Introduction

Genetic variation varies over traits, species and populations. Understanding the sources of this variation is one of the main areas of research in evolutionary genetics. Theoretically, variation in population size and natural selection are, apart from mutation, the most important factors that may change amounts of genetic variation. Quantitative genetic variance can be split into variance caused by the direct inheritance of genes from the parents (additive genetic variance, VA) and variance caused by the specific combination of genes in the offspring (nonadditive variance, mostly dominance variance VD) (Falconer, 1989; Lynch and Walsh, 1998). Consistent (natural) selection should result in the fixation of the most favourable genotype and consequent elimination of additive genetic variation (VA) (Fisher, 1930; Bulmer, 1971; Roff, 1997). Selection, however, does not reduce VD directly. Reduction in population size will also eliminate genotypes and hence reduce VA (Roff, 1997). However, VD may be converted to VA after a severe bottleneck (Willis and Orr, 1993; Wang et al, 1998), thus increasing VA after a bottleneck, although this depends on the type and number of genes involved in the nonadditive variance (Lopez-Fanjul et al, 2000; Naciri-Graven and Goudet, 2003).

In laboratory experiments, some evidence for these processes has been found. After selection for many generations, a decrease in response has been found frequently, but depletion of VA is generally not the case (overview in Roff, 1997, Chapter 4.7). Bottleneck experiments have both resulted in reductions of VA and increases in VA at the expense of VD, although the latter occurs less frequently (Bryant and Meffert, 1996; Whitlock and Fowler, 1999; Saccheri et al, 2001). Comparison of levels of VA and VD in life history traits, thought to be under consistent natural selection, with morphological traits points in the same direction. Heritabilities tend to be lower in life history traits than in morphological traits (Mousseau and Roff, 1987) although VA is not depleted in life history traits (Mousseau and Roff, 1987; Houle, 1992). Moreover, life history traits tend to harbour relatively higher amounts of dominance variance relative to other traits (Crnokrak and Roff, 1995).

Information on these processes in the field and how large a role they play is scarce. Careful analysis in populations of the collared flycatcher (Ficedula albicollis) of genetic variation and the response to natural selection has shown in one case (body condition) that there was indeed a reduction in VA after natural selection (Merila et al, 2001). However, in another case (tarsus length), no change in VA could be demonstrated despite natural selection (Kruuk et al, 2001).

During colonisation of a new area by a species, genetic variation is expected to change. Both strong selection and an initial reduction in population size (a bottleneck) will generally occur. These processes would cause a reduction of VA and a relative increase in VD (see above). In this study, we test this prediction by examining levels of additive and dominance variation for life history and morphological traits in an island population of the Speckled Wood Butterfly Pararge aegeria. This island was relatively recently colonised (Owen et al, 1986) and life history theory and comparison with mainland populations indicate that selection must have played a substantial role after colonisation (Nylin et al, 1993).

Material and methods

Study species and population

The butterflies in this study were sampled from the Atlantic island of Madeira (Portugal, 33°N, 17°W). This island was relatively recently (1970 s) colonised by Pararge aegeria (Owen et al, 1986). Despite close attention of lepidopterologists due to the presence of the unique, endemic P. xiphia, P. aegeria was never previously observed on Madeira before. Initially after colonisation only a few individuals were observed, but now P. aegeria is abundant, and more common than the endemic P. xiphia. Because of the mild climate P. aegeria can breed continuously on Madeira, unlike the rest of its range, and has approximately five (overlapping) generations per year. This suggests that the population at the time of study had gone through about 120 generations since colonisation. Although the exact source of Madeiran P. aegeria is as yet unknown, comparisons with mainland populations suggest that during and since colonisation the population has adapted to the nonseasonal climate, with considerable changes in life history (Nylin et al, 1993; Gotthard et al, 1994). Protandry (males emerging before females) has been lost, growth rate has decreased and starvation resistance has increased, and P. aegeria has thereby shifted in the direction of P. xiphia. These changes are in accordance with predictions from life history theory (Nylin et al, 1993). It is thus likely that natural selection has changed these traits.

