I. lacunosa and I. cordatotriloba are noxious weeds that are indigenous to the southeastern United States (Jones and Deonier, 1965). The two plants have different floral morphologies, with I. lacunosa typically having small white flowers, and I. cordatotriloba normally having larger, purple flowers (Abel and Austin, 1981). A recent analysis of shared genetic variation suggests that I. lacunosa and I. cordatotriloba are exchanging genetic material because they are genetically indistinguishable at neutral markers (Duncan, 2013, in press). Reflecting this genetic similarity, the two species can cross and produce viable offspring (Abel and Austin, 1981; Duncan, 2013, in press). In North Carolina (NC) and South Carolina (SC) the ranges of the two species overlap, with I. lacunosa growing along the coast as well as in the central area of the two states, while I. cordatotriloba is found predominately along the coast (Duncan, 2013, in press). In NC and SC the plants germinate in late May and begin to flower in August or early September. Flowering ceases sometime in mid to late fall and plants die at the first hard frost. Plants of each taxon are self-compatible; however, I. cordatotriloba has been described as having a mixed mating system (selfing and outcrossing), while I. lacunosa is thought to reproduce largely by self-pollination (McDonald et al., 2011).

We did not estimate selfing rates directly; instead, selfing rates were estimated indirectly from neutral markers (microsatellites). In particular, our sampling of populations and individuals allowed us to estimate selfing rates from observed heterozygosity, H. Selfing rate is related the inbreeding coefficient, F, by s = 2 F/(1 + F) (Hartl and Clark, 1997). In turn, the inbreeding coefficient is related to heterozygosity by F=(H0−H)/H0 where H₀ is the expected heterozygosity based on allele frequencies and random mating (Hartl and Clark, 1997). Combining these two equations yields the relationship between selfing rate and heterozygosity: s=2(H0−H)/(2H0−H) Estimation of selfing rates for individual populations were based on four microsatellite loci sampled from 97 I. cordatotriloba individuals from 7 populations and 110 I. lacunosa individuals from 8 populations (Duncan, 2013, in press). Observed and expected heterozygosity were calculated using the program Arlequin 3.5 (Excoffier and Lischer, 2010).

The statistical significance of variation in morphological characters among I. cordatotriloba populations was assessed using a one-way analysis of variance as implemented by the GLM procedure in SAS (SAS Institute Inc., Cary, NC, 1989–2007). The relationship between population means for selfing rate and morphological variables and their squares was assessed by regression analysis using a backward elimination procedure implemented by the REG procedure in SAS. The critical value for keeping a variable in the model was set to p = 0.05.

To determine whether trait divergence between the two species is larger than expected under neutrality, we used a modified F_ST – Q_ST approach (Whitlock and Gilbert, 2012). The standard F_ST – Q_ST approach compares populations at a single spatial level (e.g., Leinonen et al., 2008; Whitlock, 2008). However, in our analysis populations are nested within species and we are primarily concerned with examining divergence between species. There are thus two hierarchical “spatial” levels.

For each morphological trait measured, we calculated the summary statistic Q_CT using the method developed by Whitlock and Gilbert (Whitlock and Gilbert, 2012). Q_CT describes the magnitude of variation for a morphological trait between I. lacunosa and I. cordatotriloba. Under the assumption that trait divergence is due to genetic drift, the expected value of Q_CT is QCT=VC/(VC+VP+2VI)=FCT where V_C, V_P, and V_I are the components of genetic variance among species, among populations, and among individuals within populations, respectively (Whitlock and Gilbert, 2012) and F_CT is the between-species F-statistic for the microsatellite loci.

To test whether natural selection contributed to trait divergence between the two species, we tested whether the null hypothesis of no selection, corresponding to the above equation, could be rejected in favor of the hypothesis that Q_CT > F_CT. Using bootstrap sampling, we compared the distributions of Q_CT and F_CT. To generate a bootstrap sample from one of the species, we first randomly chose a population from that species. We then drew n individuals randomly with replacement from those in that population, where n was the actual number of individuals scored in that population. We continued sampling in this way until the number of populations in the sample was equal to the number actually scored. For each species, we preformed 1000 bootstrap samples.

