Experimental manipulation of flowers to determine the functional modes and fitness consequences of self-fertilization: unexpected outcome reveals key assumptions


Correspondence author. E-mail: chris.eckert@queensu.ca


  1. Reproductive assurance is often invoked to explain the repeated transition from outcrossing to selfing among plants. The most common experimental test of this hypothesis uses floral emasculation (anther removal) to manipulate the capacity of flowers for autogamous self-pollination.
  2. The coastal dune endemic Camissoniopsis cheiranthifolia exhibits striking variation in floral morphology and the mating system across its geographic range in California, and reproductive assurance is likely to have played a role in the transition from large flowers and predominant outcrossing to small flowers and predominant selfing. To compare variation in the timing of self-pollination and the extent to which it provides reproductive assurance in large- vs. small-flowered populations, we emasculated flowers at different times during anthesis and measured the effects on both seed production and the proportion of seeds self-fertilized vs. outcrossed estimated using marker genes. We also manipulated large- vs. small-flowered phenotypes within two morphologically variable populations.
  3. As predicted from floral development, small-flowered plants self-pollinated before flowers open, whereas large-flowered phenotypes did not. Comparing flowers that were emasculated before closing suggested that a seemingly well-developed mechanism for delayed selfing (anthers contacting the stigma when flowers close) does not increase seed production.
  4. Using emasculation to manipulate selfing assumes that removing anthers from flowers does not compromise outcrossing. However, while not affecting the capacity of flowers to produce seed, emasculation shortened floral longevity, thereby reducing opportunities for outcrossing. Genetic estimation of the mating system revealed a negative effect of emasculation on the proportion of seed outcrossed, even in small-flowered populations, further suggesting that flowers lacking anthers are unattractive to pollinators. These side effects of emasculation prevent a clear interpretation of the reproductive assurance value of selfing.
  5. Given that much of what we know about reproductive assurance in natural populations comes from emasculation experiments, our results suggest that the assumptions of this approach, which are rarely verified, require much more serious consideration.


An evolutionary transition from a reproductive system involving mating among unrelated individuals (outcrossing) to predominant self-fertilization has occurred thousands of times in plants. The most widely held hypothesis for why selfing evolves is that it provides reproductive assurance (RA) when pollinators and/or mates are scarce (Eckert, Samis & Dart 2006; Cheptou & Schoen 2007; Harder & Aizen 2010). The RA hypothesis can be tested experimentally by manipulating the capacity of flowers for self-pollination, usually by emasculation (anther removal) and then measuring the effect on seed production and the mating system (Schoen & Lloyd 1992; Schoen, Morgan & Bataillon 1996). The RA hypothesis is supported if seed production (F) and the proportion of seeds self-fertilized (s) are higher for intact flowers (I) capable of self-pollination than emasculated flowers (E) no longer capable of self-pollination (Eckert et al. 2010). Almost all of what we know about the selective benefits of RA in natural plant populations has come from these emasculation experiments, which reveal that selfing often provides RA in nature (Eckert, Samis & Dart 2006).

Emasculations can be performed at different stages of floral development to estimate the timing of self-pollination, which may strongly influence the costs and benefits of selfing (Lloyd 1992; Schoen & Lloyd 1992; Schoen, Morgan & Bataillon 1996). For instance, selfing that occurs at the end of floral life, after opportunities for outcrossing have passed (‘delayed selfing’), may provide RA without compromising the production of high viability outcrossed seed or outcross siring success through pollen export. In contrast, selfing occurring before opportunities for outcrossing (‘prior selfing’) or simultaneously with outcrossing (‘competing selfing’) may also increase total seed production (i.e. provide RA) but at the cost of reducing the number of outcrossed seeds produced (seed discounting) or seed sired on other plants (pollen discounting). However, few studies have varied the timing of emasculation to estimate the timing of selfing (Leclerc-Potvin & Ritland 1994; Griffin, Mavraganis & Eckert 2000; Zhang & Li 2008).

Using floral emasculation to estimate RA (i.e. RA = FI − FE) is based on the critical assumption that removing anthers from flowers only reduces the capacity of flowers for within-flower (autogamous) self-pollination. However, emasculation may also reduce seed production if it damages flowers, thereby reducing their capacity to mature seeds. It may also reduce seed production by diminishing opportunities for outcrossing by shortening floral life span via a wound-induced hastening of floral senescence (O'Neill 1997). The absence of anthers may also reduce visitation by pollinators if they avoid emasculated flowers because they seek pollen as a reward and/or because anthers are involved in pollinator attraction (Griffin, Mavraganis & Eckert 2000; Eckert & Herlihy 2004). Violations of this critical assumption overestimate RA but can be detected by comparing I vs. E flowers in terms of seed production after hand-pollination, floral life span and pollinator visitation. The later might be problematic because visitation rates are likely to be low under ecological conditions that might select for RA. However, the effect of emasculation on outcross-pollination can be inferred by comparing the proportion of seeds outcrossed (t) vs. selfed (s,= 1) estimated using genetic markers. The production of more outcrossed seeds by I flowers than E flowers (tIFI > tEFE) indicates that emasculation compromises opportunities for outcrossing. Few experimental studies that have used emasculation have addressed these assumptions (see Table S1, Supporting information).

