.0001
ΔP : O/Early P : O (or Proportional ΔP : O)0.99250.2017(−0.001 to 0.002)1/6.0163.800.0989
| ΔP : O |0.17810.4005(−3.24 to 2.97)1/6.473159.93< 0.0001
| ΔP : O |/Early P : O0.71550.3918(−0.00068 to 0.00139)1/5.5882.630.1596
(b) parviflora vs xantianaCovariate: stem biomassCovariate: number of nodes between Early and Late budsPopulation (species) 95% CI of variance componentSpecies
NDF/DDFF-ratioP > F
  1. Stem biomass and the number of nodes between sampled buds were included as covariates (only the P-values for these effects are reported here). In all cases, the Taxon × Stem biomass and the Taxon × Number of Nodes interactions were nonsignificant, so these terms were excluded from the model. Where the 95% confidence interval (CI) for the random effect (Population) overlaps zero, the Population effect is not statistically significant at α = 0.05. Significant P-values of fixed effects are in bold type; P-values between 0.05–0.10 are italicized. REML was used to estimate variance components and the Satterthwaite method (default option) was used to estimate degrees of freedom. NDF/DDF = df in the numerator/df in the denominator. P : O, pollen to ovule ratio.

Mean pollen/bud (mean of three buds)0.05610.0177(−63 915 to 254 459)1/4.75015.800.0117
Mean ovules/bud (mean of three buds)< 0.0001< 0.0001(−23.07 to 132.48)1/4.11210.150.0321
Mean P : O (mean of three buds)0.00100.0003(−26.29 to 121.15)1/4.34021.490.0081
Early bud pollen production0.07320.0008(−86 353 to 298 501)1/4.8417.220.0096
Middle bud pollen production0.09190.0922(−125 412 to 470 200)1/4.93219.230.0074
Late bud pollen production0.37140.1428(−59072 to 149013)1/5.66315.570.0085
Early bud ovule production< 0.0001< 0.0001(−28.40 to 160.66)1/4.1473.920.1165
Middle bud ovule production< 0.0001< 0.0004(−33.70 to 195.89)1/4.0945.790.0723
Late bud ovule production< 0.0001< 0.0001(−17.48 to 94.77)1/4.20617.010.0131
Early bud P : O0.0118< 0.0001(−22.43 to 84.39)1/4.55316.780.0114
Middle bud P : O0.06480.0268(−40.03 to 171.58)1/4.45618.380.0101
Late bud P : O0.02490.0002(−31.50 to 94.07)1/4.84424.930.0045
ΔP : O (Late Bud – Early Bud)0.86150.1542(−10.17 to −5.68)1/2.49316.660.0375
Δ P : O/Early P : O (or Proportional ΔP : O)0.49260.2650(−0.012 to 0.011)1/7.5134.830.0614
| ΔP : O |0.18870.0234(−8.42 to 15.43)1/5.4998.160.0319
| ΔP : O |/Early P : O0.63420.7254(−0.008 to 0.013)1/6.6550.660.4448

To compare sister taxa with respect to the magnitude of temporal effects on the P : O ratio, a repeated measures ancova was conducted within each taxon. Sampling time (Early, Middle and Late) was included as a repeated effect and node number and stem biomass were included as covariates. Sister taxa were compared with respect to the magnitude (and significance) of the F-ratio for the time effect to test the prediction that in the pollinator-dependent taxa, the F-ratio would be higher than in the autogamous taxa. All analyses were conducted using JMP statistical software (v. 6.0; SAS Institute, 2005).

Results

Differences between sister taxa in pollen and ovule production and the P : O ratio

Pollen production  Mean pollen production per bud (the mean of all sampled buds) differed significantly between sister taxa but not among populations within taxa (Tables 3, 4; Figs 2, 3). The pollinator-dependent taxa (unguiculata and xantiana) consistently produced more pollen per bud than their facultatively autogamous counterparts. Pollen production per bud was significantly higher in unguiculata than in exilis (Tables 3b, 4a; Fig. 2a), particularly among the Early buds. Among Early buds, unguiculata produced 86% more pollen per flower than exilis; among Late buds, unguiculata produced only 28% more pollen per flower (Tables 3a, 4a). The temporal convergence between species reflects the greater reduction over time in the number of pollen grains per bud in unguiculata than in exilis (Table 4a; Fig. 2a). Mean pollen production per bud was also significantly higher in xantiana than in parviflora: at each sampling time (Early, Middle, and Late buds), xantiana consistently produced 42–50% more pollen per bud than parviflora (Tables 3b, 4b, Fig. 3a).

