Pollinators, mates and Allee effects: the importance of self-pollination for fecundity in an invasive lily


Correspondence author. E-mail: rodgerjg@gmail.com


  1. Ability to self-fertilize is correlated with invasiveness in several introduced floras, and this has been attributed to its mitigating effect on fecundity when pollinator visitation and mate availability are inadequate. Cross-pollination opportunities are expected to be most limited in isolated individuals and small populations, both typical of the leading edge of an invasion. Thus, self-pollination may promote invasion in part by mitigating pollen-limitation Allee effects.

  2. We used emasculation and pollen supplementation experiments to test whether the importance of self-pollination for fecundity increased as plant abundance decreased and isolation increased, in the hawkmoth-pollinated and autonomously self-pollinating invasive lily Lilium formosanum, in its introduced range in KwaZulu-Natal, South Africa. As inbreeding depression is negligible in these populations, seed production through selfing is likely to be demographically important.

  3. In naturalized populations of L. formosanum, varying in size and degree of isolation, emasculation reduced seed production by two-thirds, indicating strong reliance on self-fertilization for fecundity due to inadequate pollinator visitation. However, this was not related to population size and was only greater for more isolated populations in one of the 3 years in which the experiment was carried out. Pollen supplementation experiments showed that pollen limitation was low – 12% on average – and significant in only one of 3 years, demonstrating that autonomous self-pollination was highly effective.

  4. In artificial arrays, consisting of plants placed inside naturalized populations or in pairs isolated (3–702 m) from populations, the effect of emasculation on fecundity was greater in isolated plants than those inside the population in one of two populations. Isolation reduced fecundity when emasculated plants were placed next to a second emasculated plant, but not when emasculated plants were partnered with an intact plant, from which they could receive pollen.

  5. We conclude that self-fertilization in L. formosanum compensates for inadequate pollinator visitation across all levels of population size and for a pollen-limitation Allee effect due to decreased mate availability in isolated plants, and may thus play an important role in invasion.


Insufficient fecundity may prevent invasion entirely or reduce rate of spread of introduced species (Parker 1997), especially when it arises as an Allee effect (Veit & Lewis 1996; Leung, Drake & Lodge 2004; Taylor et al. 2004). An Allee effect occurs when low abundance reduces fecundity or any other aspect of performance, often resulting in population growth rate becoming negative (Stephens, Sutherland & Freckleton 1999). Inadequate pollen receipt (pollen limitation) is a common cause of Allee effects in plants that cannot self-fertilize (Knight et al. 2005; Gascoigne et al. 2009). One of the main reasons for this is that smaller, sparser and more isolated patches of plants are less likely to be discovered by animal pollinators and, if discovered, are less profitable for foraging (Sih & Baltus 1987; Feinsinger, Tiebout & Young 1991; Ågren 1996; Groom 1998). Moreover, when a plant species occurs at very low density, pollinators are likely to carry less or none of its pollen, even if visitation per plant is not reduced (Duncan et al. 2004b). Thus, ability to self-pollinate may enhance invasiveness in plants by mitigating Allee effects.

Ability to self-fertilize generally reduces or eliminates pollen limitation (Kalisz, Vogler & Hanley 2004; Knight et al. 2005; Eckert, Samis & Dart 2006; Brys et al. 2011). Ability to self-fertilize should provide an ecological (and evolutionary) advantage through increased reproduction, so long as the benefits of selfing in terms of fecundity (reproductive assurance benefits) outweigh the costs of inbreeding depression. Inbreeding depression occurs when progeny arising from selfing performs less well than those from outcrossing (Jain 1976). As self-fertilization reduces the availability of ovules and pollen for outcrossing, less-fit selfed progeny may be produced at the expense of fitter outcrossed progeny (gamete discounting; Lloyd 1992; Herlihy & Eckert 2002). As a result, ability to self-fertilize is most likely to be advantageous when inbreeding depression and opportunities for outcrossing are both low.