In this study, we emasculate flowers to manipulate self-pollination and its timing, and then measure the effects on seed production and the mating system in Camissoniopsis cheiranthifolia (Onagraceae, Fig. 1). In doing so, we test for the potential side effects of emasculation that may complicate interpretation of experimental results. This species is a short-lived, herbaceous endemic of Pacific coastal dunes of western North America, from southern Baja California Mexico, through California to southern Oregon USA, and exhibits striking co-variation between floral morphology and the mating system across its geographic range (Raven 1969; Dart et al. 2012). Plants from populations in San Diego County California produce large, self-incompatible (SI) flowers (large-flowered, self-incompatible = LF-SI) that are primarily outcrossing (mean = 0·80, range = 0·62–0·99, = 3 populations, Dart et al. 2012). Further north along the mainland coast to Point Conception in northern Santa Barbara County California, plants produce large flowers that are fully self-compatible (LF-SC) and engage in a variable mixture of outcrossing and selfing (mean = 0·74, range = 0·47–0·96, = 9). North of Point Conception to the range limit in southern Oregon, in Baja California and on the Channel Islands, plants produce small, self-compatible flowers (SF-SC) that are predominantly self-fertilizing (mean = 0·24, range = 0·001–0·57, = 10).

Figure 1.

Flower of Camissoniopsis cheiranthifolia (Onagraceae). This flower is on a plant raised in the glasshouse from seed collected from a large-flowered population. Small flowers are very similar in morphology but half the diameter with little spatial separation between the stigma and dehiscing anthers. Emasculation involved removing the anthers but leaving the filaments.

Species like C. cheiranthifolia with wide mating system variation provide excellent opportunities to test predictions concerning the fitness costs and benefits of selfing, especially with respect to the RA hypothesis. Yet these species are underexploited in this regard (but see Elle & Carney 2003; Kennedy & Elle 2008). There seems to be at least two distinct mating system transitions in C. cheiranthifolia among populations along the mainland coast of California (Raven 1969). There is a transition from predominant outcrossing promoted by large SI flowers (LF-SI) in San Diego County to large but self-compatible flowers (LF-SC) and mixed mating (outcrossing + selfing) in somewhat more northerly populations. Theory suggests that the loss of SI but the retention of large flowers is likely via selection for RA (Porcher & Lande 2005). Although LF-SC flowers are self-compatible, dehiscing anthers are held well away from receptive stigmas (mean = 3·27 mm of herkogamy) throughout floral anthesis, which likely limits self-pollination before and during opportunities for outcross-pollination. However, flowers in these populations typically open for two consecutive days, and anthers and stigmas often come into contact when flowers close for the evening at the end of each day. The extent to which this occurs correlates positively with the proportion of seeds selfed (Dart et al. 2012). This potential mechanism of delayed selfing provides RA, while possibly limiting the costs of seed and pollen discounting (Lloyd 1992).

Populations further north along the California coast (north of Point Conception) have progressed further along the transition to higher s. Linsley et al. (1973) report that flowers of C. cheiranthifolia are primarily visited by pollen-collecting females and nectar-collecting males of several oligolectic species of Andrenid bees but that visitation to flowers in SF-SC populations is infrequent largely because heavy morning fog characteristic of this region impedes pollinator visitation. Hence, higher levels of selfing may have evolved to provide RA. SF-SC flowers seem to differ from LF-SC flowers in the timing of self-pollination. SF-SC flowers bear stigmas and anthers in close proximity throughout floral life such that self-pollen, whether it is shed on to stigmas autonomously or through the actions of visiting pollinators, likely competes with outcross pollen for fertilizations (competing selfing, sensu Lloyd 1992). SF-SC flowers also experience autogamous pollination before flowers open (prior selfing) because anthers dehisce shed pollen especially on the underside of the globular stigma while still in bud. This temporal shift in self-pollination to earlier in floral life may be a response to chronic outcross pollen limitation in dune habitats north of Point Conception (Lloyd 1992; Eckert et al. 2010).

We discovered two populations of C. cheiranthifolia in the transition zone just north of Point Conception that contain both LF-SC and SF-SC phenotypes that differ in floral traits and genetic estimates of s to about the same extent as LF-SC and SF-SC phenotypes in segregated populations (Dart et al. 2012). This within-population variation allowed us to use emasculations to investigate differences in the timing and fitness consequences of selfing when divergent phenotypes experience the same pollination environment.

This study compares LF-SC (hereafter LF) populations south of Point Conception to SF-SC (hereafter SF) populations to the north to address the following questions: (i) Is prior selfing that occurs before flowers open more prevalent in SF than LF populations due to self-pollination in the bud? (ii) Does autogamy occurring after flowers open provide RA by increasing seed production and s, and are these effects greater in LF populations where autogamy occurs after flowers open than in SF populations where self-pollination is expected to occur before flowers open? (iii) Does selfing occurring when flowers close increase seed production, and if so, are these effects greater in LF than SF populations? (iv) Do the differences between LF and SF populations in the timing and fitness consequences of selfing also occur between LF and SF phenotypes within populations? Addressing this question will indicate the extent to which differences are due to floral morphology and development as opposed to variation in the pollination environment and/or other extrinsic ecological factors that may differ between pure LF and SF populations. (v) Are there unintended side effects of emasculation that could complicate the interpretation of the results? In particular, we determine whether emasculation damages flowers and reduces their capacity to produce seed and reduces floral longevity and thus opportunities for outcrossing. We also use genetic estimates of self-fertilization and outcrossing to infer whether emasculation decreases cross-pollination through reduced pollinator visitation.