Table 4.  Species means, standard deviations (SD), standard errors (SE) and coefficients of variation (CV) for all focal traits; parameters estimated from maternal family means
(a) exilis vs unguiculata
 Clarkia exilisClarkia unguiculata
nMeanSDSECVnMeanSDSECV
Early pollen1833858.8736.1854.4219.081607184.61396.70110.4219.44
Late pollen1832130.7566.5141.8826.591602732.1864.0068.3131.62
ΔPollen (Late – Early pollen)183−1728.1749.6255.4143.38160−4452.51205.5295.3027.07
Early ovules183124.210.680.798.6016084.88.980.7110.59
Late ovules18390.09.590.7110.6516045.87.040.5615.36
ΔOvules (Late – Early ovules)183−34.213.741.0240.20160−38.99.400.7424.18
Early P : O18331.26.030.4519.3416085.216.911.3419.85
Late P : O18323.75.720.4224.1816059.414.911.1825.09
ΔP : O (Late – Early P : O)183−7.66.990.5292.59160−25.816.761.3365.05
Proportional ΔPollen ((Late – Early)/Early)183−0.440.150.0133.99160−0.620.100.0116.79
Proportional ΔOvules ((Late – Early)/Early)183−0.270.100.0135.57160−0.460.090.0118.99
Proportional ΔP : O ((Late – Early)/Early)183−0.220.210.0291.85160−0.290.170.0159.43
| ΔP : O |1838.505.800.4368.2516026.6115.381.2257.79
| ΔP : O |/Early P : O1830.260.150.0158.861600.300.150.0148.58
(b) parviflora vs xantiana
 Clarkia xantiana ssp. parvifloraClarkia xantiana ssp. xantiana
nMeanSDSECVnMeanSDSECV
  1. Populations were pooled within species; the mixed model anova detected no significant differences among populations (within species) for any trait (Table 3). n, Number of family means contributing to the species mean. P : O, pollen to ovule ratio.

Early pollen1642459.41223.3495.5349.741783584.71418.89106.3539.58
Middle pollen1613021.91487.20117.2149.211774525.81621.13121.8535.82
Late pollen1632040.11192.4693.4058.451782917.01373.24102.9347.08
ΔPollen (Late – Early pollen)163−433.91132.3188.69260.97178−667.61364.24102.25204.34
Early ovules16486.514.281.1216.5217776.910.500.7913.66
Middle ovules16288.113.371.0515.1917774.010.440.7914.11
Late ovules16273.413.201.0417.9817755.055.00.7518.05
ΔOvules (Late – Early ovules)162−13.114.081.11107.90177−22.010.540.7948.02
Early P : O16431.217.831.3957.1217748.620.921.5743.03
Middle P : O15835.018.311.4652.3817662.626.952.0343.04
Late P : O16031.322.321.7671.3917755.629.442.2152.92
ΔP : O (Late − Early P : O)1600.0518.891.4938.6821777.0125.621.93365.50
Proportional ΔPollen ((Late − Early)/Early)163−0.070.640.0588.982178−0.110.450.03402.78
Proportional ΔOvules ((Late − Early)/Early)162−0.140.150.01110.72177−0.280.120.0144.59
Proportional ΔP : O ((Late − Early)/Early)1600.090.630.05710.451770.260.760.06290.56
| ΔP : O |16013.513.131.0496.9817719.517.931.3591.80
| ΔP : O |/Early P : O1600.470.430.0391.291770.510.610.05119.75
Figure 2.

Population means (± 1 SE) of floral traits measured in Early vs Late buds produced on the primary stem in Clarkia exilis (selfing) and Clarkia unguiculata (outcrossing). (a) Pollen number per bud. (b) Ovule production per bud. (c) pollen to ovule (P : O) ratio per bud. (a–c: open bars, Early bud; closed bars, Late bud.) (d) The mean temporal change in the P : O ratio between Early and Late buds. Negative values indicate a temporal reduction in the mean P : O ratio. Full names of populations appear in Table 1.

Figure 3.

Population means (± 1 SE) of floral traits measured in Early, Middle and Late buds produced by the primary stem in Clarkia xantiana ssp. parviflora (selfing) and Clarkia xantiana ssp. xantiana (outcrossing). (a) Pollen number per bud. (b) Ovule production per bud. (c) Pollen to ovule (P : O) ratio per bud. (a–c: open bars, Early bud; tinted bars, Middle bud; closed bars, Late bud). (d) The mean temporal change in the P : O ratio between Early and Late buds. Negative values indicate a temporal reduction in the mean P : O ratio. Full names of populations appear in Table 1.

Ovule production  Mean ovule production per bud differed significantly between sister taxa but not among populations within taxa (Table 3). In contrast to pollen, ovule production was higher in the autogamous taxa relative to their outcrossing sister taxa (Fig. 2b; Table 4a). Among Early and Late buds, respectively, exilis produced 46% and 96% more ovules per bud than unguiculata. The increased divergence between species over time resulted from the greater proportional decline in ovule production from Early to Late buds in unguiculata than in exilis (Table 4a; Fig. 2b). Mean ovule production per bud was similarly higher in parviflora than in xantiana (Table 4b; Fig. 3b), although the difference between them was statistically significant only among Late buds (Table 3b). Among Early, Middle, and Late buds, respectively, parviflora produced 12, 19 and 33% more ovules than xantiana (Table 4b).

The P : O ratio  Sister taxa differ significantly in P : O ratio, but populations within taxa were similar (Tables 3, 4; Figs 2c, 3c). Among both Early and Late buds, the mean P : O ratio of unguiculata was more than twice that of exilis (Table 4a; Fig. 2c), although the difference was greater among Early buds. The convergence in the P : O ratio over time between the two species was caused by the greater temporal decline in the P : O ratio of unguiculata than exilis (Table 4a). The P : O ratio of all buds sampled was also significantly higher in xantiana than in parviflora (Tables 3b, 4b; Fig. 3c).

