Annals of Botany 115: 315–326, 2015 doi:10.1093/aob/mcu243, available online at www.aob.oxfordjournals.org Variation in pollen limitation and floral parasitism across a mating system transition in a Pacific coastal dune plant: evolutionary causes or ecological consequences? Sara Dart and Christopher G. Eckert* Department of Biology, Queen’s University, Kingston, Ontario, K7L 3N6 Canada * For correspondence. E-mail [email protected] Received: 25 June 2014 Returned for revision: 25 July 2014 Accepted: 24 October 2014 Published electronically: 22 December 2014 Background and Aims Evolutionary transitions from outcrossing to self-fertilization are thought to occur because selfing provides reproductive assurance when pollinators or mates are scarce, but they could also occur via selection to reduce floral vulnerability to herbivores. This study investigated geographic covariation between floral morphology, fruit set, pollen limitation and florivory across the geographic range of Camissoniopsis cheiranthifolia, a Pacific coastal dune endemic that varies strikingly in flower size and mating system. Methods Fruit set was quantified in 75 populations, and in 41 of these floral herbivory by larvae of a specialized moth (Mompha sp.) that consumes anthers in developing buds was also quantified. Experimental pollen supplementation was performed to quantify pollen limitation in three large-flowered, outcrossing and two small-flowered, selfing populations. These parameters were also compared between large- and small-flowered phenotypes within three mixed populations. Key Results Fruit set was much lower in large-flowered populations, and also much lower among large- than small-flowered plants within populations. Pollen supplementation increased per flower seed production in largeflowered but not small-flowered populations, but fruit set was not pollen limited. Hence inadequate pollination cannot account for the low fruit set of large-flowered plants. Floral herbivory was much more frequent in largeflowered populations and correlated negatively with fruit set. However, florivores did not preferentially attack large-flowered plants in three large-flowered populations or in two of three mixed populations. Conclusions Selfing alleviated pollen limitation of seeds per fruit, but florivory better explains the marked variation in fruit set. Although florivory was more frequent in large-flowered populations, large-flowered individuals were not generally more vulnerable within populations. Rather than a causative selective factor, reduced florivory in small-flowered, selfing populations is probably an ecological consequence of mating system differentiation, with potentially significant effects on population demography and biotic interactions. Key words: Beach evening primrose, Camissoniopsis cheiranthifolia, coastal dunes, floral herbivory, fruit set, geographic variation, mating system variation, microlepidoptera, parasitism, pollen limitation, reproductive assurance, self-fertilization. INTRODUCTION The evolutionary transition from large, showy, outcrossing flowers to small, less conspicuous, self-fertilizing flowers has occurred thousands of times (Stebbins, 1974), and generated sustained interest in the ecological and evolutionary causes and consequences. The most widely invoked adaptive explanation is that selfing provides reproductive assurance when pollinators and/or mates are scarce (Eckert et al., 2006). Selfing taxa occur in geographically or ecologically marginal habitats where outcross pollination may be unreliable (Baker, 1955; Jain, 1976; Lloyd, 1980; Busch, 2005) and are over-represented among invasive plants that may often experience Allee effects (Rambuda and Johnson, 2004; van Kleunen et al., 2008). Pollen limitation, which is common and often severe (Burd, 1994; Ashman et al., 2004; Knight et al., 2005), may be alleviated by traits enabling selfing (Larson and Barrett, 2000). Selection on floral traits may also arise from herbivory (Ashman, 2002; Eckert et al., 2006; Strauss and Whittall, 2006). Florivory is common and can affect fitness directly though consumption of pollen, ovules and seeds, or indirectly through consumption of attractive structures and rewards that alter pollinator visitation and foraging (Krupnick et al., 1999; McCall and Irwin, 2006; Steets et al., 2007b). Floral traits that attract pollinators can also attract herbivores (Adler and Bronstien, 2004; Teixido et al., 2011; Theis and Adler, 2012), and larger flowers may provide more resources for herbivores, resulting in indirect selection on traits that correlate with overall size. Hence, florivory may select for smaller, less conspicuous flowers (Hanley et al., 2009), indirectly promoting the evolution of selfing (Penet et al., 2009). Although herbivoremediated selection on floral traits that influence the mating system is not mutually exclusive of selection for reproductive assurance via pollen limitation, the evolutionary scenarios are fundamentally different (Eckert et al., 2006). Selection for reproductive assurance favours floral modifications that increase self-pollination (e.g. reduced herkogamy and dichogamy) and C The Author 2014. Published by Oxford University Press on behalf of the Annals of Botany Company. All rights reserved. V For Permissions, please email: [email protected] 316 Dart & Eckert — Pollen limitation, florivory and mating system variation self-fertilization (weaker self-incompatibility), with reductions in attractive and reward traits and overall flower size favoured only after selfing predominates. Selection to reduce florivory would act directly on attractive/reward traits and flower size, with higher selfing occurring as a secondary consequence. Key floral traits can vary strikingly among populations within species (Jain, 1976). This facilitates investigation of mating system evolution, as it reduces the likelihood that variation in ecological and genetic factors that cause (or at least maintain) mating system differentiation is confounded with major differences in life history and ecology. Variation among plants within populations is additionally useful, because the effect of floral variation on pollination, florivory and mating can be evaluated more directly. Such intraspecific variation has enabled analyses of covariation between key floral traits and mating (reviewed in Button et al., 2012; Dart et al., 2012), and between selfing and inbreeding depression, the main selective factor opposing selfing (Husband and Schemske, 1996; Winn et al., 2011). However, the pollination ecology of small-flowered selfing and large-flowered outcrossing populations within species (Inoue et al., 1996; Fausto et al., 2001) or phenotypes within populations (Piper et al., 1986) has rarely been studied, leaving fundamental predictions with equivocal support. For instance, the prediction that self-compatibility increases fruit or seed set has been supported in some species (Lyons and Antonovics, 1991; Busch, 2005) but not in others (Barrett et al., 1989; Hererra et al., 2001; Schueller, 2004). The prediction that selfing alleviates pollen limitation is also weakly supported (Lyons and Antonovics, 1991; Hererra et al., 2001; Schueller, 2004; Kennedy and Elle, 2008; Moeller et al., 2012). Whether variation in floral morphology and mating system affects floral antagonists or vice versa has rarely been studied. It is not known, for instance, whether smaller, less conspicuous flowers experience lower rates of florivory than larger, showier flowers. We combine multi-year surveys of fruit set and florivory with experimental pollen supplementation to investigate the causes and consequences of mating system differentiation in Camissoniopsis cheiranthifolia (Onagraceae). This is a shortlived perennial endemic to Pacific coastal dunes of western North America (Samis and Eckert, 2007) that varies widely in floral traits and mating system across its geographic range (Dart et al., 2012). Raven (1969) divided the species into two subspecies: large-flowered, often self-incompatible and putatively outcrossing ssp. suffruticosa, distributed from Point Conception, California to the southern range limit in northern Baja California, Mexico; and small-flowered, highly autogamous and putatively selfing ssp. cheiranthifolia, distributed from Point Conception to the northern range limit in southern Oregon and on the Channel Islands. Linsley et al. (1973) observed that visitation by the principle pollinators, small oligolectic Andrena bees, was much lower in smallflowered than large-flowered populations, and hypothesized that heavy morning fog typical north of Point Conception reduces bee activity, thereby selecting for selfing (and consequently small flowers) to provide reproductive assurance. Dart et al. (2012) revealed greater complexity. Flower size did vary discontinuously, such that populations could readily be categorized as large- (LF, mean corolla width approx. 