Materials and methods

Effect of emasculation and pollination on seed production

We used seven populations of C. cheiranthifolia (Table S2, Supporting information), including three LF-SC populations south of Point Conception (COR1C, CMG1C, CBV1C) two SF-SC populations north of Point Conception (CMS1C, CST1C), plus the two highly variable populations located just north of Point Conception (CGN1C, CSP1C) that include LF and SF phenotypes (Dart et al. 2012). Within each population, we randomly allocated individual plants to one of eight treatments (described in Table 1), although not all treatments were applied in all populations (sample sizes in Table S3, Supporting information). Four treatments involved open (i.e. natural) pollination, with individual flowers left intact (I) or emasculated just after opening (EO, during 05.00–08.00 h), emasculated just before closing on the first day of anthesis (EC, 15.00–18.00 h) or just before closing on the second day of anthesis (EC2, 15.00–18.00 h). Each flower contains four epipetalous and four episepalous stamens from which the anthers are easily removed without causing incidental self-pollination or any apparent damage to flowers (Fig. 1). For three additional treatments, I, EO and EC, flowers were hand-pollinated with self-pollen just after opening (+S). For the final treatment, flowers were emasculated at opening and excluded from pollinators (‘bagged’, EO + B). Exclusion bags were made of fine sheer nylon stretched across a wire frame cage to exclude even the smallest insect that we have observed visiting C. cheiranthifolia flowers. Hand self-pollination involved brushing the stigma with three dehisced anthers from the same flower. Treated flowers were marked with a spot of acrylic paint on the ovary. Mature fruits were collected 4–6 weeks after treatment, and the small filled seeds (~0·1 mg) in each fruit were counted and stored at 20–25 °C. We could not reliably count undeveloped ovules or seeds aborted during development because they were too small and difficult to distinguish from small fragments of ovary. To assess whether any of these experimental treatments altered floral longevity, we monitored individual flowers and recorded whether they re-opened for a second day of anthesis.

Table 1. Emasculation and pollination treatments used and their effects on the components of pollination and seed production (listed below). Based on the observations of floral morphology, development and display in C. cheiranthifolia, G is expected to be negligible. Small-flowered (SF) phenotypes should experience all forms of self-pollination (especially P). Large-flowered (LF) phenotypes should experience self-pollination primarily through D1 and D2. Parameters e and d capture the inadvertent side effects of emasculation on pollination and seed production capacity, respectively. Treatment combinations not applied are denoted as ‘na’
Emasculation treatmentOpen-pollinationFlowers self-pollinated by hand
  1. Seed production due to components of pollination: T, outcross-pollination; e, reduction in outcross-pollination due to emasculation at opening (subscript O) or closing on day 1 (subscript C1); P, ‘prior’ autogamous self-pollination before flowers open; S1, autogamous self-pollination during day 1; S2, autogamous self-pollination during day 2; D1, autogamous self-pollination when flowers close after day 1; D2, autogamous self-pollination when flowers close after day 2; G, geitonogamous (between-flower) self-pollination; d, reduction in seed production due to damage (see right); F, Seed production of flower exposed to a combination of hand self-pollination plus natural cross- and self-pollination; d, reduction in seed production capacity caused by damage due to emasculation when flowers open (subscript O), at the end of day 1 (subscript C1), or the end of day 2 (subscript C2).



S1 + D1 + S2 +D2 + G



At opening then bagged


− dO

At opening


(− eO) + − dO


− dO

At closing after day 1


(− eC1) + S1 + G − dC1


− dC1

At closing after day 2


S1 + D1 + S2 + G − dC2


Our questions were addressed by contrasting sets of treatments (Table 1) as follows:

  1. We evaluated the capacity for prior selfing (P in Table 1) by comparing seeds/flower between EO + B flowers (that could only engage in prior selfing) and EO + S flowers (for which seed production via selfing should be maximized) in the two SF populations, the two variable populations but only one LF population (COR1C) due to time constraints. For a subsample of flowers, we also determined whether pollen had been deposited on stigmas before anthesis by inspecting stigmas with a hand lens in flowers that were just opening. In a pilot study, we tried to more directly estimate the contribution of P by emasculating flowers in the bud stage before anther dehiscence but this clearly damaged flowers.
  2. We evaluated the contribution of selfing after flowers opened to seed production in all seven populations by contrasting I flowers (all forms of autogamous self-pollination [S1 + D1 + S2 + D2] plus geitonogamous between-flower self-pollination [G] plus outcross-pollination [T], details in Table 1) with EC flowers (all forms of autogamy for the first day of anthesis but none when the flower closes for the night or during subsequent days of anthesis [S1] plus T) and EO flowers (only prior autogamy [P] plus T).
  3. We determined the contribution of self-pollination when flowers closed for the first evening (D1) to seed production by contrasting EC flowers (S1 + G T) and EC2 flowers (all autogamy for the first and second days of anthesis but no autogamy when the flower closed for the second night: S1 + D1 + S2 + G T) in one LF population (COR1C), two SF populations and the two variable populations. This contrast overestimates the contribution of D2 because compared to EC flowers, EC2 flowers also experience S, G and T occurring during day 2. However, this did not complicate the interpretation of our results.
  4. We determined whether emasculation reduced seed production capacity by damaging flowers (d) by contrasting the number of seeds produced per flower (F) between I + S (no d), EC + S (dC1) and EO + S (dO) flowers in all populations.
  5. We investigated whether emasculation reduced opportunities for cross-pollination (e) by contrasting the proportion of flowers opening for a second day (floral longevity) between I + S (no e), EC + S (eC1) and EO + S (eO) treatments for all populations. We could not determine whether emasculation reduced pollinator attraction (another component of e) because pollinator visitation rates were extremely low in all populations we studied.