Differences between sister taxa in the magnitude of temporal change in the P : O ratio

In both pairs of sister taxa, ΔP : O and | ΔP : O | were significantly higher in the outcrossing taxon than in its selfing counterpart (Table 3; Figs 5, 6). In addition, the mean proportional change in the P : O ratio (ΔP : O/Early P : O) was higher in the outcrosser than in the selfer, but the difference did not reach statistical significance (0.10 < P < 0.05 in both cases; Table 3). The proportional change in the absolute value of the P : O (| ΔP : O |/Early P : O) did not differ significantly between sister taxa.

Figure 5.

Frequency distributions among maternal family means of the pollen to ovule (P : O) ratio (Early and Late buds pooled), the temporal change in the P : O ratio between Early and Late buds (P : O Late bud – P : O of Early bud) and the proportional change in the P : O ratio in: (a) Clarkia exilis (selfing; closed bars) and Clarkia unguiculata (outcrossing; open bars) and (b) Clarkia xantiana ssp. parviflora (selfing; closed bars) and C. xantiana ssp. xantiana (outcrossing; open bars). Negative values indicate that pollen production per bud declines more than ovule production per bud, resulting in Late buds that are more female-biased than Early buds.

Figure 6.

Frequency distributions among maternal family means of the absolute value of the temporal change in the pollen to ovule (P : O) ratio. These show the magnitude of the temporal change within maternal families irrespective of the direction of the change. (a) Clarkia exilis (selfing; closed bars) and Clarkia unguiculata (outcrossing; open bars) and (b) Clarkia xantiana ssp. parviflora (selfing; closed bars) and Clarkia xantiana ssp. xantiana (outcrossing; open bars).

Temporal change in pollen and ovule production and the P : O ratio (ΔP : O)

exilis (selfing) vs unguiculata (outcrossing)  Both species exhibited a temporal decline in pollen production (Fig. 2a; Table 4a). There was a significant Time × Species interaction term in the repeated ANCOVA (Table 5a); the temporal reduction in pollen production in exilis (44%) was less than that in unguiculata (62%). Both measures of the change in pollen production (Δpollen and Δpollen/Early bud pollen production) were significantly higher in unguiculata than in exilis (Tables 3a, 4a).

Table 5.  Repeated measures ancova to detect significant differences between species in the magnitude of temporal change in ovule production, pollen production, and the pollen to ovule (P : O) ratio
(a) exilis vs unguiculata
SourcePollen numberOvule numberP : O
NDF/DDF = 1/337NDF/DDF = 1/337NDF/DDF = 1/337
FPFPFP
Between subjects
Covariate: number of nodes between early and late buds (NN)23.39< 0.00010.120.733626.96< 0.0001
Covariate: stem biomass38.59< 0.000111.070.001019.84< 0.0001
Species373.04< 0.00012452.80< 0.00011416.37< 0.0001
Species × NN1.010.31571.080.29946.160.0136
Species × Stem biomass0.060.80110.600.43980.700.4050
Within Subjects
Time66.93< 0.000175.60< 0.000110.730.0012
Time × NN0.210.64937.090.00810.090.7682
Time × Stem biomass5.610.01841.160.28151.540.2161
Time × Species435.26< 0.00018.690.0034119.57< 0.0001
Time × Species × Stem biomass0.010.93850.170.67870.180.6714
Time × Species × NN0.600.43800.090.76070.430.5110
(b) parviflora vs xantiana
SourcePollen numberOvule numberP : O
NDF/DDF = 1/329NDF/DDF = 1/331NDF/DDF = 1/325
FPFPFP
  1. The repeated measures are based on two sampled buds per individual (Early vs Late) in exilis/unguiculata, and on three buds per individual (Early, Middle and Late) in parviflora/xantiana. NDF/DDF = Numerator df/Denominator df.

Between subjects
Covariate: number of nodes between early and late buds (NN)6.410.011823.89< 0.00019.950.0018
Covariate: stem biomass2.10< 0.000123.09< 0.00012.210.1383
Subspecies66.20< 0.0001193.77< 0.000195.49< 0.0001
Subspecies × NN0.950.331122.09< 0.00011.260.2632
Subspecies × Stem biomass1.410.235845.64< 0.00014.780.0295
Within Subjects
Time12.28< 0.00015.830.00333.750.0246
Time × NN1.220.29599.58< 0.00011.260.2838
Time × Stem biomass0.250.78190.260.76890.520.5919
Time × Subspecies4.880.008210.45< 0.00017.610.0006
Time × Subspecies × Stem biomass1.310.27030.690.50431.660.1909
Time × Subspecies × NN0.200.81613.870.02190.290.7477

Both species also exhibited a decline in ovule production between Early and Late buds (Table 4a; Fig. 2b). The proportional reduction in ovule production in exilis (27.0%) was less than that in unguiculata (46.0%). The difference between species in the decline was also reflected in the significant Time × Species interaction in the repeated measures ancova (Table 5a).