29 mm) or small-flowered (SF, mean approx. 16 mm), and common garden experiments confirmed a strong genetic basis to this. As expected, plants from SF populations (ssp. cheiranthifolia) are self-compatible and can achieve full seed set via autonomous self-pollination within buds before anthesis (Dart and Eckert, 2013). Plants from LF populations (ssp. suffruticosa) in San Diego County California are self-incompatible (SI). However, LF populations just to the north are self-compatible (SC) and can set considerable seed by autonomously selfing when flowers close, although the contribution of delayed selfing to seed production in natural populations is unclear (Dart and Eckert, 2013). Populations in Baja California, classified by Raven (1969) as ssp. suffruticosa, turned out to be SF, SC, highly autogamous and very similar to ssp. cheiranthifolia. Further, all populations engage in some self-fertilization (Dart et al., 2012). The mean proportion of seeds self-fertilized ranged from 018 in LF–SI populations and 036 in LF–SC populations to 067 in SF–SC populations north of Point Conception and 084 in SF–SC populations on the Channel Islands and Baja California. Strong geographic differentiation in floral traits but weaker differentiation in self-compatibility, autonomous selfing and mating system suggests that selective factors other than or in addition to reproductive assurance have promoted floral differentiation in C. cheiranthifolia. Interactions with herbivores might have influenced floral evolution in C. cheiranthiolia. Developing flowers are exploited by a specialist microlepidopteran herbivore that consumes anthers within buds, causing flowers to abscise before anthesis. Because buds are not consumed outright, we refer to this as bud parasitism. DNA barcoding of >200 larvae identified the parasite as a single undescribed Mompha species (Coleophoridae; Emery et al., 2009), within a genus known for species-specific associations with the Onagraceae. Large-flowered plants may be more heavily parasitized if adult moths are attracted to larger flowers, or if LF plants produce larger buds that provide more resources for larval development. We investigate the roles of pollen limitation and florivory in the differentiation of floral morphology and mating system in C. cheiranthifolia by addressing the following predictions. (1) Fruit set will be higher in SF than in LF populations due to some combination of high autonomous selfing and lower floral parasitism. (2) Because fruit set in LF populations depends on the abundance and activity of specialist pollinators and florivores, which may vary in time and space, fruit set will vary more among populations and between years within populations in LF than in SF populations. (3) A higher proportion of buds will be parasitized in LF than in SF populations. (4) Mompha may parasitize LF plants more frequently because they have larger, more attractive flowers and/or larger buds. Hence, within populations, plants with larger flowers will be parasitized more than those with smaller flowers. (5) An important assumption of prediction (4) is that larger flowers arise from relatively larger buds. (6) Selfing could have been favoured via reproductive assurance if LF phenotypes experienced chronic pollen limitation, hence outcross pollen limitation of fruit set and seed production will be greater for LF than for SF populations and for LF than for SF plants within mixed populations. Dart & Eckert — Pollen limitation, florivory and mating system variation Geographic covariation between flower size, fruit set and bud parasitism We tested whether the capacity for selfing increases fruit set by comparing 15 LF and 60 SF populations from across the geographic range from May to July 2002–2005 (Supplementary Data 1). Populations were classified as LF or SF following Dart et al. (2012) based on a large sample of flowers measured for corolla width (measured from the tip of a randomly selected petal to the tip of the opposing petal; Fig. 1A; data from Dart et al., 2012) and herkogamy (distance from the base of the stigma to the closest dehiscing anther; Dart et al., 2012). Populations were visited near the end of the flowering season so that plants had produced many flowers that were unambiguously either aborted or developing into fruit. We randomly selected a single lateral branch on 10–51 randomly selected plants per population (mean ¼ 25, total n ¼ 1851 branches) and counted the mature, maturing and aborted fruits, and calculated fruit set as the number of mature or maturing fruits divided by the total number of ovaries. We tested the prediction that bud parasitism was more frequent in LF than in SF populations, by estimating the frequency of parasitism for 41 populations (16 LF, 25 SF) across the range (Supplementary Data 1) by sampling a single bud just before opening from 20–98 plants per population (mean ¼ 42) visited once during peak flowering from May to July 2005–2010 (25 populations sampled in 2005, 11 in 2007, two in 2009, three in 2010, n ¼ 1706 buds total). We measured the length of each bud (to 001 mm), dissected it and scored it as parasitized if it contained a moth larva, larval frass or characteristic damage to 40 35 30 25 20 15 10 1·0 LF−SI LF−SC VAR SF−SC 0·8 Mean fruit set Camissoniopsis cheiranthifolia (Hornem. ex Spreng.) W.L. Wagner & Hoch produces actinomorphic flowers with four green, reflexed sepals, four bright yellow petals, a shallow hypanthium, eight stamens in two whorls and a large globular stigma presented adjacent to or 9 mm above the anthers. Flowers are produced singly in leaf axils, mostly along lateral stems extending from a central rosette. Flowers open at sunrise and close in the late afternoon or early evening. The duration of opening within a day and the consecutive days that flowers reopen are much lower in SF than in LF populations (Dart et al., 2012). Peak flowering occurs during May–July in all populations studied here (S. Dart and C.G. Eckert, unpubl. data). Mature fruits and undeveloped ovaries remain attached to stems, allowing retrospective measurement of fruit set. Because LF and SF populations are geographically segregated (Dart et al., 2012), differences in fruit and seed set, bud parasitism and pollen limitation might arise from geographic variation in the abundance and activity of mutualists and parasites in addition to or instead of variation in floral morphology. Although geographic surveys alone cannot verify a causative influence of floral morphology, LF and SF phenotypes co-occur within three populations, allowing more direct tests of associations between floral morphology, fruit set, pollen limitation and bud parasitism. 