Effects of mating system on adaptive potential for leaf morphology in Crepis tectorum (Asteraceae)

Abstract

Background and Aims

A shift from outcrossing to selfing is thought to reduce the long-term survival of populations by decreasing the genetic variation necessary for adaptation to novel ecological conditions. However, theory also predicts an increase in adaptive potential as more of the existing variation becomes expressed as homozygous genotypes. So far, relatively few studies have examined how a transition to selfing simultaneously affects means, variances and covariances for characters that might be under stabilizing selection for a spatially varying optimum, e.g. characters describing leaf morphology.

Methods

Experimental crosses within an initially self-sterile population of Crepis tectorum were performed to produce an outbred and inbred progeny population to assess how a shift to selfing affects the adaptive potential for measures of leaf morphology, with special emphasis on the degree of leaf dissection, a major target of diversifying selection within the study species.

Key Results

Three consecutive generations of selfing had a minor impact on survival, the total number of heads produced and the mean leaf phenotype, but caused a proportional increase in the genetic (co)variance matrix for foliar characters. For the degree of leaf dissection, the lowest 50th percentile of the inbred progeny population showed a disproportionate increase in the genetic variance, consistent with the recessive nature of the weakly lobed phenotype observed in interpopulation crosses. Comparison of inbreeding response with large-scale patterns of variation indicates a potential for selection in a (recently) inbred population to drive a large evolutionary reduction in degree of leaf dissection by increasing the frequency of particular sibling lines.

Conclusions

The results point to a positive role for inbreeding in phenotypic evolution, at least during or immediately after a rapid shift in mating system.

INTRODUCTION

Natural selection is a major force driving evolutionary change and underlies much of the phenotypic diversity observed within and among species (Darwin, 1859; Turesson, 1922; Primack and Kang, 1989). It has long been recognized, however, that selection can interact with other processes to produce adaptive change and divergence (Wright, 1978; Cohan, 1984). One classical example concerns the possibility for random drift to enhance the adaptive potential of outcrossing populations by converting non-additive to additive (selectable) genetic variance (Robertson, 1952; Cheverud et al., 1999). Another example – and the topic of the present paper – concerns the interaction between selection and mating system, or, more specifically, the effects of inbreeding on the adaptive potential of populations that switch from outcrossing to partial or complete selfing (Allard et al., 1968; Cockerham, 1983; Kelly, 1999a, b; Kelly and Williamson, 2000).

Selfing has evolved repeatedly and therefore occurs in a variety of plant lineages (Stebbins, 1957; Barrett and Eckert, 1990; Takebayashi and Morrell, 2001). The transition to selfing is usually attributed to selection, driven by, among other things, poor pollination conditions (Jain, 1976; Kalisz et al., 2004) and the selective removal (purging) of those alleles that depress fitness under inbreeding (Lande and Schemske, 1985). Selfers are often regarded as ‘evolutionary dead ends’ (Stebbins, 1957; Takebayashi and Morrell, 2001) because of their expected tendency to lose genetic variation or undergo long-term accumulation of deleterious mutations due to random drift (Charlesworth et al., 1993; Charlesworth and Charlesworth, 1995; Glémin and Ronford, 2013). However, there is long-standing evidence to suggest that genetic variation and adaptive potential can remain high for some time after a shift in mating system (Allard and Jain, 1962; Allard et al., 1968; Schoen and Brown, 1991). In fact, many selfers have evolved floral traits adapted for self-pollination (Stebbins, 1957; Ornduff, 1969) and (or) have undergone local genetic adaptation in response to spatial variation in the environment (Hereford, 2010; Novy et al., 2013). Obviously, there must be factors that oppose, or at least retard, the loss of adaptive potential predicted for selfing lineages.

