The majority of plant species and many animals are hermaphrodites, with individuals expressing both female and male function. Although hermaphrodites can potentially reproduce by self-fertilization, they have a high prevalence of outcrossing. The genetic advantages of outcrossing are described by two hypotheses: avoidance of inbreeding depression because selfing leads to immediate expression of recessive deleterious mutations, and release from drift load because self-fertilization leads to long-term accumulation of deleterious mutations due to genetic drift and, eventually, to extinction. I tested both hypotheses by experimentally crossing Arabidopsis lyrata plants (self-pollinated, cross-pollinated within the population, or cross-pollinated between populations) and measuring offspring performance over 3 years. There were 18 source populations, each of which was either predominantly outcrossing, mixed mating, or predominantly selfing. Contrary to predictions, outcrossing populations had low inbreeding depression, which equaled that of selfing populations, challenging the central role of inbreeding depression in mating system shifts. However, plants from selfing populations showed the greatest increase in fitness when crossed with plants from other populations, reflecting higher drift load. The results support the hypothesis that extinction by mutational meltdown is why selfing hermaphroditic taxa are rare, despite their frequent appearance over evolutionary time.

Many species of plants and animals are hermaphrodite organisms expressing both female and male sexes (Jarne and Charlesworth 1993; Jarne and Auld 2006). Individuals of most of these species can potentially self-fertilize, but they nevertheless typically reproduce by outcrossing or mixed mating (Goodwillie et al. 2005; Igic and Kohn 2006; Jarne and Auld 2006). It has been estimated that predominant outcrossing occurs in about 65% of hermaphrodite plants (Igic and Kohn 2006) and in 35% of hermaphrodite animals (Jarne and Auld 2006). A mixture of outcrossing and selfing occurs in about 24% of hermaphrodite plants and 47% of hermaphrodite animals, leaving only a minority of hermaphrodites that predominantly self-fertilize.

There are two genetic hypotheses for the rarity of selfing. The first states that self-fertilization has immediate negative genetic consequences caused by segregating recessive deleterious mutations, also called inbreeding load. Models of mating system evolution in hermaphrodites propose that substantial inbreeding depression, defined as the decline in fitness due to the expression of inbreeding load under self-fertilization, can drive the evolution of outcrossing (Maynard Smith 1978, p. 125–130; Lande and Schemske 1985; Uyenoyama 1989). Although evidence for substantial inbreeding load in outcrossing taxa is compelling (Husband and Schemske 1996; Winn et al. 2011), there is only mixed empirical support for the hypothesis that selfing populations have reduced inbreeding load. Meta-analyses reveal lower inbreeding load in selfing taxa compared to outcrossing taxa (Husband and Schemske 1996; Winn et al. 2011), but the difference disappears after accounting for phylogenetic relationships (Byers and Waller 1999). This implies that inbreeding load may not be the sole driver of mating system evolution. In addition, it contradicts the expectation that a shift to selfing initiates purging of recessive deleterious alleles, and that purging in turn will further reduce inbreeding load (Lande and Schemske 1985). One problem with such comparisons is that even related species differ in their history, ecology, and life history, and these factors may affect levels of inbreeding load. The best data are therefore provided by comparisons within species or lower taxonomic entities (Johnston and Schoen 1996; Vogler et al. 1999; Busch 2005; Goodwillie and Knight 2006; Willi 2009).

