SEARCH

SEARCH BY CITATION

Keywords:

  • Camissoniopsis cheiranthifolia ;
  • geographical variation;
  • inbreeding coefficient;
  • inbreeding depression;
  • mating system;
  • outcrossing;
  • self-fertilization;
  • self-incompatibility

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Theory predicts that inbreeding depression (ID) should decline via purging in self-fertilizing populations. Yet, intraspecific comparisons between selfing and outcrossing populations are few and provide only mixed support for this key evolutionary process. We estimated ID for large-flowered (LF), predominantly outcrossing vs. small-flowered (SF), predominantly selfing populations of the dune endemic Camissoniopsis cheiranthifolia by comparing selfed and crossed progeny in glasshouse environments differing in soil moisture, and by comparing allozyme-based estimates of the proportion of seeds selfed and inbreeding coefficient of mature plants. Based on lifetime measures of dry mass and flower production, ID was stronger in nine LF populations [mean δ = 1−(fitness of selfed seed/fitness of outcrossed seed) = 0.39] than 16 SF populations (mean δ = 0.03). However, predispersal ID during seed maturation was not stronger for LF populations, and ID was not more pronounced under simulated drought, a pervasive stress in sand dune habitat. Genetic estimates of δ were also higher for four LF (δ = 1.23) than five SF (δ = 0.66) populations; however, broad confidence intervals around these estimates overlapped. These results are consistent with purging, but selective interference among loci may be required to maintain strong ID in partially selfing LF populations, and trade-offs between selfed and outcrossed fitness are likely required to maintain outcrossing in SF populations.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Mating system evolution depends on a dynamic interplay between ecological and genetic factors. Whether self-fertilization is favoured over outcrossing will depend on aspects of pollination ecology that determine the prospects of gaining male and female fitness through self- vs. outcross-pollination (Lloyd, 1992), as well as the extent to which the expression of genetic load reduces the fitness of progeny produced through self- compared with outcross-fertilization (inbreeding depression, Charlesworth & Charlesworth, 1987). Self-fertilization is potentially advantageous because it increases the maternal plant's genetic representation in its offspring and because it may allow offspring production when pollinators and/or mates are scarce (reproductive assurance, Eckert et al., 2006). Given that these advantages can be substantial, a strong selective force is likely required to prevent the spread of selfing. Hence, inbreeding depression (hereafter ID), because it is ubiquitous and usually quite strong, is generally viewed as the major selective factor maintaining outcrossing. However, ID is not a static selective force because it is expected to covary with selfing (Lande & Schemske, 1985).

Inbreeding depression is primarily caused by the expression of at least partially recessive deleterious mutations in the highly homozygous progeny produced by selfing (Charlesworth & Charlesworth, 1987; Charl-esworth & Willis, 2009). As homozygosity increases through selfing recessive mutations should be exposed to selection and ‘purged’ from populations, thereby reducing the magnitude of ID. Early theoretical work modelling joint evolution of the mating system and ID predicted that purging should destabilize mixed mating systems thereby yielding two evolutionarily stable end points: predominant selfing associated with weak ID and predominant outcrossing associated with strong ID (Lande & Schemske, 1985; Charlesworth et al., 1990). However, individuals engage in substantial amounts of both selfing and outcrossing (mixed mating) in > 40% of plant species (Goodwillie et al., 2005). Moreover, evidence for purging derived from experimental studies comparing the performance of self-fertilized vs. outcrossed progeny is mixed. Husband & Schemske (1996) demonstrated the expected negative correlation between ID and the proportion of seeds self-fertilized (s) among plant species, although the relation was not strong (r2 = 0.18 among all 44 species; r2 = 0.14 among 35 angiosperms species). Byers & Waller (1999) reviewed 52 studies that compared ID between species, populations and/or lineages that differed in inbreeding history and concluded that purging is inconsistent and rarely efficient enough to reduce ID. A recent comprehensive meta-analysis of experimental results from 100 populations of 58 species (Winn et al., 2011) found only a weak and nonsignificant negative correlation between s and ID measured across the life cycle (r2 = 0.032, = 56 populations). Moreover, ID measured as δ = 1−(ωs/ωx), where ωs and ωx represent mean fitness of selfed and outcrossed progeny, respectively, was just as strong in populations that engaged in mixed mating (0.2 < < 0.8, mean δ = 0.58) as in predominantly outcrossing populations (< 0.2, mean δ = 0.54). Only populations that engaged in very high levels of selfing (> 0.8) exhibited relatively weak ID (mean δ = 0.26).

In addition to experimental comparisons of selfed vs. outcrossed progeny, ID can be inferred from the relation between s and the inbreeding coefficient (F) of reproductively mature plants estimated using genetic markers (Ritland, 1990). In the absence of ID (δ = 0), F at equilibrium should be a simple function of s [Fe = s/(2−s)]. However, data for a large sample of plant species show that, in general, F << Fe, indicating strong ID even for species that exhibit high s (Goodwillie et al., 2005; Scofield & Schultz, 2006). Here again, evidence for purging is weak, especially for large and/or long-lived species (Scofield & Schultz, 2006). There also appears to be a high threshold level of selfing (~ 0.80) below which ID is not effectively purged (Lande et al., 1994; Scofield & Schultz, 2006).