0·6 0·4 0·2 0 1·0 Proportion parasitized Study species Corolla width (mm) MATERIALS AND METHODS 317 0·8 0·6 0·4 0·2 0 30 32 34 36 38 40 42 44 Latitude (°N) FIG. 1. Geographic variation in flower size (corolla width), fruit set and the proportion of buds parasitized among natural populations of Camissoniopsis cheiranthifolia across its geographic range. The dotted vertical lines indicate the locations of the Mexico–USA border (325 N) and Point Conception (345 N). Filled triangles are large-flowered self-incompatible populations (LF–SI); filled circles are large-flowered self-compatible populations (LF–SC); open circles are small-flowered self-compatible populations (SF–SC); and diamonds indicate the three phenotypically mixed populations just north of Point Conception (VAR). In subsequent figures and all analyses, LF–SI and LF–SC are pooled as LF populations. Note the two clusters of SF–SC populations south of Point Conception: one on the Channel Islands (33–34 N) and the other in northern Baja California (30–31 N). The flower size data are from Dart et al. (2012). floral organs. In analyses below, the frequency of bud parasitism varies both within and among flowering seasons, hence there is undoubtedly uncontrolled temporal variation in parasitism among the populations in our survey that may obscure the difference between LF and SF populations. Generalized linear mixed models (GLMMs) evaluated whether fruit set or bud parasitism differed between LF and SF populations, with fruit set (number of flowers setting fruit or not) and bud parasitism (bud parasitized or not) as binomial responses, population flower size as a fixed categorical predictor and population as a random effect nested within flower size. Models using the logit link function were fit using the glmer command in the package lme4 (version 1.0-5, http://cran. r-project.org/package ¼ lme4, accessed 13 November 2013) for the R statistical computing environment (version 3.0.2; R Core Team, 2013). Fruit set data were overdispersed, so we fit models using quasi-likelihood estimation (Zuur et al., 2013). A mixed-effects linear model implemented using the lmer command (lme4 package) tested the prediction that buds were longer in LF than in SF populations, with population flower size as a fixed predictor and population as a nested random effect. For all analyses, likelihood ratio tests evaluated the significance of potential predictors. 318 Dart & Eckert — Pollen limitation, florivory and mating system variation We tested whether fruit set and bud parasitism are more variable among LF than among SF populations using Levene’s test (Schultz, 1985; leveneTest command, car package, version 2.0-19, http://cran.r-project.org/web/packages/car/index.html, accessed 13 November 2013). Given that fruit set and the proportion of buds parasitized are both binomial variables, their statistical variance should be highest when mean ¼ 05. More variation among LF than among SF populations would be expected if, as predicted, the grand mean of LF populations was closer to 05 than that of SF populations. Simulations were used to determine whether Levene’s test detected additional variation among LF populations due to stochasticity of pollination or parasitism (above and beyond differences in binomial sampling error) (Supplementary Data 2). Absolute year to year differences in fruit set and bud parasitism were also compared between LF and SF populations (Supplementary Data 3). To supplement categorical comparisons of fruit set, parasitism frequency and bud length between LF and SF populations, generalized linear models (GLMs; glm command in R) analysed the regression between these response variables and corolla width as a continuous predictor, with fruit set and proportion parasitized as binomial responses (logit link) and bud length as a Gaussian response (identity link). Each analysis was done for all populations and for LF and SF populations separately to determine if the large-scale regressions were observed within population flower size groups. At the population level, fruit set was represented by the total sampled flowers setting fruit vs. not setting fruit. Similarly, the proportion of buds parasitized was represented by the number of sampled buds parasitized vs. not parasitized. GLMs also analysed regressions between proportion parasitized and fruit set (binomial response) and bud length and proportion parasitized (binomial response). Data for the response variables fruit set and proportion parasitized were overdispersed, so we used quasi-likelihood estimation. For each analysis, we report the regression coefficients back-transformed to the original scale of the data using the predict function in R (Zuur et al., 2009, p. 250). Covariation between flower size, fruit set and bud parasitism within populations For a more direct test of the prediction that smaller flowers and higher selfing are associated with increased fruit set, we compared LF and SF phenotypes within three phenotypically mixed populations (CGN1C, CSP1C and CMN1C). In 2005, plants in each population were classified as LF (corolla width >20 mm) or SF (20 mm) following Dart et al. (2012). We estimated fruit set on 30 plants per phenotype (total n ¼ 263 plants). A GLM with quasi-likelihood estimation modelled fruit set as a binomial response (logit link), with population, floral phenotype (LF vs. SF) and their interaction as potential predictors. Levene’s test with simulations evaluated whether fruit set was less variable among SF than among LF plants (Supplementary Data 2). We tested the prediction of higher parasitism on plants with larger flowers in two ways. In 2005, we sampled the most mature bud from LF and SF plants within each mixed population (n ¼ 230 buds), measured its length and scored it as parasitized or not. We compared LF vs. SF plants for bud length using linear models, and for frequency of parasitism (binomial response) using GLM (logit link), with population, floral phenotype and their interaction as potential predictors. GLM (logit link) evaluated the association between parasitism and bud size with population and bud length as predictors. Secondly, we sampled a single flower and the most mature adjacent bud for a large sample of plants in three LF populations (CCO1C, CIV1C and CBV1C) during 2007. We randomly sampled 60 plants on each of five occasions at CCO1C, two at CIV1C and one at CBV1C. The corolla width of the sampled flower was measured and the adjacent bud was measured and scored as parasitized or not. For each population, GLM modelled parasitism as a binomial response (logit link), with corolla width (or bud length) and sampling date as predictors. Pollen supplementation experiment To test the prediction that small flower size and consequent selfing reduces pollen limitation (PL), supplemental pollinatatuion was provided to flowers in three LF and two SF populations in 2004 and 2006 (Supplementary Data 1). We randomly selected 34–115 plants per population, each with at least one open flower, and randomly assigned 16–63 plants to each of two treatments: cross-pollen supplementation (þX) or open pollination (OP). Pollen limitation is usually quantified by comparing naturally pollinated and cross-supplemented flowers, whereas the extent to which selfing might alleviate PL must be determined by comparing naturally pollinated flowers with those supplemented with self pollen (Eckert et al., 2010); therefore, we added a self-supplemented treatment (þS) in 2006. For each supplemented plant, we hand-pollinated every open flower (mean approx. 3, range ¼ 1–10) on a single day during peak flowering by brushing pollen from three anthers across the stigma, using one anther from each of three donor plants located 3 m away for þ X flowers and three anthers from another flower on the same plant for þ S flowers. We randomly selected a single treated flower on each plant as the focal flower, marked it with acrylic paint on the ovary, recorded whether it matured a fruit and counted the filled seeds within. To avoid confounding effects of flower position and phenology (Wesselingh, 2007), all focal flowers were centrally located on their stems and open during peak flowering. Because the number of seeds produced per focal flower is the product of a Bernoulli variable (fruit set) and a zerotruncated Poisson variable (seeds per mature fruit), we modelled the effect of pollination treatment, population flower size and their interaction using the ASTER model in R (Shaw et al., 2008; version 0.8-27, http://cran.r-project.org/web/packages/aster/index.html, accessed 13 November 2013). We followed Geyer et al. (2007) to estimate the expectation and 95 % confidence intervals of seeds per focal flower. Population could not be included as a nested random effect because too few populations within flower size