For each bootstrap sample, we calculated Q_CT or F_CT from the variance components from a standard nested ANOVA. Because V_I is the additive genetic variance within populations and could not be calculated for the traits from our data, we instead examined two extremes: heritability = 1 and heritability = 0. For heritability = 1, V_I was set equal to the within-population component of variation from the nested ANOVA for the trait (equal to the within-population phenotypic variance); for heritability = 0, we set V_I = 0. Because both approaches led to similar results, we report only the results for V_I = within population variance component. This approach is conservative because it produces a smaller Q_CT.

The eight populations of I. lacunosa we sampled uniformly exhibited very high selfing rates (mean s = 0.955 ± 0.018 Table 1). By contrast, I. cordatotriloba populations were more variable across populations [mean s = 0.511 ± 0.182 (negative estimates considered 0); Table 1]. This difference is statistically significant (one-tailed t-test: t = 2.43, df = 13, p < 0.025). It thus appears that selfing rate is on average substantially higher for I. lacunosa than for I. cordatotriloba.

Because in many plant species, variation in selfing rate is associated with variation in floral morphological characteristics, we examined the relationship between selfing rate and morphological variables corolla length and width, anther-stigma distance and style length and their squares for the combined data from the two species (Table 2). In a backward-elimination multiple regression involving these variables, all variables except anther-stigma distance and the square of style length were eliminated from the regression. The regression involving these two variables explained 79% of the variation in selfing rate (Figure 2). Anther-stigma distance and style length were both highly significant [F_{(1, 9)} = 11.63 and 31.36, p < 0.0077 and 0.0003, respectively]. Moreover, each of these variables differed significantly between the two species [Anther-stigma distance: F_{(1, 10)} = 20.14, p = 0.0012; Square of style length: F_{(1, 10)} = 64.93, p < 0.0001]. It thus appears that much of the variation between species is due to the joint effects of these two morphological variables.

As remarked above, I. cordatotriloba populations are substantially variable for selfing rate. Among just these populations, a multiple regression of selfing rate on anther-stigma distance and the square of style length yielded a significant effect of the latter variable [F_{(1, 4)} = 8.40, p = 0.0442]. The effect of anther-stigma distance was not quite significant [F_{(1, 4)} = 1.98, p = 0.118], though this is perhaps not surprising given the small number of populations. Overall, the square of style length explains 52% of the variation in selfing rate among I. cordatotriloba populations and anther-stigma distance explains another 27%. Moreover, the estimated regression coefficients are similar for I. cordatotriloba and the combined 2-species sample (anther-stigma distance: 0.72 ± 0.361 vs. 0.81 ± 0.24, respectively; square of style length: −0.0070 ± 0.0024 vs. −0.0053 ± 0.00096, respectively). It thus seems likely that, as for the overall difference in selfing rates between the two species, these variables also account for much of the variation in selfing rate among I. cordatotriloba populations.

To test whether divergence in floral morphological traits between I. cordatotriloba and I. lacunosa is consistent with neutral divergence, we conducted a hierarchal Q_CT − F_CT analysis. If Q_CT > F_CT, divergence in the trait is typically inferred to be caused by selection. By contrast, if Q_CT = F_CT, then the data is consistent with divergence by genetic drift. We performed a comparison of Q_CT vs. F_CT for the measured floral traits, as well as for leaf traits that served as a control.

Q_CT values calculated for flower length, flower width, and flower length-width ratio are extremely differentiated from F_CT values (Figure 3). For flower length and width, the distribution of bootstrap values of F_CT does not overlap the distribution for Q_CT, while for the ratio of floral length/width, there is only minimal overlap. In all three cases the significance of the difference is p < 0.001, indicating divergence is inconsistent with neutral expectations. Because these traits are moderately correlated (Table 3), these tests may not be independent and divergence in these traits may reflect selection on a single, composite developmental character. Style length and anther-stigma distance are less correlated with corolla length and width and with each other than are corolla length and width to each other (Table 3). Thus, style length and anther-stigma distance presumably represent separate developmental modules, while corolla length and width do not. However, for these characters there is little or no overlap between the bootstrap distributions of Q_CT and F_CT (Figure 4). Both of these differences are significant (style length, p < 0.001; anther-stigma distance, p = 0.002), again indicating inconsistency with neutral expectations.