Herbert Baker proposed that plants that can self-fertilize should be better colonists than those that cannot, because selfing would allow single individuals, isolated from mates and pollinators by long distance dispersal, to found new populations (Baker 1955, 1967). This principle, known as Baker's law or rule, has subsequently been expanded to include invasive species. As introduced plants have to reproduce successfully at low abundance along the leading edge of an invasion, reproductive assurance through self-fertilization should increase their likelihood of becoming invasive. In other words, selfing may contribute to invasiveness by mitigating pollen-limitation Allee effects (van Kleunen, Fischer & Johnson 2007; Ward, Johnson & Zalucki 2012). This idea is consistent with evidence that ability to self-fertilize is positively correlated with invasive status and size of invaded range among introduced species in several floras (van Kleunen & Johnson 2007; van Kleunen et al. 2008; Hao et al. 2011; Pyšek et al. 2011).

As invasive plants are frequently visited by pollinators in the novel range (Richardson et al. 2000; Memmott & Waser 2002; Pyšek et al. 2011), it is not clear to what extent those that can self-fertilize actually depend on this ability for their reproduction. Very few studies have assessed the benefits of selfing in the introduced range, particularly in relation to plant abundance (Knight et al. 2005; Eckert, Samis & Dart 2006 although see van Kleunen, Fischer & Johnson 2007). It is therefore not generally known whether animal pollination of invasive plants is less reliable in small founder populations, such that plants in these populations rely more on selfing. While mate availability and pollinator visitation may decline at different rates as plant abundance decreases, their effects on cross-pollen receipt are seldom distinguished (although see Kunin 1993; Duncan, et al. 2004b; Elam et al. 2007). A functional approach incorporating both these processes would allow us to understand why some plants are more vulnerable to pollen-limitation Allee effects than others, and predict in which plants self-pollination is most likely to be important for fecundity and, in the introduced range, invasiveness.

Lilium formosanum Wallace (Fig. 1) is an autonomously self-pollinating and hawkmoth-pollinated geophyte that is invasive in South Africa. We explored the contributions of self-fertilization and pollinators to fecundity of this species in its introduced range in South Africa, asking the following specific questions: (i) What is the magnitude of reproductive assurance benefits derived from self-fertilization? (ii) Are reproductive assurance benefits greater in smaller and more isolated populations of L. formosanum? (iii) Are reproductive assurance benefits higher in isolated plants than those in continuous patches? (iv) Do reproductive assurance benefits increase with distance from continuous patches? (v) Is any increase in reproductive assurance benefits with isolation attributable to reduced pollinator visitation or mate availability?

Figure 1.

Lilium formosanum growing in a disturbed habitat (a), being pollinated by the hawkmoth Agrius convolvuli (b) and stigma showing hawkmoth scales and pollen (c). Scale bars are 87 mm (a), 27 mm (b) and 1·3 mm (c).

Materials and methods

Study species

Lilium formosanum (Fig. 1) is a bulbous perennial plant with erect, annual stems. Each stem terminates in an inflorescence of 1–8 white, nocturnally scented, trumpet-shaped flowers (Rodger, van Kleunen & Johnson 2010). In South Africa, its principal pollinator is the native hawkmoth Agrius convolvuli (Fig. 1b; Rodger, van Kleunen & Johnson 2010). Populations vary in self-compatibility in its native range of Taiwan (Sakazono et al. 2012), and it is completely self-compatible and autonomously self-pollinating in its introduced range in Japan (Inagaki 2002) and South Africa (Rambuda & Johnson 2004). A molecular-marker study in the native range showed that fixation indices (Fis) of populations range from 0·032 to 0·901, suggesting variation among populations in mating system (Hiramatsu et al. 2001). As no inbreeding depression is evident in progeny up to 3 years of age in the introduced range (Rodger, van Kleunen & Johnson 2010; Rodger 2012), the reproductive assurance benefits attained through selfing are likely to be important for population growth and invasive spread.

Population size and isolation study

Study region and populations

Experiments were conducted from January to March in 2005, 2006 and 2007 in naturalized populations in KwaZulu-Natal, South Africa, 10–1700 m above sea level (Table A1, Supporting information). Observations of hawkmoth scales on stigmas (Fig. 1c) indicated that visitation of L. formosanum by these insects occurs throughout the study region (J.G. Rodger, unpublished results). Populations were mainly in disturbed grassland adjacent to exotic tree plantations or on grassy road verges, with a few in exotic forests or in otherwise pristine natural grasslands and indigenous forests. Population size was taken as the number of flowering stems. We used 50 m as the minimum separation distance between populations, allowing us to span a large range of isolation from almost no separation to over 15 km from the nearest population. An index of population isolation was calculated as the log10 of the mean distance to the nearest three populations. Data were obtained from 37 populations in 2005, 20 populations in 2006 and 22 populations in 2007 (Table A1, Supporting information). Most populations were accessible or available in only one of the three study years, although eight populations were studied in 2 or 3 years (Table A1, Supporting information).