A second hypothesis for the prevalence of outcrossing in hermaphrodites states that selfing has long-term negative genetic effects. Selfing populations have an effective population size (Ne) only half that of outcrossing populations (Pollack 1987; Nordborg 2000). In addition, directional selection and recurrent extinction/recolonization of local demes can diminish Ne in selfing populations even more than they do in outcrossing populations (Charlesworth and Wright 2001; Ingvarsson 2002). Therefore, according to this hypothesis, selfing populations are more vulnerable to the accumulation and fixation of deleterious mutations due to enhanced genetic drift, called drift load (Whitlock 2000). In fact, theory predicts that obligately selfing organisms are as vulnerable to mutation accumulation as asexual organisms, leading to so-called “mutational meltdown”: a spiraling reduction in individual fitness and population size, leading eventually to population extinction (Lynch et al. 1995a). Predominant selfing combined with low population size is also predicted to be vulnerable to mutation accumulation (Charlesworth et al. 1993) and potentially to mutational meltdown. Indeed, a recent comparative study of plants in the family Solanaceae discovered that extinction occurs at higher rates in species that can potentially self compared to those that are obligate outcrossers, and this reduces net diversification despite a higher speciation rate (Goldberg et al. 2010). However, few studies have estimated both inbreeding and drift load in the context of mating system variation (Busch 2006).

The two hypotheses of high inbreeding load in outcrossing populations—expressed as inbreeding depression under selfing—or increased drift load after a switch to predominant selfing may explain the rarity of selfing in hermaphrodites. I tested two associated sets of predictions: (1) selfing is rare because of high inbreeding load, in which case selfing lineages are less likely to evolve from outcrossing ones. Where outcrossing is predominant we expect high inbreeding load, and where selfing predominates we expect low inbreeding load. (2) Drift load leads to increased extinction of selfing lineages. In this case, we expect low drift load where outcrossing is predominant and high drift load where selfing is predominant. I estimated the two kinds of genetic load in replicate natural populations of the plant Arabidopsis lyrata subsp lyrata. This species has a genetically determined sporophytic self-incompatibility system that effectively prevents self-fertilization (Schierup et al. 2001; Prigoda et al. 2005). However, a few populations in the Great Lakes region of North America have switched reproductive mode from outcrossing, the ancestral state, to derived selfing in several independent shifts (Foxe et al. 2010; Willi and Määttänen 2010). The Great Lakes were covered by the Laurentide ice sheet, and this implies a recent history of postglacial colonization. The species is a diploid, short-lived perennial that occurs in relatively open habitats, either on sandy substrate or rocky ledges. Thus, these populations of A. lyrata are well suited for comparing inbreeding and drift load because they differ in mating system but are otherwise similar in their population history, ecology, and life history.

The two types of genetic load were assessed with a crossing experiment involving four predominantly selfing and one mixed mating population (hereafter, S), and 13 predominantly outcrossing populations (O). Inbreeding load was calculated from the difference in performance between self-pollinations and within-population crosses (Dudash 1990). Drift load for each population was defined as the difference in performance between offspring produced by crossing plants from different populations and within the same population. The latter comparison assesses drift load via heterozygote advantage or heterosis due to overcoming the drift load. Both kinds of load were estimated for an integrated measure of individual fitness over three reproductive seasons, and—to ascertain the life stages at which inbreeding load and heterosis are expressed—for performance components at particular life stages. For some performance traits, differences between selfing and outcrossing populations could also reflect adaptive divergence in life history associated with mating system. These might include investment in male reproductive function or shifts in the timing of reproduction and senescence. I accounted for this possibility by testing for average differences between mating systems in various components of male and female performance.



In summer 2007, I collected seed material from 15 populations of A. lyrata in the Great Lakes region of North America (Appendix Table S1). At each site, ripe fruits (siliques) were collected from 30 plants at 5-m intervals along three parallel transects separated by 5 m (grid area 10 × 45 m). If no plant with ripe fruits was within 2.5 m of a grid point, a replacement was found along a 5-m extension of one of the transects. Transect sampling could not be applied in three smaller populations on rocky outcrops, because plants grew in patches. There, plants were sampled such that distances within patches of occurrence were maximized and the combined surface area of the patches sampled was about 450 m2. Seeds from three additional populations at Long Point and Rondeau, Ontario, Canada, and Toledo, Ohio, USA, were kindly provided by B. Mable. Her material was collected in a comparable manner, but over a somewhat larger area. The study included a total of 18 populations.