classes were used to estimate its random variance accurately. Supplementary analysis substituting population for flower size revealed no heterogeneity among populations within size categories that would complicate interpretation. Data for 2004 and 2006 were analysed separately because different populations were used. Dart & Eckert — Pollen limitation, florivory and mating system variation To determine whether PL might cause the lower fruit set observed in LF than in SF populations in our geographic survey (see the Results), GLM (logit link) analysed the fruit set component of seed per flower, with pollination treatment, population flower size and their interaction as predictors. Pollen limitation was compared between LF and SF phenotypes within two mixed populations (CGN1C and CSP1C). As above, randomly chosen plants were randomly assigned to þX or OP treatments in 2004, and þX, OP and þS treatments in 2006. ASTER and GLM evaluated the effect of population (CGN1C vs. CSP1C), pollination treatment, flower size and their interactions on seeds per flower and fruit set. Providing supplemental pollination to only a sample of flowers on individual plants and comparing these with flowers on open-pollinated plants may overestimate PL (Knight et al., 2006) because preferential allocation of resources to supplemented flowers may elevate their fruit and seed set at the expense of unsupplemented flowers. We evaluated this bias experimentally and found no evidence that seed set of supplemented flowers was boosted by resource reallocation in any population (Supplementary Data 4). Depression of fruit set by bud parasitism alone Comparing the fruit set of flowers that had been potentially exposed to parasitism vs. those that had not estimated how much bud parasitism alone depresses fruit set. Elimination of parasitism by pesticide application was not permitted in our study populations, so we adopted the following approach. First, we retrospectively estimated fruit set for all flowers on the focal stem on all plants selected for the 2004 pollen supplementation experiment before experimental treatments were applied. These ‘retrospective’ flowers were exposed to natural pollination and bud parasitism, with fruit set potentially depressed by resource limitation, pollen limitation and/or parasitism. The open-pollinated ‘focal’ flowers used in the supplementation experiment were exposed to natural pollination, but not bud parasitism (parasitism causes abscission before opening). The difference in fruit set between ‘retrospective’ and ‘focal’ flowers estimates fruit set lost to bud parasitism. This assumes that background fruit abortion does not vary temporally, because retrospective flowers tended to open earlier than focal flowers. The increase in abortion during the flowering sequence typical of many plants (Diggle, 1995) would cause the effect of bud parasitism to be underestimated. However, our surveys suggest extremely infrequent parasitism in SF populations, which should reveal any temporal reduction in fruit set alone. We monitored fruit set of retrospective and focal flowers on 360 plants in three LF (COR1C, CBV1C and CCO1C) and two SF populations (CGN3C and CMS1C). Directly comparing fruit set of retrospective vs. focal flowers was not possible because, for each plant, the former involved a sample of flowers, whereas the latter involved a single flower. Instead, we calculated and compared maximum likelihood estimates of the expected value and 95 % confidence intervals for fruit set (binomial response) of each type of flower for each population using a GLM (logit link). 319 RESULTS Geographic covariation between flower size, fruit set and bud parasitism The mean proportion of flowers setting fruit (fruit set) ranged from 023 to 100 among 75 populations surveyed (n ¼ 1851 plants; Fig. 1) and was, as predicted, lower for 15 LF populations (mean 6 s.e. ¼ 053 6 0045) than for 60 SF populations (097 6 0009; likelihood ratio v2 ¼ 6782, d.f. ¼ 1, P < 0001). Mean fruit set varied more among LF than among SF populations (Levene’s test F1,73 ¼ 1919, P < 0001). The difference in variance was not quite significantly larger than that expected from differences in binomial sampling variance alone (Psim approx. 0057) (Supplementary Data 2). Between-year difference in fruit set was not greater within the nine LF than within the 31 SF populations surveyed in > 1 year (Supplementary Data 3). Overall, 23 % of the 1706 buds were parasitized, ranging from 000 to 090 among 41 populations surveyed (Fig. 1) and, as predicted, this was higher for 16 LF (041 6 0056) than for 25 SF populations (005 6 0013; v2 ¼ 3438, d.f. ¼ 1, P < 0001). The proportion of buds parasitized varied more among LF than among SF populations (F1,73 ¼ 1274, P < 0001) but not more than expected based on binomial sampling variances (Psim approx. 012) (Supplementary Data 2). Between-year differences in parasitism were higher for seven LF than for six SF populations (Supplementary Data 3). Bud length ranged from 266 to 1535 mm (n ¼ 1293 buds) and was 40 % shorter in 25 SF (617 6 012 mm) than in ten LF populations (1068 6 027 mm; v2 ¼ 8313, d.f. ¼ 1, P < 0001). Across all populations, increasing corolla width was associated with decreasing fruit set, increasing proportion parasitized and increasing bud length (Fig. 2A, B, C, respectively; Table 1). However, none of these regressions was significant among LF or SF populations analysed separately (Table 1). Similarly, fruit set covaried negatively with proportion parasitized (n ¼ 35 populations; Fig. 2D) but only the regression among SF populations neared significance. Proportion parasitized covaried positively with bud length (n ¼ 35) but not within flower size categories (Table 1). Covariation between flower size, fruit set and bud parasitism within populations In three mixed populations, fruit set ranged from 017 to 10 (mean ¼ 082, n ¼ 263 plants; Fig. 3A) and varied significantly among populations (likelihood ratio v2 ¼ 1010, d.f. ¼ 2, P ¼ 00064). As predicted, fruit set was lower for LF than for SF phenotypes (v2 ¼ 12532, d.f. ¼ 1, P < 0001; no interaction v2 ¼ 299, d.f. ¼ 2, P ¼ 022). Within each population, fruit set varied more among LF than among SF plants (Levene’s test: CGN1C F1,91 ¼ 2191, P < 0001; CMN1C F1,58 ¼ 946, P ¼ 00032; CSP1C F1,108 ¼ 3901, P < 0001). However, the difference in variance exceeded that expected from binomial sampling variance in CGN1C (Psim approx. 0044) and CSP1C (Psim approx. 00072) but not in CMN1C (Psim approx. 026) (Supplementary Data 2). Overall, 19 % of 230 buds were parasitized, and parasitism varied among populations (v2 ¼ 857, d.f. ¼ 2, P ¼ 0014), and was higher for LF than for SF plants Dart & Eckert — Pollen limitation, florivory and mating system variation 320 (v2 ¼ 399, d.f. ¼ 1, P ¼ 0046). Although the interaction was not quite significant (v2 ¼ 496, d.f. ¼ 2, P ¼ 0084), the difference between floral phenotypes was only evident in one of three populations (Fig. 3B). Bud length did not vary among populations (v2 ¼ 088, d.f. ¼ 2, P ¼ 064) but was 40 % shorter on SF than on LF plants (v2 ¼ 57462, d.f. ¼ 1, P < 0001), and this difference varied in magnitude but not in direction among populations (v2 ¼ 2481, d.f. ¼ 2, P < 0001), being somewhat smaller in CSP1C (31 %) than in CGN1C (42 %) or CMN1C (47 %). There was no association between bud length and parasitism across these populations (v2 ¼ 0049, d.f. ¼ 1, P ¼ 082). In the three LF populations where bud parasitism and bud length were measured along with corolla width of the adjacent flower, 38 % of 458 buds were parasitized. Bud length covaried positively with corolla width (range of r among populations and sampling dates ¼ 037–059, all P < 001, mean r ¼ þ 048). Although parasitism varied among populations and sampling dates, parasitism did not vary with corolla width in any populations (Table 2). Similarly, bud length and parasitism did not correlate within populations (CCO1C v2 ¼ 140, d.f. ¼ 1, P ¼ 023; CIV1C v2 ¼ 112, P ¼ 029; CBV1C v2 ¼ 083, P ¼ 036). A 1·0 Mean fruit set 0·8 0·6 0·4 0·2 Large-flowered Small-flowered 0 B Proportion parasitized 1·0 0·8 0·6 0·4 0·2 