Artificial selection on the shape of reaction norms for eyespot size in the butterfly Bicyclus anynana: direct and correlated responses

Transcription

1 Artificial selection on the shape of reaction norms for eyespot size in the butterfly Bicyclus anynana: direct and correlated responses P. J. WIJNGAARDEN,* P. B. KOCH &P.M.BRAKEFIELD* *Institute of Evolutionary and Ecological Sciences, Leiden University, The Netherlands Department of General Zoology and Endocrinology, University of Ulm, Germany Keywords: artificial selection; butterfly eyespots; constraints; correlated responses; ecdysteroids; phenotypic plasticity; reaction norms. Abstract The tropical butterfly Bicyclus anynana shows phenotypic plasticity in its ventral wing pattern as an adaptive response to wet-dry seasonality. Wet season form individuals have large eyespots, whereas individuals of the dry season generation have small eyespots. In the laboratory these forms can be obtained by rearing larvae at high and low temperatures, respectively. To study the extent to which the shape of the nearly linear reaction norms for eyespot size can be changed we applied four generations of sib selection by rearing full-sib families across three temperatures. In addition, we measured ecdysteroid titres shortly after pupation in the final generation. Although phenotypic variation in shape was present initially, the experiment yielded lines with reaction norms with similar shapes but different elevations. High, positive genetic correlation across temperatures can explain this lack of response. Differences in ecdysteroid titres did not readily relate to differences in eyespot size. Introduction Coping with environmental heterogeneity is a major challenge for many organisms. Phenotypic plasticity, the ability of some genotypes to develop into different phenotypes in different environments, can be a solution to the problems posed by heterogeneous environments (Levins, 1968). It enables these genotypes to produce a better phenotype environment match than if their expression were limited to only one phenotype. Plasticity is a feature of the reaction norm, the set of phenotypes expressed by a single genotype across a range of environments. Plasticity may evolve when genotypes differ in their reaction norms for a trait and this variation translates into fitness differences. Reaction norm evolution will be constrained, however, when genetic variation in plasticity is absent, or when the same genes are Correspondence: Pieter J. Wijngaarden, Laboratory of Genetics, Wageningen University, Dreijenlaan 2, NL-6703 HA Wageningen, The Netherlands. Tel.: +31 (0) , fax: +31 (0) , pieter.wijngaarden@genetics.dpw.wau.nl expressed across environments, thereby creating a genetic correlation. Even when genetic variation is available and genetic correlation are near zero, plasticity may still not evolve because it is costly or because of limits to its benefits (Van Tienderen, 1991; DeWitt et al., 1998). Seasonal polyphenism is an extreme form of phenotypic plasticity; it refers to the occurrence of distinct phenotypes in different seasonal generations (Shapiro, 1976). The African butterfly Bicyclus anynana exhibits a striking polyphenic wing pattern as an adaptive response to wet-dry seasonal environments (Brakefield & Larsen, 1984; Windig et al., 1994; Brakefield, 1997). The species has a wet season form (WSF) with large eyespots on the ventral side of the wings and a dry season form (DSF) in which the size of these eyespots is greatly reduced. The eyespots of WSF individuals are thought to act as deflection devices, whereas DSF individuals rely on crypsis (Brakefield & Larsen, 1984). These two forms reflect the different challenges of the two seasons: surviving predator attacks in the wet season (when food plants for the larvae are widely available and adults show active, reproductive behaviour), and escape from detec- 290

2 Artificial selection on reaction norm shape 291 tion in the dry season (when there are no opportunities for reproduction because the vegetation has died off). The results from field experiments support this adaptive scenario (N. Reitsma & P. M. Brakefield, unpublished). Temperature, especially during the final, fifth larval instar, is the main factor controlling the difference between the alternative forms (Kooi & Brakefield, 1999): in the laboratory a high rearing temperature (>23 C) gives rise to the WSF, whereas the DSF results from rearing larvae at a low temperature (<19 C). Rearing at intermediate temperatures in the laboratory reveals that the WSF and DSF are points on a continuous, almost linear reaction norm. Temperature has its effect on wing pattern in Bicyclus mainly by lengthening or shortening development time (Windig, 1993). Development is tightly controlled by the hormonal system so hormones [in insects especially juvenile hormone and ecdysteroids (Nijhout, 1994, 1999b)] are likely candidates for regulating phenotypic plasticity. It has indeed been shown in Bicyclus that ecdysteroid hormones are involved in the difference in elevation of reaction norms for eyespot size: lines selected for different sizes of their ventral eyespots (which are plastic) showed correlated changes in ecdysteroid titres, whereas lines selected for different sizes of their dorsal eyespots (which are nonplastic) did not show such a correlated response (Koch et al., 1996; Brakefield et al., 1998). Moreover, injections of 20-hydroxyecdysone into early pupae of the line selected for small ventral eyespots gave adults with larger eyespots (although not as large as those of unselected WSF individuals). Here we report on a continuation of our studies on the genotype phenotype relationship for wing pattern in B. anynana. As only the outer ends of