One plant from each of 30 field-collected fruits per population was raised in a greenhouse (tubs 7 × 7 × 8 cm; substrate 1:1 sand:peat; 16:8 light:dark cycle at ∼120 μE/m2s; temperature 20°C:16°C day:night; RH 50–70%). Twelve individuals per population were chosen at random as target plants for use in three types of crosses: self-pollination, outcrossing with a haphazardly chosen plant from the same population (WP), and outcrossing with a haphazardly chosen plant from a randomly chosen different population (between-population, BP; crossing design illustrated in Appendix Table S2). I performed hand pollinations at the bud-stage, when self-incompatibility is not yet (fully) expressed and can therefore be overcome. Immature anthers were first removed, and then anthers with pollen from another flower/plant were rubbed carefully over the stigma. Contamination was avoided by cleaning forceps after each pollination by holding over a flame and dipping into alcohol. I verified the absence of contamination by comparing the multilocus microsatellite genotypes of 100 offspring from 73 selfed and WP crosses with those of their parents. Failed crosses were repeated—outcrossings often with a different pollen donor—until a healthy looking fruit was produced. Failed crosses were those in which the ovaries did not develop fully; hence, there was neither fruit elongation nor much seed development. Success rate was similar among cross-types in selfing populations (mean ± SE for S-self: 0.79 ± 0.05, S-WP: 0.79 ± 0.04, S-BP: 0.84 ± 0.03). Failure was most likely due to physical damage. For outcrossing populations, reduced success rate of selfed crosses and to some extent WP crosses was probably due to partial expression of self-incompatibility at the bud stage (O-self: 0.48 ± 0.03, O-WP: 0.73 ± 0.03, O-BP: 0.85 ± 0.02).


Performance of offspring from the three kinds of crosses was assessed over 3 years in an outdoor garden experiment. When fruits were mature, I haphazardly chose three seeds from each seed family for measurement of seed length under a dissecting scope, and then transferred the seeds to a dark room at 3°C for 7 days. Four seeds per cross were germinated in each of two tubs, split into two blocks in separate air-conditioned greenhouse cabinets (16:8 light:dark at 100 μE/m2s; temperature 18°C:16°C; RH > 50%). Proportion of seeds germinated was scored 34–37 days after sowing.

On 27–29 April 2009, 3 weeks after sowing, I removed all except for one randomly chosen seedling per tub, and transferred the tubs to an outdoor garden at the University of Zürich. Seedlings that germinated later were also removed unless there was no earlier seedling. The two tubs per cross were distributed over two garden beds, with positions within beds assigned at random. During the first summer the beds were covered with 50% shade cloth, watered daily, and provided with slug repellent and occasional insecticides. In July, infection by the white rust pathogen, Albugo candida, occurred, and affected plants were excluded from analyses of later life stages, except for multiplicative performance II (defined below).

Male and female contributions to fitness were estimated for every plant. During the first summer, I scored bolting and flowering three times per week. Male reproductive output was estimated from pollen counts conducted on two of the earliest, freshly opened flowers. The flower was freed from sepals and petals, the receptacle squeezed so that the stamens with their anthers bent outward, and the flower was dried at 60°C for 40 h. The flower was then combined with 12 mL of an isotonic solution and shaken in an ultra sound bath for 3 min. Pollen grains were counted with a CASY TTC Cell Counter (Schärfe System GmbH, Reutlingen, Germany), set to record three serial aliquots of 400 μl. I considered particles within the size range of 14–27μm to be pollen; larger or smaller items were broken flower pieces or unidentified microparticles. Siring success was not considered here.

Flower and female reproductive output were recorded on several occasions. At the end of the flowering period, on 17 August 2009, I counted the number of all flower pedicels for each plant, which reflected the total number of flowers produced over the season. Fruit production had been scored by weekly cutting and counting fully developed fruits. In 2010, timing of flowering was recorded once a week, and the number of flowers and fruits that had been produced was counted toward the end of the flowering period in May. In 2011, I assessed only flower and fruit production.