Population genetic theory suggests that adaptive potential for phenotypic traits can increase after a shift to selfing: more of the genetic variation becomes visible to selection as alternative alleles segregate into homozygotes; furthermore, previously hidden (rare) recessives become exposed at the phenotypic level (Robertson, 1952; Falconer, 1989; Charlesworth, 1992; Glémin and Ronfort, 2013). However, a selfing population is not likely to have a greater adaptive potential than its outbreeding ancestor if the increase in genetic variance is accompanied by reductions in direct components of fitness such as viability and fecundity (inbreeding depression; Charlesworth and Charlesworth, 1987). For suites of traits related by development, e.g. different measures of leaf morphology, it is also necessary to consider changes in the matrix of genetic variances and covariances (the G matrix), given the key role played by the G matrix in determining short-term adaptive potential (Lande, 1979) and the perturbed covariance structure that might result from inbreeding (Phillips et al., 2001; Andersson et al., 2010). So far, relatively few studies have been performed to assess how a shift to selfing simultaneously affects components of fitness and genetic (co)variances for ecologically important characters (Shaw et al., 1998; Lynch et al., 1999; Kelly and Arathi, 2003; Holeski and Kelly, 2006; Andersson, 2012).

Angiosperm leaves vary extensively in morphology, from entire to finely dissected or compound, with many genera, and even some species, showing the full range of variation between extreme phenotypes (Jones et al., 2009; Nicotra et al., 2011). Many lineages have undergone rapid and putatively adaptive shifts in leaf morphology (Jones et al., 2009; Geeta et al., 2012; Schmerler et al., 2012), driven by changes in critical physiological functions such as hydraulic efficiency, thermoregulation and light interception (Gates et al., 1968; Parkhurst and Loucks, 1972; Givnish, 1987; Nicotra et al., 2011). Consistent with these large-scale patterns, there is evidence for past or ongoing selection on foliar characters in present-day species (Lewis, 1969; Gurevitch, 1988; Willis, 1996; Bright and Rausher, 2008). However, despite evidence that leaves of some species become smaller under inbreeding (Willis, 1996; Rao et al., 2002; Ellmer and Andersson, 2004), it is still uncertain whether a shift to selfing can promote adaptive change in leaf morphology (Rao et al., 2002; Ellmer and Andersson, 2004).

Populations of Crepis tectorum (Asteraceae) have diverged for a number of characters, including mating system and degree of leaf dissection (Fig. 1A; Andersson, 1989a, b, 1991). The most finely dissected leaves are produced by C. tectorum subsp. pumila (Liljebl.) Sterner, a morphologically distinct, self-sterile ecotype adapted to the dry, treeless outcrop areas (‘alvars’) on the Baltic island of Öland. Deeply dissected leaves have a large capacity to dissipate excess heat (Givnish, 1987; Nicotra et al., 2011) and probably evolved within subsp. pumila to avoid overheating or transpiration under the hot, sunny conditions prevailing in the alvar habitat (Andersson, 1989a, 1991). In fact, dissected leaves still represent a major target of selection in present-day populations of this ecotype (Andersson, 1992). Since weakly lobed leaves behave as a recessive trait in interpopulation hybrids (Andersson, 1991, 1995) and tend to be a feature of self-fertile, autogamous populations adapted to relatively shady and mesic habitats, e.g. arable fields (Andersson, 1989a, b), it becomes meaningful to ask whether a transition to selfing could promote adaptive change toward more simple leaves.

Fig. 1.

Histograms showing the distribution of leaf dissection in Crepis tectorum, based on population means (data from Andersson, 1991), or means of progeny derived by random outcrossing or three generations of selfing in a single population of subsp. pumila (this study). The dissection index represents the ratio of minimum to maximum leaf width (more deeply lobed leaves toward the left).

In a recent study of C. tectorum, the mating regime of a population of subsp. pumila was experimentally manipulated to determine how selection and inbreeding interact to drive the evolution of small, autogamous flowers within this species (Andersson, 2012). The shift to selfing was found to promote evolutionary reduction of flower size, both directly, by causing inbreeding depression in mean phenotype, and indirectly, by exposing more of the underlying genetic variation to selection. In the present work, we extend the analysis to a suite of foliar traits with particular focus on degree of leaf dissection. In addition to assessing the effects of selfing on means and variances, we considered changes in the G matrix, based on the tight developmental links previously observed for foliar traits in this study system (Andersson, 1989a). To evaluate the role of inbreeding in foliar evolution, we related the foliar inbreeding responses to large-scale patterns of variation in leaf morphology (cf. Lynch et al., 1999; Rao et al., 2002; Andersson, 2012) and levels of inbreeding depression in two direct components of fitness (survival and total head number).