The P : O ratio of all populations in both species declined significantly from Early to Late buds (Fig. 2c,d); Late flowers were phenotypically more female than Early flowers. The P : O ratio was more stable in the exilis than in the outcrossing unguiculata, particularly when measured in absolute terms; Fig. 2d vs Fig. 4a); the mean change in the P : O ratio among exilis maternal families was –7.6 while the mean change among unguiculata families was –25.8 (Table 4a). Moreover, the repeated ancova detected a significant Time × Species interaction effect on the P : O ratio (Table 5a). In addition, the repeated ancovas conducted on each species separately indicated that the strength of the temporal effect on the P : O ratio in unguiculata was much greater than that in exilis (the F-ratios for the temporal effects were 9.49 vs 1.11, respectively; Table 6a). As a proportion of the P : O ratio of the Early buds, the change in the P : O was not significantly different between taxa at the P < 0.05 level (P > 0.09, Table 3a; Fig. 4a).

Figure 4.

Population means (± 1 SE) for the proportional change in the pollen to ovule (P : O) ratio between Early and Late buds for each pair of sister taxa. (a) Clarkia exilis (selfing) and Clarkia unguiculata (outcrossing). (b) Clarkia xantiana ssp. parviflora (selfing) and Clarkia xantiana ssp. xantiana (outcrossing). Negative values indicate a proportional decline in the P : O ratio. Full names of populations appear in Table 1.

Table 6.  Repeated measures ancova within each taxon to detect differences in the strength of the temporal change in the pollen to ovule (P : O) ratio
(a) exilis vs unguiculata
SourceClarkia exilisClarkia unguiculata
Between-SubjectsNDF/DDFF ratioP > FNDF/DDFF ratioP > F
Covariate: Number of nodes between early and late buds (NN)1/18013.160.00041/15719.90< 0.0001
Covariate: Stem biomass1/18021.42< 0.00011/15711.080.0011
Within-Subjects
Time1/1801.110.29341/1579.490.0024
Time × NN1/1800.180.67611/1570.320.5709
Time × Stem biomass1/1800.810.37071/1571.150.2845
(b) parviflora vs xantiana
SourceClarkia xantiana ssp. parvifloraClarkia xantiana ssp. xantiana
Between-SubjectsNDF/DDFF-ratioP > FNDF/DDFF-ratioP > F
  1. The F-value for the Time effect is higher in unguiculata (outcrossing) than in exilis (selfing) and in xantiana (outcrossing) than in parviflora (selfing). NDF/DDF = numerator degrees of freedom/denominator degrees of freedom for the F-statistic. Repeated measures are based on two sampled buds per individual (Early vs Late) in exilis/unguiculata and on three buds per individual (Early, Middle, and Late) in parviflora/xantiana. Bold type denotes P-values < 0.05.

Covariate: Number of nodes between early and late buds (NN)1/15211.86< 0.00071/1731.900.1695
Covariate: Stem biomass1/1527.010.00901/1730.340.5584
Within-Subjects
Time2/1510.840.43562/1724.170.0170
Time × NN2/1510.020.97552/1722.610.0761
Time × Stem biomass2/1510.580.56102/1720.400.6716

In summary, the magnitude of the mean temporal decline in the P : O ratio was greater in the outcrossing unguiculata than in exilis; this was because of the particularly sharp reduction in pollen production in the former (Fig. 2a). The proportional decline in the P : O ratio did not differ as sharply between these sister species (Fig. 4a).

parviflora (selfing) vs xantiana (outcrossing)  Both subspecies exhibited a decline in pollen production between the Early and Late bud (Fig. 3a; Table 4b). The repeated ancova detected a significant Time × Subspecies interaction term (Table 5b); the temporal reduction in pollen production between Early and Late buds in the selfing parviflora (7.2%) was less than that in outcrossing xantiana (11.4%).

Both subspecies also exhibited a decline in ovule production between Early and Late buds (Table 4b; Fig. 3b). The proportional reduction in ovule production in parviflora (14%) was less than that in xantiana (28.0%). The difference between subspecies in the proportional decline in ovules per bud was also reflected in the significant Time × Subspecies interaction in the repeated measures ancova (Table 5b).

In contrast to unguiculata and exilis, the P : O ratio of parviflora and xantiana increased significantly from Early to Late buds (Tables 3b, 4b; Figs 3d, 4b); Late flowers were phenotypically more male than Early flowers. The P : O ratio was more stable within the selfing parviflora than in the outcrossing xantiana when measured in absolute terms (Table 4b, Fig. 3c). The mean (± 1 SE) increase in the P : O ratio among parviflora maternal families was 0.049 ± 1.49 while xantiana families increased by 7.01 ± 1.93 (Table 4b). Accordingly, the repeated ancova detected a significant Time × Subspecies interaction effect on the P : O ratio (Table 5b). In addition, the repeated ancovas conducted on each subspecies separately indicated that the strength of the temporal effect on the P : O ratio in the outcrossing xantiana was much greater than that in the selfing parviflora (the F-ratios were 4.17 and 0.84, respectively; Table 6b). The proportional change in the P : O, however, did not differ significantly between taxa (P = 0.0614; Table 3b; Fig. 4b), although xantiana consistently exhibited a greater change in the proportional P : O ratio than parviflora.