The main dependent variable was multiplicative performance, calculated for each plant as the proportion of seeds that germinated in the tub times the sum of flowers produced in the first, second, and third years (multiplicative performance I = proportion germination × sum of flowers in 2009, 2010, and 2011, without diseased plants) or fruits produced in the first, second, and third years (multiplicative performance II = proportion germination × sum of fruits in 2009, 2010, and 2011). Although not equivalent to individual fitness, the two measures of multiplicative performance integrated survival and fecundity over 3 years, and are therefore likely to be correlated with lifetime fitness. Seed set was not estimated, but prior information from the parental generation showed that female fertility was similar between mating systems: 11–18 seeds per fruit for selfing populations (95% confidence interval), 12–17 for outcrossing populations.

Multiplicative performance and its individual components were first analyzed with hierarchical mixed models using REML, with tub (for germination) or plant nested within cross-identity, maternal plant, and maternal population at the first level, then cross-identity nested within maternal plant and population at the second level, maternal plant nested within population at the third level, and population at the fourth level (MIXED procedure in SAS; Singer 1998; SAS Institute 2002; SAS code in Appendix Table S3). For germination, pollen number, and pollen size, the replicate within tub represented a further level in the random part of the model. Germination, pathogen infection, and 2009 flower and fruit production were binary dependent variables, and cumulative flower and fruit numbers and multiplicative performance were analyzed assuming a Poisson distribution (GLIMMIX procedure). Seed size, cumulative fruit production, and multiplicative performance II were ln-transformed. Cross-type was a fixed effect on the level of the cross-identity. The random interaction involving cross-type was included at levels higher than cross-identity, and I assumed a “banded main diagonal” variance–covariance matrix (SAS Institute 2002). Mating system was a fixed effect on the level of the population. Mating system was treated as a classed variable because, with one exception, populations were either mostly selfing or mostly outcrossing (Willi and Määttänen 2010, 2011). The single mixed mating population was considered to be selfing because the population is derived from an ancestral state of outcrossing, which implies that it experiences conditions that favor selfing. Block, where applicable, was a fixed effect.

Next, inbreeding load was estimated for each population by (WWP - Wself, F= 0.5)/WWP, where WWP is the mean performance of WP crosses, and Wself, F= 0.5 is the mean performance of selfed crosses, assuming the inbreeding coefficient (F) of selfed crosses was 0.5. However, especially plants of selfing populations had inbreeding coefficients that were already substantial (Willi and Määttänen 2011), here calculated as Fe= (1 - tm)/(1 +tm) (Brown 1979), where tm is the multilocus outcrossing rate of the population based on progeny array analysis (Appendix Table S1). Therefore, Fe was assessed on seedlings that later served as parents in the crossing experiment. Expected Wself, F= 0.5 was extrapolated between the two cross-types, WP crosses and selfed crosses, for which F= 0 and F= (Fe+ 0.5 × (1 - Fe)), respectively:


Drift load was calculated as the extent to which heterosis led to an increase in performance due to BP outcrossing (WBP), (WBPWWP)/WWP. Higher values corresponded to weaker performance for pathogen infection and timing of flowering, so infection was given the value of 0 (now reflecting resistance) and timing of flowering was expressed as the number of days to the population-cross-type combination that flowered last. Loads for cumulative fruit production were based on ln-transformed data. Estimates of population-level inbreeding load and heterosis were tested in a general linear model with mating system as a categorical fixed effect.


Performance of replicate offspring from each cross was measured from the seed stage to reproductive output over three years, integrating proportion of seeds germinated, survival, and total flower production (multiplicative performance I) or total fruit production (multiplicative performance II). Mixed model analysis of multiplicative performance I over the 3 years showed that there were significant effects of cross-type, mating system, and their interaction (Table 1). Multiplicative performance II revealed effects of similar magnitude (N= 1269, CT: F= 16.23, P < 0.001; MS: F= 14.25, P < 0.01; CT × MS: F= 3.54, P < 0.05). The crossing experiment demonstrated that both types of genetic load were detectable in selfing populations, but that drift load was far more important than inbreeding load (Table 1). The inbreeding load, calculated as the proportional decline in performance due to selfing, was no higher in outcrossing populations than in selfing populations (about 0.18 in populations of both mating systems; Table 1; Fig. 1A). Drift load was estimated from the proportional improvement in performance due to interpopulation outcrossing, also called heterosis. Selfing populations showed significant heterosis, around 2.9, and this was substantially higher than heterosis in outcrossing populations (Table 1; Fig. 1A). The comparison of WP crosses between the two mating systems revealed a 69% performance reduction in selfing compared to outcrossing populations, further corroborating their higher drift load (Fig. 1B).