MATERIALS AND METHODS

Plant material

Crepis tectorum L. is a diploid annual (2n = 8), geographically differentiated into a widespread weed ecotype and a series of geographically restricted outcrop forms in north-west Eurasia, the most distinctive being the Öland endemic subsp. pumila (Babcock, 1947; Andersson, 1993). Plants of C. tectorum usually behave as winter annuals, germinating in early autumn, overwintering as rosettes and flowering in the following summer. The yellow, ligulate flowers are arranged in heads (capitula) and develop into one-seeded, indehiscent fruits (hereafter ‘seeds’). Flowers of subsp. pumila are self-sterile and require outcross pollen to set seed, as opposed to the self-fertile and autogamous flowers normally found in C. tectorum (Andersson, 1989b).

The plants used in this study are descendants of seeds originally collected in a population of C. tectorum subsp. pumila, situated approx. 1·5 km south of the village of Vickleby on southern Öland. This population is genetically variable for measures of leaf morphology (Andersson, 1999) and has been found to contain a low frequency (approx. 5 %) of fully self-fertile genotypes (Andersson, 2006). A series of pairwise crosses between self-fertile and normal (self-sterile) individuals, followed by three generations of random outcrossing, were used to establish a large base population with an elevated frequency of self-fertile individuals (approx. 40 %; for details, see Andersson, 2012). The base population was used to produce two sets of fullsib families (progeny populations), one derived by random crosses between pairs of individuals (‘outbred progeny’) and one derived by three generations of selfing, each founded from a randomly selected (self-fertile) maternal parent in the base population (‘inbred progeny’). Seeds for the outbred progeny population were stored in separate paper bags until the seeds for the inbred progeny population became available. Under the assumption of no prior inbreeding in the base population, the coefficient of inbreeding (F) should be 0 and 0·875 for the outbred and inbred population, respectively (Falconer, 1989). Some unintended selection may have taken place during the production of inbred progeny, as only surviving, self-fertile plants could contribute selfed seeds to the next progeny generation.

Cultivation experiment

In November 2010, we established 84 outbred and 80 inbred progeny families to obtain data on leaf morphology. Two seeds per family were planted into each of seven 70 cm³ cells (filled with standard peat soil) in a series of plug trays (connected side by side to reduce edge effects) on a single bench in a semi-automated greenhouse at the Department of Biology, Lund University. Seeds from a particular family were randomized across the whole planting area. If both seeds in a cell germinated, the slowest seedling to emerge was removed and (if necessary) used to replace a missing seedling in another cell assigned to that family. Plants were watered when needed and given additional light during cloudy periods (12 h day lighting), but no extra fertilizer was added. When all plants had attained maximum leaf dissection (April 2011), the youngest fully developed leaf in each rosette was pressed. All leaves were collected within a 2 d period.

Measurements

Fitness was assessed by recording whether or not a plant survived to leaf collection (scored as 1 or 0, respectively) and by counting the heads on each plant after the flowering season. The following measures were scored on the pressed leaves: length, maximum width, minimum width (between two lobes), tip distance (distance from the leaf tip to the widest point on the lamina), the number of teeth along the leaf margin and the degree of leaf dissection (the ratio of minimum to maximum width). All leaf traits have diverged within C. tectorum and show significant heritability within the study population (Andersson, 1989a, 1991, 1999). Leaf data were obtained for a total of 442 outbred plants and 414 inbred plants.