Pollen supplementation experiment 0 C Bud length (mm) 12·5 10·5 8·5 6·5 4·5 10 20 30 Corolla width (mm) 40 0·2 0·4 0·6 0·8 Proportion parasitized 1·0 D 1·0 Mean fruit set 0·8 0·6 0·4 0·2 0 0 FIG. 2. Relationships of fruit set, bud parasitism and bud length to corolla width (A–C) and of fruit set to bud parasitism (D) among natural populations of Camissoniopsis cheiranthifolia. Open circles are small-flowered populations, and filled circles are large-flowered populations. Analysis based on these data is given in Table 1. In both 2004 and 2006, supplemental pollination increased the seeds produced per flower more in LF than in SF populations (Fig. 4), although the interaction between population flower size and pollination treatment was not quite significant in 2004 (Supplementary Data 5). For both years, the effect of pollination treatment was significant for LF populations (2004, v2 ¼ 5304, d.f. ¼ 1, P < 0001; 2006, v2 ¼ 4052, d.f. ¼ 1, P < 0001) but not SF populations (2004, v2 ¼ 015, d.f. ¼ 1, P ¼ 070; 2006, v2 ¼ 262, d.f. ¼ 1, P ¼ 027). Cross-pollination did not increase seeds per flower more than self-pollination in 2006. In fact, the trend was in the opposite direction (Fig. 4B), though not quite significant (v2 ¼ 378, d.f. ¼ 1, P ¼ 0052). Supplemental pollination increased seeds per flower in LF populations primarily by increasing seeds per fruit rather than fruit set. Fruit set was generally high in both years (2004 ¼ 893 %; 2006 ¼ 870 %). Although fruit set was higher in SF than in LF populations (2004, LF ¼ 823 %, SF ¼ 963 %; 2006, LF ¼ 766 %, SF ¼ 979 %), supplemental pollination did not increase fruit set, nor did the effect of pollination treatment and population flower size interact (Supplementary Data 5). Effects of supplemental pollination in the two mixed populations were complex (Fig. 5). In 2004, the effects of population, flower size and pollination treatment interacted (Supplementary Data 6) but not in a way that indicated consistently stronger pollination effects for LF than for SF plants. Instead, supplemental cross-pollination increased seed production of SF but not LF plants in population CGN1C and decreased seed production of SF but not LF plants in CSP1C. For instance, population and pollination treatment interacted for SF (v2 ¼ 1039, d.f. ¼ 1, P ¼ 00013), but not LF plants (v2 ¼ 008, d.f. ¼ 1, P ¼ 078). In 2006, all two-way interactions but not the three-way interaction were significant (Supplementary Data 6) but, again, not in a way that indicated stronger effects of pollination for LF plants. Instead, the interactions seem to arise from the Dart & Eckert — Pollen limitation, florivory and mating system variation 321 TABLE 1. Associations among flower size (as measured by corolla width), bud length, fruit set and the proportion of buds parasitized by Mompha among populations of Camissoniopsis cheiranthifolia Response (distribution) Predictor Fruit set (binomial) Corolla width Proportion parasitized (binomial) Corolla width Bud length (Gaussian) Corolla width Fruit set (binomial) Proportion parasitized Proportion parasitized (binomial) Bud length All populations LF populations only SF populations only 0031 (43) 1561 (P < 0001) 0034 (24) 273 (P < 0001) 0293 (19) 1215 (P < 0001) 076 (35) 281 (P < 0001) 0056 (35) 323 (P < 0001 0028 (11) 32 (P ¼ 0074) 00048 (11) 0040 (P ¼ 084) 0074 (6) 037 (P ¼ 054) 0011 (14) 035 (P ¼ 055) 00069 (10) 064 (P ¼ 042) 00021 (32) 003 (P ¼ 057) 00094 (13) 0059 (P ¼ 044) 016 (13) 226 (P ¼ 013) 034 (21) 381 (P ¼ 0051) 00031 (25) 167 (P ¼ 020) Responses were modelled as binomial variables except for bud length, which was considered Gaussian. Analyses were conducted for all populations and for large-flowered (LF) and small-flowered (SF) populations separately. The upper level of each cell contains the back-transformed regression slope (change in the response per unit change in the predictor) with the number of populations sampled in parentheses. The lower level contains the v2 value from a likelihood ratio test (with the P-value). Plots of the first four relationships are given in Fig. 2. 1·0 A Fruit set 0·9 0·8 Large-flowered 0·7 and the effects of population and pollination interacted (v2 ¼ 1141, d.f. ¼ 2, P ¼ 00033). In contrast, for SF plants, only the interaction was marginally significant (v2 ¼ 654, d.f. ¼ 2, P ¼ 0038) and the effect of pollination was not (v2 ¼ 479, d.f. ¼ 2, P ¼ 0091). Variation in treatment effects among floral phenotypes in these mixed populations resulted from variation in seeds per fruit rather than fruit set. Fruit set was high for all treatments in both 2004 (935 %, n ¼ 387) and 2006 (920 %, n ¼ 138). The GLM detected only a difference in fruit set between populations and only in 2004 (Supplementary Data 6). Small-flowered 63 0·6 30 80 30 30 30 B For all three LF populations, retrospective flowers exposed to bud parasitism set fewer fruit than focal flowers for which losses could not have involved parasitism, and in two populations there was no overlap in the 95 % confidence intervals (Fig. 6). In contrast, fruit set was consistently very high (>90 %) in both SF populations and did not differ between retrospective and focal flowers. 0·4 Proportion buds parasitized Depression of fruit set by bud parasitism alone 0·3 0·2 0·1 DISCUSSION 0 Covariation between the mating system and pollen limitation 49 19 CGN1C 64 34 CSP1C 34 30 CMN1C Population FIG. 3. Comparison of mean (6 95 % confidence interval) fruit set (A) and the proportion of buds parasitized (B) between large-flowered (LF) and smallflowered (SF) phenotypes within three mixed populations of Camissoniopsis cheiranthifolia. The numbers of plants sampled for each combination of population and flower size phenotype are indicated. difference between self-supplementation vs. both natural pollination and cross-supplementation, being somewhat larger for LF than for SF plants. Analysing LF plants alone, the effect of pollination was significant (v2 ¼ 1052, d.f. ¼ 2, P ¼ 00052) Reproductive assurance is the most widely accepted adaptive explanation for the evolution of self-fertilization (Cheptou, 2004; Morgan et al., 2005), yet some of its most fundamental predictions have rarely been tested, and the few existing studies produced mixed results. The prediction of higher fruit or seed set in selfing vs. outcrossing populations is supported in some species (Primula vulgaris, Piper et al., 1986; Helleborus foetidus, Herrera et al., 2001; Leavenworthia spp., Lyons and Antonovics, 1991; Busch, 2005) but not others (Eichhornia paniculata, Barrett et al., 1989; Nicotiana glauca, Schueller, 2004). At face value, our results lend additional support to this prediction. Fruit set was much higher, though not biologically less variable, in SF vs. LF C. cheiranthifolia populations. Although LF and SF populations are geographically Dart & Eckert — Pollen limitation, florivory and mating system variation 322 TABLE 2. Analysis testing for an association between flower size, bud length and bud parasitism among plants within three largeflowered populations of Camissoniopsis cheiranthifolia Population CCO1C CIV1C CBV1C Summary statistics Generalized linear model (binomial response ¼ parasitized) Generalized linear model (Gaussian response ¼ bud length) Proportion of buds parasitized Corolla width (mm) Bud length (mm) Date (D) Corolla width (C) DC Date (D) Corolla width (C) DC 0431 (028–066) 0287 (028–029) 0300 (NA) 3500 (2113–4592) 3337 (2181–4396) 3487 (2128–4396) 1119 (410–1918) 1150 (740–1506) 1169 (708–1498) 2141 (4) <0001 00019 (1) 096 NA 013 (1) 072 068 (1) 041 002 (1) 090 362 (4) 046 225 (1) 013 NA 1246 (4) 0014 340 (1) 0067 NA 8913 (1) <0001 2581 (1) <0001 1649 (1) <0001 518 (4) 027 035 (1) 055 NA Summary statistics cells contain mean and range (in parentheses) among sampling dates for parasitism among flowers measured for corolla width and among buds measured for bud length. Results of generalized linear models include the likelihood ratio v2 values (above), P-values (below) and degrees of freedom (in parentheses). Population CCO1C was sampled five times, CIV1C twice and CBV1C only once (hence variation among sampling dates was