Table 1.  Summary of hierarchical mixed-effects linear models testing for differences in fitness components among mating systems and cross-types, and estimates of inbreeding load and heterosis for selfing and outcrossing populations.
Dependent variable N Fixed effectsInbreeding loadHeterosis
CTMSCT × MSδSδOCoefficientHSHOCoefficient
  1. Fixed effects in the model were cross-type (CT: selfed, within-population, and between-population), mating system (MS: selfing and mixed mating populations [0] or outcrossing populations [1]), and their interaction. Block was included but the results are not shown here. Random effects were population, maternal genotype within population, their interactions with cross-type, then cross-identity within maternal genotype and population, and where applicable, tub/plant within cross-identity, maternal genotype and population. The table shows F-values for fixed effects; df are 1,16 for MS and 2,29–32 for CT and CT × MS. Inbreeding load (δ= inbreeding depression due to one generation of selfing) and heterosis (H) were calculated as described in section Methods, and reported as least squares means (LSM) with SE in parentheses for selfing (S) and outcrossing (O) populations. The difference between mating systems (S compared to O), here called the coefficient, was tested with a general linear model on population estimates (N= 18, except for reproductive traits 2009 N= 16, 17 and for pollen traits N= 16). Significance is indicated: *P < 0.05, **P < 0.01, ***P < 0.001.

Multipl. performance I 1078 23.05*** 12.82** 3.52*  0.18 (0.07)*  0.18 (0.05)**  0.00 (0.09)  2.89 (0.83)**  0.47 (0.52)  2.42 (0.98)*
Seed size1812 7.62** 0.025.57** 0.01 (0.02)−0.01 (0.01) 0.01 (0.02) 0.07 (0.02)** 0.01 (0.01) 0.06 (0.02)*
Germination 4419  8.90***  0.00 1.71  0.03 (0.02)  0.01 (0.01)  0.02 (0.03)  0.16 (0.09)  0.03 (0.06)  0.14 (0.11)
Pathogen infection1252 2.06 4.79*0.25 0.01 (0.04)−0.03 (0.03) 0.04 (0.05) 0.14 (0.06)* 0.06 (0.04) 0.08 (0.07)
Flowering time 2009 705  1.43  4.49* 0.54  0.08 (0.13) −0.09 (0.07)  0.17 (0.15) −0.07 (0.17) −0.03 (0.09) −0.04 (0.20)
Flowers 20091054 2.32 4.380.01 0.18 (0.09) 0.13 (0.05)* 0.05 (0.10) 1.31 (0.64) 0.06 (0.35) 1.25 (0.73)
Fruits 2009 1054  3.78*  4.72* 0.06  0.14 (0.13)  0.14 (0.06)*  0.01 (0.15) −0.05 (0.17)  0.16 (0.08) −0.21 (0.19)
Flowering time 201088212.62*** 3.362.04 0.07 (0.08) 0.17 (0.05)**−0.10 (0.09) 0.54 (0.93) 0.71 (0.58)−0.17 (1.10)
Flowers up to 2010 1054 16.84*** 13.21** 2.63  0.11 (0.08)  0.15 (0.05)** −0.03 (0.09)  2.06 (0.61)**  0.39 (0.38)  1.67 (0.72)*
Fruits up to 2010105413.39***11.79**2.67 0.21 (0.08)* 0.18 (0.05)** 0.03 (0.09) 1.12 (0.37)** 0.23 (0.23) 0.89 (0.43)
Pollen number 1310  0.80  3.32 2.98 −0.01 (0.06)  0.09 (0.03)** −0.10 (0.07)  0.23 (0.08)** −0.06 (0.04)  0.29 (0.09)**
Pollen size131010.76*** 1.910.17−0.00 (0.01)−0.01 (0.00)* 0.01 (0.01)*−0.01 (0.01)−0.01 (0.00)−0.01 (0.01)
Figure 1.