Statistical analyses

The effects of selfing on adaptive potential for leaf morphology was evaluated by contrasting the outbred and inbred progeny population with respect to genetic (co)variance structure. As a first step, we partitioned the phenotypic variance (V_total) in each leaf trait into its among- and within-family components (V_fam, V_within) and then compared the genetic variance (measured by V_fam) and its standardized value (H² = V_fam/V_total, termed broad-sense heritability) between the two progeny populations. These parameters quantify adaptive potential as the ‘visibility’ of the genetic variation to selection at the phenotypic level. This relatively simple approach does not distinguish between additive and non-additive sources of genetic variation (cf. Shaw et al., 1998; Kelly and Arathi, 2003) and relies on the premise that much of the non-additive variance becomes available to selection under rapid inbreeding (Falconer, 1989). For degree of leaf dissection, we tested for a particularly large increase in V_fam and H² toward the recessive (weakly lobed) phenotype by contrasting the lower and upper 50th percentile of the two progeny populations. The 50th percentile was determined separately for each distribution, based on approximately the same median value for the two progeny populations (approx. 0·08). Point estimates and 95 % confidence intervals (CIs) of relevant parameters were obtained using restricted maximum-likelihood procedures (Lynch and Walsh, 1998).

As a complementary approach, we used common principal components (CPCs) analysis to assess the effect of mating system on the G matrix. This technique compares two or more (co)variance matrices [estimated with one-way analyses of covariance (ANCOVAs)] in a hierarchical way, starting with unrelated structure, progressing through a series of models in which the matrices share an increasing number of principal components, and ending with models in which the matrices are identical or differ by a simple constant of proportionality (Phillips and Arnold, 1999). Following the comparison of the outbred and inbred G matrix, we tested for long-term stability in the (co)variance structure for the foliar traits by comparing the two G matrices with an estimate of the among-population (co)variance matrix, based on data from a previous comparison of 54 greenhouse-grown populations (each represented by leaves from approx. 16 plants; Andersson, 1991). To avoid mathematical redundancy, we excluded the leaf dissection index (the ratio of minimum to maximum leaf width) in all CPC analyses.

To quantify changes in mean leaf phenotype and total head number, we estimated the least-square mean and its 95 % CI for each trait and progeny population using a general linear model analysis of variance (ANOVA) with ‘family’ as a random effect. This approach accounted for family structure and eliminated the heteroscedasticity that would result from combining data from the two progeny populations in the same ANOVA (S. Andersson, University of Lund, Sweden, unpubl. res.). The grand means were used to estimate the traditional inbreeding-depression coefficient (termed ∂ to conform with common usage), obtained as one minus the ratio of the inbred mean to the outbred mean (Charlesworth and Charlesworth, 1987). For foliar traits, we also scaled the difference between the outbred and inbred mean by the current range of population mean phenotypes, calculated from previous greenhouse data (see above). The resulting parameter (termed ∂_range) expresses inbreeding response in units of evolutionary change under the assumption that plants in this and the previous study experienced similar growth conditions. For traits that had to be log transformed to meet parametric assumptions (see the Results), we estimated ∂ and ∂_range from back-transformed means. To make estimates directly comparable with data from other studies (e.g. Husband and Schemske, 1996), the means of the inbred progeny population were adjusted by linear interpolation to reflect an F-value of 0·5 (corresponding to a single generation of selfing).

The effect of mating system on survival rate (a binary trait) was tested for significance using a contingency test (data pooled across families).

All statistical procedures were performed with SPSS version 19 (IBM, Tulsa OK, USA), except for the estimation and comparison of G matrices, which were carried out with H2boot (Phillips, 1998a) and CPCrand (Phillips, 1998b), respectively.

RESULTS

The among-family variance (V_fam) was significantly greater than zero for all measures of leaf morphology (CIs excluding 0 in all cases, Table 1). When expressed as a proportion of the total variance (H²), the among-family component of variance increased from 12–30 % in the outbred base population to 36–68 % after three generations of selfing. The estimates of H² were greatest for minimum leaf width and degree of leaf dissection, for both the outbred and inbred progeny population (Table 1).

Table 1.