Breeding

In all, 24 wild caught females were caught and allowed to oviposit on grass leaves in small containers at the end of October 1997. Females of the speckled wood butterfly have never been found to pair more than once (based on spermatophore counts), so their offspring can be considered full sibs (Wiklund and Forsberg, 1991). Upon emergence, the offspring were transferred to grass tufts standing in water. Individuals were raised solitarily at a constant temperature of 23°C and a day length of 14 h. In the second laboratory generation, offspring of 24 pairs of unrelated, first-generation individuals were raised. Each of these full sib families consisted of 12 individuals.

Traits

Two larval traits, three pupal traits and two adult traits were measured. Both life history traits and morphological traits were included. Two juvenile periods were measured: larval development time, days from egg laying until pupation and pupal development time, days from pupation until adult emergence. Both traits were log transformed prior to analysis to obtain normally distributed data. They have probably been under intense natural selection after colonisation because their means diverge from means in other populations (Nylin et al, 1993). Pupal weight, measured 2 days after pupation, probably has not been under such intense selection since it appears not to have changed much since colonisation. Larval growth rate was calculated as the quotient of pupal weight and larval time (assuming negligible variation in egg weight). Adult size was measured as the total area of hind wings in mm2, determined with an image analyser. Details of measurement procedures can be found in Windig and Nylin (2000).

Two morphological traits measured were pupal spot and adult colour. The pupal spot was a small brown spot on the segment of the pupa that covers the fore leg. Its width was measured with a microscope fitted with a micrometer at a magnification of × 50 (details in Windig and Nylin, 2002). Its function is unknown, although it might help to disrupt the shape of the pupa, making it less easy to detect for predators. Natural selection on its width has probably been small or absent. Adult colour was measured as the average darkness of the base of the forewings. The darkness was determined with an image analyser on a grey scale from 0 (=black) to 255 (=white). The colour plays a role in thermoregulation and, because of that, also in mate-searching behaviour in males (Van Dyck and Matthysen, 1998). Pale males more often adopt territorial behaviour and dark males patrolling behaviour. Since climatic conditions are different on Madeira from anywhere else in the range of P. aegeria, some natural selection since colonisation seems likely.

Genetic analysis

Breeding produced a mixture of parents, partly being full sibs, and offspring consisting of full sibs and half cousins. A residual maximum likelihood (REML) analysis was used to analyse this data set. Such a procedure takes into account all relationships in the data set and, given an appropriate data set, can be used to estimate additive as well as dominance variance components (Lynch and Walsh, 1998). It proceeds by calculating a likelihood for a particular set of (co-)variances, and repeating the calculation in a iterative procedure until the maximum likelihood (ML) is found.

Analyses were carried out with the software program ASREML (Gilmour et al, 2000). Generation and sex were added as fixed effects to the model to eliminate their effects on genetic variation. Both additive and dominance variance components were estimated. For the estimation of the dominance variance components, the inverse of the matrix of dominance coefficients of relationship was computed and used as input for ASREML. These coefficients are easily derived from the coancestries of the parents of two individuals (Falconer 1989, equation (9.12)). Coancestries were computed following Meuwissen and Luo (1992). Analyses were also performed for sexes separately to determine if levels of genetic variation differ between the sexes. The relationship coefficients for the sex-specific analyses included relationships of the other sex so that they remain accurate.

All analyses were run unconstrained, so that negative variance components were possible. This is important because, due to sampling errors, negative variance components are expected. Calculating averages over traits of, for example, heritabilities with constrained variances may bias results if only downward error, resulting in negative variance components, is eliminated. Significance testing of variance components is best performed by comparing the ML of an unconstrained analysis to the ML in an analysis where the variance component is constrained to 0 (Shaw, 1987; Lynch and Walsh, 1998). Significance tests were performed for VA and VD separately and for VA and VD combined. If VA (VD) is constrained separately, most of its variance will end up in VD (VA), since its covariance structure is more similar to it than the covariance structure of the residual variance. This test thus indicates whether a significant portion of the genetic variance is due to additive (dominance) effects. When both VA and VD are constrained at the same time, the analysis provides a test of whether a significant portion of the variance is genetic. Covariances and genetic correlations could not be reliably estimated for this data set due to the limited sample size.