Inbreeding load and drift load estimated via heterosis in outcrossing and selfing populations of Arabidopsis lyrata subsp. lyrata revealed by three kinds of experimental crosses and assessment of multiplicative performance I (A). Panels B–D show for the three kinds of crosses multiplicative offspring performance I over 3 years, total flower production per plant over the first 2 years, and total fruit production per plant over the first 2 years. The three kinds of crosses are selfed, within-population crossed (WP), and between-population crossed (BP). Symbols are least squares means ± SE.

It might be argued that some differences between selfing and outcrossing populations were due to adaptive differences in life history between mating systems rather than to drift load. This appears unlikely. Inspection of separate fitness components suggests lower absolute fitness—rather than a life-history shift—in selfing populations. In Table 1, the effect of mating system employs data from all three cross-types to compare selfing and outcrossing populations; results were similar when only self and WP crosses were included. Key performance differences between S and O populations appeared after an initial growth phase; there was no difference between mating systems in seed size or germination (Table 1; Fig. 2A, B). Survival to the end of the first reproductive period was very high and therefore not analyzed separately (>96%; 40 of 1252 plants died). An oomycetous pathogen infected some plants in the first season, and these were significantly more likely to be offspring from S populations (Fig. 2C). First year flower production did not vary with mating system, but timing of flowering was significantly later and fruit production was lower in selfing populations (Fig. 2D, E); analysis on self and WP crosses revealed no significant difference between mating systems for flowering time (P > 0.1) and only a trend for fruit production (F1,16= 4.45, P= 0.051). By the second year, S populations had lower cumulative flower and fruit production (Fig. 1C, D). However, timing of flowering in the second year, pollen numbers per flower and pollen grain size did not significantly differ between mating systems (Fig. 2F–H). None of the variables showed any evidence of adaptation to selfing or outcrossing; instead, the significant transitions that I observed represent enhanced drift load.

Figure 2.

Fitness components differing between three kinds of experimental crosses of outcrossing and selfing populations of Arabidopsis lyrata subsp. lyrata: seed size (A), germination (B), infection by the oomycete Albugo candida in 2009 (C), fruit production in 2009 (0/1 data) (D), timing of flowering in 2009 in days since germination (E), timing of flowering in 2010 in days since start of the year (F), pollen number per flower (G), and mean pollen size of a flower (H). Symbols in panels A–H represent offspring from selfed crosses, within-population crosses (WP), and crosses between populations (BP). Symbols are least squares means ± SE. For pathogen infection and flowering time, inbreeding load and drift load are shown here when selfed > WP and WP > BP, respectively.

At what point in the life cycle was genetic load expressed?Table 1 reports the inbreeding loads and drift loads for S and O populations, along with significance tests for all life stages. Estimation of load is necessary even in the absence of a cross-type-by-mating system interaction. The linear model evaluates differences in absolute performance measures whereas genetic loads are calculated from proportional changes in performance. The two analyses should therefore be similar, but they need not be identical. Inbreeding load appeared late in life, at flower and fruit production in the first 2 years and more systematically for O populations. Outcrossing populations also showed significant inbreeding load for pollen production, while virtually no inbreeding load was expressed in the pollen of S populations.