Estimates of genetic variance (measured as V_fam) with 95 % confidence intervals (CI) for measures of leaf morphology obtained from outbred and inbred progeny of C. tectorum subsp. pumila

	Outbred progeny			Inbred progeny
Variable	Vfam	CI	H2	Vfam*	CI	H2
Leaf length	20·63	10·41–30·84	0·25	37·31	22·08–52·55	0·44
Maximum leaf width	0·96	0·42–1·50	0·20	2·42	1·36–3·49	0·36
Minimum leaf width†	0·13	0·07–0·19	0·29	0·31	0·20–0·42	0·68
Tip distance	3·17	0·79–5·55	0·12	11·12	6·30–15·95	0·37
Number of teeth	1·67	0·61–2·73	0·16	6·53	3·93–9·14	0·48
Degree of leaf dissection†	0·16	0·09–0·23	0·30	0·37	0·24–0·50	0·67

H² denotes the broad-sense heritability (V_fam/V_total).

n = 442 and 414 for outbred and inbred plants, respectively.

* A value in bold is significantly greater than the corresponding value for outbred progeny (no overlap between CIs).

^† Analyses and statistics based on log-transformed data.

Figures 1 and 2 illustrate the variation observed for the degree of leaf dissection after collapsing the data into family means. Both the outbred and inbred population showed a frequency distribution skewed to the right, i.e. with a tail directed toward the recessive (simple-leaved) phenotype (Fig. 1B, C). The tail of simple-leaved families was most pronounced after selfing: the frequency distribution of inbred families spanned almost all of the foliar variation seen at the among-population level (Fig. 1A, C). In agreeement with this observation, the increase in V_fam and H² due to selfing was particularly large for the upper 50th percentile of the frequency distribution (8- to 9-fold vs. ≤2-fold for the lower percentile; Table 2).

Fig. 2.

Silhouettes of leaves from progeny families representing the extremes for degree of leaf dissection (both from the inbred progeny population).

Table 2.

Estimates of genetic variance (measured as V_fam) with 95 % confidence intervals (CI) for degree of leaf dissection obtained from outbred and inbred progeny of plants of C. tectorum subsp. pumila classified as weakly or strongly dissected (representing the upper or lower 50th percentile, respectively)

	Weakly dissected			Highly dissected
Plant material	Vfam	CI	H2	Vfam	CI	H2
Outbred progeny	0·021	–0·019 to 0·061	0·07	0·018	0·000–0·035	0·16
Inbred progeny	0·205*	0·111–0·300	0·60	0·027	0·010–0·044	0·32

H² denotes the broad-sense heritability (V_fam/V_total).

Analyses were based on log-transformed data.

* Significantly higher than all other estimates (no overlap between CIs).

The CPC analysis that compared the G matrix of the two progeny populations rejected the equality model (P = 0·012), but provided support for the model that the two matrices differed by a simple proportional constant (P = 0·463). A pairwise comparison of each G matrix with the among-population (co)variance matrix, as well as a simultaneous analysis of all matrices, showed the same pattern: little or no support for the equality model (P≤0·002) and a failure to reject the proportionality model (P = 0·104–0·493). In agreement with these test results, the matrices differed greatly in the magnitude of (co)variance, with the largest estimates for the among-population matrix and the lowest for the outbred G matrix (Supplementary Data Table S1). We identify three major patterns that contribute to the shared structure between matrices: (1) leaf length, width and tip distance covaried positively; (2) negative covariances always involved minimum leaf width; and (3) the number of lobes and teeth varied independently of other variables (Supplementary Data Table S1).

The selfing phase caused a slight increase in survival rate (67·3 % vs. 53·0 % for outbred progeny; χ² = 24·4, d.f. = 1, P<0·001) but did not significantly influence the number of heads produced (mean 6·87, CI = 6·43–7·34 vs. 6·98, CI = 6·31–7·73 for outbred progeny). The inbred population had smaller leaves with fewer teeth relative to the outbred population, while the difference in minimum leaf width and degree of leaf dissection failed to reach significance (Table 3). Estimates of ∂ were low (≤0·060), especially for degree of dissection (0·005, Table 3), head number (0·010) and survival rate (–0·155). Scaling by the range of population means resulted in somewhat higher estimates of inbreeding response for leaf length (∂_range=0·043) and tip distance (0·063) than for other leaf traits (≤0·031, Table 3).

Table 3.