Results

Phenotypic level

Average phenotypic values for traits are similar to those found in previous research on Madeiran P. aegeria (Nylin et al, 1993). For all traits, differences between males and females are significant (Table 1). Larval time was longer for females than for males, but pupal time was shorter in females. In a previous research (Nylin et al, 1993), this difference was not significant, probably due to smaller sample size in their study. Larval growth rate was higher in females. There is no significant difference between males and females in the total length of juvenile period (t-test −0.596, P=0.551), contrary to Spanish mainland butterflies (Nylin et al, 1993). Pupal weight is higher in females as is adult size. Females have a relatively smaller pupal spot and males have a darker adult colour.

Table 1 Average value for the traits measured, and the significance of the difference between males and females
Full size table

Genetic level

Additive variance components did not differ significantly from 0, except for pupal spot (Table 2). Dominance variance components, on the other hand, were significantly different from 0 for two traits (larval time and adult colour), and approached significance for a third (pupal weight). For all traits, total genetic variance (VA+VD) was significantly different from 0 (Table 2). For all traits except PUPAL SPOT, VD was larger than VA.

Table 2 Estimated variance components and their significance (P, in brackets) as estimated by REML analysis
Full size table

Heritabilities (VA/VP) were rather low (average 0.078 and less than 0.115, except for pupal spot which was 0.272, Table 3). Dominance variances as a fraction of total variance (VD/VP), on the other hand, are comparable to what is usually found. Values range from 0.101 (pupal spot) to 0.659 (adult colour) and are on average 0.313 (Table 3). VD as a fraction of total genetic variation is rather high; values range from 0.699 to 1.162, except for pupal spot (0.271).

Table 3 Relative size of variance components estimated by REML analysis
Full size table

The analysis of the sexes separately does not change the overall picture. With the exception of pupal spot, h2's are low (<0.1), VD/VP's are intermediate (>0.25) and VD/Vg's are high (>0.7). The largest difference between males and females is found between VD/VP for adult colour, which is 0.94 for males and 0.38 for females. Due to the relatively large standard errors, none of the differences are significantly different between the sexes. Moreover, not all analyses converged for both sexes. The average of VD/VP was somewhat higher for the sexes analysed separately than for the analysis of all butterflies together (0.58 and 0.51 vs 0.31).

Discussion

The island population of P. aegeria on Madeira shows low levels of VA relative to levels of VD. This is in accordance with predictions from theory and laboratory experiments that VA, but not VD, decreases after a bottleneck and strong selection. Unfortunately, no estimates of VA and VD in mainland populations of P. aegeria are available for direct comparison. Only in a Belgian population have h2's been estimated (Van Dyck et al, 1998). These are, however, broad sense heritabilities (h2bs) and cannot be compared directly with strict sense heritabilities (h2ss), as estimated in this paper. The minimum value for h2ss given a h2bs can be found assuming all variation is genetic (h2ss+VD/VP=1.00). Given that h2bs=h2ss+1/2VD/VP, the h2ss is then 2h2bs–1.00. In the Belgian population, a h2bs of 0.63 for adult colour and 0.51 for adult size was found. This implies that for both traits total genetic variation is smaller in Madeira than in the Belgian population. Morover, for adult colour, h2ss is necessarily smaller in Madeira than in Belgium (minimum estimate in Belgium=0.26 vs 0.07 for Madeira).