not analysed; NA). separated, the difference in fruit set is probably not due to confounding geographic variation in abiotic factors (e.g. temperature or precipitation), as we consistently observed high fruit set in all three geographically disjunct groups of SF populations: north of Point Conception, Baja California and the Channel Islands. Moreover, SF plants set more fruit than LF plants within three mixed populations where divergent phenotypes experience similar environments. Our results also support the prediction that selfing increases seed production because it alleviates outcross pollen limitation. Again, results from previous pollen supplementation experiments are mixed, with lower pollen limitation observed in selfing than in outcrossing populations of Primula vulgaris, but not of Leavenworthia crassa (Lyons and Antonovics, 1991), Helleborus foetidus (Herrera et al., 2001), Nicotiana glauca (Schueller, 2004) or Collinsia parviflora (Kennedy and Elle, 2008; see also Goodwillie, 2001). This disparity may partially reflect the fact that selfing is not the only evolutionary response to pollen limitation. Outcrossing populations might evolve other compensating mechanisms, such as increased floral attraction, reward or longevity, depending on the ecological conditions affecting the strength of pollen limitation and other selective factors such as inbreeding depression (Eckert et al., 2010). Although stronger pollen limitation in outcrossing vs. selfing populations may linger after mating system differentiation, implicating pollen limitation as a causal selective force (as may be the case in C. cheiranthifolia), selfing and outcrossing populations might diverge in multiple compensating mechanisms, yielding low pollen limitation and high fruit set regardless of the mating system. Higher pollen limitation in LF than in SF populations is consistent with the conjecture of Linsley et al. (1973) that selfing (and consequentially smaller flowers) evolved in C. cheiranthifolia to provide reproductive assurance. If pollen limitation is common in LF populations south of Point Conception, as seems the case, decreased pollinator visitation in the colder and foggier environment to the north could tip the selective balance to favour higher selfing. However, we recognize two caveats: first, pollen limitation cannot explain the much lower fruit set in LF vs. SF populations north and south of Point Conception, as pollen supplementation only increased seeds per fruit (not fruit set) in LF populations. Instead, the lower fruit set in LF populations is probably caused by differential Mompha parasitism (see below). Secondly, verifying that pollen limitation of LF phenotypes would be higher north of Point Conception requires experimental transplants (e.g. Geber and Eckhart, 2005; Moeller and Geber, 2005), for which we could not obtain permits. The mixed LF/SF populations north of Point Conception provided a ‘natural experiment’ but, contrary to expectations, pollen limitation was not higher for LF than for SF phenotypes. However, naturally mixed populations are an imperfect analogue of transplant experiments, as they may experience conditions (e.g. frequent pollinator visitation) that permit the persistence of LF phenotypes (indeed, LF plants have probably existed in these populations for at least 45 years; Raven, 1969, p. 266). Thus LF phenotypes could suffer much stronger pollen limitation if transplanted into pure SF populations north of Point Conception. Mompha parasitism as a selective force on floral morphology, development and the mating system By directly interfering with reproduction, florivory is a strong potential selective force (Hanley et al., 2009), but only if florivory induces covariation between fitness and floral traits. Mompha larvae can reduce fruit set dramatically by causing C. cheiranthifolia flowers to abscise before opening, although we could not estimate this effect directly since experimental florivore exclusion (e.g. Galen, 1999; Herrera et al., 2002) was not permitted. Mixed populations provided equivocal results, as lower parasitism was associated with higher fruit set in only one of three populations, although this might be because parasitism was measured only once per flowering season whereas fruit set integrates reproductive success across the season. Florivore effects on fitness were more clearly revealed by lower fruit set of retrospective flowers exposed to both pre- and post-anthesis losses compared with focal flowers exposed to post-anthesis losses only (Fig. 6). The negligible difference in Dart & Eckert — Pollen limitation, florivory and mating system variation 60 A 2004 80 323 A 2004 Natural Selfed Crossed Seeds per flower 70 Seeds per flower 50 40 50 40 30 Natural 30 Selfed Crossed 60 102 90 98 93 B 2006 50 20 48 46 80 B 2006 20 25 80 66 58 64 7 10 8 16 15 16 14 15 14 Large Small 70 Seeds per flower 20 Seeds per flower 60 60 50 40 30 40 20 7 7 9 Large 30 Small CGN1C CSP1C Population and flower size 20 82 84 Large 82 79 78 79 Small Population flower size FIG. 4. Comparisons of mean (6 95 % confidence interval) seeds per flower between open-pollinated (open circles), cross-pollinated (filled circles) and self-pollinated (filled triangles; 2006 only) flowers in large-flowered (LF) and small-flowered (SF) populations of Camissoniopsis cheiranthifolia in 2004 (A) and 2006 (B). The numbers of plants sampled for each combination of treatment and population are indicated. fruit set between retrospective and focal flowers in SF populations confirms that Mompha parasitism is the only appreciable cause of pre-anthesis fruit abortion. Lack of experimental proof notwithstanding, we conclude that Mompha parasitism reduces fruit set, contributing strongly to the much lower fruit set in LF than in SF populations. Evidence that Mompha florivory favours smaller, less conspicuous flowers is weaker. The positive correlation between flower size and parasitism among populations was not evident among LF or SF populations analysed separately, and LF plants suffered more frequent parasitism in only one of three mixed populations. Flower size and parasitism also did not correlate in any of the three LF FIG. 5. Comparisons of mean (6 95 % confidence interval) seeds per flower between open-pollinated (open circles), cross-pollinated (filled circles) and self-pollinated (filled triangles; 2006 only) flowers on large-flowered (LF) and small-flowered (SF) phenotypes in two mixed populations of Camissoniopsis cheiranthifolia in 2004 (A) and 2006 (B). The numbers of plants sampled for each combination of treatment, phenotype and population are indicated. populations analysed. Female Mompha may not preferentially oviposit on larger flowers if there is no benefit to doing so or little opportunity to distinguish flower size. The more abundant nectar in large flowers may not attract females if their feeding proboscises are not functional (Emery et al., 2009), and floral traits may not cue oviposition if Mompha adults are strictly nocturnal, when C. cheiranthifolia flowers are closed (Dart et al., 2012). As we could not reliably observe the oviposition behaviour of these nocturnal microlepidopterans (see also Artz et al., 2010), it remains uncertain whether plants with less conspicuous flowers would avoid parasitism. Alternatively, higher florivory in LF populations may arise if they support larger Mompha populations. Individual larva may require >1 bud for development (C.G. Eckert, unpubl. data), and LF buds are 40 % larger than SF buds, providing more resources. If moving among buds exposes larvae to predation or parasitism, larger buds could increase survival by reducing bud to bud movements. Higher mitochondrial DNA haplotype diversity in Mompha populations south of Point Conception Dart & Eckert — Pollen limitation, florivory and mating system variation 324 1·0 0·9 Fruit set 0·8 0·7 0·6 0·5 Retrospective 0·4 0·3 Focal 27 40 COR1C CBV1C 114 97 82 CCO1C CGN3C CMS1C Population FIG. 6. Comparison of mean (6 95 % confidence interval) fruit set in flowers exposed to bud parasitism (retrospective flowers: pre- and post-anthesis losses) vs. flowers in the