Grand means with 95 % confidence intervals (CI) for measures of leaf morphology obtained from outbred and inbred progeny of C. tectorum subsp. pumila

	Outbred progeny		Inbred progeny
Variable	Mean	CI	Mean*	CI	∂	∂range
Leaf length (mm)	59·60	58·79–60·42	55·92	55·22–56·62	0·035	0·043
Maximum leaf width (mm)	12·36	12·16–12·57	11·75	11·54–11·97	0·028	0·023
Minimum leaf width (mm)†	1·13	1·06–1·19	1·05	1·01–1·09	0·039	0·006
Tip distance (mm)	22·00	21·51–22·50	19·68	19·24–20·13	0·060	0·063
Number of teeth	15·06	14·75–15·37	13·63	13·35–13·90	0·054	0·031
Degree of leaf dissection†	0·093	0·087–0·099	0·092	0·088–0·096	0·005	0·001

∂ and ∂_range denote the response to inbreeding, scaled by the mean of outbred progeny and the range of population means, respectively.

n = 442 and 414 for outbred and inbred plants, respectively.

* A value in bold is significantly lower than the corresponding value for outcrossed progeny (no overlap between CIs).

^† Analyses and statistics (back-transformed) based on log-transformed data.

DISCUSSION

In the present investigation of C. tectorum, we have compared patterns of variation in leaf morphology before and after a transition to selfing, with particular focus on the degree of leaf dissection, a major target of selection within the study species. Our analyses showed a positive effect of inbreeding on the level of genetic variance for measures of leaf morphology but revealed little change in mean phenotype and the pattern of covariance among the foliar characters. Consistent with the pattern of dominance observed in earlier crossing experiments, there was a disproportionate increase in the level of genetic variance for the 50th percentile of individuals that had the least dissected leaves in the inbred progeny population. These and other observations indicate that inbreeding can enhance adaptive potential for ecologically important traits, at least during or immediately after a shift in mating system.

Adaptive potential

Results of this study confirm previous observations that populations of C. tectorum subsp. pumila exhibit genetic variation in leaf morphology (Andersson, 1999) but also show that more of this variation becomes exposed to selection after a shift to selfing. The proportion of variance due to family increased from 12–30 % in the base population to 36–68 % after three generations of selfing, consistent with the expected conversion of within- to among-family variation as alternative alleles segregate into homozygotes (Falconer, 1989). Such changes can be expected to accelerate the response to selection, although the magnitude of this effect also depends on allele frequencies, patterns of dominance and levels of linkage disequilibrium between loci (Robertson, 1952; Allard et al., 1968; Charlesworth, 1992; Kelly, 1999a, b; Kelly and Williamson, 2000).

Judging from the results of the CPC analyses, the genetic (co)variances largely increased in a proportional manner, i.e. without major changes in the principal component structure. Thus, the genetic associations between different foliar characters remained essentially the same after the transition in mating system. Interestingly, the G matrices of the outbred and inbred progeny population shared a substantial portion of their principal component structure with the among-population (co)variance matrix, extending previous observations of close, persistent associations among foliar traits in C. tectorum and other plant species (Andersson, 1991, 1999; Klingenberg et al., 2012). Given these observations and the strong impact of the G matrix on the short-term trajectory of evolution by selection (Lande, 1979), we infer that outbred and (recently) inbred populations of subsp. pumila would respond similarly in direction (but not in magnitude) to the same selection pressure on leaf morphology.