Reproductive assurance benefits

Emasculation experiments can be used to distinguish between the importance of self-pollination vs. pollinator-mediated cross-pollination (Lloyd 1992; Eckert, Samis & Dart 2006). The reduction in fecundity experienced by emasculated relative to intact flowers is a measure of reproductive assurance benefits – in other words dependence on self-pollination for fecundity (Schoen & Lloyd 1992; Kalisz & Vogler 2003). Lilium formosanum flowers were emasculated by opening buds and removing anthers with alcohol-sterilized forceps. For naturally pollinated controls, buds were opened and forceps inserted. We considered it unlikely that emasculation would affect pollinator visitation to L. formosanum, as hawkmoths do not forage for pollen and A. convolvuli readily visits emasculated flowers (J.G. Rodger, pers. obs). This was confirmed by data that showed that the presence of lepidopteran scales (Fig. 1c), a measure of pollinator visitation, did not differ between stigmas of emasculated and intact flowers in three populations in 2006 and in four populations in 2007 (7–17 flowers per treatment per population, J.G. Rodger, unpublished results).

A single bud was emasculated on each of three (2005 and 2007) or 10 (2006) plants per population, and the same number of flowers was similarly allocated as controls, except in four populations for which we needed measures of within-population variation for a separate study in 2007 (Table A1, Supporting information). Control flowers were on separate plants to emasculated flowers in 2005 and on the same plants in other years. Because we emasculated only a single flower per plant, pollinator-mediated geitonogamy could have contributed to fecundity of emasculated flowers, making our estimates of reproductive assurance benefits conservative. However, this is unlikely to be important as preliminary analyses indicated that fecundity of emasculated flowers was not positively related to number of flowers per plant (J.G. Rodger, unpublished results).

Thirty-four populations were used in 2005, 15 in 2006 and 22 in 2007. We chose low levels of replication within populations to avoid a bias in sampling effort against small populations, first because analysis of variance is less robust to unequal variance and non-normality when data are unbalanced (Quinn & Keough 2002) and secondly because the effective replicate for a relationship between population attributes and plant performance is the population, so statistical power is likely to be increased by maximizing the number of populations at the expense of sample size per population (Quinn & Keough 2002). We calculated the overall reproductive assurance benefit of selfing for each year as the proportional reduction in fecundity caused by emasculation: RA = 100 × (1 − emasculated/control) (Eckert, Samis & Dart 2006) with fecundity defined as seeds per flower (percentage fruit set × mean seeds per fruit). The fecundity values used were grand means of population means.

Pollen limitation

Pollen supplementation experiments were used to test for pollen limitation in L. formosanum as autonomous self-pollination is not necessarily sufficient to fertilize all ovules (Rodger, van Kleunen & Johnson 2010). The same populations and the same sample-size regimes were used as for emasculations, but different plants (Table A1, Supporting information). Supplementation consisted of saturating the stigma with outcross pollen from a plant at least 5 m away in the same population. Plants sometimes allocate resources preferentially to flowers that have more fertilized ovules, which can lead to overestimation of pollen limitation in pollen supplementation experiments (Knight, Steets & Ashman 2006). However, as pollen limitation was generally low in this study, any overestimation would be quite small.

We calculated pollen limitation from seed per flower as 100 × (1 − control/supplemented) (Larson and Barrett 2000). Conducting emasculation and supplementation in the same populations also allowed us to assess how pollen limited L. formosanum would have been, had it lacked the ability to self-fertilize. This is termed pollinator failure and is calculated as 100 × (1 − emasculated/supplemented) (cf Kalisz & Vogler 2003).

Fruit and seed scoring

Fruits were harvested for seed counting at maturity, 10–12 weeks after flowering. Seeds were counted if they contained an embryo that was at least half the length of the seed, excluding the wing. For each fruit, we measured the mass of the entire contents and the mass and number of seeds in a random subsample containing approximately 50 seeds and used this information to calculate seeds per fruit. All seeds were counted in fruits containing fewer than 50 seeds. Seeds per flower data (fruit set × seeds per fruit) were zero inflated as there were many flowers that did not set fruit. Fruit set and seeds per fruit were therefore analysed separately.