The strength of ID acting at different stages of the life cycle should coevolve with the mating system. Strongly, deleterious mutation can build up in outcrossing populations if they are highly recessive, and these mutations, when made homozygous by inbreeding, are expected to cause marked declines in fitness (lethal or nearly so) very early in the life cycle. Hence, they are readily purged from partially selfing populations. In contrast, mildly deleterious mutations can build up even if only partly recessive, may often cause inbreeding depression expressed later in the life cycle, and remain recalcitrant to purging in the face of chronic selfing (Lande & Schemske, 1985; Charlesworth & Charlesworth, 1987; Charlesworth et al., 1990; Lande et al., 1994). Several theoretical models have shown that ID being expressed more strongly ‘predispersal’ (during seed maturation) than ‘post-dispersal’ (from germination through reproduction) might set the stage for the evolutionary maintenance of mixed mating (Porcher & Lande, 2005a; Harder & Routley, 2006; Aizen & Harder, 2007; Harder et al., 2008). However, empirical support for this assumed contrast between pre- and post-dispersal ID is mixed. Husband & Schemske (1996) found that, as predicted, predispersal ID was strong for outcrossing populations and weak for selfing populations, whereas inbreeding depression expressed later in the life cycle was wither relatively weak (germination and juvenile survival) or did not differ between outcrossing and selfing populations (growth/reproduction). In contrast, a more recent meta-analysis of data from a larger number of species (Winn et al., 2011) found that the difference in predispersal ID between outcrossing and selfing taxa was the result of extremely strong ID for seed set among gymnosperms. Predispersal ID did not covary with the mating system among angiosperms.

Broad interspecific surveys have failed to consistently support some of the fundamental predictions concerning how ID should coevolve with the mating system. More direct tests of theory are provided by species that exhibit wide variation in s among populations (e.g. Busch, 2005). In these species, variation in the mating system is less likely to be confounded with differences in life history and long-term demography that might also affect the magnitude of ID independently of the mating system (Bataillon & Kirkpatrick, 2000; Scofield & Schultz, 2006). However, taking together the relatively small sample of species (= 9) for which both ID and s have been directly estimated for more than one population does not support purging (Winn et al., 2011). ID correlates negatively with s for some species (Clarkia tembloriensis – Holtsford & Ellstrand, 1990; Eichhornia paniculata – Toppings, 1989; Barrett & Husband, 1990; Leptosiphon jepsonii – Goodwillie, 2000; Goodwillie & Knight, 2006) but not others (Lupinus perennis – Shi, 2004; Michaels et al., 2008; Salvia pratensis – van Treuren et al., 1993; Ouborg & van Treuren, 1994). However, firm conclusions are not possible because relatively few studies have estimated these parameters for a large sample of populations that vary widely in the mating system. Even fewer studies have inferred ID from the relation between F and s among populations, and the scant results do not suggest reduced ID in chronically selfing populations (Eckert & Herlihy, 2004).

In this study, we quantify variation in the magnitude and timing of ID among populations of Camissoniopsis cheiranthifolia (Hornemann ex Sprengel) W.L. Wagner & Hoch (Onagraceae), a near-annual species that exhibits broad variation in the mating system across its geographical range along the Pacific coastal dunes of western North America (Raven, 1969; Samis & Eckert, 2007; Dart et al., 2012). Based on previous analyses of floral variation and outcrossing estimated using genetic markers, populations can be organized into three groups: (1) large-flowered, gametophytically self-incompatible populations that occur exclusively in San Diego County, California (LF-SI, mean corolla width = 29.6 mm, mean proportion of seeds self-fertilized (s) = 0.20, range = 0.01–0.38); (2) large-flowered, self-compatible populations that occur north of San Diego County to Point Conception in northern Santa Barbara County, California (LF-SC, mean corolla width = 27.7 mm, mean = 0.26, range = 0.04–0.53); and (3) small-flowered, self-compatible populations (SF-SC, mean corolla width = 15.6 mm, mean = 0.86, range = 0.43–0.99) that occur north of Point Conception to the northern range limit in Coos Bay Oregon, on the Channel Islands, and in Baja California. Based on results from common-garden experiments in the glasshouse and field observations over several years, there is no appreciable variation in life history or population dynamics associated with striking differentiation in the mating system (C.G. Eckert & S. Dart, unpublished).

We compare the fitness of selfed and outcrossed progeny produced by hand pollination for a large sample of populations to test the predictions that small-flowered (SF), predominantly selfing populations should exhibit weaker ID than large-flowered (LF), predominantly outcrossing populations. We also test the prediction that the difference in ID between LF and SF populations should be most pronounced early in the life cycle during seed maturation (predispersal) than later in life after seed dispersal because this is a key assumption in several theoretical models of mating system evolution.

In addition, we compare the performance of selfed vs. outcrossed progeny in two glasshouse environments that differ strongly in soil moisture. It is generally expected that ID will be expressed more strongly in more challenging environments, and the dependence of ID on the environment might significantly affect the outcome of mating system evolution (Lloyd, 1980; Cheptou & Donohue, 2011). However, this prediction has received only mixed support (Armbruster & Reed, 2005; Willi et al., 2007; Waller et al., 2008; Fox & Reed, 2011). As is typical of the Mediterranean biomes, there is strong seasonal variation in precipitation across the geographical range of C. cheiranthifolia, and the species exhibits classic adaptations to seasonal drought (e.g. deep tap roots, densely pubescent leaves). Rainfall occurs from November to March, followed by a prolonged dry season during which soil moisture in the dunes can rarely be detected with standard instruments and drought stress appears to cause significant mortality and reduced growth of C. cheiranthifolia (S. Dart & C.G. Eckert, personal observation). We tested whether simulated drought enhances the expression of ID because geographical variation in drought may influence the evolution of the mating system in this species. The duration and intensity of drought increases from north to south along the Pacific coast, such that a concomitant increase in ID could favour the maintenance of large flowers and predominant outcrossing in southern California while selfing may be more likely to evolve in populations occurring in moister habitat north of Point Conception and on the Channel Islands.

Finally, we complement our experimental analysis of ID by comparing s to the inbreeding coefficient (F) of mature plants within and among populations to estimate ID following Ritland (1990). Population-genetic estimates of ID reflect processes that occur across much of the life cycle under natural conditions, yet very few studies have compared population-genetic estimates with those obtained experimentally (Eckert & Barrett, 1994; Kohn & Biardi, 1995).