the reaction norms are thought to be exposed to selection in the field, we examined how this has affected an intermediate, hidden part of the reaction norm. More specifically, we ask whether an intermediate point on the reaction norm is free to evolve independently of the end points, i.e. to what extent the shape of the reaction norms can be changed by selection. The underlying physiology may act as a constraint by limiting the range of possible shapes. Our approach is similar to the one used by Koch et al. (1996) and Brakefield et al. (1998): we looked for correlated responses in ecdysteroid titre after selecting a pair of selection lines for contrasting shapes of their reaction norms for eyespot size. For a third line we selected for horizontal reaction norms at an intermediate eyespot size (in our case the mean eyespot size across temperatures of the base population). Brakefield et al. (1996) selected on eyespot size at one, initially intermediate temperature and found that the resulting lines had lost most of their ability to express the alternative seasonal forms. Their lines had also lost some (HIGH line) or most (LOW line) plasticity, but as the eyespot sizes in these lines are at their upper and lower limits, respectively, the apparent lack of plasticity may be an artefact. A previous effort (using truncation selection) to obtain horizontal reaction norms at an intermediate eyespot size did not yield a response (Wijngaarden & Brakefield, 2001) so we used the more sensitive method of sib selection (a form of family selection) in combination with a more variable base population. From the last generation we took haemolymph (blood) samples shortly after pupation to measure their ecdysteroid content. Because families were split over three temperatures we were able to study both the genetic and the environmental causes of (co)variation in eyespot size and ecdysteroid titre. Materials and methods The study species Our laboratory stock was established from about 80 gravid females collected at a single locality in Malawi. Several hundred breeding adults are reared in each generation so that high levels of heterozygosity are maintained (Saccheri & Bruford, 1993). Larvae are raised on young maize plants in climate rooms with controlled temperature, high relative humidity, and a 12 h : 12 h light : dark cycle. Adult butterflies feed on mashed banana. The selection procedure The aim of selection is shown in Fig. 1. We tried to set up a line with reaction norms for eyespot size that show a steep slope between 19.5 and 22 C, and a horizontal slope between 22 and 24 C (the SH line, first Steep then Horizontal reaction norms). For the second line we selected for reaction norms with a horizontal slope between 19.5 and 22 C, and a steep slope between 22 Temperature Fig. 1 The shapes of the reaction norms we selected for. The dotted line represents the average reaction norm for eyespot size in the unselected stock. The solid lines are the target reaction norms of the SH line (squares), the H line (diamonds), and the HS line (triangles).

3 292 P. J. WIJNGAARDEN ET AL. and 24 C (the HS line, first Horizontal then Steep reaction norms). In terms of a second degree polynomial we selected the SH and HS lines for negative and positive values, respectively, of the coefficient of the quadratic term. In both lines we tried to keep the line means at 19.5 and 24 C the same across generations to avoid changes in elevation. For the third line we selected for horizontal reaction norms at an intermediate eyespot size across the three temperatures (the H line, with zero slope and curvature). This intermediate eyespot size was calculated as the mean across temperatures of the four subpopulations that made up the base population (see below) and turned out to be We based our decisions about which families to use for the next generation on visual inspection of the reaction norms: those families were selected that showed the closest correspondence to the target reaction norms of the respective lines. 2 Following Holloway et al. (1993) and Holloway & 3 Brakefield (1995) the selected character was the diameter of the black ring of the largest (fifth) eyespot on the ventral hindwing relative to forewing length. Eyespot size and forewing length were both measured with high repeatability (Brakefield & Breuker, 1996) using a binocular microscope fitted with a micrometer. Because of the limited amount of time available between generations only females were measured. For the final generation both sexes were measured. We selected on the basis of the family mean, a procedure that is particularly useful when the character selected has a low heritability, when there is little variation because of common environment, and when families are large 4 (Falconer & Mackay, 1996). Under these conditions the mean phenotypic value of a family comes close to being an estimate of its genotypic value. Strictly speaking, we applied sib selection rather than family selection because we used full-sibs of measured individuals as parents of the next generation (see below). Sib selection and family selection are fully equivalent only when families are (very) large (Falconer & Mackay, 1996). A drawback of our selection method is that the rate of inbreeding tends to increase with the intensity of selection. As we used individually marked individuals as parents we were able to construct pedigrees and evaluate the effects of inbreeding. In the final generation the SH line, H line, and HS line had mean inbreeding coefficients (SD) of (0.113), (0.166) and (0.057), respectively. Analyses with the inbreeding coefficient as a covariate gave no evidence for effects of inbreeding on eyespot size, development time or pupal weight. Van Oosterhout et