In each pair of sister taxa, the magnitude of the change in the P : O ratio was significantly greater in the outcrossing than in the selfing taxon when measured as ΔP : O (Figs 2d, 3d, 5). The proportional change in the P : O ratio was not significantly different between sister taxa at the 0.05 significance level, although in both pairs of taxa, the change was lower in the autogamously selfing taxon than in the pollinator-dependent taxon (Fig. 4). We found no evidence that the selfers exhibited less stable P : O ratios as a result of a (putative) lack of homeostasis.

The frequency distributions of the P : O ratio and the absolute and proportional temporal change in the P : O among maternal families offer indirect evidence that stabilizing selection on the P : O ratio and on ΔP : O may be stronger in autogamous selfers than in pollinator-dependent taxa. First, the ranges of the mean P : O ratio, the ΔP : O, and the | ΔP : O | among maternal families was greater in unguiculata than in exilis (Figs 5, 6). Second, the mode of the frequency distribution of proportional ΔP : O deviated from zero more in unguiculata than in exilis (Fig. 5a). Xantiana exhibited greater variation among maternal families than parviflora in the | ΔP : O ratio | (Fig. 6), but the subspecies do not differ markedly with respect to variation in the other parameters.

Discussion

Most of our observations support the conclusion that the P : O ratios of autogamous taxa are more stable than those of their pollinator-dependent, outcrossing sister taxa, at least between the very first and last buds produced on the primary axis. In addition, we found no evidence that the facultatively autogamously selfing maternal lineages are more unstable than the maternal families of their outcrossing counterparts.

Three lines of evidence support our prediction that selfers evolve to exhibit less intraindividual variation among flowers in the P : O ratio. First, mixed model ancovas (Table 3) detected significant differences between sister taxa (in both pairs studied here) in the magnitude of the temporal change in the P : O ratio (ΔP : O ) and in its absolute value (| ΔP : O |), and nearly statistically significant (0.07 < P < 0.10) differences between them in the proportional change in the P : O ratio (ΔP : O/Early bud P : O). In both pairs of sister taxa, the selfer exhibited lower values of ΔP : O and | ΔP : O | between the first and last bud sampled than its outcrossing counterpart. Second, the repeated measures ANOVAs detected significant Time × Taxon interactions for the P : O ratio within both pairs of sister taxa, demonstrating that sister taxa differed in the magnitude of the change in P : O (Table 5; Figs 2d, 3d). Inspection of the data (Figs 2c,d, 3c,d, 6) confirm that the autogamously selfing taxa exhibited smaller temporal changes in the P : O ratio relative to their outcrossing counterparts, and that this difference tends to persist (although more weakly) even when controlling for differences between taxa in the mean values of the P : O ratio (Fig. 4). Third, the F-ratios associated with the time effect within each taxon (Table 6) show that the pollinator-dependent outcrossers exhibited much stronger time effects than either autogamous species.

Theoretical predictions concerning temporal variation in sex allocation state that protandrous outcrossers (such as C. unguiculata and C. xantiana ssp. xantiana) should exhibit temporal increases in the P : O ratio among flowers produced sequentially along the primary axis (Brunet & Charlesworth, 1995). This pattern was observed in ssp. xantiana, but unguiculata exhibited a temporal decline in the P : O ratio between the first and last buds produced. In the following sections we discuss possible reasons for this anomaly.

Mean gametophytes per flower and the P : O ratio in selfing vs outcrossing sister taxa

The sister taxa observed here have diverged phenotypically in a manner consistent with sex allocation theory. For example, the insect-pollinated outcrossing unguiculata and xantiana exhibit higher P : O ratios than their sister taxa (see also Vasek & Weng, 1988; Delesalle et al., 2008). Given that pollen transfer is often less efficient in animal-pollinated outcrossers (because of the loss or consumption of pollen) than in autogamous selfers that are morphologically adapted to ensure high rates of pollination, more pollen per ovule is likely to be necessary to achieve the same male reproductive success per unit of resources allocated to pollen production in outcrossers (Charnov, 1982). Accordingly, the higher P : O ratios observed in outcrossing relative to autogamous Clarkia mirror those seen in other outcrossing vs selfing congeners and in large comparative studies examining the relationship between pollination mechanism and P : O ratios (Cruden, 1977; Spira, 1980; Thomas & Murray, 1981; Schoen, 1982; Preston, 1986; Ritland & Ritland, 1989; Mione & Anderson, 1992; Armbruster et al., 2002; Jürgens et al., 2002; Parachnowitsch & Elle, 2004; Götzenberger et al., 2008). Exilis and parviflora also produce less pollen per flower, and smaller and more short-lived flowers, than their outcrossing sister taxa.

Similar to the patterns observed in field populations (Delesalle et al., 2008), glasshouse populations of the facultatively autogamous exilis and parviflora produce more ovules in both Early and Late buds than their outcrossing sister taxa (Figs 2, 3). This greater ovule production is coupled with the development of smaller seeds (e.g. individual exilis seeds are about one-half the mass of those of unguiculata; Knies et al., 2004). The relatively low pollen production, small petals and short-lived flowers of the selfers relative to the outcrossers (Runions & Geber, 2000; Mazer et al., 2007) are consistent with the interpretation that selfers have evolved to allocate resources not used for pollinator rewards or attraction to additional ovule production per flower. Conversely, small seeds may be adaptive for other ecological reasons in the selfing taxa, and an ensuing size/number trade-off may result in higher ovule production per flower.