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Experimental analysis of inbreeding depression in contrasting glasshouse environments

We sampled open-pollinated maternal seed families from five LF-SI, 11 LF-SC and 23 SF-SC populations across the species' geographical range (Supporting Information Table S1). The main goal of this study was to contrast ID between population groups that differ in mating system. Hence, we maximized the number of populations included in the experiment at the cost of using a modest number of families from each population (~5/population). We sowed eight seeds from each of 28 LF-SI, 59 LF-SC and 130 SF-SC families in a common glasshouse environment and grew these plants to flowering (Supporting Information Methods S1). From September 2004 to March 2005, we performed paired self- and outcross-pollinations on 217 maternal plants from a total of 16 LF and 23 SF populations (mean = 5.4 plants/population). We collected fruits when mature and counted the filled seeds in each. Fruits were defined as containing ≥ 1 filled seed and fruit set was calculated as the proportion of pollinated flowers maturing a fruit.

Analysis of fruit set – Results indicated quite strong self-incompatibility among plants from the five LF-SI populations, and hence these populations were excluded from the analysis of ID for fruit set and seeds/fruit. Observations of pollen tube growth have not indicated post-pollination, prezygotic barriers to seed production in LF-SC populations (Y. Chang, A. Lopez & C.G. Eckert, unpublished). Hence, we used the data for paired self- and cross-pollinations where at least one flower set a fruit on 180 maternal plants from nine LF-SC and 23 SF-SC populations (Supporting Information Table S1). We contrasted relative fruit set after self-pollination (S) vs. outcross-pollination (X) between LF and SF populations using a generalized linear model (GLM) with fruit set as a binomial response variable and pollination treatment and population flower size and their interaction as categorical predictors. We tested the significance of each term using likelihood ratio tests contrasting models with and without the term in question. All analyses were performed using the statistical software R (version 2.13.0, R Development Core Team 2011).

Analysis of seeds per fruit – From the 217 plants pollinated, we chose the maternal plants where both self- and cross-pollination yielded a fruit for a total of 110 maternal plants from eight LF-SC and 17 SF-SC populations (Supporting Information Table S1). We contrasted seed number in S vs. X fruits between LF and SF populations using a GLM with pollination treatment and population flower size and their interaction as categorical predictors. Modelling the data with Poisson errors resulted in over-dispersed residual deviance. Hence, we modelled a negative binomial error distribution using the glm.nb command in the MASS R package (following Zuur et al., 2009, pp. 234–235). Significance was determined using likelihood ratio tests as above.

Analysis of progeny performance in glasshouse environments with contrasting soil moisture – In September 2006, we individually sowed about eight randomly selected seeds for each pollination treatment (S and X) from each of 107 maternal families from 10 LF and 15 SF populations (Supporting Information Table S1; 1647 seeds sown, growth conditions in Supporting Information Methods S1). We randomly assigned half of the seeds from each family to each of two soil moisture environments (hereafter ‘moist’ and ‘dry’). To simulate seasonal variation in soil moisture, we imposed the soil moisture treatment 65 days after sowing which roughly corresponds to the onset of summer drought in Pacific coastal dune habitat. Plants in the moist environment experienced soil moisture previously determined to be optimal for growth and flowering of C. cheiranthifolia in the glasshouse (Supporting Information Methods S1). Plants in the dry environment were watered only half as often. This simulated drought treatment dramatically reduced plant growth and flowering (see below).

After sowing, pots were checked daily to record seedling emergence. The survival of emerged seedlings was scored 35 and 65 days after sowing. Plants of C. cheiranthifolia typically start to senesce after four months of growth in a glasshouse environment, which was the case in this experiment. We harvested plants 125 days after sowing, scored whether each had survived to harvest, whether each had flowered, if so, how many flowers each produced, and quantified aboveground biomass (to 0.001 g) after drying at 60°C for 2 weeks. We measured belowground dry mass by washing the roots free of soil and drying them as above. However, roots could not be extricated from the potting medium for all surviving plants, thus we analyse aboveground mass only. The two components of total dry mass correlated strongly among the plants for which we obtained both measures (r = +0.86, < 0.0001, = 1204) suggesting that aboveground mass captures much of the variation in total biomass. Aboveground plant size is a consistent and reliable predictor of estimated per plant fruit production in natural populations of this short-lived plant (C.G. Eckert & S. Dart, unpublished based on 1645 plants sampled from 68 populations).

We calculated two fitness measures: aboveground biomass produced per seed sown (hereafter ‘dry mass’), and total flowers produced per seed sown (hereafter ‘flower number’). As a substantial proportion of seeds did not emerge as seedlings, did not survive to harvest or did not flower, the data were zero-inflated and strongly over-dispersed when analysed using a GLM with Poisson or negative binomial errors. Hence, we analysed these data using the ASTER model in R (Shaw et al., 2008), which uses a maximum likelihood algorithm accounting for final fitness resulting from the compounding of Bernoulli variables (emergence, survival, flowering) with variables that roughly exhibit zero-truncated Poisson distributions [flower number and dry mass (converted to integer units of 10 mg)]. ASTER analyses multiplicative fitness, from sowing to senescence, and makes use of data from all the 1647 individuals in the experiment in a single analysis. We tested the significance of pollination treatment, soil moisture treatment, population flower size and their interactions using likelihood ratio tests as above. We also analysed survival from sowing up to soil moisture treatment application (day 65) using ASTER with pollination treatment, population flower size and their interaction as predictors. Population was not included in this analysis or any of the analyses described above because ASTER cannot accommodate random effects, and because each population was represented by relatively few families to maximize the number of replicate populations per flower size category. The mean and standard error are not accurate measures of central tendency and reliability for skewed data; hence, we used ASTER to estimate their unconditional expectations and 95% confidence intervals for each fitness measure following Geyer et al. (2007). Cumulative post-dispersal ID, from sowing to harvest, was calculated from each of the two final fitness measures as δ = 1−(ωs/ωx), where ωs and ωx are the expectations for selfed and outcrossed progeny, respectively. Predispersal fitness components (fruit set and seeds/fruit) were not included in the calculation of cumulative ID because these variables did not differ between outcross- and self-pollination (see below) and because a somewhat different set of populations and families were used to quantify post-dispersal progeny performance (see above). R scripts for all analyses are available by request from the corresponding author.