al. (2000) also found no significant effect of inbreeding on development time in B. anynana. We have almost certainly selected against inbred families, however, because egg hatching in B. anynana is very sensitive to inbreeding (Saccheri et al., 1996; Van Oosterhout et al., 2000) and females had to produce a sufficient number of hatched larvae in order to be included in the experiment. The base population We started the experiment with a base population consisting of families from four subpopulations: 35 families from the unselected stock plus families from three F 2 populations of crosses between selection lines. These F 2 populations were primarily used as additional sources of phenotypic variation, but as their history of selection is known they may also give additional information about the genetic basis of shape differences. One F 2 population was derived from a cross between lines selected for large (HIGH line) and small (LOW line) ventral eyespots, respectively, at a single, initially intermediate temperature (Brakefield et al., 1996). The other F 2 populations were derived from crosses between lines selected for high plasticity (HP) and low plasticity (LP) at 18.5 and 23.5 C (Wijngaarden & Brakefield, 2001). The first HP line was crossed with the first LP line, whereas the second replicate HP line was crossed with the second, replicate LP line. The four subpopulations will be referred to as the S, HL, P1 and P2 lines, respectively. Evaluation of plasticity We assessed plasticity by partitioning the full-sib progeny of single-pair matings over three rearing temperatures (19.5, 22 and 24 C). The temperatures 19.5 and 24 C are comparable to average temperatures in the field during periods of larval development in the dry and wet season, respectively (Brakefield & Mazzotta, 1995). A fourth cohort was reared at 16 C; development takes so long at this temperature that in the meantime butterflies from the other temperatures can be measured and decisions about which families to select can be made. Adults from the 16 C cohort can thus be used as parents of the next generation. In the final generation each family was replicated twice at 19 and 24 C, and four times at 22 C. These replicate cages enabled us not only to obtain an impression of common environment effects, but also to obtain two independent estimates of the slopes of the reaction norms between 19 and 22 C and between 22 and 24 C. For logistical reasons it was not possible to rear families from the unselected stock as a control in each generation. Only in the final generation did we rear families from the stock alongside the families of the selection lines. Collection of haemolymph and measurement of ecdysteroid titre In the final generation, we collected two pupae from every cage, giving totals of four pupae per family at 19.5

4 Artificial selection on reaction norm shape 293 and 24 C, and eight pupae per family at 22 C. However, because of mortality of some sampled pupae, loss of some haemolymph samples, and (occasionally) excessive sampling these numbers are often slightly different from the intended numbers. Collecting took place during the peak period of pupation. Haemolymph was sampled at about one-fourth of the pupal stage (i.e. 40 h after pupation at 24 C, 50 h after pupation at 22 C, and 70 h after pupation at 19.5 C) because previous experiments had shown that differences between selection lines are most marked then (Brakefield et al., 1998). Ecdysteroid titres were determined by radioimmunoassay (Koch et al., 1996; Brakefield et al., 1998). Pupae were weighed shortly before their haemolymph was sampled. We used these weights to determine the sex of the pupae using discriminant function analysis. Both the sex and the pupal weights of the full-sibs of the sampled individuals were known in most cases so that these data could be used to derive discriminant functions (using the DISCRIM procedure in SAS) for each line and each temperature. This calibration information was then applied to the weights of the pupae used for sampling haemolymph. Applying the discriminant functions to the data from which they were derived resulted in estimated misclassification rates of 2 5% at 19.5 C, 6 8% at 22 C and 5 12% at 24 C. We will assume throughout that males and females have been classified correctly in the discriminant procedure. Data analyses Plastic responses can be viewed from two perspectives (Via et al., 1995): as consequences of selection for different trait means in different environments (the character state approach), or as functions (e.g. polynomials) that describe phenotypic change across environments (the reaction norm approach). These perspectives are interchangeable under some conditions 5 (De Jong, 1995; Van Tienderen & Koelewijn, 1994). Bicyclus exhibits two forms (character states) in the field, but a continuous range of forms (a reaction norm) can be obtained in the laboratory. We will therefore adopt both approaches. For the character state approach this involves the calculation of heritabilities within temperatures and genetic correlations across temperatures. For the reaction norm approach we fit a second degree polynomial to obtain estimates of the intercept, slope and curvature. We have rescaled temperature by subtracting the mean temperature (21.83) to reduce correlations among regression coefficients. General We used the GLM and MIXED procedures of the SAS package for data analyses. We considered line and temperature as fixed effects and family (nested within line) as a random effect. The final generation contained cage (nested within family) as an additional random effect, but because the cages showed neither a significant main effect nor interaction with temperature in any of the analyses we pooled