By contrast with the floral characters, distributions of Q_CTand F_CT values broadly overlap for leaf length, leaf width, leaf length-width ratio (Figure 5) and differences are not statistically significant (leaf length, p = 0.80; leaf width, p = 0.64; leaf length-width ratio, p = 0.44). This pattern provides no evidence for selection causing divergence in these traits and indicates they are diverging neutrally.

Our results verify that I. lacunosa is highly selfing whereas I. cordatotriloba appears to have a mixed mating system. This difference in selfing rates is explainable by reduced anther-stigma distance in I. lacunosa, which in other plant species increases autogamy (Chang and Rausher, 1998; Schueller, 2004; Takebayashi et al., 2006), as well as by a reduction in style length.

I. cordatotriloba populations are much more variable in estimated selfing rate than I. lacunosa populations. Some populations exhibit selfing rates nearly as high as those in I. lacunosa, whereas others are estimated to be largely outcrossing (Table 1). Although spatial variation in the availability of pollinators could conceivably contribute to this variation (Barrett et al., 1996), most of it appears to be explainable by variation in floral morphology. In particular, selfing rate seems to be determined largely by anther-stigma distance and style length. While reduced anther-stigma distance typically increases selfing rates in plants (Motten and Stone, 2000), it is unclear how reduced style length contributes to increased selfing in the two Ipomoea species. The existence of between-population variation in these characters in I. cordatotriloba suggests that standing genetic variation in these characters may have facilitated even further reduction in anther-stigma distance and style length in I. lacunosa.

As described in the introduction, Category 1 traits are traits for which variation directly affects selfing rate and that are likely targets of selection acting to alter selfing rate. Our analyses suggest that both anther-stigma distance and also style length are likely Category 1 traits. Both of these traits are highly correlated with selfing rates at the population level.

Our a priori expectation was that divergence in Category 1 traits would likely be caused by natural selection rather than genetic drift because it is difficult to imagine that variation in selfing rate would be selectively neutral. In order to be neutral, there would have to be precise tradeoffs among inbreeding depression, pollen discounting, and the benefits of reproductive assurance (Holsinger, 1988) that yielded no fitness differences among genotypes with different character values of these traits, which would require an improbable balance of effects. This expectation was in fact realized: both anther-stigma distance and style length exhibited a Q_CT between species that was substantially greater than F_CT, indicating that divergence is inconsistent with neutral processes.

Although a number of advantages have been suggested for increased selfing, including pollinator uncertainty, avoidance of interspecific hybridization, and increased genetic transmission (Fisher, 1941; Smith and Rausher, 2007), we currently have no information that would allow us to distinguish among these possibilities. However, it is likely that the evolution of high levels of selfing set the stage for the subsequent evolution of other selfing-syndrome characters.

Our results indicate that divergence between I. lacunosa and I. cordatotriloba in floral size was driven by natural selection. One possible explanation for this divergence is that it is the result of indirect selection due to correlations between anther-stigma distance and corolla dimensions and selection on the former. We believe, however, that this explanation is unlikely. Although we did not measure genetic correlations among these traits, the phenotypic correlations between anther-stigma distance and corolla dimensions are very low and not significantly different from 0. We suspect these reflect underlying low genetic correlations that would essentially preclude the evolution of reduced corollas due to indirect selection.

It also seems unlikely that direct selection to reduce selfing rates acted to reduce corolla size in I. lacunosa. Because variation in anther-stigma distance and style length accounts for approximately 80% of the variation in selfing rates, while corolla dimensions are not significantly related to selfing rate, there is little scope for this type of selection. Instead, we suspect that reduction in corolla size in I. lacunosa resulted from selection to redirect resources from petals to other fitness-enhancing traits (Brunet, 1992). Final resolution of this issue, however, must await functional analysis of selection on these traits.

Even though floral size and mating systems have been found to be extremely correlated, a systematic survey of outcrossing and selfing species found that there is a much weaker correlation between flower size and vegetative tissue dimensions, such as leaf length or width (Ashman and Majetic, 2006). We also found very weak phenotypic correlations between floral traits and leaf size, suggesting that evolution of leaf traits should not be influenced by indirect selection on floral traits. Our results are consistent with this expectation. In particular, we found no significant differences between F_CT and Q_CT for leaf dimensions, indicating that divergence in leaf traits are primarily due to genetic drift. Thus, as has been found in other systems (Ashman and Majetic, 2006), evolution of leaf size appears to be independent of the evolution of floral characters.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.