For heterosis, outcrossing populations were relatively unaffected and selfing populations showed impacts at various stages throughout life. Selfing populations produced especially large seeds from BP crosses. After seedling establishment, significant heterosis was expressed in infection probability, and cumulative flower and fruit production through 2010. Pollen production of plants from WP crosses was reduced compared to BP crosses (Fig. 2G). This may reflect incipient adaptation to selfing (Sicard and Lenhard 2011), and therefore heterosis in pollen production cannot be interpreted as drift load. Table 1 suggests that overall divergence in pollen production was weak because there was no significant effect of mating system or its interaction with cross-type (also when only self and WP crosses were considered). The difference in heterosis between S and O populations was significant for seed size, cumulative flower production through the second year and pollen number, apart from multiplicative performance.

Population means of raw data for multiplicative performance and life stage components are listed in Appendix Table S4.


Why is self-fertilization in hermaphrodites relatively rare? One hypothesis is that inbreeding load must be low to enable the evolution of selfing (Maynard Smith 1978; Lande and Schemske 1985). Once selfing has evolved, inbreeding load is exposed to purging and should further decline. This is especially true for load caused by recessive mutations with large deleterious effects because of an increase in overall homozygosity levels (Lande and Schemske 1985). These two ideas together predict that selfing populations should have reduced inbreeding load, well below 0.5 (Lande and Schemske 1985). My results for A. lyrata do not support this prediction. Although the inbreeding load was indeed below 0.5 in selfing populations, it was equally low in outcrossing populations.

This finding of low inbreeding load in both selfing and outcrossing populations leads to three important conclusions. First, purging must have been an important force in this system in the past, leading to generally low inbreeding load. Indeed, population bottlenecks that would favor purging (Kirkpatrick and Jarne 2000) were probably frequent as this species recolonized the Great Lakes region during the postglacial period. Second, for populations that have switched their mating system to selfing, further purging of recessive deleterious mutations may be inefficient (Byers and Waller 1999). This could be tied to inbreeding load arising from recessive deleterious mutations of relatively weak effect, which are not easily purged (Lande and Schemske 1985; Lynch et al. 1995a). Third, inbreeding load does not seem to be an effective driver of mating system evolution in A. lyrata, because if it were more populations would have shifted to self-compatibility and selfing. The potential to evolve toward selfing is present in nearly all outcrossing populations because all have some partially or fully self-compatible individuals that at least partially self (Willi and Määttänen 2010). Other outcrossing species also show evidence that, despite low inbreeding load, self-compatibility or selfing increases in frequency only during periods of low pollen or mate availability (Kalisz et al. 2004; Willi 2009).

Inbreeding depression is known to be environment dependent. Two studies report that inbreeding depression is somewhat higher in nature than in an outdoor garden (Dudash 1990, Eckert and Barrett 1994), and a review article suggests that stressful conditions have a variable effect on inbreeding load, but on average increase load by about 50% (Armbruster and Reed 2005). The type of stress is important because inbreeding depression is no higher when plants are subjected to increased competition (Willi et al. 2007a). In any event, even if inbreeding load was underestimated by up to about twofold in my experiment, it would still be below 0.5, a critical value suggested by theory of mating system evolution (Lande and Schemske 1985). More important from the perspective of this study were the similar values of inbreeding depression in selfing and outcrossing populations. Moreover, the relatively high values of fixed drift load suggest that benign conditions did not suppress the expression of load.

A recent meta-analysis showed that about half the published estimates of inbreeding depression in outcrossing plants are below 0.5 (Winn et al. 2011). Assuming that most taxa included in this analysis were diploid (polyploids suffer reduced inbreeding depression; Lande and Schemske 1985), this raises the question of what prevents the spread of selfing in the many outcrossing taxa with low inbreeding load. There are of course many ecological advantages associated with the increased effective recombination afforded by sexual reproduction (Busch et al. 2004, Hartfield and Keightley 2012), especially for sessile organisms in heterogeneous environments (Lenormand and Otto 2000). Also, there may be additional barriers to the evolution of selfing, related to movement of pollen to the stigmas, for example, or related to the many integrated adaptations that promote outcrossing, such as flower morphology and color, floral scent, and pollinator rewards (Lloyd and Schoen 1992).