the cage sums of squares for each family. The appropriate degrees of freedom have been determined using Satterthwaite s approximation. Because of unequal family sizes the mean squares for eyespot size are based on type III sums of squares; because of some missing cells type IV sums of squares are used for ecdysteroid titres. Differences in means or mean slopes between lines have been tested using the ESTI- MATE statement in the MIXED procedure. The sequential Bonferroni procedure (Rice, 1989) was used to account for multiple testing. Response to selection Responses to selection can be evaluated by calculating the deviation of the observed mean eyespot size at 22 C from the expected mean eyespot size at that temperature. The expected value is based on a linear reaction norm and is calculated as [19.5/( )] mean eyespot size at 19.5 C + [24/( )] mean eyespot size at 24 C. In the SH and the HS lines this deviation from linearity should increase in opposite directions in the course of the selection experiment, whereas in the H line it should decrease. Moreover, in the H line the difference between mean eyespot sizes at 19.5 C and 24 C should become smaller. We will also provide a visual means of assessing the response to selection. For each line, the mean eyespot sizes at the three temperatures are plotted against generation. When the changes in eyespot size we selected for occur, the mean eyespot size at 22 C should approach the mean eyespot size at 24 C (in the SH line) or 19.5 C (in the HS line); the eyespot sizes at 24 C and 19.5 C should remain the same. For the H line the mean eyespot sizes at the three temperatures should converge on the value (see above). Genetic correlations Estimates of genetic correlations across temperatures were obtained in two ways. The product moment correlation of family means is both conceptually and computationally simple, but this method is potentially biased because of the presence of within-family variance in the variance and covariance terms (Via, 1984). The second method is based on the mixed-model analysis of variance (ANOVA) (Fry, 1992). For each line, separate ANOVAs for each pair of temperatures were run with family as the random effect and temperature as the fixed effect. Next, separate ANOVAs for each temperature were run. Estimates of the genetic correlation are obtained by dividing the variance component associated with the family main effect from the mixed-model ANOVA by the geometric mean of the between-family variances in each of the two temperatures. We used restricted maximum likelihood (REML) to estimate variance components. Because the REML algorithm sets negative variance

5 294 P. J. WIJNGAARDEN ET AL. component estimates to zero we replaced zero estimates by estimates obtained from equating expected and observed mean squares when the product moment correlations were negative. Confidence intervals were calculated using the z-transformation (Sokal & Rohlf, 1995). Heritabilities Heritabilities were calculated as two times the intraclass correlation coefficient. This yields estimates that are potentially biased as a result of dominance, additive additive interaction and common environment effects (Falconer & Mackay, 1996). The estimates of betweenfamily and within-family variance components have been obtained by REML. The significance levels reported are those from the family main effect in ANOVAs per line and per temperature. Results Reaction norms in the base population The reaction norms for eyespot size (Fig. 2) suggest that we have been successful in raising the variability in shapes by supplementing the stock with F 2 families. The corresponding ANOVA (Table 1a) indicated that there were highly significant differences in eyespot size because of family, temperature and family temperature interaction effects (see Web material for ANOVAs per subpopulation). There was progressively less variation among families at higher temperatures (Table 1b); estimates of heritabilities (SE) were 1.07 (0.090) at 19.5 C, 1.01 (0.100) at 22 C and 0.64 (0.093) at 24 C. Standardizing the covariances for pairs of temperatures yielded very similar estimates (SE) of genetic correlation across temperatures: 0.84 (0.060) for C, 0.87 (0.054) for C, and 0.86 (0.056) for C. When comparing the parameters of a quadratic regression we found that most variation occurred in the intercept; (very) little variation was found in slope and curvature (Table 1c). Temperature Fig. 2 Reaction norms for eyespot size in females of the four subpopulations (see text for nomenclature) that make up the base population. Table 1 Characteristics of the base population. Source d.f. MS F-value P-value (a) Analysis of variance for eyespot size in the base population. Temperature is a fixed effect, family is a random effect Family (F) Temperature (T) F T Error (b) Genetic covariance matrix for eyespot size in the pooled base population at the three different temperatures ( C) (c) Genetic covariance matrix for the parameters of a quadratic regression Intercept Slope Curvature Intercept Slope Curvature P < Response to selection The numbers of families selected in each generation are given in Table 2 together with the differences between the mean slopes of the selected families and all families in a line (i.e. the selection differentials). Figure 3 shows the mean reaction norms for eyespot size in successive generations; the corresponding ANOVAs are summarized in Table 3. Only in generation 2 was there (weak) evidence for line temperature interaction, indicating that lines differed in their response to temperature. Within lines, families differed in their response to temperature in all generations, although the evidence is weak in the Table 2 The number of families (N) of each line that contributed to the next generation (from generation 1 onwards), and the selection differentials, i.e. the differences in slope between the selected families and all families of a line. S 1 : selection differentials of slopes between 19.5 and 22 C. S 2 : selection differentials of slopes between 22 and 24 C. Also given are the selection intensities i 1 and i 2, i.e. the selection differentials divided by the phenotypic standard deviations. Gen. Line N 1S 1 i 1 S 2 i 2 1Þ2 SH H HS Þ3 SH H HS Þ4 SH H HS