Although these patterns match those observed in field populations of unguiculata and exilis, there are some interesting differences (Table 7; see also Vasek & Weng, 1988). First, in the two field populations of each of these species surveyed by Delesalle et al. (2008), all but one population (exilis) exhibited temporal increases in the P : O ratio. By contrast, all six glasshouse populations of unguiculata and exilis studied here exhibited temporal declines in the P : O ratio. Second, in both unguiculata and exilis, the glasshouse-grown populations produced fewer pollen grains per flower but more ovules per flower than wild populations (Delesalle et al., 2008). This resulted in P : O ratios that were lower in the glasshouse populations reported here (particularly among the Late buds) than those observed in our field populations or by Vasek & Weng (1988: mean P : O among populations: exilis 52.5; unguiculata 158.5). The low pollen production in the glasshouse is somewhat surprising given that plants were fertilized and well watered, and it suggests that the water-rich growing conditions provided here may have induced higher ovule production while suppressing pollen production.

Table 7.  Comparison of significant temporal changes in sex allocation and its components between field and glasshouse populations of Clarkia taxa with contrasting mating systems
TaxonMating systemField vs Glasshouse population (F or G)Ovules per budPollen per budΔP : O ratio
  1. Temporal changes refer to the direction of change from Early- to-Late-sampled buds. Glasshouse results from Delesalle et al. (2008). Boldface entries indicate where field and glasshouse populations exhibited qualitatively distinct temporal changes. Field and glasshouse populations differ in the direction or magnitude of the pollen to ovule (P : O) ratio only when the magnitudes of the temporal changes in pollen or ovule production were unequal; the directions of change for ovule and pollen production never differed qualitatively. The P : O ratio increased over time when the temporal decline in ovule production exceeded the decline in pollen production; it declined when the reduction in pollen production exceeded the reduction in ovule production.

unguiculataOutcrossingFDeclineStableIncrease
unguiculataOutcrossingGDeclineDeclineDecline
exilisFacultatively autogamousFDeclineDeclineIncrease
exilisFacultatively autogamousGDeclineDeclineDecline
xantianaOutcrossingFDeclineDeclineStable
xantianaOutcrossingGDeclineDeclineIncrease
parvifloraAutogamousFDeclineDeclinePopulation-dependent
(may decline or increase)
parvifloraAutogamousGDeclineDeclineStable

Our glasshouse populations of the xantiana and parviflora also differed from sampled field populations. Both field and glasshouse populations exhibited temporal reductions in ovule and pollen production. However, the field populations exhibited no consistent temporal change in the P : O ratio (Delesalle et al., 2008), while all three glasshouse populations of xantiana exhibited significant temporal increases. Our glasshouse populations also exhibited lower P : O ratios than the field populations sampled by Vasek & Weng (1988: mean P : O ratios: parviflora 36.1; xantiana 136.6).

Two outcomes of this experiment are striking and difficult to explain. First, the two outcrossing taxa, both of which are protandrous, exhibited qualitatively different patterns of temporal change in the P : O ratio in the glasshouse; the P : O ratio declined in unguiculata but increased in xantiana. Second, the ΔP : O values differed between glasshouse-grown maternal families and field-sampled individuals. In unguiculata, the P : O increased in the field but declined in the glasshouse; in xantiana, the P : O remained constant in the field but increased in the glasshouse (Table 7). Although theory predicts that selection will favor temporal increases in the P : O ratio in protandrous outcrossing species (Brunet & Charlesworth, 1995), this may not apply if flowers become so highly pollen-limited towards the end of an individual's flowering period that selection favors autogamous self-pollination and lower P : O ratios. We have initiated a study of pollen limitation in field populations of unguiculata and xantiana to determine whether this process may be operating differently in these taxa.

Could temporal variation in floral nutrient status explain the differences between sister taxa observed in the glasshouse?

Several observations support the inference that the differences between the glasshouse-grown sister taxa in the magnitude and direction of temporal change in the P : O ratio are genetically based rather than environmentally induced. Sister taxa were grown simultaneously and in a common environment, minimizing the abiotic environmental factors that could have affected them differently. Moreover, sister taxa were sampled at the same developmental stages. The first buds of all individuals were sampled when plants were robust, when interfloral competition for resources was relatively low, when the maximum number of mature and healthy leaves on the primary axis was available to provision the developing bud, and when there was no interfruit competition; the last buds were sampled just before the termination of flower production, at which time leaves had abscised or were senescing and all plants were garnering insufficient resources to produce additional flowers.