Seeds/flower was highly skewed because some flowers failed to set fruit, which commonly occurs, especially in LF populations (Dart & Eckert in press). Hence, we contrasted treatments using the ASTER model (Shaw et al. 2008) implemented in the R statistical software (version 2.13.0, R Core Development Team 2011). ASTER uses a maximum likelihood algorithm that accounts for seeds per flower consisting of a Bernoulli variable (whether a flower set a fruit) compounded with a zero-truncated Poisson variable (number of seeds per mature fruit). The analysis uses all the data, including flowers that failed to set a fruit (zero seeds/flower) and allows estimation of the unconditional expectations (analogue of the mean) and 95% confidence intervals for each treatment (Geyer, Wagenius & Shaw 2007). Population was not included as a random factor nested within flower size class (LF vs. SF) because ASTER does not accommodate random factors (Shaw et al. 2008). However, supplementary analyses (not shown) including population as a fixed factor rather than population flower size did not reveal heterogeneity among populations within flower size categories that would complicate the interpretation of population flower size effects. Floral longevity, as measured by whether a flower opened for the second day, and our observations of whether pollen had been deposited on stigmas at anthesis were analysed as binary response variables using generalized linear models (GLM) with a binomial error distribution implemented in R.

Our analyses of seeds/flower reported below indicated higher average seed production for SF than LF populations. We could not directly determine whether this was due to flowers from SF populations containing more ovules because total ovule number could not be reliably estimated for the flowers used in our experiment. To investigate differences in ovule number/flower between LF and SF populations, we sampled 2–10 freshly opened flowers (mean = 5) from each of eight LF and 17 SF populations within the latitudinal range of the populations used in this study (33·6–36·9°N; = 165 flowers total), preserved the flowers in 70% ethanol and then dissected the ovary of each flower and counted the ovules within. Ovule number was averaged across the flowers from each population, and these means were contrasted between LF and SF populations using Welch's t-test. On average, ovaries of flowers from LF populations contained 15% more ovules (mean ± 1 SD = 85·75 ± 7·42) than those from SF populations (74·45 ± 11·88; = 3·37, d.f. = 21, = 0·0029). This difference does not complicate the interpretation of our results because we are primarily interested in the effects of emasculation and pollination treatments within flower size classes, not the average differences between flower size classes.

For both GLM and ASTER models, we initially included population flower size (LF, SF), emasculation/pollination treatment (Table 1) and their interaction as potential predictors. For models contrasting treatments between LF and SF phenotypes within the two variable populations, we initially included population, floral phenotype (LF, SF), treatment and all possible interactions. We tested the significance of each effect using likelihood ratio tests to compare models with and without the effect in question, with degrees of freedom equal to the difference in number of parameters between competing models. We tested interactions between main effects by comparing models with and without the interaction in question. Individual main effects were then tested by comparing models without the main effect to a model with all main effects but no interaction.

Effect of emasculation on the mating system

The proportion of seeds self-fertilized (s) was estimated for I and EO flowers in two LF populations (COR1C and CBV1C), two SF populations (CMS1C and CST1C) and for LF and SF phenotypes within two variable populations (CGN1C and CSP1C) from segregation of seven polymorphic allozyme loci in open-pollinated progenies assayed as seedlings following Dart et al. (2012). Using the mixed mating model as implemented by the maximum likelihood (ML) program MLTR (Ritland 2002), we estimated s simultaneously for each treatment group within each population by assuming that all plants within a population experienced a common outcross pollen pool. Pollen and ovule allele frequencies were constrained to be equal, parental genotypes were inferred from progeny genotypes, and Newton–Raphson iteration was used to find ML values. Ninety-five percentage of confidence limits for the ML estimates were derived as the 2·5- and 97·5 percentile of the distribution of 1000 bootstrap values generated using the seed family as the unit of re-sampling. For all populations, estimates of s converged on a single ML value regardless of initial parameter values. We contrasted estimates of s between I and EO flowers by calculating the proportional overlap between bootstrap distributions, which is roughly equivalent to a P-value. ML estimates were considered significantly different if < 0·025.


Capacity for prior selfing revealed by emasculation and bagging

Flowers emasculated at opening and then excluded from pollinators (EO + B) set fewer seed than emasculated flowers self-pollinated by hand (EO + S, Fig. 2a), but the difference was much larger in the LF than the two SF populations as revealed by a significant interaction between population flower size and treatment (= 0·010, see Table S4A, Supporting information). In SF populations, EO + B flowers produced, on average, 60% as much seed as EO + S flowers. In contrast, EO + B flowers in the one LF population (COR1C) set very few seed. We also subjected a small sample of plants to the EO + B treatment in the two other LF populations, and only two of 12 EO + B flowers set fruit, one fruit with a single seed and the other with four seeds. Among the 120 stigmas examined in three LF populations, only 9·2% bore pollen at anthesis compared to 83·6% of 220 stigmas sampled in two SF populations (GLM effect of flower size: likelihood ratio LR = 213·3, d.f. = 1, < 0·00001).