Data analyses

Fruit set was analysed as a binomial response variable in generalized linear models incorporating a logit link function. Separate analyses of the effects of emasculation and pollen supplementation were carried out for each year, as most populations were used in only 1 year. Fruit set did not need to be analysed for the supplementation experiment in 2007 as there was 100% fruit set in both treatments. Significance was assessed from quasi-F-statistics in sequential analysis of deviance, analogous to F-statistics in anova with type I sums of squares (Payne 2011). Models included floral manipulation (emasculation or supplementation) as a fixed factor, population as a random factor, log10 population size and log10 population isolation as covariates and population size-by-floral manipulation and population isolation-by-floral manipulation interactions. A type I approach was used because of the hierarchical structure of the data, with replicates occurring within populations and population size and isolation measured at the population level. Terms were entered in the same order as they appear in Tables 1 and 2. The order in which terms are entered may affect their significance in sequential analyses. Nevertheless, reversing the order of population size and isolation, and the population size × floral manipulation and population isolation × floral manipulation interactions gave very similar results to those presented here and did not affect the conclusions drawn from them (J.G. Rodger, unpublished results). Population size and isolation were tested against population, and other terms were tested against the residual. Where models were not overdispersed (i.e. where residual deviance ≤ residual d.f.) we assumed residual mean deviance = 1 for the purposes of calculation of quasi-F-ratios, and when models were overdispersed we used the model-calculated residual mean deviance (Payne 2011). Model validation consisted of checking plots of residuals against fitted values for patterns (Zuur et al. 2009).

Table 1. Significance levels from generalized linear models for fruit set and REML analysis for seeds per fruit in emasculation experiments across a range of populations differing in size and isolation. Full tables are in Appendix S1 (see Supporting information). Test statistics are Quasi-F-statistics (ratios of mean changes in deviance) for fruit-set analyses, Wald F-statistics for fixed effects in seed-set analyses and change in deviance (tested against the chi-squared distribution) for random effects (P, P × E) in the seed-set analyses. Residual mean deviance shown in brackets for fruit-set analyses
EffectFruit setSeeds per fruit
d.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statistic
  1. PS, population size; PI, population isolation; P, population; E, emasculation.

  2. < 0·1; * < 0·05; ** < 0·01; ***< 0·001.

  3. †Random effects.

PS1,300·031, 110·441, 184·241, 186·191, 100·61, 192·61
PI1, 301·231, 114·211, 188·75**1, 220·491, 100·011, 182·85
P30, 301·6011, 111·6418, 170·6713·14127·97***115·79***
E1, 3011·45**1, 115·00*1, 176·81*1, 16113·13***1, 77·33*1, 1654·26***
PS × E1, 300·761, 110·031, 173·941, 168·31*1, 100·291, 190·02
PI × E1, 305·72*1, 117·14*1, 173·221, 204·291, 62·761, 170·02
P × E      10·1811·9415·17*
Residual30(1·20)11(0·43)17(2·1 × 10−5)      
Table 2. Significance levels from generalized linear models for fruit set and REML analysis for seeds per fruit in pollen supplementation experiments across a range of populations differing in size and isolation. Full tables are in Appendix S1 (see Supporting information). Test statistics are Quasi-F-statistics (mean change in deviance) for fruit set analyses, Wald F-statistics for fixed effects in seed-set analyses and change in deviance (tested against the chi-squared distribution for random effects in the seed-set analyses. Residual mean deviance shown in brackets for fruit-set analyses. 2007 data were not analysed for fruit set as all replicates of both treatments set fruit in this experiment
EffectFruit setSeeds per fruit
d.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statistic
  1. PS, population size; PI, population isolation; P, population; S, supplementation.

  2. *< 0·05; **< 0·01; ***< 0·001; †, random effects.