In other butterfly species, h2bs has been estimated for the same or similar traits as in this study. For example, for adult colour, h2bs's have been estimated in Bicyclus anynana (Windig, 1994a), Pieris occidentalis (Kingsolver and Wiernasz, 1991), Araschnia levana (Windig and Lammar, 1999) and Inachis io (Windig, 1999). These estimates range from 0.599 to 0.885, all substantially higher than the h2bs of 0.40 found in this study and implying a larger h2ss. The same conclusion holds for larval time (estimates for the same species range from 0.511 to 0.671). For pupal time, growth rate and adult size, h2bs range from 0.303 to 0.525, all larger than those found in the Madeiran population, but whether h2ss is larger cannot be determined for these traits. Paulsen (1996) estimated h2ss for a number of traits related to ADULT SIZE in Precis coenia and P. evarete. She found an average h2ss of 0.349 and 0.440, respectively, both substantially larger than what is found in the Madeiran population (0.085). Thus, in other butterfly species, h2bs is clearly larger for all traits discussed above and h2ss is larger for at least some traits, but possibly for all traits. There is one exception to this pattern. In the peacock butterfly (Inachis io), the h2bs was estimated for pupal spot as 0.264 (Windig, 1998), somewhat smaller than the value of 0.322 found in this study. This exception may be explained, since it is likely that there has been no substantial natural selection on this trait.

Roff and co-workers have studied the overall pattern of relative amounts of VA and VD, as reported in literature. Both the average h2 of 0.26 for life history traits and of 0.46 for morphological traits (Mousseau and Roff, 1987) are larger than the average of 0.078 found in this study. The picture for VD/VP is more complicated. Average values of 0.31 for life history traits and 0.13 for morphological traits (Crnokrak and Roff, 1995) for VD/VP are smaller than the values found in this study if females and males are analysed separately. But when the sexes are analysed jointly, the average of 0.331 is comparable to the overall value for life history traits. Thus, the value of VD/VP seems to be elevated, but not as clearly as the h2. Average values in literature for VD/Vg are 0.54 (life history traits) and 0.17 (morphological traits), while in this study an average value of 0.808 was found. Overall, the conclusion is that the relative amount of VA in the Madeiran population is remarkably lower than what is usually found, while it is larger for VD at least relative to VG.

Nylin et al (1993) showed that Madeiran butterflies develop more slowly than Spanish butterflies at 17°. Gotthard et al (1994) showed that this coincides with a greater resistance to starvation, and proposed a physiological trade-off between growth rate and starvation resistance. Stronger resistance to starvation and a slower development are even more strongly accentuated in the Madeiran endemic P. xiphia. It thus seems likely that at least larval time and growth rate have been under strong natural selection since colonisation.

For the other traits, with the exception of pupal spot, we know that they are related to fitness, but it is not clear to what extent they have been under the influence of natural selection since colonisation. For example, adult females in Madeira are larger and colour is paler in females as in other populations of P. aegeria (Van Dyck and Matthysen, 1998), and indeed other satyrine butterflies such as Bicyclus anynana (Windig, 1994b). We know that adult colour plays an important role in thermoregulation and that it is related to territorial behaviour in males in Belgium. Thus, it is likely that natural selection has played a role since colonisation, at least in males, but strong evidence for it is lacking.

VA seems to have been reduced since colonisation, whereas VD seems not. To what extent this difference is due to a bottleneck in population size and to what extent to natural selection after colonisation cannot be determined in this study. An increase in VA due to conversion of VD does not seem to have played an important role. Another possible effect of the bottleneck and subsequent selection may have been the purging of recessive deleterious alleles. In the laboratory, it was noted that there were no detectable effects of inbreeding (unpublished results). Full sib matings and several generations of strong selection did not affect the fitness of the descendants of the butterflies used in this study. This result is in contrast with the situation in the related butterfly Bicyclus anynana, where inbreeding effects are pronounced (Saccheri et al, 1996). Wang et al. (1999) have investigated under which situations purging of inbreeding depression might occur. They concluded that it would only be important during a rapid increase after a population bottleneck. Miller and Hedrick (2001) found evidence for such an effect in laboratory populations of Drosophila. The population of P. aegeria in this study has likewise experienced a rapid increase of population size after introduction of one or a few individuals (Owen et al, 1986). Consequently, this population may be another example of purging of deleterious alleles in the field, as was found for a specific population of cattle in England (Visscher et al, 2001). Finding the original source of the Madeiran population and careful analysis of the genes lost since colonisation could help verify this interpretation.