Figure 2.

Variation in the capacity for self-pollination in the bud stage (prior selfing) in Camissoniopsis cheiranthifolia. Unconditional expectations of seeds per flower with 95% confidence intervals calculated using the ASTER model are contrasted between flowers that were emasculated at opening and self-pollinated by hand (EO + S) and flowers that were emasculated at opening and then excluded from pollinators by bagging (EO + B). Panel (a) contrasts treatment effects between one large-flowered and two small-flowered populations (= 234 flowers). Panel (b) contrasts large- and small-flowered phenotypes within two variable populations (= 122 flowers). Analyses of data summarized in this figure are in Table S4 (Supporting information).

The results from the two variable populations were similar (Fig. 2b). EO + B flowers set far fewer seed than EO + S flowers but only among LF phenotypes. However, the difference between treatments was larger in CGN1C than in CSP1C where EO + B LF flowers set considerable seed, resulting in a three-way interaction plus two-way interactions involving treatment (all < 0·003, Table S4B, Supporting information). Among the 327 stigmas examined on flowers from LF phenotypes, 18·9% bore pollen at anthesis compared to 78·7% of 225 stigmas from SF flowers. Again, however, the difference between phenotypes was smaller for CSP1C (31·7% vs. 80·2%) than CGN1C (4·7% vs. 76·9%; GLM interaction between population and flower size: LR = 21·2, d.f. = 1, < 0·00001).

Effect of emasculation after anthesis on seed production of open-pollinated flowers

Given that SF flowers are self-pollinated in the bud, selfing occurring after flower opening should make a larger contribution to seed production for LF than SF plants. Consistent with this prediction, emasculation at flower opening reduced seed production to a much greater extent in LF than SF populations (Fig. 3a, interaction < 0·00001, Table S5A, Supporting information). In contrast, emasculation just before flowers closed at the end of the first day did not affect seed production in LF or SF populations. Analysis of just I and EC flowers did not detect an effect of treatment (LR = 0·21, d.f. = 1, = 0·65) but did reveal a small but significant interaction (LR = 11·83, d.f. = 1, = 0·00058) that resulted from EC flowers producing fewer seeds than I flowers in SF but not LF populations.

Figure 3.

Comparison of seed production between intact and emasculated flowers in Camissoniopsis cheiranthifolia. Unconditional expectations of seeds per flower with 95% confidence intervals calculated using the ASTER model are contrasted between intact flowers (I), those emasculated at opening (EO) and those emasculated just before closing on the first day of anthesis (EC). Panel (a) contrasts treatment effects between three large-flowered and two small-flowered populations (= 483 flowers). Panel (b) contrasts large- and small-flowered phenotypes in two variable populations (= 143 flowers). Analyses of data summarized in this figure are in Table S5 (Supporting information).

Again, the pattern of variation for the two variable populations was more complex as indicated by a significant three-way interaction between population, flower size and treatment (Fig. 3b, = 0·0010, Table S5B, Supporting information). Both emasculation treatments reduced seed production but more so among LF than SF plants, and the interaction between flower size and treatment was particularly strong (= 0·00038). Analysing LF and SF plants separately revealed strong effects of population (LR = 10·28, d.f. = 1, = 0·0013), treatment (LR = 28·75, d.f. = 1, < 0·00001) and an interaction between these effects (LR = 17·94, d.f. = 1, = 0·00013) for LF plants, whereas none of these terms were significant for SF plants (all > 0·15).

Effects of emasculation after 1 vs. 2 days of anthesis

Seed production was generally higher in SF populations (unconditional expectation = 54·39 seeds/flower, 95% confidence interval = 55·64–53·15) than the LF population (27·85, 31·81–23·88 seeds/flower, < 0·00001) despite ovaries from SF populations containing fewer ovules (mean = 74·4) than those from LF populations (mean = 85·7, see 'Materials and methods'). However, there was no difference (= 0·26) between flowers emasculated at the closing of the first day (EC) and those emasculated at the closing of the second day (EC2), and no interaction between population flower size and treatment (= 0·15, Fig. 4a, Table S6A, Supporting information).

Figure 4.

Comparison of seed production between flowers emasculated at the end of the first day of anthesis (EC) vs. the second day of anthesis (EC2) in Camissoniopsis cheiranthifolia. Unconditional expectations of seeds per flower with 95% confidence intervals calculated using the ASTER model are presented. Panel (a) contrasts treatment effects between one large-flowered and two small-flowered populations (= 187 flowers). Panel (b) contrasts large- and small-flowered phenotypes in two variable populations (= 91 flowers). Analyses of data summarized in this figure are in Table S6 (Supporting information).

For the variable populations, there was a strong interaction between flower size and treatment (< 0·00001, Table S6B, Supporting information): EC2 flowers produced more seed than EC flowers among LF plants but a weaker and opposite trend occurred among SF plants (Fig. 4b). Analysing LF and SF plants separately revealed a strong effect of treatment (LR = 29·70, d.f. = 1, < 0·00001) and an interaction between population and treatment (LR = 11·43, d.f. = 1, P = 0·00072) for LF plants, whereas none of these terms were significant for SF plants (all > 0·07).