However, two differences between the sister taxa in this experiment could have differentially affected the resource status of the last buds sampled in the outcrossers relative to the selfers. First, the outcrossers produced more pollen per flower than their selfing sister taxa, particularly among the early buds. If the production of this additional, nitrogen-rich pollen caused faster nitrogen-depletion over time in the outcrossers than in the selfers, outcrossers might necessarily show larger declines in the P : O ratio, as observed in unguiculata relative to exilis. If, however, the production of extra flowers by unguiculata necessarily accounts for the greater declines in the P : O ratio than seen in exilis, then we should observe a positive correlation between the number of flowers produced between the Early and Late buds and the magnitude of the decline in the P : O ratio. This correlation was, however, nonsignificant among maternal family means in all four taxa (unguiculata, R2 = 0.0, P > 0.9741, n = 160; exilis, R2 = 0.001, P > 0.2788, n = 183; xantiana, R2 = 0.009, P > 0.1134, n = 177; parviflora, R2 = 0.0, P > 0.5812, n = 160). Moreover, we would expect to see the same kind of temporal change in xantiana as that detected in unguiculata. However xantiana exhibited a significantly temporal increase in pollen production.

A second difference between the sister taxa studied here is that the pollinator-dependent taxa were not hand-pollinated and therefore exhibited lower fruit set than their selfing counterparts (50% in unguiculata vs 95.5% in exilis; 20% in xantiana vs 98% in parviflora (measured on the primary axis)). Consequently, each (aborted) flower produced by the outcrossers was potentially a weaker nutrient sink than each flower (including its developing fruit) was in the selfers. Accordingly, unguiculata produced an average of three more flowers between the sampled buds on the primary stem than exilis (least squares mean (LSM) number of flowers between the Early and Late buds: unguiculata, 21.04 flowers (SE = 0.33); exilis, 18.10 flowers (SE = 0.31); one-way anova, F1,342 = 42.57; P < 0.0001). If pollen production by these additional flowers resulted in greater nitrogen depletion over time in the outcrossers, we would expect unguiculata to exhibit greater declines in the P : O ratio than exilis, which is consistent with our results. Nevertheless, the absence of a correlation between flower production on the primary stem and the magnitude of the decline in the P : O ratio (described above) argues against this explanation.

The differences between xantiana and parviflora in pollen production per flower and in flower production do not explain the difference between them in ΔP : O either. First, although pollen production per flower is higher in xantiana than in parviflora, predicting faster nitrogen depletion in the former, xantiana exhibited a temporal increase in the P : O ratio while the parviflora P : O remained constant. Second, in contrast to unguiculata, xantiana did not produce more flowers between the first and last buds than parviflora (xantiana, LSM = 7.90 flowers (SE = 0.18); parviflora, LSM = 7.63 flowers (SE = 0.19); F1,338 = 1.16; P > 0.28). This similarity in flower production reduces the potential that the late buds of xantiana were more nitrogen-depleted than those of parviflora, but it does not predict the temporal increase in the P : O ratio observed in xantiana. In summary, the two outcrossing taxa displayed opposite directions of change in the P : O ratio, which cannot be attributed simply to greater nitrogen depletion in the outcrossers owing to higher pollen and flower production.

Although we sampled all four taxa at developmentally equivalent points on the primary axis (i.e. when buds first appeared, and as the final buds were produced), nevertheless, it is possible that the outcrossers’ gametophyte production became limited over time by different resources than the selfers’ as a result of higher pollen production per flower in the former and higher fruit production in the latter. Similarly, given that unguiculata had higher fruit set and higher pollen production per flower than xantiana, the Late buds produced by these two outcrossers may have been limited by different resources. If so, pollen and ovule production in the Late buds may have been differentially affected in the selfers vs the outcrossers and in unguiculata vs xantiana. Accordingly, one may speculate that differences between our focal sister taxa and/or between the outcrossers in the rate at which gender-specific resources were depleted over time could have contributed to differences between them in the direction or magnitude of ΔP : O (e.g. Fig. 3d vs Fig. 4d). This is a matter of empirical investigation, which we hope this work will encourage.

The Brunet & Charlesworth (1995) models were not designed to predict how sex allocation per flower should change over time given variation in patterns of nutrient depletion. Future experiments to test the predictions evaluated here would benefit from the manipulation of fruit set and specific resources (e.g. nitrogen, water and carbon) in order to measure their effect on ΔP : O, and by the sampling of more buds along the primary axis. To equalize the fruit set of the selfers and outcrossers in an experiment of this size would have been an immense challenge. Preventing fruit set in the selfers would have required emasculating tens of thousands of flowers in the bud, while generating full fruit set in the outcrossers would have required pollinating tens of thousands of flowers. We recognize, however, that given the differences between unguiculata and xantiana in both the direction of temporal change in the P : O ratio and the level of fruit set, the interpretation of this experiment would be affected if the rate or type of nutrient depletion was known to differ among our focal taxa.

Temporal change towards more phenotypically female flowers

The significant temporal decline in the P : O ratio in unguiculata was unexpected under the mating opportunity hypothesis (Brunet & Charlesworth, 1995; Brunet, 1996), which predicts that in outcrossing protandrous species (e.g. unguiculata and xantiana), flowers produced late in the season are likely to be pollen-limited when in their female phase, resulting in selection favoring higher pollen production among late-blooming flowers in the male phase. Temporal change towards male-biased flowers is also predicted when seed production becomes increasingly resource-limited or architecturally constrained towards the end of the flowering season or at branch tips. As the ability of maternal plants to deliver resources to developing seeds declines over time, selection may favor genotypes that invest disproportionately in male function over female function (Lloyd, 1980; Lloyd et al., 1980; Thomson, 1989; Herrera, 1991; Jennersten, 1991; Diggle, 1995; Ashman & Hitchens, 2000). The temporal change towards more female-biased buds observed in exilis and unguiculata under the glasshouse conditions used here cannot be explained by either of these hypotheses. By contrast, the temporal increase in the P : O ratio observed in xantiana is consistent with these hypotheses.