Dissected leaves have been observed to behave as a dominant trait in crosses between C. tectorum subsp. pumila and simple-leaved populations of the same species (Andersson, 1991, 1995). To the extent that similar dominance effects exist for loci that segregate within natural (nonhybrid) populations, the genetic variance after selfing should be highest in the recessive direction, as the phenotypic effects of previously hidden (rare) recessives become exposed by inbreeding (Robertson, 1952; Falconer, 1989). Our results conform to this prediction: selfing extended the tail of weakly dissected outlier families and caused a disproportionate increase in the genetic variance toward the recessive, simple-leaved phenotype. Similarly, in a previous study of subsp. pumila, outbred plants were more likely to produce weakly dissected leaves if their parents originated from the same 0·25 m² plot in the field than if they came from plots separated by ≥10 m, a result attributed to the increased expression of recessive alleles in the short-range crosses (biparental inbreeding; Andersson, 1999). As found in the present study, the most simple-leaved progeny family in the selfed population had almost the same mean for the leaf dissection index as the most simple-leaved population in a previous survey of within-species variation (Fig. 1). Thus, given a shift toward autogamy and conditions favouring more simply shaped leaves, it is easy to imagine how selection could favour a large, rapid reduction in leaf dissection through the spread of simple-leaved sibling lines, a scenario also invoked for the evolution of smaller flower size within the same study system (Andersson, 2012).

Patterns of inbreeding depression

The proportional increase in genetic variance that resulted from three generations of selfing was not accompanied by large changes in the mean phenotype. Inbreeding did reduce tooth number and measures of leaf size, but all ∂ estimates were small (≤0·060) and fall within the low range normally documented for foliar traits (Willis, 1996; Rao et al., 2002; Ellmer and Andersson, 2004) and for morphological traits in general (DeRose and Roff, 1999). Similarly, and unlike observations from other plants (Charlesworth and Charlesworth, 1987; Husband and Schemske, 1996), there was little evidence for inbreeding depression in survival rate and head number (∂≤0·010 in both cases). In fact, the inbred progeny population showed a slight (though significant) increase in survival rate relative to the outbred progeny population, which resulted in a negative ∂ value for this variable (–0·155).

The increase in survival rate is exactly opposite to that expected for direct components of fitness (Charlesworth and Charlesworth, 1987; Falconer, 1989) and has at least three possible explanations: (1) that the self-fertile plants used to establish the experimental population carried few deleterious recessives because of past purging effects (Lande and Schemske, 1985); (2) that inbreeding depression was purged during the forced selfing that was necessary to produce the inbred population; and (3) that the selfed lines were unintentionally selected for increased survival and reproductive success in the greenhouse environment. Consistent with the potential for purging or environmentally induced selection, slightly more than half of the inbred lines were lost during the three-generation selfing phase (90 out of 169 initial self-pollinations), because of either low survival or reduced selfing capacity (S. Andersson, University of Lund, Sweden, unpubl. res.). In this regard, it is worth noting that family-level analyses revealed weak, non-significant relationships between leaf morphology and survival rate, both for the outbred and for the inbred subpopulation (Spearman r approx. –0·10, P>0·36; S. Andersson, University of Lund, Sweden, unpubl. res.). Thus, it seems unlikely that selection (on survival) would reduce the foliar inbreeding responses and thus counteract the increased adaptive potential for leaf morphology inferred from the variance component analyses.

The present study is likely to underestimate lifetime inbreeding depression, not only because of unintentional selection effects (see above), but also because of our focus on relatively ‘late’ traits, scored on plants raised under favourable greenhouse conditions. Inbreeding responses in early fitness components such as seed set and germination can make a significant contribution to lifetime inbreeding depression, especially in outcrossing populations (Husband and Schemske, 1996). As for the possibility of bias arising from the use of a benign cultivation environment (Dudash, 1990), the ∂ value for total multiplicative fitness, based on head number and survival rate, was positive when estimated from progeny of subsp. pumila grown in an outdoor garden (∂=0·52; Andersson, 2012), as opposed to the negative value obtained in the present study (∂=–0·25; S. Andersson, University of Lund, Sweden, unpubl. res.). The garden-based ∂ estimate is slightly higher than the threshold inbreeding depression needed to maintain outcrossing (∂=0·5; Lande and Schemske, 1985). This finding, together with the possibility of reductions in unmeasured fitness traits, helps to explain why selfing genotypes remain rare in present-day populations of subsp. pumila (Andersson, 2012). However, given the high selfing capacity of C. tectorum plants adapted to weedy conditions or the smaller and more island-like outcrop sites outside of Öland (Andersson, 1989b), we infer that inbreeding depression in ancestral populations has been too weak – or too easily purged – to act as a long-term constraint on the spread of selfing within the study species.