Population-genetic analysis of inbreeding depression

We jointly estimated the proportion of seeds self-fertilized (s) and the inbreeding coefficient of reproductively mature plants (F) using the mixed mating model as implemented by the maximum likelihood (ML) program MLTR (Ritland, 2002) for a subset of 16 populations from across the geographical range that represented the range of variation in floral morphology (same populations studied by Dart et al., 2012; Supporting Information Table S1). We assayed 10 newly germinated seedlings for each of ~30 seed families per population for variation at seven polymorphic allozyme loci following Dart et al. (2012). Pollen and ovule allele frequencies were constrained to be equal, parental genotypes were inferred from progeny genotypes, Newton-Raphson iteration was used to find ML estimates and 95% confidence limits for ML estimates were derived as the 2.5- and 97.5-percentile of the distribution of 1000 bootstrap values generated with the seed family as the unit of re-sampling. We only analysed s and F estimates for which the bootstrap distribution estimates did not exhibit substantial irregularity.

ID was calculated from the ML estimates of s and F using Ritland's (1990) equilibrium estimator: δ = 1−[2F(1−s) / s(1−F)]. Like experimental measures of δ, this estimator can be < 0 if selfed progeny exhibit higher survival than outcrossed progeny, but unlike experimental estimates it can be > 1 when F < 0 (i.e. more heterozygotes than expected at Hardy–Weinberg equilibrium). This method is most appropriate for use on populations that practice a mixture of both selfing and outcrossing (i.e. 0.1 < < 0.9; Ritland, 1990), hence only populations for which ML estimates of s were within this range and F estimates seemed reliable were included (see above). We contrasted ML estimates of s, F and δ between LF and SF populations by averaging randomly paired bootstrap estimates across populations within groups, and calculating the proportional overlap between averaged bootstrap distributions between LF and SF groups, which is roughly equivalent to a P-value. The mean for one group was considered significantly different from another group if P was < 5% because all comparisons involved 1-tailed tests.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Predispersal inbreeding depression for fruit set and seeds per fruit

Among the 360 flowers pollinated on 180 maternal plants, fruit set was 87.2% and did not differ between self- vs. cross-pollination or between plants from LF vs. SF populations (Fig. 1, Table 1). The number of seeds per fruit ranged 1–116 and averaged (± 1 SE) 47.5 ± 1.7 among the 110 maternal plants that produced both selfed and outcrossed fruits, but did not differ between pollination treatments or between LF vs. SF populations (Fig. 1, Table 1).

Table 1. Analysis of variation in inbreeding depression expressed for fruit set and seeds per fruit between predominantly outcrossing large-flowered and predominantly selfing small-flowered populations of Camissoniopsis cheiranthifolia. Values are likelihood ratios comparing models with and without the term in question (d.f. = 1) with P-values in parentheses. No model including any of these predictors accounted for a significant component of variance for either response variable. The likelihood ratio comparing the full model with a model including only an intercept (d.f. = 3) was 1.69 (= 0.64) for fruit set and 0.77 (= 0.83) for seeds per fruit. The data on which these analyses are based are summarized in Fig. 1
Response variableError distributionFlower size (LF vs. SF)Treatment(S vs. X)Interaction
Fruit setBinomial0.20 (0.65)0.40 (0.53)1.86 (0.17)
Seeds per fruitNegative binomial0.62 (0.70)0.17 (0.70)0.01 (0.93)
image

Figure 1. Comparison of fruit set and seeds per fruit between experimental self- and cross-pollinations in large-flowered vs. small-flowered populations of Camissoniopsis cheiranthifolia. Points are expected values from generalized linear models and bars are 95% confidence intervals. Fruit set was estimated from paired self- and cross-pollinations on 180 maternal plants representing nine large- and 23 small-flowered populations. Seeds per fruit were estimated from paired selfed and outcrossed fruits on 110 maternal plants from eight large- and 17 small-flowered populations. Analysis of these data is in Table 1.

Download figure to PowerPoint

Post-dispersal inbreeding depression before soil moisture treatment

Of 1647 seeds sown, 51.3% emerged as seedlings of which 65.7% survived to day 35 and 62.6% survived to day 65 (32.1% of seed sown). Analysis using the ASTER model (Fig. 2) revealed higher cumulative survival to day 65 among seeds from LF populations (42.5%, 95% confidence interval = 38.7–46.2%) compared to SF populations (25.5%, 22.9–28.2%; likelihood ratio = 56.28, d.f. = 1, < 0.00001) and slightly higher survival of outcrossed seed (34.3%, 31.2–37.5% compared to selfed seed (29.9%, 26.9–33.0%; likelihood ratio = 4.0, d.f. = 1, = 0.046). Although our data suggest somewhat higher ID in LF than SF populations (Fig. 2), the interaction term was not significant (likelihood ratio = 2.5, d.f. = 1, = 0.11). Inbreeding depression based on the survival to day 65 estimated using ASTER was 0.128.

image

Figure 2. Comparison of survival from sowing to the application of the soil moisture treatment (day 65) of selfed and outcrossed seed from large- vs. small-flowered populations of Camissoniopsis cheiranthifolia. Points are expected values and bars are 95% confidence intervals estimated from ASTER models. The number of selfed and crossed seeds sown was 322 and 321, respectively, for large-flowered populations and 514 and 495 for small-flowered populations. See text for analysis of these data.

