Heredity 73 (1994) 459—470 Received 11 October 1993 Genetical Society of Great Britain

Genetic correlations and reaction norms in wing pattern of the tropical

JACK J. WINDIG* Institute of Evolutionary and Ecological Sciences, Section of Evolutionary Biology, University of Leiden, Sche/penkade 14A, 2313 ZT Leiden, the Netherlands

Geneticcorrelations (rg) within and across environments, were determined in the tropical, dry-wet seasonal polyphenic butterfly Bicyclus anynana, over four temperatures, for larval DEVELOP- MENT time (plastic), pupal WEIGHT (less plastic) and two wing pattern characters: SEASONAL FORM (plastic) and THERMAL FORM (less plastic). The rgs for SEASONAL FORM were weak, making it relatively independent across seasons. The rgs for WEIGHT were intermediate between THERMAL and SEASONAL FORM. Negative rgs were present for DEVELOPMENT. The reaction norms for DEVELOPMENT time clearly crossed at an intermediate temperature, whereas the others did not. This implies that selection for fast growers in one season has an opposite effect in the other season. TgS between WEIGHT and the other characters remained constant over temperatures, as did the correlation between DEVELOPMENT and THERMAL FORM. Both the correlation between DEVELOPMENT and SEASONAL FORM and between THERMAL FORM and SEASONAL FORM showed a sign change across temperatures. Reaction norms confirmed and clarified these sign changes. The sign change for DEVELOP- MENT-SEASONAL FORM might reflect underlying physiological processes. The sign change for THERMAL FORM-SEASONAL FORM might be caused by different trade-offs in the different seasons.

Keywords:geneticcorrelations, jackknife, , plasticity, reaction norms, wing pattern.