6 Artificial selection on reaction norm shape 295 a Temperature Fig. 3 Reaction norms for eyespot size of the successive generations of the SH line (squares), H line (diamonds), HS line (triangles) and, in generation 4, the unselected stock (open circles: females; closed circles: males). b Table 3 F-values and mean square errors obtained from analyses of variance for eyespot size of females in generations 1 4. The stock has been omitted from generation 4 to enable comparison with previous generations. Line and temperature are fixed effects; family is a random effect and nested within lines. Source Line 2.04 NS Family Temp L T 2.15 NS NS 0.34 NS F T Error NS: not significant, *P < 0.05, P < c final generation. Estimates of both genetic correlations across temperatures and heritabilities of eyespot size within temperatures were positive and high in most cases (see Web material). The deviations from linearity in the SH, H and HS lines were nonsignificant in generations 1, 3 and 4 (t-tests not shown). In generation 2 the observed mean eyespot sizes differed significantly from the expected values (SH line: t 153 ¼ 3.09, P ¼ ; H line: t 147 ¼ 3.15, P ¼ ; HS line: t 124 ¼ 8.55, P < ), but this was probably caused by an environmental effect (see below). A slight decrease in the differences between the mean eyespot sizes at 19.5 and 24 C in the H line was found [estimates (SE) for generations 1 4: (0.0027), (0.0018), (0.0022), and (0.0022)], but these discrepancies remain highly significantly different from zero. Estimates of parameters of a polynomial regression are given as additional Web material. Changes in eyespot size across generations become more visible when plotted as in Fig. 4. In the SH line, we aimed at keeping eyespot sizes at 19.5 and 24 C constant whereas eyespot sizes at 22 C were selected to approach those at 24 C. We clearly did not succeed in achieving Generation Fig. 4 Mean eyespot sizes across generations of females of (a) the SH line, (b) the H line and (c) the HS line. The solid symbols of generation 0 indicate the grand means of the base population; they are therefore the same for the three lines. The solid line connects eyespot sizes at 24 C, the long dashed line connects eyespot sizes at 22 C and the dashed line connects eyespot sizes at 19.5 C. these objectives (Fig. 4a), although eyespot size at 24 C remained fairly constant. In the H line, a response to selection would show up as eyespot sizes converging on the value This, too, did not occur (Fig. 4b). In the HS line, we aimed at keeping eyespot sizes at 19.5 and 24 C constant whereas eyespot sizes at 22 C should approach those at 19.5 C. Again, no response to selection was observed (Fig. 4c). Mean eyespot sizes at 19.5

7 296 P. J. WIJNGAARDEN ET AL. and 22 C showed a drop in the second generation; that all three lines have been affected suggests that slightly lower temperatures in the 19.5 and 22 C climate rooms were the cause. Because of the magnitude of this environmental effect the cumulative selection response and realized heritability were not calculated. Because in the final generation each family was replicated twice at 19.5 and 24 C, and four times at 22 C, we are able to compare independent estimates of slopes between 19.5 and 22 C and between 22 and 24 C. Neither in males nor in females did the two slopes differ in their means (males: F 1,33.6 ¼ 0.11; P ¼ NS; females: F 1,34.1 ¼ 0.03; P ¼ NS), or did lines show variation in the difference between the two slopes (males: F 3,33.7 ¼ 1.95; P ¼ NS; females: F 3,34.4 ¼ 1.33; P ¼ NS). Correlated responses in ecdysteroid titre The average reaction norms for ecdysteroid titre are shown in Fig. 5 (family reaction norms are given as additional Web material). The corresponding ANOVAs are given in Table 4 (note that the collection of pupae has [Ecdysteroids] Temperature Fig. 5 Reaction norms for ecdysteroid titre (lg ml )1 ). Each line connects the means (calculated as the weighted mean of family means) of a line. Circles: stock. Squares: SH line. Diamonds: H line. Triangles: HS line. Open symbols: males. Closed symbols: females. been considerably female biased). In both sexes a highly significant temperature effect was found; in females, but not in males, did lines differ in their response to temperature. There was weak evidence for a line effect in both sexes and, in males, for differences among families. In Fig. 6 the reaction norms for ecdysteroid titre and eyespot size are combined. At both 22 and 24 C ecdysteroid titre and eyespot size seemed to be positively correlated, although the order in mean eyespot size matched the order in mean ecdysteroid titre only for males at 24 C. At 19.5 C a negative correlation between ecdysteroid titre and eyespot size seemed to occur in females, but in males there was no evidence for a correlation. Within temperatures, we found only one significant product moment correlation between family means of eyespot size and ecdysteroid titre (males SH line at 24 C, r ¼ 0.68, n ¼ 12, P ¼ ). Discussion Direct response The rate and direction of reaction norm evolution are determined by the genetic variances within environments, in combination with the degree of independence of the character states expressed in the different