PS1, 324·19*1, 121·371, 250·771, 121·211, 140·21
PI1, 320·241, 12 3·111, 272·011, 100·541, 152·87
P†32, 322·37**12, 120·7817·70**119·18***128·09***
S 1, 320·031, 120·021, 745·86*1, 1381·711, 1340·26
PS × S1, 324·33*1, 120·771, 752·041, 1392·131, 1340·03
PI × S1, 321·211, 122·361, 760·021, 1380·481, 1340·98
P × S    10·0010·0010·00
Residual32(20·29)12(3·7 × 10−4)      

Seeds per fruit was analysed in restricted maximum likelihood (REML) analysis of variance to accommodate differences in sample size between populations. REML analysis of variance used the same statistical design as the generalized linear model for fruit set except they also included the population-by-floral manipulation interaction as a random effect. Significance was evaluated using Wald F-statistics for the fixed terms. For random terms, the change in deviance in the models when a term was dropped was compared with a chi-squared distribution with one degree of freedom (Payne, Welham & Harding 2011). Residual plots were examined to check whether assumptions were met.

Array experiment

Array layout

To test whether reproductive assurance was greater for plants isolated from continuous patches and, if so, whether this was due to decreased visitation or mate availability, we created arrays of emasculated and intact plants transplanted either into central patches of L. formosanum or similar grassland habitat that was isolated from the patches. Plants used were sourced from the same populations. Two populations with discrete patches of L. formosanum in open habitat (mainly natural grassland) were selected for experiments in February and March 2009. At Baynesfield (29 45·162S, 30 21·377E, Alt. 810 m), there was a population consisting of a single large patch of 748 plants. In the Karkloof population (29 20·229, 30 17·527, Alt. 1100 m), four patches of 67–610 plants were used. The array experiments were conducted from 31 January to 14 February 2009 at Baynesfield and from 28 February to 10 March 2009 at Karkloof. We obtained data from 87 plants at Baynesfield and 59 at Karkloof.

Emasculated and intact plants were placed singly inside continuous patches or in isolated pairs outside of these patches (Fig. C1, Supporting information). Isolated pairs consisted either of two emasculated plants or an emasculated plant and an intact plant, 1 m apart, to distinguish between effects of isolation on reproductive assurance benefits through reductions in pollinator visitation vs. mate availability. This approach is original to this study. Distances between successive pairs were chosen randomly from increasing intervals of the log2 scale (2–4, 4–8, 8–16…), so that as distance away from the central patch increased, density decreased as well. Distance from central patches ranged from 3 to 702 m at Baynesfield and 3 to 561 m at Karkloof. After flowering, all transplanted individuals were re-excavated and brought back to the University of KwaZulu-Natal Pietermaritzburg campus and maintained in plant pots until fruits were mature.


To test whether reproductive assurance compensated for reduced cross-pollen receipt in isolated plants, we compared fecundity in emasculated and intact plants placed inside central patches and in isolated pairs. Statistical analyses of fruit set and seeds per fruit were again conducted separately. Reproductive assurance indices were calculated for plants inside patches. Fruit set was analysed in generalized linear models as before, but with number of flowers (per plant) as the binomial total in an events/trials structure (Payne 2011). Using a type I analysis allowed us to test for the effect of distance from central patch after accounting for the effect of isolation (inside vs. outside patches). Terms in order of entry were isolation, distance from patch (log10 transformed), emasculation (intact vs. emasculated), emasculation-by-isolation, emasculation-by-distance. Distance was scored as zero for plants inside patches. For seeds per fruit, mean values were calculated for each plant, log10-transformed to improve homogeneity of variance and analysed in REML analysis of variance as sample sizes were unbalanced. The same model was used as described previously for fruit set. Terms were sequentially added to a model, and the significance of these terms was evaluated from Wald F-statistics.

Mate availability

To distinguish between effects of reduced mate availability vs. pollinator visitation on reproductive assurance benefits in isolated plants, we compared fecundity of emasculated plants inside populations, isolated and paired with another emasculated plant or isolated and paired with an intact plant as a test for the effect of mate availability. Fruit set and seeds per fruit were analysed using generalized linear models and REML analysis of variance as described above. Analyses included mate presence as a fixed factor, distance as a continuous variable and the mate presence-by-isolation distance interaction. As appreciable heterogeneity of variance remained for seeds per fruit, even after transformation, we conducted pairwise comparisons between groups with Mann–Whitney U-tests. Although corrections for multiple comparisons are sometimes applied for pairwise comparisons, we have not done so because in this case each comparison tests a different hypothesis, so the multiple comparisons do not inflate type 1 error.

Scale and pollen deposition

We also addressed the question of whether isolated plants experienced decreased pollinator visitation or mate availability by scoring emasculated flowers for the presence of lepidopteran scales, an indication of visitation, and presence of pollen on stigmas, an indication of successful pollination, using a 20X hand lens. Each plant was scored once, 3–4 days after transplanting, for all flowers that had been open for at least one night. Scale and pollen deposition were analysed in general linear models for binomial data with a logit link function including isolation as a fixed factor and distance as a covariate.