Download figure to PowerPoint

Cumulative post-dispersal inbreeding depression across soil moisture treatments

Of the 1647 seeds sown, 31.6% survived to harvest and aboveground dry mass ranged 0.11–7.83 g and averaged 3.15 ± 0.06 g among the survivors. ASTER analysis revealed that cumulative aboveground dry mass (in 10 mg units) per seed sown was higher under moist conditions (estimated value, 95% confidence interval = 128.81, 118.20–139.41) than dry conditions (70.05, 61.08–79.02), higher for LF populations (136.59, 124.44–148.74) than SF populations (75.95, 67.62–84.29) and higher for outcrossed progeny (111.90, 101.49–122.31) than selfed progeny (87.50, 77.92–97.09; Fig. 3, Table 2). There was also a significant interaction between the effects of population flower size and pollination treatment indicating stronger cumulative ID in LF than SF populations. ID estimates based on expected values of dry mass calculated using ASTER were 0.300 for LF and 0.087 for SF populations. The interaction between soil moisture treatment and pollination treatment was not significant (Table 2).

Table 2. Analysis of variation in inbreeding depression expressed for two measures of lifetime fitness in two soil moisture environments for large-flowered vs. small-flowered populations of Camissoniopsis cheiranthifolia. Likelihood ratio test results (d.f. = 1) are presented for the effects of population flower size (large- vs. small-flowered), pollination treatment (self vs. cross) and soil moisture treatment (moist vs. dry) plus the interactions between these main effects. The data on which these analyses are based are summarized in Fig. 3
Fitness measure/term in modelLikelihood ratio P
Aboveground dry mass per seed sown
Population flower size (F)68.51<0.00001
Pollination treatment (P)11.470.00071
Soil moisture treatment (S)69.27<0.00001
F × P6.170.013
F × S0.160.69
P × S0.670.41
F × P × S0.370.54
Flowers produced per seed sown
Population flower size (F)23.03<0.00001
Pollination treatment (P)0.420.52
Soil moisture treatment (S)56.23<0.00001
F × P5.870.015
 F × S0.180.67
P × S1.300.25
F × P × S0.280.60
image

Figure 3. Comparison of two measures of lifetime fitness of selfed vs. crossed seed grown in contrasting soil moisture environments for large-flowered vs. small-flowered populations of Camissoniopsis cheiranthifolia. Points are expected values and bars are 95% confidence intervals estimated from ASTER models. Number of seed sown is indicated in Fig. 2. The number of selfed and crossed progeny surviving to harvest was 108 and 143, respectively, for large-flowered and 134 and 135 for small-flowered populations. The number of selfed and outcrossed progeny flowering was 34 and 58 for large-flowered and 72 and 65 for small-flowered populations. Analysis of these data is in Table 2.

Download figure to PowerPoint

Of the 520 plants that survived to harvest, 44% flowered and the number of flowers produced ranged 1–124 and averaged 20.7 ± 1.6. ASTER analysis revealed that total flowers produced per seed sown were higher for SF plants (3.52, 3.03–4.02) than LF plants (1.85, 1.39–2.31) and higher under moist conditions (4.17, 3.59–4.76) than dry conditions (1.57, 1.20–1.94; Fig. 3, Table 2). Although there was no overall effect of pollination treatment, the interaction between pollination treatment and population flower size was, again, significant with LF populations expressing more ID than SF populations (Table 2). Estimates of ID based on expected values of flower number were 0.467 for LF and –0.030 for SF populations. Again, the interaction between soil moisture treatment and pollination treatment was not significant (Table 2).

Population-genetic analysis

The proportion of seeds self-fertilized (s) ranged 0.01–1.00 among the 16 populations assayed, and was significantly higher for SF than LF populations (Table 3). The inbreeding coefficient of parental plants (F) ranged –0.13 to 0.51 and correlated positively although not quite significantly with s (Pearson r = 0.424, = 12, 1-tailed = 0.065; Fig. 4). F was, on average, twice as high for SF than LF populations, although the comparison was only marginally significant (Table 3). There was considerable variation in F within population flower size groups, and most maximum likelihood estimates had large 95% confidence intervals.

Table 3. Maximum likelihood estimates of the proportion of seeds self-fertilized (s), the inbreeding coefficient of mature plants (F) and inbreeding depression (δ) for 16 populations of Camissoniopsis cheiranthifolia. Population sampled include seven large-flowered (LF) [four self-compatible (SC) and three self-incompatible (SI)], seven small-flowered (SF) and two phenotypically variable (Var). Ninety-five percent confidence intervals for each estimate are in parentheses. Estimates of F for CBV1C and CMK1C are not presented (na) because bootstrap distributions were highly irregular. δ was only calculated for populations with reliable estimates of F, moderate estimates of s (0.1–0.9) and reasonable bootstrap distributions. Means for LF and SF populations and a comparison between them (P-value) appear in the bottom three rows of the table
PopulationType s F δ
CTP1CLF-SI0.01 (–0.03 to 0.03)0.28 (0.13 to 0.43)na
CCB1CLF-SI0.16 (0.00 to 0.33)–;]?>0.13 (–0.20 to 0.05)2.16 (–3.35 to 7.55)
CSO1CLF-SI0.38 (0.19 to 0.52)–;]?>0.01 (–0.20 to 0.19)1.04 (0.13 to 1.60)
CDW2CLF-SC0.26 (0.05 to 0.40)–;]?>0.10 (–0.20 to 0.17)1.51 (–3.07 to 2.06)
COR1CLF-SC0.53 (0.38 to 0.63)0.31 (0.16 to 0.48)0.21 (–1.47 to 0.65)
CMG1CLF-SC0.11 (0.02 to 0.18)0.37 (0.20 to 0.66)na
CBV1CLF-SC0.53 (0.37 to 0.68)nana
CGN1CVVar-SC0.67 (0.56 to 0.76)0.36 (0.25 to 0.48)0.45 (–0.16 to 0.73)
CSP1CVVar-SC0.58 (0.43 to 0.72)0.47 (0.27 to 0.77)–;]?>0.23 (–3.76 to 0.54)
BES1CSF-SC0.84 (0.72 to 0.96)0.33 (0.12 to 0.51)0.81 (0.50 to 0.97)
CSC1CSF-SC0.85 (0.66 to 0.95)0.25 (0.02 to 0.49)0.88 (0.44 to 0.99)
CMS1CSF-SC0.37 (0.15 to 0.59)0.20 (–0.04 to 0.44)0.12 (–4.94 to 1.07)
CST1CSF-SC0.68 (0.57 to 0.79)0.34 (0.01 to 0.58)0.52 (–0.47 to 1.00)
CMC1CSF-SC0.87 (0.77 to 1.00)0.09 (–0.2 to 0.23)0.97 (0.89 to 1.01)
CMK1CSF-SC0.43 (0.23 to 0.60)nana
ONO1CSF-SC1.00 (0.81 to 0.99)0.51 (0.20 to 0.80)na
MeanLF0.28 (0.22 to 0.32)0.12 (0.06 to 0.21)1.23 (–2.75 to 2.59)
MeanSF0.72 (0.66 to 0.77)0.23 (0.15 to 0.32)0.66 (–0.47 to 0.90)
LF vs. SFP-value< 0.0010.0500.158
image