Introduction Within a single organism TgS can also differ consider- ably between environments (e.g. Giesel et a!., 1982; Geneticcovariance, or its standardized form genetic Service & Rose, 1985; Gebhardt & Stearns, 1988; correlation (rg), can have a strong influence on the Holloway et al., 1990). In other words the genetic result of natural selection and consequently play an relationship between characters can be plastic. Plasti- important role in multivariate evolution (Lande, 1982). city can be analysed with the help of TgS. In this case it is rgs between two characters reflect the number of genes, not TgS between two characters that are used, but rgs or linked genes, that influence both characters, and between the expressions of a single character in two also the distribution of relative strength of effects of the environments (Via & Lande, 1985). The r5 between genes (Falconer, 1989). rgs tend to be stronger between two characters, a and b, indicates the strength of the characters that are developmentally related (e.g. simultaneous effect on b, when selection occurs on a. Cowley & Atchley, 1990) and/or functionally related The Tgwithina character in two environments, x and y, (e.g. Kingsolver & Wiernasz, 1991). Negative rgs are indicates the effect on the character in y, when selec- expected in the case of trade-offs (Stearns, 1992). rgs tion occurs in x. between similar characters frequently differ (reviewed Via & Lande (1985) used the relationship between in Stearns et al., 1991), both between species (e.g. genotype by environment interaction (g x e) and the rg Lofsvold, 1986), and between populations (e.g. Dingle between environments to analyse plasticity. They etal., 1988). modelled the evolutionary trajectory of a (plastic) character when different optima are favoured in two *Present address and correspondence: Department of Biology, University of Antwerp (UTA), Universiteitsplein 1, B-26 10, Wilrijk, environments, i.e. selection for plasticity. The evolution Belgium. towards a joint optimum can be slowed down by strong 459 460 J. J. WINDIG correlations, but only correlations of +1or —I (no Materials and methods g Xe)can prevent the realization of a genotype which is optimal in both environments. G Xehas played an Study system arid rearing important role hitheanalysis of plasticity (Schlichting, 1976; Schemer & Lyman, 1989). Many studies have Bicyclusis a species-rich of estimated gXe (e.g. Zuberi & Gale, 1976; Groeters & (Condamin, 1973) and occurs throughout Africa south Dingle, 1987; Newman, 1988; Wade, 1990; Hughes, of the Sahara. Most members display a conspicuously 1992) in order to examine the evolutionary potential of dry—wet season polyphenism. The wing pattern of wet organisms in relation to variable environments. season forms has conspicuous elements (e.g. eyespots One rg within a character between two environments and a white median band) (Windig et a!., 1994). These may not be enough to characterize plasticity. Many elements are thought to deflect predator attacks away plastic characters show continuous responses to from the body, or to disrupt the shape of the wing. The changes in the environment. Even discrete responses, wing pattern of the dry season form is more uniformly or discrete phenotypes caused by discrete environ- brown and thought to be cryptic against brown, dead ments, often have underlying continuous reaction leaves (Brakefield & Larsen, 1984). The butterflies norms (Windig, 1992). Reaction norms can be used to only reproduce in the wet season. At the end of the wet analyse such characters (Thompson, 1991). The effects season, dry season butterflies appear, mainly resting on of reaction norms on rgs were modelled by de Jong the ground covered with dead brown leaves. In the next (1989. 1990a, b). wet season they are the first reproducing generation Genetic correlations within one environment (Brakefield & Reitsma, 1991). Bicyclus anynana is one between two characters can also be reflected in reac- of the most widespread members of the genus, and tion norms. Reaction norms can also be presented as occurs in savannah and at the edges of forests. plots of one character against another (Stearns, 1992) A laboratory population of butterflies originated with different genotypes and environments indicated. from a sample of over 80 gravid females from a In such plots rgs are reflected by regression lines population at Nkatha Bay, Malawi, with a highly through the points within one environment. These seasonal climate. A total of 43 pairings in two experi- regression lines will run parallel if rgs do not change ments were derived from this stock. Initially 21 families over environments. Their slopes will be different if the were obtained; the experiment was then repeated once rgschange. to obtain more families. Males and females were The aim of this study is to analyse a system, which is allowed to pair only once, so all families consisted of adaptively plastic, with the help of reaction norms and full-sibs. Offspring of each family were split over four rgs. The study system used is the tropical butterfly temperatures: 17°C (dry season temperature), 28°C Bicyclus anynana which has different, temperature (wet season temperature) and 20°C and 23°C (interme- induced, wing patterns in the dry and wet season diate temperatures). The number reared successfully (Brakefield & Reitsma, 1991). A continuous range of differed between experiments and temperatures (Table dry to wet wing patterns can be obtained in the labora- 1). Relative humidity was around 90 per cent and the tory by raising the butterflies at different temperatures light/dark regime was 12/12 h in all temperatures. (Windig, 1992). The genetics of the wing pattern Larvae were raised on a mixture of one of their natural changes across temperatures and there is substantial food plants (the grass Oplismenus compositus) and genetic variation for the plasticity itself (Windig, 1993, young maize (Zeamays).Details of the breeding proce- 1994). Artificial selection on one wing character (size dure are described byWindig(1994). of an eyespot) influences many other characters at the same temperature (Holloway et a!., 1993a). In this Measurements study the following questions will be addressed. I What form do the bundles of reaction norms have? Measurementsof the wing pattern were made with an 2 Are the rgswithincharacters, between tempera- image analyser (Windig, 1991), to an accuracy of tures, significant? Are there negative rgs? around 1 per cent. Eight characters of the wing pattern 3 Are the rgs between characters, within tempera- were measured. They were selected in the closely tures, significant? If so, are they constant, or do they related B. safitza for efficiency in indicating the wing change sign between temperatures? pattern and the accuracy of their measurement (for 4 Are the estimated rgs consistent with the reaction details see Windig, 1991, 1993). norms? To reduce the number of characters to be evaluated a principal component analysis (PCA) was used to summarize all wing characters into two components REACTION NORMS AND GENETIC CORRELATIONS 461

Table 1 Number of butterflies (Bicyclus anynana), mean number per family and total number of families (in parentheses) for temperatures, sexes and experiments

Experiment 1 Experiment 2

Males Females Males Females

Temperature n Mean(Fams) n Mean(Fams) n Mean(Fams) n Mean(Fams)