All statistical analyses were performed in Genstat 12.1 (VSN International, Hemel Hempstead, UK).


Population size and isolation study

Reproductive assurance benefits

Emasculation significantly reduced fruit set and number of seeds per fruit in naturalized populations in all 3 years, with a mean reduction in total fecundity (reproductive assurance benefits) of 67%: 90% in 2005, 45% in 2006 and 66% in 2007 (Table 1; Fig. B1, Supporting information). The effect of emasculation was not greater in smaller or more isolated populations except that there was a greater effect of emasculation in more isolated populations for fruit set in 2005 (Table 1; Figs 2 and 3). In other cases where there were significant population size-by-emasculation and isolation-by-emasculation interactions, these were not attributable to fruit set or seeds per fruit declining more for emasculated than control flowers as population size decreased or isolation increased (Table 1; Figs 2 and 3).

Figure 2.

Fruit set (model adjusted, a–c) and seeds per fruit (d–f) of emasculated and intact, naturally pollinated plants in relation to population size for 3 years. PS, population size, E, emasculation; ns, nonsignificant; < 0·1; *< 0·05; **< 0·01; ***< 0·001. Circles represent predicted values for populations for fruit set (adjusted for population isolation) and mean populations values for seeds per fruit. Regression lines shown for seeds per fruit.

Figure 3.

Fruit set (model adjusted, a–c) and seeds per fruit (d–f) of emasculated and intact naturally pollinated plants in relation to population isolation for 3 years. PI, population isolation [mean distance (m) to nearest three populations], E, emasculation; ns, nonsignificant; < 0·1; *< 0·05; **< 0·01; ***< 0·001. Circles represent predicted values for populations for fruit set (adjusted for population size) and mean populations values for seeds per fruit. Curves for fruit set were fit in generalized linear models using logit-transformed data and back-transformed.

Pollen limitation

Pollen supplementation increased fecundity (indicating pollen limitation) by an average of 12% (16% in 2005, 11% in 2006 and 10% in 2007), but this was only significant in 2005 (Table 2; Fig. B1, Supporting information). A significant supplementation-by-population size interaction in this year showed that supplementation increased fruit set only in smaller populations (Table 2; Fig. B2, Supporting information). There was no evidence for any effect of population isolation on pollen limitation as the interaction between population isolation and pollen supplementation was never significant (Table 2; Fig. B3, Supporting information). Pollinator failure was estimated as 92% in 2005, 48% in 2006 and 72% in 2007.

Array experiment


Fruit set and seeds per fruit were significantly lower in emasculated than in intact plants for both the Baynesfield and Karkloof sites (Table 3; Fig. 4). Indices of reproductive assurance were 75% for plants inside continuous patches and 96% for isolated plants at Baynesfield; 80% inside patches and 84% for isolated plants at Karkloof. At Baynesfield, emasculation reduced seeds per fruit more strongly in isolated plants than those in populations [significant isolation-by-emasculation interaction – with a nonsignificant trend in the same direction for fruit set (Table 3; Fig. 4a,b)]. However, at Karkloof, the effect of emasculation on fruit set and seeds per fruit was not related to isolation (Table 3; Fig. 4c,d). The effect of emasculation on fruit set and seeds per fruit did not increase with distance of isolation in either population (Table 3; Fig. C2, Supporting information). Although the distance-by-emasculation interaction was significant for seeds per fruit at Baynesfield, the effect of emasculation actually decreased with distance due to an outlier (Fig. C2, Supporting information).