pollen supplementation experiment not exposed to bud parasitism (focal flowers only post-anthesis losses) in five populations of Camissoniopsis cheiranthifolia. Retrospective and focal flowers were on the same set of plants. Populations COR1C, CBV1C and CCO1C are large flowered. Populations CGN3C and CMS1C are small flowered. The numbers of plants sampled in each population are indicated. (Emery et al., 2009) also suggests that LF C. cheiranthifolia populations maintain larger Mompha populations. This proposed causal link between flower size and parasite abundance would explain their geographic association, which is unlikely to reflect a coincidental match in latitudinal trends because Mompha abundance was low in geographically disjunct groups of SF populations. Although we cannot rule out the possibility that Mompha abundance reflects unmeasured environmental factors that also correlate with C. cheiranthifolia flower size, abrupt, parallel shifts in flower size and parasitism make a causal role for flower size (or closely correlated trait) a more parsimonious explanation. CONCLUSIONS This study provides a rare example of the reduced pollen limitation in selfing vs. outcrossing populations generally predicted by theoretical models and broad interspecific comparative analyses. Our results are consistent with a role for reproductive assurance in the transition from outcrossing to selfing, a hypothesis that requires further testing with transplant experiments. Large flowers and other traits promoting outcrossing correlate strongly with florivory by the larvae of a microlepidopteran Mompha species, which appears to account more directly than pollen limitation for the much lower fruit set in large-flowered populations of C. cheiranthifolia. However, the available evidence suggests that differential floral parasitism is probably a consequence, rather than a cause, of mating system evolution. The evolution of selfing may fundamentally change the ecology of plant species by influencing key demographic parameters (Lloyd, 1980; Cheptou, 2004). In particular, selfing may allow species to colonize marginal habitats and persist under low-density conditions in the absence of pollinators (Baker, 1955; Morgan et al., 2005; Eckert et al., 2006). Further, plants in small, isolated populations may avoid herbivory (e.g. Kery et al., 2001), resulting in an indirect association between selfing and reduced herbivory. In C. cheiranthifolia, however, SF populations are not smaller than LF populations (Samis and Eckert, 2007), suggesting a more direct effect of floral variation on herbivory that may commonly occur in flowering plants. Evolutionary shifts to selfing are associated with reduced floral and display size (Goodwillie et al., 2010) and hence resources for florivores. Reduced florivore fitness and, in turn, abundance may increase the ecological success of nascent selfing lineages. For instance, SF C. cheiranthifolia populations outnumber and are more widely distributed than LF populations (Samis and Eckert, 2007; Dart et al., 2012). This ‘release from florivory’ hypothesis could be evaluated with demographic analysis (Steets et al., 2007a). Reduced florivory may, in turn, select upon the types, levels and deployment (constitutive vs. inducible) of plant chemical defences, a possibility for which there is evidence at both micro- and macro-evolutionary scales (Campbell and Kessler, 2013). These speculations highlight the potential for relatively unexplored evolutionary feedbacks between pollination and herbivory (Hanley et al., 2009). Although Mompha florivory may not have caused the evolution of higher selfing in C. cheiranthifolia, it may still influence the mating system, especially in LF populations (e.g. Penet et al., 2009). High Mompha-induced flower loss might strongly reduce mate availability and pollinator visitation (Krupnick et al., 1999), and thus outcrossing opportunities. This may contribute to the substantial variation in outcrossing observed among LF populations, which remains unexplained by floral morphology (Dart et al., 2012), and might ultimately influence the maintenance of mixed selfing and outcrossing in largeflowered populations (Steets et al., 2007b). SUPPLEMENTARY DATA Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following. 1: list of study populations, their locations and the type of data collected from each. 2: simulation analysis of variation in fruit set and bud parasitism. 3: analysis of between-year variation in fruit set and floral parasitism. 4: analysis testing for resource reallocation between flowers within plants following pollen supplementation. 5: analysis of pollen supplementation’s effect on fruit set and seed production among large- and small-flowered populations. 6: analysis of pollen supplementation’s effect on fruit set and seed production among large- and small-flowered phenotypes in two mixed populations. ACKNOWLEDGEMENTS We thank Karen Samis, Emily Austen, Sarah Yakimowski, Colleen Inglis and Kyle Lauersen for help in the field; Alisa Yokum, Virginia Emery, Jeffrey Lam, Johanna McGlaughlin, Adam Kwok and William Mi for help in the lab; Peter Raven, Peter Hoch, Dave Hubbard and Jenny Dugan for logistic Dart & Eckert — Pollen limitation, florivory and mating system variation support and advice; Anna Hargreaves and Mick Hanley for comments on the manuscript; and the United States National Park Service, California State Parks and Oregon State Parks for research permits and logistic support. This work was supported by an Ontario Graduate Scholarships to S.D. and the Natural Sciences and Engineering Research Council of Canada through a Discovery Grant to C.G.E. LITERATURE CITED Adler LS, Bronstein JL. 2004. Attracting antagonists: does floral nectar increase leaf herbivory? Ecology 85: 1519–1526. Artz DR, Villagra CA, Raguso RA. 2010. Spatiotemporal variation in the reproductive ecology of two parapatric subspecies of Oenothera cespitosa (Onagraceae). American Journal of Botany 97: 1498–1510. Ashman TL. 2002. The role of herbivores in the evolution of separate sexes from hermaphroditism. Ecology 83: 1175–1184. Ashman TL, Knight TM, Steets JA, et al. 2004. Pollen limitation of plant reproduction: ecological and evolutionary causes and consequences. Ecology 85: 2408–2421. Baker HG. 1955. Self-compatibility and establishment after ‘long-distance’ dispersal. Evolution 9: 347–348. Barrett SCH, Morgan MT, Husband BC. 1989. The dissolution of a complex genetic polymorphism: the evolution of self-fertilization in tristylous Eichhornia paniculata (Pontederiaceae). Evolution 43: 1398–1416. Burd M. 1994. Bateman’s principle and plant reproduction: the role of pollen limitation in fruit and seed set. Botanical Review 60: 83–139. Busch JW. 2005. The evolution of self-compatibility in geographically peripheral populations of Leavenworthia alabamica (Brassicaceae). American Journal of Botany 92: 1503–1512. Button L, Villalobos AL, Dart SR, Eckert CG. 2012. Reduced petal size and color associated with transitions from outcrossing to selfing in Camissoniopsis cheiranthifolia (Onagraceae). International Journal of Plant Sciences 173: 251–260. Campbell SA, Kessler A. 2013. Plant mating system transitions drive the macroevolution of defense strategies. Proceedings of the National Academy of Sciences, USA 110: 3973–3978. Cheptou PO. 2004. Allee effect and self-fertilization in hermaphrodites: reproductive assurance in demographically stable populations. Evolution 58: 2613–2621. Dart S, Eckert CG. 2013. Experimental and genetic analyses reveal that inbreeding depression declines with increased self-fertilization among populations of a coastal dune plant. Journal of Evolutionary Biology 26: 587–599. Dart S, Samis KE, Austen E, Eckert CG. 2012. Broad geographic covariation between floral traits and the mating system in Camissoniopsis cheirantifolia (Onagraceae): multiple stable mixed mating systems across the species’ range? Annals of Botany 109: 599–611. Diggle PK. 1995. Architectural effects and the interpretation of patterns of fruit and seed development. Annual Review of Ecology and Systematics 26: 531–552. Eckert CG, Samis KE, Dart S. 2006. Reproductive assurance and the evolution of uniparental reproduction in flowering plants. In: Harder LD, Barrett SCH, eds. Ecology and evolution of flowers. Oxford: Oxford University Press, 183–203. Eckert CG, Kalisz S, Geber MA, et al. 2010. Plant mating systems in a changing world. Trends in Ecology and Evolution 25: 35–43. Emery VJ, Landry J-F, Eckert CG. 2009. Combining DNA barcoding and morphological analysis to identify specialist floral parasites (Lepidoptera: Coleophoridae: Momphinae: Mompha). Molecular Ecology Resources 9: 217–223. Fausto JA Jr, Eckhart VM, Geber MA. 2001. Reproductive assurance and the evolutionary ecology of self-pollination in Clarkia xantiana (Onagraceae). American Journal of Botany 88: 1794–1800. Galen C. 1999. Flowers and enemies: predation by nectar-thieving ants in relation to variation in floral form of an alpine wild flower, Polemonium viscosum. Oikos 85: 426–434. Geber M, Eckhart V. 2005. Experimental studies of adaptation in Clarkia xantiana. II. Fitness variation across a subspecies border. Evolution 59: 521–531. 