Side effects of emasculation: floral longevity and seed production

When flowers were self-pollinated by hand at opening to control the timing and intensity of pollination, seed production, although higher (< 0·00001) in SF populations (57·66, 59·09–56·23 seeds/flower) than LF populations (48·18, 50·81–45·56 seeds/flower), did not differ between treatments (= 0·60, Fig. 5a, Table S7A, Supporting information). Results were complicated within the two variable populations by a two-way (< 0·00001) and a three-way interaction (= 0·0010), but there was no consistent reduction of seeds/flower associated with either emasculation treatment (Fig. 5b, Table S7B, Supporting information).

Figure 5.

Comparison of seed production (a, b) and floral longevity (c, d) among treatments applied to flowers of Camissoniopsis cheiranthifolia. Both response variables are contrasted between intact flowers (I + S) and those emasculated at opening (EO + S) or just before closing on the first day (EC + S) that were self-pollinated by hand at opening. Panels (a) and (c) contrast treatment effects between three large-flowered and two small-flowered populations (seeds/flower: = 489 flowers; longevity: = 275). Panels (b) and (d) contrast large- and small-flowered phenotypes in two variable populations (seeds/flower: = 144 flowers; longevity: = 320). Points are expected values (means) and error bars are 95% confidence intervals back-transformed from the statistical model used to analyse the data. Analyses of data summarized in this figure are in Tables S7 and S8 (Supporting information).

In contrast, the longevity of flowers pollinated by hand at opening was lower in SF than LF populations and was reduced substantially among EO + S compared to I + S or EC + S flowers even though all flowers in these treatment groups were hand-pollinated at the same stage of floral development (Fig. 5c, Table S8A, Supporting information, < 0·00001). Although emasculation seemed to reduce floral longevity more in LF than SF populations, the interaction term was not significant (= 0·67). We observed similar differences between treatments and between SF and LF phenotypes within the two variable populations, although the magnitude of the difference between floral phenotypes differed between populations (interaction = 0·022, Fig. 5d, Table S8B, Supporting information).

Effect of emasculation on proportion of seeds self-fertilized

Emasculating flowers should reduce the capacity for autogamous self-pollination and thus will generally reduce the proportion of seeds self-fertilized (s). In contrast, maximum likelihood estimates of s were higher for emasculated (EO) than intact (I) flowers in one LF and two SF populations, and the difference was significant in both SF populations but not the LF population (Table 2). A similar unexpected result occurred in one of the two variable populations (Table 2). EO flowers had higher estimates of s than I flowers for both LF and SF phenotypes in CGN1C, and the difference for SF flowers neared significance (= 0·067). In CSP1C, estimates of s were slightly but not significantly lower for EO than I flowers. We used these estimates of s to partition total seed production into outcrossed and self-fertilized seeds for both I and EO flowers separately (Fig. 6). In all but one case (SF plants in population CSP1C), I flowers made more outcrossed seed than EO flowers and, in some populations, many more.

Table 2. Comparison of estimated proportion of seeds self-fertilized (s) between intact flowers (I) and those emasculated as flowers opened (EO) in two large-flowered (LF), two small-flowered (SF) and two phenotypically variable populations of Camissoniopsis cheiranthifolia
PopulationFlower size categoryFamilies, progenyProportion seeds self-fertilized (s) P
Intact (I)Emasculated (EO)
  1. Sample sizes in terms of seed families and progeny assayed for seven polymorphic allozyme loci are presented along with maximum likelihood estimates of s with 95% confidence limits in brackets and a P-value comparing proportional overlap of bootstrap distributions. Estimates for which < 0·025 are significantly different.

COR1CLF37, 2810·456 (0·33 to 0·60)0·332 (0·14 to 0·49)0·091
CBV1CLF23, 1670·200 (−0·01 to 0·50)0·339 (0·15 to 0·46)0·16
CMS1CSF60, 4920·381 (0·13 to 0·60)0·911 (0·79 to 0·99)<0·001
CST1CSF62, 4800·312 (0·17 to 0·76)0·838 (0·71 to 0·95)0·011
CGN1CVariable LF36, 2610·046 (0·55 to 0·91)0·416 (0·70 to 0·94)0·17
Variable SF35, 2380·722 (−0·2 to 0·20)0·817 (0·02 to 0·65)0·067
CSP1CVariable LF19, 1460·134 (0·01 to 0·32)0·128 (−0·20 to 0·75)0·41
Variable SF26, 2030·874 (0·75 to 0·95)0·759 (0·62 to 0·95)0·15
Figure 6.

Estimated numbers of outcrossed vs. self-fertilized seed produced by intact (I) vs. emasculated (EO) flowers in Camissoniopsis cheiranthifolia. Panel (a) contrasts two large-flowered (LF) and two small-flowered (SF) populations. Panel (b) contrasts large- and small-flowered phenotypes within two variable populations. The mean number of seeds produced per flower for each treatment in each population was partitioned into self-fertilized vs. outcrossed using the maximum likelihood estimates of proportion seeds self-fertilized (s, proportion seeds outcrossed = 1 − s) from Table 2.


The interpretation of results from experimental emasculations is based on the assumption that emasculation only reduces the capacity for autogamous self-pollination and does not reduce seed set by damaging flowers or compromising opportunities for outcrossing. Our results suggest that although emasculation did not affect the capacity of C. cheiranthifolia flowers to mature seeds, emasculating flowers when they opened (EO) significantly reduced floral longevity, which may have reduced outcross pollen receipt. Moreover, emasculation did not reduce the proportion or absolute number of seeds self-fertilized. In fact, the opposite was observed, suggesting again that emasculation reduced opportunities for outcrossing. Later we discuss the conclusions that can be safely drawn from our results and identify those for which these unwanted side effects of emasculation complicate interpretation.