Table 3. Significance levels from generalized linear models for fruit set, scale deposition and pollen deposition and REML analyses for seeds from array experiments on Lilium formosanum. Test statistics are Quasi-F-statistics (mean change in deviance) for fruit-set, pollen and scale analyses, Wald F-statistics for fixed effects in seed-set analyses and change in deviance (tested against the chi-squared distribution) for random effects in the seed-set analyses. Residual mean deviance shown in brackets for fruit-set, pollen and scale analyses
EffectFruit setSeeds per fruit
d.f.Test statisticd.f.Test statisticd.f.Test statisticd.f.Test statistic
  1. *< 0·05; **< 0·01; ***< 0·001

Emasculation analyses
Isolation1, 464·49*1, 350·231, 320·371, 300·02
Distance1, 464·93*1, 350·241, 320·31, 301·34
Emasculation1, 4680·41***1, 3516·37***1, 3231·07***1, 3023·32***
E × I1, 461·361, 350·041, 327·31*1, 300·15
E × D1, 460·021, 350·3051, 326·54*1, 300·01
Mate availability analyses
Distance2, 465·28**2, 300·842, 170·382, 107·09*
Donor presence1, 460·061, 300·581, 175·67*1, 100·03
DP × D1, 460·001, 300·601, 170·031, 100·17
Pollen and scale analyses
Isolation1, 260·031, 221·081, 2625·88***1, 220·01
Distance1, 260·511, 220·021, 260·001, 221·31
Figure 4.

Fruit set and seeds per fruit for emasculated and intact plants in array experiment at Baynesfield (a, b) and Karkloof (c, d). For fruit set (a, c), bars represent means of fruit-set values for individual plants. For seeds per fruit (b, d), back-transformed means and error bars are plotted. Numbers above bars are numbers of plants.

Mate availability

Mate availability had a significant effect on fruit set at Baynesfield (Table 3): isolated emasculated plants with no mates available (i.e. with an emasculated partner) had significantly lower fruit set than those inside populations (two-tailed t-tests; t = 2·96, d.f. = 48, = 0·009) or isolated with a potential mate (intact partner, t = 2·07, d.f. = 48, = 0·044). Isolated plants with intact partners did not differ significantly from those in continuous populations (t = 1·08, d.f. = 48, = 0·286). In isolated plants at Baynesfield, seeds per fruit was not affected by mate availability although it did increase with isolation distance (Table 3; Fig. 5), contrary to the expectation of decreased pollen transfer in more isolated plants (Table 3; Fig. C3, Supporting information). At Karkloof, mate availability had a significant effect on seeds per fruit (Table 3; Fig. 5d): emasculated plants with no mates available (emasculated partner) had fewer seeds per fruit than those inside populations (Mann–Whitney U-test u = 0·5, = 0·019, n = 4, 6) or isolated with a potential mate (u = 0·0, = 0·016, n = 4, 5). Isolated plants with intact partners did not differ significantly from those in continuous populations (u = 13, = 0·792, n = 6, 5).

Figure 5.

Fruit set (a, c) and seeds per fruit (b, d) of emasculated plants at different levels of mate availability in array experiment at Baynesfield (a, b) and Karkloof (c, d). For fruit set, bars represent means of fruit set values for individual plants. For seeds per fruit, back-transformed means and standard errors of plant means on log2 scale shown. Numbers above bars are numbers of plants. For seeds per fruit, all fruits had at least one seed.

Scale and pollen deposition

Scale deposition was not related to isolation or distance at either Baynesfield or Karkloof (Table 3). Isolated plants had significantly lower pollen receipt than those in the main patch at Baynesfield (Table 3), but not at Karkloof (Table 3).


These results show that L. formosanum relies heavily on self-fertilization for fecundity even though it has an effective hawkmoth pollinator in its invasive range (Rodger, van Kleunen & Johnson 2010). On average, reproductive assurance benefits from self-pollination, as assessed by floral emasculations, accounted for 67% of the total fecundity of naturally occurring plants, but this did not vary according to population size, contrary to expectations from other studies that show component Allee effects via a decrease in animal-mediated pollination in small populations (Ågren 1996; Groom 1998; Brys et al. 2011). The 71% average estimate of pollinator failure shows that L. formosanum would be highly pollen limited if it was self-incompatible. Although we have previously documented variation in the ability of L. formosanum to self-pollinate autonomously in the study region, the fact that pollen supplementation increased fecundity by only 12% on average indicates that most populations have high levels of autofertility.