Temporal changes toward female function are less common than temporal shifts towards male function (Delesalle et al., 2008) but have been documented in at least five species (Damgaard & Loeschcke, 1994; Ashman et al., 2001; Huang et al., 2004; Kliber & Eckert, 2004; Guitian, 2006). Among hermaphrodites of the gynodioecious Fragaria virginiana, small plants exhibit declines in the P : O ratio as they progress from primary to quaternary inflorescences while large plants exhibit the reverse. In Helleborus foetidus, stamen number declines while ovule number increases between first and last flowers in four-flowered inflorescences, consistent with the prediction for protogynous species (Brunet & Charlesworth, 1995). In Aquilegia yabeana, flowers become increasingly female-biased, a pattern that may be adaptive given higher pollen availability for late-blooming protogynous flowers (Huang et al., 2004). By contrast, in Aquilegia canadensis, high levels of herbivory on early flowers have been proposed to explain a similar increase in female allocation (Kliber & Eckert, 2004). Wild populations of the four Clarkia taxa observed here are highly vulnerable to fruit predation by sphingid moth larvae (Mazer & Delesalle, pers. obs.), and the role of this predation in natural selection on life history, phenology and sex allocation merits further investigation.

Explanations for variable P : O ratios among maternal families of exilis and parviflora

While the mean change in the P : O ratio and its absolute value was low in both selfing taxa, variation among maternal families was quite high (Figs 5, 6), particularly in parviflora. At least two factors may contribute to this variation. First, some maternal lineages among these facultative selfers may occasionally outcross, particularly in exilis, in which a mixed mating system may be common (Vasek & Harding, 1976; for a discussion of this possibility in parviflora see Eckhart & Geber, 1999). If so, these genotypes may evolve to exhibit changes in floral sex allocation in accordance with the predictions described above. Second, under conditions of occasional outcrossing, some lineages may accumulate a genetic load that causes developmental instability. Developmental instability has been induced by inbreeding in the perennial herb, Scabiosa canescens (Dipsacaceae) (Waldmann, 2001; but see Waldmann, 2002). A previous study of Clarkia tembloriensis, however, found that populations with different selfing rates exhibited similar levels of developmental stability, particularly among floral traits (Sherry & Lord, 1996), so mating system is not necessarily correlated with instability. To evaluate whether homozygosity generates within-individual or within-family variation in the ΔP : O requires experimental manipulation of selfing rates or inbreeding and then comparing lineages to determine the effects of homozygosity on ΔP : O ratios.

Conclusions

The modularity of flower production in hermaphroditic plants enables them to alter sex allocation over time in response to biotic or abiotic environmental variation (Lloyd et al., 1980; Charnov, 1982; Brunet & Charlesworth, 1995). Temporal variation among flowers in sex allocation is not necessarily adaptive, however, and few studies have aimed to isolate the selective, genetic, or ecological factors that generate within-individual temporal changes in sex expression (Medrano et al., 2000; Guitian et al., 2004; Huang et al., 2004). Previously, we proposed that the constancy of sex allocation among flowers within individual plants is a trait that may respond to natural selection (cf. Fenster & Galloway, 1997), and that autogamously selfing vs outcrossing taxa should evolve different degrees of constancy (Mazer & Delesalle, 1998). The results presented here (cf. Delesalle et al., 2008) provide qualified support for this prediction in Clarkia, while rejecting the contrasting hypothesis that selfers must generally exhibit higher developmental instability in the P : O ratio. Additional studies are needed in other genera and families to determine whether this is a general phenomenon that can be detected in both field and glasshouse conditions.

Acknowledgements

This work was supported by grants from the National Science Foundation (DEB 9816256 to V.A.D. and 9815300 to S.J.M.) and from Gettysburg College. We thank UCSB's Committee on Research, the Division of Mathematical, Life and Physical Sciences, and the Faculty Research Assistance Program for additional financial support. We thank Barron Rugge for glasshouse management and undergraduates Kendal Allman, Kerry Apostolo, Dawn Baron, Guy Carmelli, Andrea Censullo, Chia-En Chang, Cynthia Clark, Ashley Cooley, Elaine Delorimier, Huy Do, Anna I. Erickson, Naomi Fujita, Steven Hardee, Todd Lemein, Lindsey Hohmann, Natalie Hohmann, Benjamin Malcolm, Meron Meshesha, Jalil Mousallam, Regina Ng, Cathy Nguyen, Vy Nguyen, Andrea Ruhsam, Jessica Sanford, Robert Tacke, Benjamin Tan, Vivian Tran, and Jessie Wang for glasshouse and laboratory assistance. We also thank Alisa Hove and John Damuth, who provided insightful comments on earlier drafts of the manuscript.

Ancillary