Figure 4. Comparison of maximum likelihood estimates of the inbreeding coefficient of parental plants (F) and the proportion of seeds self-fertilized (s) for 14 populations of Camissoniopsis cheiranthifolia. Symbols indicate population flower size and self-compatibility as follows: LF-SI = large-flowered and self-incompatible, LF-SC = LF and self-compatible, SF-SC = small-flowered and SC, Variable = phenotypically variable. 95% confidence intervals for these estimates are in Table 3. The solid line indicates the expected parental F at equilibrium with no inbreeding depression (δ = 0); and the dashed line indicates the expected F with δ = 0.5.

Download figure to PowerPoint

Of the 14 populations for which F could be reliably estimated (see 'Materials and methods'), ML estimates were lower than that expected with no ID for 11 populations and lower than expected if the relative fitness of selfed progeny (ωs) was half that of outcrossed progeny (ωx) for 8 populations (Fig. 4). As a result, ID (δ = 1−ωs/ωx) calculated using Ritland's (1990) equilibrium estimator was typically strong, although individual population estimates were associated with very wide 95% confidence intervals, and the mean across all populations (δ = 0.78) had 95% confidence intervals that overlapped both zero and one (95% CI = –0.92 to 1.19). Hence, the almost 2-fold difference in maximum likelihood estimates of δ for LF compared to SF populations was not significant (Table 3).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The ‘purging’ of segregating genetic load that causes inbreeding depression (ID) is a fundamental component of most genetically realistic theoretical models investigating mating system evolution (Goodwillie et al., 2005; Eckert et al., 2006). Although the most recent interspecific analyses do not support the prediction that ID consistently declines with increasing s, the proportion of seeds self-fertilized (Winn et al., 2011), species like C. cheiranthifolia that exhibit wide variation in s among populations provide a more direct test of this prediction. Our results from an experimental comparison of selfed and outcrossed offspring in two glasshouse environments as well as a population-genetic analysis support the prediction of higher ID in LF predominantly outcrossing populations than SF predominantly selfing populations. These results are consistent with the purging of ID associated with a shift from outcrossing to selfing. Below, we compare our results from experimental vs. population-genetic analyses, discuss what our results say about the timing of ID in outcrossing vs. selfing populations and the effect of environmental stress on the expression of ID, challenge the extent to which the difference in ID between LF and SF populations is a result of inbreeding history alone and consider the implications for the maintenance of mating system variation in this species.

Experimental vs. population-genetic estimates of inbreeding depression

Our understanding of the expression of ID in plant populations is based largely on comparisons of performance between selfed vs. outcrossed progeny produced by hand pollination (Husband & Schemske, 1996; Winn et al., 2011). However, inferences from these experiments are usually limited because experimental material is derived from a small sample of populations and fitness is assayed under artificial conditions that may underestimate ID. This is a problem for researchers investigating mating system evolution because they are typically interested in an accurate estimate of ID (whether δ > 0.5 or δ < 0.5, see below) rather than just whether ID is expressed (δ > 0; see Eckert & Herlihy, 2004). Population-genetic estimates derived from s and F can complement experimental studies because they should reflect differential mortality operating across much of the life cycle under natural conditions (Ritland, 1990). The use of both experimental and population genetic approaches in this study provides somewhat independent lines of evidence for the predicted negative correlation between ID and s among populations of C. cheirathifolia. Estimates of ID for LF and SF populations averaged 0.39 and 0.03, respectively, in the glasshouse and 1.23 and 0.66, respectively, based on population-genetic analysis. Our results, along with other studies that have used both approaches, suggest that ID is underestimated under artificial conditions (e.g. Eckert & Barrett, 1994) perhaps because field environments present plants with a greater diversity of challenges compared with relatively simple glasshouse experiments.