Variation in the timing of selfing among populations

Our results yield two conclusions concerning variation in the timing of selfing among populations of C. cheiranthifolia that are not complicated by side effects of emasculation. First, our comparison of EO + B with EO + S flowers and analysis of pollen on stigmas at anthesis confirmed qualitative observations of floral morphology and development by showing that flowers in SF populations have a much greater capacity for prior selfing than those from LF populations. The same was largely true for SF vs. LF phenotypes within variable populations, although both the bagging treatment and examination of pollen on stigmas suggest that LF phenotypes in population CSP1C have some capacity for prior selfing. The only other study on a species exhibiting the kind of broad floral variation seen in C. cheiranthifolia also found that small-flowered phenotypes engaged in more prior selfing due to earlier stigma receptivity and reduced herkogamy (Elle et al. 2010). These results suggest that the evolutionary differentiation in flower size involved a major change in the timing of self-pollination. Theory suggests that selfing earlier in floral life is only favoured under conditions of chronic outcross pollen limitation because prior selfing is likely to incur the costs of seed and pollen discounting (Lloyd 1992). Linsley et al. (1973) suggested that pollinators of C. cheiranthifolia (mostly oligolectic solitary bees) are scarce north of Point Conception California because the heavy morning fog characteristic of that coastal region severely reduces bee foraging. It is conceivable that these conditions result in chronic pollen limitation that may have played a role in spurring selection to favour the shift to prior selfing. However, as we argue in greater detail later, although pollinator-mediated outcrossing may be, on average, less frequent north of Point Conception, plants sometimes experience periods of substantial cross-pollination.

Contact between stigmas and anthers at flower closing is common in LF C. cheiranthifolia and correlates positively with the proportion of seeds selfed (Dart et al. 2012). It seemed likely that this could be a mechanism of delayed selfing. Delaying selfing until after opportunities for outcrossing should be favoured under a wide range of conditions because it provides RA while largely avoiding the costs of seed and pollen discounting (Lloyd 1992; Schoen, Morgan & Bataillon 1996). In other words, it provides the ‘best of both worlds’ by not interfering with outcross fertilization when pollinators and mates are present while providing RA when they are scarce (Goodwillie, Kalisz & Eckert 2005). Hence, delayed selfing mechanisms may often evolve in populations that retain traits like large flowers typically associated with outcrossing (Cruden & Lyon 1989). However, results from the two studies that have experimentally measured the fertility contribution of a delayed selfing mechanism in natural populations are mixed. Ablating the mechanism of delayed selfing did not reduce seed set in Mimulus guttatus (Leclerc-Potvin & Ritland 1994) and only affected fertility in Bulbine vagans during inclement weather (Vaughton & Ramsey 2010).

Our comparison of EC with I and EC with EC2 flowers likely provides a reliable assessment of the fertility consequences of delayed selfing because the EC treatment did not reduce the capacity for seed production or floral longevity. Given this, our results do not support a fitness benefit to delayed selfing in LF populations of C. cheiranthifolia. In SF populations, however, EC and EC2 flowers set fewer seed than I flowers suggesting an unexpected though small benefit of delayed selfing. The results from the two variable populations contrast somewhat with those from pure LF and SF populations. For instance, neither EC nor EC2 treatments reduced seed production when applied to SF phenotypes, and both EO and EC treatments reduced seed production among LF phenotypes. In population CSP1C, EC2 flowers did not produce fewer seeds than I flowers, suggesting that autogamy during day 2 of anthesis, which occurred in I and EC2 but not EC flowers, may have contributed to seed production. In population CGN1C, both EC and EC2 flowers produced fewer seed than I flowers, suggesting a contribution of selfing at the end of floral life (flowers typically open for only 2 days in these variable populations, Dart et al. 2012). Although, the fertility consequences of stigma-anther contact during flower closing seemed to vary with the combination of floral morphology and ecological context in variable populations, delayed selfing did not provide detectable RA in pure LF populations where we expected to find it.

Reproductive assurance?

Most manipulative experiments contrasting I vs. E flowers have supported the RA hypothesis. In Table S1 (Supporting information), we summarize the results of the 47 studies involving 49 species from 25 families published to date. Emasculation reduced seed production in at least one if not all of study populations for 48 of 49 species. At face value, our comparison of seed production between I and EO flowers in C. cheiranthifolia suggests that selfing occurring after anthesis provides RA in LF populations and increases seed production to a greater extent in LF than SF populations where prior selfing likely predominates. However, contrasting floral life span and genetic estimates of selfing between I and EO flowers also suggests that emasculating flowers of C. cheiranthifolia right when they open compromised opportunities for outcrossing in addition to reducing autogamous self-pollination. As a result, it is unrealistic to derive estimates of RA or seed discounting for C. cheiranthifolia.