environments (Via & Lande, 1985). Throughout our selection experiment genetic variation for eyespot size was available at each of the three temperatures; estimates of genetic correlations between temperatures were generally positive and high. These findings can explain why we were not successful (at least in the short term) in creating selection lines that differ in the shape of their reaction norms; the changes we did observe were mainly in the elevation of the reaction norms. Estimates of regression parameters of the base population already hinted at this outcome (although parameter estimates of a population may not accurately reflect the evolutionary potential of that population, see below). We therefore conclude that phenotypic variation in reaction norm shape mainly consists of nonadditive or environmental variation. The drop in eyespot size that occurred in generation 2 is most probably an environmental effect as it affected all lines. Unfortunately, we did not raise families from the Males Females Source d.f. MS F P d.f. MS F P Line Family NS Temp L T NS F T NS NS Error Table 4 Analyses of variance of ecdysteroid titres (cuberoot transformed). Line and temperature are fixed effects; family is a random effect and nested within lines. NS: not significant, P < 0.05, P <

8 Artificial selection on reaction norm shape 297 Eyspot size Eyspot size a b [Ecdysteroids] Fig. 6 Reaction norms for ecdysteroid titres (lg ml 1 ) at about onefourth of the pupal stage and relative eyespot size. Each line connects the means (calculated as the weighted mean of family means) of a line for a single temperature ( C). Circles: stock. Squares: SH line. Diamonds: H line. Triangles: HS line. (a) Females, and (b) males. unselected stock as a control so we cannot correct for this effect. We realize that because of the small numbers of families this negative result could be the consequence of a lack of statistical power. Our results are consistent, however, with the findings of Holloway & Brakefield (1995) and Brakefield et al. (1996), who conducted more powerful selection experiments, although not targeted at shape per se. The results of these three experiments do not agree with Windig s (1994) conclusions. He raised full-sib families from the unselected B. anynana stock across four temperatures (17, 20, 23 and 28 C). Heritabilities of the linearity of plasticity of the black ring of the fifth eyespot of 0.49 (males) and 0.45 (females) suggest that a relatively rapid response to selection is possible should selection favour differently shaped reaction norms. Moreover, he found that cross temperature correlation for seasonal form (the principal component reflecting plastic traits) were nearly all insignificant and about half of them are positive and the other half negative. His conclusion was therefore that natural selection in one season for a wetter or drier phenotype will cause no or only small changes in the wetness of the wing pattern in the other season. Based on data from a selection experiment on a single, key trait and from a narrower, more natural range of temperatures, we conclude that selection on eyespot size at one temperature will lead to a correlated response in the same direction at other temperatures. Hence, mainly changes in elevation are possible. A small number of families will especially affect estimates of quantitative genetic parameters that are based on (co)variances. This may account for the estimates of heritabilities and genetic correlation that were >+1 (although with confidence intervals embracing values of less than unity). For the same reason tests of significance had little power. When interpreting estimates of heritabilities and genetic correlation it may therefore be more useful to evaluate the qualitative impression they make (e.g. a positive and high correlation ) rather than the precise values of the estimates. Both the heritabilities and the genetic correlation reported here are also likely to be biased. Because they are based on a full-sib design the estimates of heritabilities are potentially inflated because of dominance, additive additive interaction and common environment effects (Falconer & Mackay, 1996). The latter effect seems of little importance as no differences between replicate cages were found. The occurrence of dominance at a single temperature has been shown in an analysis of crosses between lines selected for large and small ventral eyespots (Wijngaarden & Brakefield, 2000). Correlated response Hormonal correlates of polyphenic development have been demonstrated in a number of insect species [see Nijhout (1994, 1999a) and Dingle & Winchell (1997) for examples], but many questions remain even in such well studied cases as wing dimorphism in Gryllus crickets [see Zera (1999) and references therein]. In Bicyclus, the role of ecdysteroid hormones has been established for differences in elevation of reaction norms for eyespot size: the line selected for large ventral eyespots showed higher ecdysteroid titres shortly after pupation than the line selected for small eyespots (Koch et al., 1996; Brakefield et al., 1998). We set out to do the same for differences in reaction norm shape, but our results are difficult to interpret (apart from the lack of response). The difficulties mainly stem from uncertainties in the timing of the collection of pupae and the sampling of haemolymph. Because males develop faster than females this might imply that the male pupae we collected are drawn from the late pupating part of the distribution, whereas the female pupae are representative of the early pupating part of the distribution. Relating ecdysteroid titres of pupae to the eyespot sizes of their full-sibs may therefore result in attributing larger eyespots to males and smaller eyespots to females