There are three drawbacks of marker-based methods. First, they do not incorporate early acting ID when offspring genotypes are assayed at the seedling stage. This has probably not biased estimates in our study because experimental analysis indicates very weak ID for fruit set and seeds per fruit and supplementary analysis (not shown) did not detect an expression of ID during germination and seedling establishment. Second, marker-based estimates of ID often have large confidence intervals (Ritland, 1990), as is the case here. Confidence intervals for LF and SF populations overlapped each other, estimates from the glasshouse and zero. This may have resulted from substantial variation in estimates of s and F among populations within LF and SF groups rather than sampling effort, which was high in this study (16 populations, 552 progeny arrays, 4410 progeny, 7 polymorphic loci). Third, population-genetic estimates are based on assumptions (Ritland, 1990; Charlesworth, 1991), especially that populations are at inbreeding equilibrium with little fluctuation in s between years within populations. The marked variation in s among populations with similar floral morphology as well as observations of variation in pollinator visitation within and among closely situated sites (Linsley et al., 1973) suggests that this assumption may not be met in C. cheiranthifolia. Violation of inbreeding equilibrium will not systematically bias ID estimates unless there is a rapid and concerted shift in s across populations. However, it can contribute to the statistical uncertainty of the estimates by causing random mismatches between s and F within populations. Although our genetic estimates of ID concur with experimental estimates in suggesting stronger ID in LF than SF populations, we treat inferences of the absolute strength of ID based on genetic markers with caution.

Variation in the expression of inbreeding depression across the life cycle and in different environments

Theory predicts differential purging of deleterious alleles that affect fitness early vs. late in the life cycle (Charlesworth & Charlesworth, 1987), and the specific contrast between pre- and post-dispersal ID has become a key assumption of some theoretical models that seek to explain the widespread occurrence of mixed mating (Porcher & Lande, 2005a; Harder & Routley, 2006; Aizen & Harder, 2007; Harder et al., 2008). Stronger pre- than post-dispersal ID in predominantly outcrossing populations allows partially selfing individuals to potentially compensate for the disproportionate death of selfed embryos by replacing them with more vigorous but less ‘genetically valuable’ outcrossed progeny. Hence, mixed mating can maximize both the number of seeds produced as well as maternal genetic representation in surviving seed (Harder & Routley, 2006). However, our results do not suggest stronger predispersal than post-dispersal ID in LF populations or that LF and SF populations differ more for pre- than post-dispersal ID. Although we detected much stronger ID in LF than SF populations after seed maturation, we did not detect ID for fruit set or seeds per fruit for either LF or SF populations. As such, our results agree with recent meta-analysis showing that particularly strong predispersal ID, although common in outcrossing gymnosperms, does not distinguish outcrossing from selfing populations among angiosperms (Winn et al., 2011). We also did not detect a difference between LF and SF populations for ID expressed after seed dispersal but early in the life cycle. Significant however weak ID was detected for cumulative survival to day 65, which includes germination and juvenile survival, but there was no difference between LF and SF populations. Supplementary analysis of germination alone using generalized linear models with a binomial error distribution (not shown) did not detect significant ID for germination (= 0.46) or any difference between LF and SF populations (= 0.28). A similar analysis of cumulative survival to day 35 detected weak ID (δ = 0.157, = 0.014) but, again, no difference between LF and SF populations (= 0.12). Hence, the expression of ID differs most between LF and SF populations for later life stages.

Our results also failed to support the widely held expectation of stronger ID in more stressful environments (Roff, 1997; Cheptou & Donohue, 2011). We compared the expression of ID between soil moisture treatments because soil moisture is thought to strongly limit plant growth in dune habitats (Maun, 1994), and the duration of Mediterranean summer drought varies strongly across the geographical range of C. cheiranthifolia in a way that might have influenced mating system evolution (see 'Introduction'). As drought in arid habitats like costal dunes may involve several environmental factors (soil moisture, wind, insolation), it is near impossible to replicate under glasshouse conditions. Although our simulated drought was simplistic in this regard, restricting soil moisture strongly reduced the growth and flowering in our glasshouse experiment but did not enhance ID (Fig. 3).

Although some meta-analyses suggest that effects of stress on ID are inconsistent (Armbruster & Reed, 2005; Willi et al., 2007), recent analysis by Fox & Reed (2011) suggests that the likelihood of detecting increased ID depends on the severity of experimental stress imposed. Enhanced ID is generally not detected until the imposed stress is sufficient to decrease fitness by 25% among outcrossed individuals. In our experiment, restricted soil moisture reduced dry mass per outcrossed seed sown from 2050 to 1170 mg for LF populations (a 43% reduction) and 1060 to 530 mg for SF populations (50% reduction), and reduced the number of flowers produced per seed sown from 3.35 to 1.44 for LF (57% reduction) and 4.63 to 2.05 for SF populations (56% reduction). The stress generated by our soil moisture manipulation was well within the range expected to enhance ID in most species studied to date.

Waller et al. (2008) suggested that stress may sometimes fail to enhance the expression of ID if it reduces overall phenotypic variability and hence opportunity for selection against inbred individuals. For instance, at the limit a very severe stress may kill or severely impair all individuals, greatly restricting opportunities for any fitness differential between selfed and outcrossed progeny (Armbruster & Reed, 2005). Waller et al. (2008) provide experimental evidence from Brassica rapa that the strength of inbreeding depression increases with the amount of phenotypic variation expressed in an environment. Perhaps, our simulated drought treatment did not enhance ID because it severely limited phenotypic variability. However, phenotypic variability for both our measures of fitness, calculated as the trait variance divided by the square of the trait mean (following Waller et al., 2008), was higher under dry conditions (dry mass = 3.29; flowers = 21.37) than moist conditions (dry mass = 2.19, flowers = 12.31). Levene's test of relative variation (centred on either the mean or median) indicated significantly higher phenotypic variation under simulated drought for both variables (all < 0.00001). There is no obvious explanation for why drought did not lead to more pronounced ID in C. cheiranthifolia.

Maintenance of inbreeding depression despite chronic selfing?

LF and SF populations of C. cheiranthifolia differ markedly in syndrome of floral traits that typify mating system differentiation (corolla width, herkogamy, floral longevity, floral display size, capacity for spontaneous self-fertilization, Dart et al., 2012), and this is reflected in average genetic estimates of s being much higher for SF populations (mean = 0.72) than LF populations (mean = 0.28). However, there is much variation in s within both population groups. Some LF populations exhibit substantial selfing (> 0.50, populations COR1C and CBV1C), and significant selfing was even detected in two of three self-incompatible populations. This suggests that all populations of C. cheiranthifolia might experience regular opportunities for purging. Although this may explain the absence of strong ID for fruit set and seeds per fruit, especially in LF populations (see discussion above), it also raises the possibility that the difference in ID between LF and SF populations is not simply accounted for by differences in inbreeding history.