325 Geyer CJ, Wagenius S, Shaw RG. 2007. Aster models for life history analysis. Biometrika 94: 415–426. Goodwillie C. 2001. Pollen limitation and the evolution of self-compatibility in Linanthus (Polemoniaceae). International Journal of Plant Science 162: 1283–1292. Goodwillie C, Sargent RD, Eckert CG, et al. 2010. Correlated evolution of mating system and floral display traits in flowering plants and its implications for the distribution of mating system variation. New Phytologist 185: 311–321. Hanley ME, Lamont BB, Armbruster WS. 2009. Pollination and plant defence traits co-vary in Western Australian Hakeas. New Phytologist 182: 251–260. Herrera CM, Sanchez-Lafuente AM, Medrano M, Guitian J, Cerda X, Rey PJ. 2001. Geographical variation in autonomous self-pollination levels unrelated to pollinator service in Helleborus foetidus (Ranunculaceae). American Journal of Botany 88: 1025–1032. Herrera CM, Medrano M, Rey PJ, et al. 2002. Interaction of pollinators and herbivores on plant fitness suggests a pathway for correlated evolution of mutualism- and antagonism-related traits. Proceedings of the National Academy of Sciences, USA 99: 16823–16828. Husband BC, Schemske DW. 1996. Evolution of the magnitude and timing of inbreeding depression in plants. Evolution 50: 54–70. Inoue K, Maki M, Masuda M. 1996. Evolution of Campanula flowers in relation to insect pollination on islands. In: Lloyd DG, Barrett SCH, eds. Floral biology. Studies on floral evolution in animal-pollinated plants. New York: Champman & Hall, 377–400. Jain SK. 1976. The evolution of inbreeding in plants. Annual Review of Ecology and Systematics 7: 69–95. Kennedy BF, Elle E. 2008. The reproductive assurance benefit of selfing: importance of flower size and population size. Oecologia 155: 469–477. Kery M, Matthies D, Fischer M. 2001. The effect of plant population size on the interactions between the rare plant Gentiana cruciata and its specialized herbivore Maculinea rebeli. Journal of Ecology 89: 418–427. van Kleunen M, Manning JC, Pasqualetto V, Johnson SD. 2008. Phylogenetically independent associations between autonomous selffertilization and plant invasiveness. American Naturalist 171: 195–201. Knight TM, Steets JA, Vamosi JC, et al. 2005. Pollen limitation of plant reproduction: pattern and process. Annual Review of Ecology, Evolution and Systematics 36: 467–497. Knight TM, Steets JA, Ashman TL. 2006. A quantitative synthesis of pollen supplementation experiments highlights the contribution of resource reallocation to estimates of pollen limitation. American Journal of Botany 93: 271–277. Krupnick GA, Weis AE, Campbell DR. 1999. The consequences of floral herbivory for pollinator service to Isomeris arborea. Ecology 80: 125–134. Larson BMH, Barrett SCH. 2000. A comparative analysis of pollen limitation in flowering plants. Biological Journal of the Linnean Society 69: 503–520. Linsley EG, MacSwain JW, Raven PH, Thorp RW. 1973. Comparative behavior of bees and Onagraceae V. Camissonia and Oenothera bees of cismontane California and Baja California. University of California Publications in Entomology 71: 1–68. Lloyd DG. 1980. Demographic factors and mating patterns in angiosperms. In: Solbrig OT, ed. Demography and evolution in plant populations. Oxford: Blackwell, 67–88. Lyons EE, Antonovics J. 1991. Breeding system evolution in Leavenworthia – breeding system variation and reproductive success in natural populations of Leavenworthia crassa (Cruciferae). American Journal of Botany 78: 270–287. McCall AC, Irwin RE. 2006. Florivory: the intersection of pollination and herbivory. Ecology Letters 9: 1351–1365. Moeller DA, Geber MA. 2005. Ecological context of the evolution of selfpollination in Clarkia xantiana: population size, plant communities, and reproductive assurance. Evolution 59: 786–799. Moeller DA, Geber MA, Eckhart VM, Tiffin P. 2012. Reduced pollinator service and elevated pollen limitation at the geographic range limit of an annual plant. Ecology 93: 1036–1048. Morgan MT, Wilson WG, Knight TM. 2005. Plant population dynamics, pollinator foraging, and the selection of self-fertilization. American Naturalist 166: 169–183. 326 Dart & Eckert — Pollen limitation, florivory and mating system variation Penet L, Collin CL, Ashman TL. 2009. Florivory increases selfing: an experimental study in the wild strawberry, Fragaria virginiana. Plant Biology 11: 38–45. Piper JG, Charlesworth B, Charlesworth D. 1986. Breeding system evolution in Primula vulgaris and the role of reproductive assurance. Heredity 56: 207–217. R Core Team. 2013. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www. R-project.org/ Rambuda TD, Johnson SD. 2004. Breeding systems of invasive alien plants in South Africa: does Baker’s rule apply? Diversity and Distributions 10: 409–416. Raven PH. 1969. A revision of the genus Camissonia. Contributions of the United States National Herbarium 37: 161–396. Samis KE, Eckert CG. 2007. Testing the abundant center model using rangewide demographic surveys of two coastal dune plants. Ecology 88: 1747–1758. Schueller SK. 2004. Self-pollination in island and mainland populations of the introduced hummingbird-pollinated plant, Nicotiana glauca (Solanaceae). American Journal of Botany 91: 672–681. Schultz BB. 1985. Levene’s test for relative variation. Systematic Zoology 34: 449–456. Shaw RG, Geyer CJ, Wagenius S, Hangelbroek HH, Etterson JR. 2008. Unifying life-history analyses for inference of fitness and population growth. American Naturalist 172: E35–E47. Stebbins GL. 1974. Flowering plants: evolution above the species level. Cambridge, MA: Belknap Press. Steets JA, Knight TM, Ashman TL. 2007a. The interactive effects of herbivory and mixed mating for the population dynamics of Impatiens capensis. American Naturalist 170: 113–127. Steets JA, Wolf DE, Auld JR, Ashman TL. 2007b. The role of natural enemies in the expression and evolution of mixed mating in hermaphroditic plants and animals. Evolution 61: 2043–2055. Strauss SY, Whittall JB. 2006. Non-pollinator agents of selection on floral traits. In: Harder LD Barrett SCH, eds. Ecology and evolution of flowers. Oxford: Oxford University Press, 120–138. Teixido AL, Mendez M, Valladares F. 2011. Flower size and longevity influence florivory in the large-flowered shrub Cistus ladanifer. Acta Oecologica 37: 418–421. Theis N, Adler LS. 2012. Advertising to the enemy: enhanced floral fragrance increases beetle attraction and reduces plant reproduction. Ecology 93: 430–435. Wesselingh RA. 2007. Pollen limitation meets resource allocation: towards a comprehensive methodology. New Phytologist 174: 26–37. Winn AA, Elle E, Kalisz S, et al. 2011. Analysis of inbreeding depression in mixed-mating plants provides evidence for selective interference and stable mixed mating. Evolution 65: 3339–3359. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM. 2009. Mixed effects models and extensions in Ecology with R. New York: Springer. Zuur AF, Hilbe JM, Ieno EN. 2013. A beginner’s guide to GLM and GLMM with R. Newburgh, UK: Highland Statistics Ltd.
© Copyright 2026 Paperzz