9 298 P. J. WIJNGAARDEN ET AL. (because of the negative correlation between time to pupation and eyespot size). Finally, the collection of pupae during the peak period of pupation has been strongly female-biased. Shortly after pupation ecdysteroid titres begin to increase and, after a peak has been reached, the titres slowly decrease. Koch et al. (1996) and Brakefield et al. (1998) showed that differences between selection lines are especially pronounced during the rising phase, hence our efforts to sample haemolymph around this time. Because ecdysteroid titres are rising sharply, even small differences in the time of sampling may result in substantial differences in ecdysteroid content of the samples. Differences among lines in ecdysteroid titre at about one-fourth of the pupal stage may thus partly reflect differences in stage of development. The mean ecdysteroid titres in the unselected stock show the same pattern in both sexes: a peak value at 22 C and lower values at 19.5 and 24 C (with concentrations at 19.5 C being about two times higher than those at 24 C). An indication for this pattern was already apparent in Fig. 3b in Brakefield et al. (1998) where the stock reaches a peak concentration at 27 C of about 5 lg ml 1 48 h after pupation, whereas at 20 C this concentration is reached at 96 h after pupation when the peak is yet to come. The peak at 22 C is less pronounced (or even absent) in the selection lines so that the highest titres in these lines occur at 19.5 C. Most significant differences occur at 22 C, which is what we should expect as we aimed at changing eyespot sizes in the SH and HS lines at that temperature while keeping eyspot sizes at 19.5 and 24 C the same. This suggests that ecdysteroid titres at extreme temperatures are less malleable by selection than those at an intermediate temperature. Potential implications for adaptive responses The lack of response on the shape of the underlying norm of reaction thus contrasts sharply with a ready ability to change the elevation of the complete bundle of norms of reaction. Our stock does not appear to have the necessary standing additive genetic variation to enable any substantial response to selection on the shape of reaction norms. This may reflect an inability of the underlying mechanisms involved in sensitivity to the environment and in hormonal regulation to produce either a wider or a tighter phenotypic response over a narrow environmental range. Such a phenomenon would then reflect a genetic or physiological constraint at least with respect to short-term adaptive evolution, for example, in response to particular patterns of climatic change. It might also reflect an inability of certain populations to produce an optimal phenotypic response to a highly seasonal environment in terms of visual selection but one involving only a narrow temperature range. Such a scenario would favour genotypes yielding both cryptic, dry season forms completely lacking all ventral eyespots, and wet season forms expressing a fully effective deflecting series of large, conspicuous eyespots. Similarly, populations might be unable to respond adaptively to a hypothetical change in visual predation where this is coupled with selection favouring an intermedate expression of ventral eyespots in each seasonal environment. The 80 or so species of Bicyclus in subsaharan Africa can live in very divergent environments, seasonal or aseasonal and individual species may have extensive distributions extending across different climatic regimes (Roskam & Brakefield, 1999). If different scenarios about adaptive evolution are represented in natural environments, it will be interesting to explore whether certain populations have been able to break through the constraint by, for example, some form of key innovation in the mechanisms controlling the phenotypic plasticity. Acknowledgments Els Schlatmann and colleagues secured a steady supply of maize for the caterpillars. Fanja Kesbeke (Leiden) and Marga Weigmann-Lenz (Ulm) showed PJW how to collect haemolymph and how to carry out radioimmunoassay, respectively. We are most grateful to two anonymous referees for their constructive criticism. Supplementary Material The following material is available from Fig. S1 Reaction norms for ecdysteroid titers (lg ml 1 ) at about one-fourth of the pupal stage and relative eyespot size. Each line connects the means (calculated as the weighted mean of family means) of a line for a single temperature ( C). Circles: stock, squares: SH line, diamonds: H line, triangles: HS line. (a) Females, and (b) males. Table S1 Analyses of variance for eyespot size in the four subpopulations that make up the base population [S ¼ unselected stock, HL ¼ cross between HIGH and LOW lines (Brakefield et al., 1996), P1 and P2: crosses between replicates of lines selected for high and low plasticity (Wijngaarden & Brakefield, 2001)]. Family is a random effect; temperature is a fixed effect. Table S2 Genetic correlations between eyespot sizes at 19.5 and 22 C for each generation (generation 0 is the base population), line and sex (F ¼ female, M ¼ male). PMC: the product moment correlation of family means, MMA: estimates obtained from mixed-model ANOVA (see text, indicates zero between-family variance at 19.5 or 22 C). 95 % confidence intervals are given when P < 0.05 and the estimate <1; intervals between brackets indicate estimates that become nonsignificant when the sequential Bonferroni procedure is applied within lines.