Theory suggests that strong ID might be maintained in partially selfing LF populations by selective interference (Lande et al., 1994). If deleterious alleles at different loci have synergistic effects across loci, that is they decrease fitness to a greater extent than if they affected fitness independently (as opposed to multiplicative interactions, see Charlesworth et al., 1991), ID may be so strong that selfed progeny are never recruited thereby preventing purging until some high threshold level of s occurs (Lande et al., 1994). Meta-analyses of ID based on both experimental (Winn et al., 2011) and population genetic data (Scofield & Schultz, 2006) support the existence of a threshold s below which strong ID is maintained. However, the comparative evidence for selective interference is strongest for large and/or long-lived plant species (Scofield & Schultz, 2006) and quite weak for small, short-lived species like C. cheiranthifolia. Moreover, selective interference requires near complete death of selfed progeny before reproduction, yet estimates of the inbreeding coefficient of mature plants (F) were above zero for three of six LF populations of C. cheiranthifolia indicating recruitment of inbred individuals into the reproductive population.

It is possible that the derivation of SF populations of C. cheiranthifolia north of Point Conception from LF populations to the south involved a single evolutionary transition from predominant outcrossing to predominant selfing during northward coastal colonization (Raven, 1969). Hence, historical or contemporary differences in population demography between LF and SF populations may have contributed to the reduction in ID along with or independent of a change in the mating system (Busch, 2005; Pujol et al., 2009). However, theory has provided mixed predictions about how population demography might influence the amount of segregating genetic load that causes ID. Bottlenecks that may commonly occur during colonization can reduce the frequency of strongly deleterious alleles, but the effects on inbreeding depression will typically be weak and transitory (Kirkpatrick & Jarne, 2000). More prolonged periods of small population size may reduce ID; however, it remains unclear whether the reduction will be greater for strongly recessive alleles that tend to be expressed early in life (Glémin, 2003) or the later-acting component of ID due to mildly deleterious alleles (Bataillon & Kirkpatrick, 2000), the predominant component of ID in C. cheiranthifolia. Moreover, episodes of small population size may reduce the purging effect of selfing, which is typically stronger than the effect of population size (Glémin, 2003). In general, the few existing empirical studies have failed to demonstrate an effect of contemporary population size on ID (Kennedy & Elle, 2008; Coutellec & Caquet, 2011; but see Paland & Schmid, 2003). Surveys of plant density in contemporary populations across the geographical range of C. cheiranthifolia have not revealed any consistent difference in size or density between SF and LF populations (Samis & Eckert, 2007). Only SF populations from Baja California exhibited lower density than LF populations, and these SF populations were not included in our glasshouse analysis of ID expressed beyond seed set (Supporting Information Table S1). Population-genetic analysis of neutral polymorphisms is required to provide insight into the long-term demography of populations across the mating system spectrum in this species (Foxe et al., 2009).

The role of inbreeding depression in maintaining mating system variation

All else being equal, δ ≥ 0.5 can prevent the spread of alleles that increase selfing, hence estimates of δ based on lifetime performance in glasshouse environments approach the threshold required to maintain outcrossing in LF populations of C. cheiranthifolia (δ = 0.467 for flower production and 0.300 for aboveground dry mass). Population-genetic estimates for some of these same populations were higher (mean δ > 1.0), albeit with broad confidence intervals. A variety of theoretical models predict that predominant outcrossing may be maintained in LF populations even if δ is somewhat less than 0.5 (reviewed in Goodwillie et al., 2005; Schoen & Busch, 2008). However, the maintenance of mixed mating in SF populations of C. cheiranthifolia is more enigmatic. These populations often exhibit significant and sometimes high levels of outcrossing, yet we did not detect significant ID in the glasshouse.

One explanation for the occurrence of outcrossing despite negligible ID is that these SF populations are in the process of evolving higher levels of selfing. Although we cannot reject this hypothesis, there is no obvious reason to suspect that these populations are not at equilibrium with respect to mating system evolution. If they are at equilibrium, what prevents the spread of alleles that increase s further by reducing flower size and herkogamy (both of which affect the levels of selfing and outcrossing, Dart et al., 2012)? Other closely related species of Camissoniopsis have progressed further down the evolutionary road to complete self-fertilization (Dart et al., 2012). In fact, we found one SF population of C. cheiranthifolia in Mendocino County California with high frequencies of plants that produce only cleistogamous, obligately selfing flowers (D. Denley, S. Dart, L. Doubleday & C.G. Eckert, unpublished). Although the maintenance of substantial outcrossing without significant ID seems to occur in other taxa (e.g. Collinsia verna, Kalisz et al., 2004) few theoretical models explain this. Those that do involve trade-offs between s and outcross fitness through female and/or male function (Porcher & Lande, 2005b; Johnston et al., 2009). Evaluating these trade-offs in natural populations poses a major challenge en route to a better understanding of mating system evolution.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The authors thank D. Kristensen, J. Plett, B. Dart and R. Dart for help in the glasshouse; K. Samis and E. Austen for help with collecting the seed families from the field; Thomas Flatt for comments on the manuscript; the U.S. National Park Service, California State Parks and Oregon State Parks for research permits; and the Natural Sciences and Engineering Council of Canada for a Discovery Grant to CGE.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
jeb12075-sup-0001-TableS1-MethodS1.docWord document108K

Table S1 Populations of Camissoniopsis cheiranthifolia sampled for this study, and the type of data collected for plants from each.

Methods S1 Glasshouse conditions and methods for experimental crosses.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.