The foliar inbreeding responses would make small but noticeable contributions to large-scale patterns of variation within C. tectorum: the calculated change in mean phenotype after one generation of selfing spans <7 % of the current range of population means (|∂_range| ≤0·063), and this quantity remains relatively small (<14 %) after doubling each value of ∂_range to account for the additional change that might result from a longer history of selfing (F = 1 rather than 0·5). These observations contrast with the large inbreeding effect observed for flower size in the same study system: one or two generations of selfing caused a >50 % reduction of mean flower size relative to the decline inferred to have taken place during the shift toward autogamy within C. tectorum (Andersson, 2012). Given this difference and the variety of inbreeding effects observed for vegetative and floral features in Brassica cretica (|∂_range| = 0·05–0·49; Rao et al., 2002) and other plant species (Ellmer and Andersson, 2004), the direct contribution of inbreeding to evolution seems trait specific, similar to what has been observed for levels of inbreeding depression in primary components of fitness (Husband and Schemske, 1996).

Limitations of the study

The present study simulates a very rapid increase in inbreeding (F = 0 to 0·875 within three generations) and only concerns short-term effects, based on standing genetic variation in an initially outcrossing population. Furthermore, our approach does not account for genetic phenomena that might arise during early or intermediate stages of inbreeding or in populations that possess stable mating systems involving both selfing and outcrossing (Goodwillie et al., 2005). Partial inbreeding can interact with dominance to generate novel components of non-additive genetic variance, and these ‘inbreeding components’ have potentially large effects on the evolutionary response to selection (Cockerham, 1983; Kelly and Williamson, 2000). Inferring evolutionary potential under partial inbreeding requires a more complex crossing design than that used in the present investigation (Shaw et al., 1998; Kelly and Arathi, 2003; Holeski and Kelly, 2006).

Our results are relevant to a scenario in which the spread of selfing genotypes coincides with a period of selection for more simply shaped leaves, e.g. a population of C. tectorum subsp. pumila (or some other outcrossing ancestor) that colonizes a site with poor pollinator availability and low light levels. Comparative evidence suggests that such selection pressure may have been important in the past history of C. tectorum: self-fertile and highly autogamous flowers tend to be a feature of putatively shade-adapted (simple-leaved) populations, such as those growing in arable fields or on partly shaded outcrops along the eastern coast of Sweden and on the Baltic island of Gotland (Andersson, 1989a, b). Overall, the potential for a parallel shift in mating system and non-floral traits should be great whenever an environmental change toward more extreme conditions simultaneously selects for reproductive assurance (due to low mate or pollinator availability) and morphological attributes that allow individuals to survive under the novel conditions. This possibility deserves further investigation, given the frequent association between high selfing rates and adaptation to stressful or marginal habitats found in other plant species (Levin, 2010, and references therein).

Concluding remarks

Empirical studies have shown that selfing populations can harbour considerable amounts of genetic diversity (e.g. Schoen and Brown, 1991) or provided direct evidence that many selfers retain the capacity to undergo adaptive genetic change (e.g. Allard and Jain, 1962; Ornduff, 1969; Hereford, 2010; Novy et al., 2013). As found in the present study, inbreeding has considerable potential to play a direct positive role in adaptive evolution, both generally, by increasing the overall genetic variance in ecologically important traits, and specifically, by exposing potentially advantageous recessive alleles to selection. More work is needed to determine how inbreeding effects interact with selection to produce adaptive change and divergence under variable and stressful field conditions, and whether such interactions oppose or retard the loss of long-term adaptability predicted for selfing lineages (Stebbins, 1957; Charlesworth et al., 1993; Charlesworth and Charlesworth, 1995; Takebayashi and Morrell, 2001; Glémin and Ronfort, 2013).

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxford-journals.org and consist of the following: Table S1: variance–covariance matrices for measures of leaf morphology obtained from outbred and inbred progeny of C. tectorum subsp. pumila (this study) and a comparison of 54 greenhouse-grown populations of C. tectorum (Andersson, 1991).