10 Artificial selection on reaction norm shape 299 Table S3 Genetic correlations between eyespot sizes at 22 and 24 C for each generation (generation 0 is the base population), line and sex (F ¼ female, M ¼ male). PMC: the product moment correlation of family means, MMA: estimates obtained from mixed-model ANOVA (see text, indicates zero between-family variance at 22 or 24 C). 95% confidence intervals are given when P < 0.05; intervals in parentheses indicate estimates that become nonsignificant when the sequential Bonferroni procedure is applied within lines. Table S4 Heritabilities of eyespot size per generation (generation 0 is the base population), line sex (F ¼ female, M ¼ male) and temperature. The significance levels reported are those from the family main effect in ANOVAs per line and temperature. Table S5 Estimates of parameters (SE) of a polynomial regression for eyespot size in females in generation 1 4. For t-tests d.f. ¼ 1 in all cases. References Brakefield, P.M Phenotypic plasticity and fluctuating asymmetry as responses to environmental stress in the butterfly Bicyclus anynana. In: Environmental Stress, Adaptation and Evolution (R. Bijlsma & V. Loeschcke, eds), pp Birkhäuser-Verlag, Basel. Brakefield, P.M. & Breuker, C.J The genetical basis of fluctuating asymmetry for developmentally integrated traits in a butterfly eyespot pattern. Proc. R. Soc. Lond., B 263: Brakefield, P.M., Gates, J., Keys, D., Kesbeke, F., Wijngaarden, P.J., Monteiro, A., French, V. & Carroll, S.B Development, plasticity and evolution of butterfly eyespot patterns. Nature 384: Brakefield, P.M., Kesbeke, F. & Koch, P.B The regulation of phenotypic plasticity of eyespots in the butterfly Bicyclus anynana. Am. Nat. 152: Brakefield, P.M. & Larsen, T.B The evolutionary significance of dry and wet season forms in some tropical butterflies. Biol. J. Linn. Soc. 22: Brakefield, P.M. & Mazzotta, V Matching field and laboratory environments: effects of neglecting daily temperature variation on insect reaction norms. J. Evol. Biol. 8: De Jong, G Phenotypic plasticity as a product of selection in a variable environment. Am. Nat. 145: DeWitt, T.J., Sih, A. & Wilson, D.S Costs and limits of phenotypic plasticity. Trends Ecol. Evol. 13: Dingle, H. & Winchell, R Juvenile hormone as a mediator of plasticity in insect life histories. Arch. Insect Biochem. Physiol. 35: Falconer, D.S. & Mackay, T.F.C Introduction to quantitative genetics, 4th edn. Longman, Harlow. Fry, J.D The mixed-model analysis of variance applied to quantitative genetics: biological meaning of the parameters. Evolution 46: Holloway, G.J. & Brakefield, P.M Artificial selection of reaction norms of wing pattern elements in Bicyclus anynana. Heredity 74: Holloway, G.J., Brakefield, P.M. & Kofman, S The genetics of wing pattern elements in the polyphenic butterfly, Bicyclus anynana. Heredity 70: Koch, P.B., Brakefield, P.M. & Kesbeke, F Ecdysteroids control eyespot size and wing color pattern in the polyphenic butterfly Bicyclus anynana (Lepidoptera: Satyridae). J. Insect Physiol. 42: Kooi, R.E. & Brakefield, P.M The critical period for wing pattern induction in the polyphenic tropical butterfly Bicyclus anynana (Satyrinae). J. Insect Physiol. 45: Levins, R Evolution in Changing Environments. Princeton University Press, Princeton, NJ. Nijhout, H.F Insect Hormones. Princeton University Press, Princeton, NJ. Nijhout, H.F. 1999a. Control mechanisms of polyphenic development in insects. Bioscience 49: Nijhout, H.F. 1999b. When developmental pathways diverge. Proc. Natl. Acad. Sci. USA 96: Rice, W.R Analyzing tables of statistical tests. Evolution 43: Roskam, J.C. & Brakefield, P.M Seasonal polyphenism in Bicyclus (Lepidoptera: Satyridae) butterflies: different climates need different different cues. Biol. J. Linn. Soc. 66: Saccheri, I.J., Brakefield, P.M. & Nichols, R.A Severe inbreeding depression and rapid fitness rebound in the butterfly Bicyclus anynana (Satyridae). Evolution 50: Saccheri, I.J. & Bruford, M.W DNA fingerprinting in a butterfly, Bicyclus anynana (Satyridae). J. Hered. 84: Sokal, R.R. & Rohlf, F.J Biometry, 3rd edn. W.H. Freeman, New York. Van Oosterhout, C., Zijlstra, W.G., Van Heuven, M.K. & Brakefield, P.M Inbreeding depression and genetic load in laboratory metapopulations of the butterfly Bicyclus anynana. Evolution 54: Van Tienderen, P.H Evolution of generalists and specialists in spatially heterogeneous environments. Evolution 45: Van Tienderen, P.H. & Koelewijn, H.P Selection on reaction norms, genetic correlations and constraints. Genet. Res., Camb. 64: Via, S The quantitative genetics of polyphagy in an insect herbivore. II. Genetic correlations in larval performance within and among host plants. Evolution 38: Via, S., Gomulkiewicz, R., De Jong, G., Scheiner, S.M., Schlichting, C.D. & Van Tienderen, P.H Adaptive phenotypic plasticity: consensus and controversy. Trends Ecol. Evol. 10: Via, S. & Lande, R Genotype environment interaction and the evolution of phenotypic plasticity. Evolution 39: Wijngaarden, P.J. & Brakefield, P.M The genetic basis of eyespot size in the butterfly Bicyclus anynana: an analysis of line crosses. Heredity 85: Wijngaarden, P.J. & Brakefield, P.M Lack of response to artificial selection on the slope of reaction norms for seasonal polyphenism in the butterfly Bicyclus anynana. Heredity 87: Windig, J.J The genetic background of plasticity in wing pattern of Bicyclus butterflies. PhD Thesis, Leiden University, Leiden.

11 300 P. J. WIJNGAARDEN ET AL. Windig, J.J Genetic correlations and reaction norms in wing pattern of the tropical butterfly Bicyclus anynana. Heredity 73: Windig, J.J., Brakefield, P.M., Reitsma, N. & Wilson, J.G.M Seasonal polyphenism in the wild: survey of wing patterns in five species of Bicyclus butterflies in Malawi. Ecol. Entomol. 19: Zera, A.J The endocrine genetics of wing polymorphism in Gryllus: critique of recent studies and state of the art. Evolution 53: Received 5 July 2001; accepted 19 November 2001