A second genetic explanation for the rarity of self-fertilization in hermaphrodites is that, because selfing populations experience enhanced genetic drift, they gradually accumulate drift load from slightly deleterious mutations (Charlesworth and Wright, 2001). My data support this hypothesis. Enhanced drift load in selfing populations of A. lyrata is associated with low effective population size. Estimated values of Ne for selfing populations are on average a few hundred individuals, 2–3 times lower than Ne in outcrossing populations even after accounting for increased levels of inbreeding via selfing (Willi and Määttänen 2011). These values are high enough for extinction under mutation accumulation to be a gradual rather than a rapid process (Lynch et al. 1995b). Models of asexual populations predict mean extinction times in the range of 1000 generations under intermediate selection on deleterious mutations, when the upper population size is 500 (Gabriel et al. 1993). This prediction should apply to A. lyrata as well, although some outcrossing in predominantly selfing populations alleviates the accumulation of load to some extent (Charlesworth et al. 1993). Assuming that selfing populations of A. lyrata switched their reproductive mode within the past 8000–14,000 years, during recolonization of the Great Lakes region after retreat of the Laurentide ice sheet (Larson and Schaetzl 2001), only a few thousand generations have passed and (formerly) larger selfing populations should still persist.

High drift load in selfing A. lyrata populations agrees with a previous study of Leavenworthia alabamica that revealed substantial load in one geographically isolated, self-fertilizing population compared with several large outcrossing populations and two small but less isolated selfing populations (Busch 2006). Increased drift load in selfing populations also supports recent results of phylogenetic analyses that suggested a higher extinction rate for self-compatible lineages within solanaceous plants (Goldberg et al. 2010). One factor that may accelerate the accumulation of drift load is population bottlenecks. Bottlenecks during range expansions may simultaneously favor the establishment of selfing taxa and exacerbate their drift load (Kirkpatrick and Jarne 2000).

This study illustrates two general points regarding mating system evolution. First, forces opposing the spread of self-incompatibility and selfing must exist in outcrossing populations apart from inbreeding load. This conclusion is based on my results, but also on the recent meta-analysis showing that half of outcrossing plant taxa have inbreeding depression after selfing <0.5 (Winn et al. 2011). Research on mating system evolution should extend beyond genetic explanations for the maintenance of recombination. Second, a mating system shift toward selfing reduces effective population size and will lead to mutational meltdown in the long run, and this can result in eventual extinction. The probability of extinction would be further enhanced if the decline in Ne were associated with reduced genetic variation for ecologically relevant traits (Van Buskirk and Willi 2006; Willi et al. 2006; Willi et al. 2007b) or restricts the ability to adapt to environmental challenges (Willi and Hoffmann 2009). Hence, the time to extinction should depend on factors such as Ne and the rate of environmental change or the importance of coevolutionary arms races (Morran et al. 2009; Morran et al. 2011). Mutational meltdown is therefore a likely explanation for the increased extinction rate of selfing taxa, and must be an important cause for the bias toward outcrossing in hermaphrodites.

Associate Editor: M. Johnston


I thank the many people who helped with crossing and assessing plants: J. Brunner, R. Duenner, A. von Ins, D. Lang, K. Määttänen, M. Vance, E. Willi, J. Winkler, and C. Winteler. M. Willi helped extract pollen data. J. Van Buskirk, J. Edwards, B. Mauch-Mani, and four anonymous reviewers made helpful comments on the manuscript. Collection permits were granted by the Palisades Interstate Park Commission, the Nature Conservancy of eastern New York, the New York State Office of Parks, the Commonwealth of Pennsylvania, the U.S. National Park Service, the Illinois Department of Natural Resources (DNR), the Michigan DNR, and John Haataja. Thanks to B. Mable for collecting seeds from three populations. I was supported by the Swiss National Science Foundation (31003A-116270, PP00P3–123396/1), the Genetic Diversity Centre of ETH Zürich, the Institute of Evolutionary Biology and Environmental Studies of the University of Zürich, the Botanical Garden of the University of Neuchâtel, and the Fondation Pierre Mercier pour la Science, Lausanne.