Although generally recognized (Schoen & Lloyd 1992), unwanted side effects of emasculation are rarely quantified. Of the 47 studies in Table S2 (Supporting information), the authors of only 27 (57%) acknowledged these side effects and only 22 studies (47%) quantified the effect of emasculation on at least one of seed production capacity, floral longevity or pollinator visitation. The effect on seed production capacity was most commonly acknowledged and tested (30 acknowledged, 21 tested), probably because fruit and seed set are the typical response variables in these experiments. Consistent with our results from C. cheiranthifolia, 20 of these 21 studies failed to detect an effect of emasculation on the seed set of hand-pollinated flowers. Only nine studies acknowledged the potential effect of emasculation on floral life span, only five tested for it, and none found that emasculation decreased floral life span. This contrasts with our results showing that emasculation at flower opening (but not flower closing) reduced the probability that a flower would open on subsequent days. The potential effect of emasculation on pollinator visitation was acknowledged in 25 studies but was tested in only 11, purportedly because of low pollinator visitation rates. Infrequent pollinator visitation also prevented us from directly testing this assumption in the populations of C. cheiranthifolia we studied. This may often be the case in species for which selfing is suspected of being selected because it provides RA. However, only one of 11 studies detected discrimination against emasculated flowers by flower visitors. Although we could not observe visitors directly, such discrimination seems likely in C. cheiranthifolia based on unexpectedly higher selfing (lower outcrossing) by EO than I flowers in some populations. Reduced outcrossing by EO flowers in LF populations might be adequately explained by reduced floral longevity. However, strongly reduced outcrossing also occurred in SF populations where the effect of emasculation on floral longevity seemed much smaller because relatively few SF flowers open for >1 day. Given the propensity for prior selfing by SF flowers, anything that impedes the prompt delivery of outcross pollen to stigmas may dramatically reduce the proportion of seeds outcrossed.

Reduced pollinator visitation of emasculated flowers would be expected if pollinators seek pollen as a reward or use anthers as an attractive cue and anthers are obviously exerted from flowers, as in C. cheiranthifolia. Linsley et al. (1973) report that pollen-collecting females and nectar-collecting males of several species of Andrenid bees are the most common visitors to flowers of C. cheiranthifolia, and it is quite possible that these visitors are sensitive to the absence of anthers in emasculated flowers. For instance, Miller (1981) found that pollen-collecting bumble bees (Bombus spp.) avoided emasculated flowers of Aquilegia caerulea, while nectar-seeking moths did not. In contrast, Eckert (2000) did not detect any discrimination between I and E flowers by bumble bees foraging for both nectar and pollen on flowers of Decodon verticillatus. Additional studies that link pollinator discrimination to reward type are required to make generalizations here.

Pollinator-mediated outcrossing evident in small-flowered populations

There has been much debate and theoretical investigation concerning whether mixed mating systems, where individual engage in a mix of outcrossing and selfing, are evolutionarily stable (reviewed in Goodwillie, Kalisz & Eckert 2005). Although a small proportion of C. cheiranthifolia populations exhibit what is generally classified as predominant outcrossing (> 0·8) or predominant selfing (> 0·8), most populations exhibit mixed mating systems, and there is much variation in the mating system among both LF and SF populations (Dart et al. 2012). Of particular interest is the occurrence of often substantial outcrossing in some SF populations. More than 60% of seeds were outcrossed in I flowers in the two SF populations studied here (mean = 0·65). In fact SF flowers, as a consequence of their higher total seed production (in spite of containing 15% fewer ovules than flowers from LF populations, see 'Materials and methods'), typically produce more outcrossed seeds than LF flowers (Fig. 6). Unexpectedly, the proportion of seeds outcrossed dropped to <20% in emasculated flowers (mean = 0·13), and we have suggested that this was due to pollinators avoiding emasculated flowers. Despite prior self-pollination, there seems to be considerable opportunity for outcrossing in SF populations as long as pollinators deliver outcross pollen promptly, which apparently does not occur when flowers are emasculated. Hence, the results of this study add to evidence that SF populations of C. cheiranthifolia exhibit an evolutionarily stable mixed mating system involving predominant but not complete self-fertilization. This hypothesis is also supported by an analysis of variation in floral morphology within the genus. Most of the other species of Camissoniopsis seem to have progressed further down the evolutionary pathway to full self-fertilization than the SF populations of C. cheiranthifolia (Dart et al. 2012).

What maintains outcrossing in SF populations of C. cheiranthifolia? Inbreeding depression is widely viewed as the primary selective factor preventing the transition to complete selfing (Goodwillie, Kalisz & Eckert 2005; Winn et al. 2011). However, greenhouse experiments comparing lifetime performance of selfed vs. outcrossed progeny did not detect any inbreeding depression for SF populations of C. cheiranthifolia, even under simulated drought conditions (Dart & Eckert in press). In the absence of inbreeding depression, almost all theoretical models suggest that populations should evolve towards very high levels of self-fertilization. The few models that provide an explanation for how any outcrossing can be maintained in populations with little or no inbreeding depression invoke trade-offs between self-fertilization and outcrossing through female or male function (Porcher & Lande 2005; Johnston et al. 2009). Episodes of substantial outcross-pollination in SF populations of C. cheiranthifolia may set the stage for such trade-offs.


The authors thank Kyle Laursen, Jeffrey Lam, Johanna McLaughlin, Adam Kwok and William Mi for help; Peter Raven, Peter Hoch, Dave Hubbard and Jenny Dugan for logistic support and advice; Elizabeth Elle for very helpful comments on the manuscript; California State Parks for research permits and logistic support; and the Natural Sciences and Engineering Council of Canada for a Discovery Grant to C.G.E.