There was no evidence for an effect of population size on reproductive assurance benefits in the survey of natural populations, indicating that population size did not affect pollinator visitation (Table 1; Fig. 2). However, reproductive assurance mitigated a detectable Allee effect for isolated plants lacking nearby mates in the array experiment (Fig. 5). Reproductive assurance benefits were greater for isolated plants than those placed in continuous populations at only one of the two sites (Baynesfield, Fig. 4a,b). However, in both populations, emasculated plants that were isolated with no potential mate (i.e. placed next to another emasculated plant) had lower fecundity than those isolated with a single intact plant nearby, or placed in continuous populations. This shows that the greater reproductive assurance benefits for isolated plants at Baynesfield were due to decreased mate availability, not reduced pollinator visitation (Table 3, Fig. 5a,d). Findings of lower stigmatic pollen deposition on isolated than on nonisolated plants at Baynesfield and the lack of effect of isolation on lepidopteran scale deposition on stigmas (Table 3) support this conclusion.

As pollen limitation of self-incompatible plants is generally higher in the introduced than in the native range (Burns et al. 2011), it can be expected that invasive plants obtain substantial reproductive assurance benefits from selfing. However, no reproductive assurance benefits were found in hummingbird pollinated Nicotiana glauca plants invasive in North America (Schueller 2004), while large reproductive assurance benefits were found in hawkmoth-pollinated Lilium formosanum (RA = 67%, this study) and Datura stramonium (RA = 83%, van Kleunen, Fischer & Johnson 2007). Clearly, more studies spanning a range of pollination systems, geographic areas and life forms are needed before it will be possible to assess the importance of selfing for fecundity of introduced plants generally.

Mitigation of increased mate limitation by selfing in isolated plants, as demonstrated in the array experiment (Fig. 5), could be especially important for invasion of L. formosanum, given that reproduction by isolated individuals should have a dramatic impact on the invasion process (Kot, Lewis & Driessche 1996; Clark, Lewis & Horvath 2001). We are not aware of any previously published studies showing that selfing mitigates mate-limitation Allee effects in invasive species, and only one for a native species (Brys et al. 2011). Indirect evidence from some studies of the effect of plant abundance on fecundity suggests that mate limitation may generally be more important than reduced pollinator visitation in reducing cross-pollen receipt of isolated individuals (Kunin 1993; Duncan, et al. 2004b; Elam et al. 2007). This is one of the first studies to distinguish between mate availability and pollinator visitation components of Allee effects, yet this approach is essential for deriving the functional understanding that would allow us to predict which plants should be most vulnerable to Allee effects, and to allow more refined predictions about the effects of reproductive assurance and pollen limitation on invasiveness.

The absence of a detectable effect of plant abundance on hawkmoth visitation to L. formosanum is consistent with some other studies of hawkmoth-pollinated plants (e.g. Johnson, Torninger & Ågren 2009) and contrasts with results for plants with other pollinators (Ågren 1996; Groom 1998; Brys et al. 2011). This could be because hawkmoths are more nomadic in their movements and opportunistic in their foraging than other pollinators (as suggested by Johnson, Torninger & Å 2009) or because foraging primarily by olfactory rather than visual cues renders them less capable of assessing population size prior to arrival in populations.


We have used a functional approach to assess the relationship between plant abundance and reproductive assurance benefits in L. formosanum, distinguishing between effects of pollinator visitation and mate availability as well as between isolation and population size. Although we found no evidence that pollinator visitation was related to abundance, our finding that selfing mitigated decreased mate availability in isolated plants is, to the best of our knowledge, the first evidence that selfing may contribute to invasiveness by mitigating an Allee effect. Because of this finding, because reproductive assurance benefits are high even in the absence of Allee effects, and because progeny trials have revealed almost no evidence for inbreeding depression in L. formosanum in South Africa (Rodger, van Kleunen & Johnson 2010; Rodger 2012), reproductive assurance benefits may well translate into a demographic advantage. This makes it likely that ability to self-pollinate contributes to the invasiveness of L. formosanum. Demographic analysis will be required to assess the effect of selfing on invasiveness, and the relative importance of its compensating for generally inadequate pollinator visitation vs. mate limitation in isolated individuals.


Thanks to Dalton Nyawo for assistance with seed counting, Wade Shrives for assistance with floral manipulations and Ben Khumalo for help setting up the arrays. We are grateful to Craig Morris, Mike Ramsey and Lawrence Harder for statistical advice and to Karl Duffy, Chris Eckert, Elizabeth Elle, Taina Witt and Lorne Wolfe for comments on previous drafts of this manuscript. We also thank Baynesfield Estates, the Engelbrechts and the Shaws for permission to work on their properties and UKZN Botanical Garden for space to maintain plants. This study was supported by the DST-NRF Centre of Excellence for Invasion Biology (CIB).