17° Reared 78 4.5(18) 60 4.0(15) 125 7.4(17) 115 6.7(17) Used 75 5.0(15) 58 4.5 (13) 121 8.1 (15) 111 7.4(15) 20° Reared 73 5.2(14) 74 5.7(13) 151 10.1 (15) 145 9.7(15) Used 72 5.5 (13) 73 6.1(12) 151 10.1 (15) 145 9.7 (15) 23° Reared 67 5.6(12) 71 5.9(12) 259 11.8(22) 257 12.2(21) Used 65 6.5(10) 67 6.7(10) 254 12.7 (20) 254 12.7 (20) 28° Reared 108 5.4(20) 116 5.5(21) 298 14.9(20) 273 13.7(20) Used 107 5.6(19) 112 5.9(19) 294 16.3(18) 269 14.9(18)

Reared, total number of butterflies reared; Used, familes that were (sometimes) considered outliers subtracted.

cONTRAST

PCi -SEASONAL FORM OUTER RING

Li-0 -J 0e.lC Fig. I Wing pattern traits measured, rw and composition of principal compo- N I- nents, in butterflies Bicyclus anynana. -0.5

(Fig. 1). Characters were transformed to normal (Windig, 1992). Small dark forms probably need less distributions where necessary (Windig, 1994) and time to heat up and can fly faster and at lower tempera- standardized before PCA was applied. Plastic charac- tures (Dennis, 1993). ters mainly contributed to the first principal compo- In addition to the wing characters, two other nent (PC 1, 47 per cent) which can be interpreted as an characters were measured. Larval development time index of seasonal form (henceforth SEASONAL (henceforth DEVELOPMENT) was measured as the FORM). Wing size and colour were the most important number of days from egg hatching till emergence of the characters for the second component (PC2, 32 per butterflies from pupae and was log transformed to cent), which can be interpreted as an index of thermal obtain a normal distribution. Pupae were weighed, with form and sex (henceforth THERMAL FORM), the an accuracy of 10- 4g, 2 days after pupation when their small, dark butterflies (e.g. males) having lower values cuticle had fully hardened. Pupal weight (henceforth 462 J. J. WINDIG

WEIGHT) was log transformed to obtain a normal between-family variance: distribution. The overall difference between tempera- tures was small for this character. In summary four Vm = characters are analysed in this study: two wing pattern VFamily+ (_)V0 characters and two life history characters, one of each Several other ways of estimating reliable rgs exist (Via, is plastic and the other not or only slightly so. 1984) but to attach standard errors to them is often a problem. Here a jackknife procedure (Holloway et al., 1990, 1994; Via, 1991) was used to calculate the Reactionnorms COVg and rg with standard errors. The jackknife is an Strictlyspeaking a reaction norm is the response of a iterative procedure (Arvesen & Schmitz, 1970; Miller, single genotype to changes in the environment. Here, 1974) and the calculations during each iteration are families were used to examine reaction norms rather carried out using n —1individuals (or k —1groups of than genotypes. Within a family only a limited number individuals). Here rg was jackknifed with each family of genotypes is present, and taking the mean of the omitted once so that the total number of iterations was observed (full-sib) values calculates a mean genotypic equal to the number of families. To take into account value for that family (e.g. Schemer & Lyman, 1989). the difference in the number of individuals per family, Calculating the means within environments and individuals were regressed on the mean of their family connecting them is then equivalent to constructing a in another temperature: reaction norm (Gebhardt & Stearns, 1992, used the same method for inbred isofemale lines). This ______COVi1,m2 COVm1,12 approach was used here and reaction norms were ______andrg2= compiled by connecting the means of the families in the ( V1)(p12) different temperatures. whereCOVI1m2 is the covariance of the individuals raised in one temperature with the means of their families raised in a second temperature, and V11 is the Calculationof genetic correlations variance of individuals (=phenotypicvariance) in the Analysisof data was performed on experiments and first temperature. rgs were z -transformed and normal- sexes separately because means, variances and h2s ized distributions of the jackknife estimates (Arvesen & varied between sexes and experiments, within tempera- Schmitz, 1970). The estimates are generally similar but tures (Windig, 1994). rgs were estimated for the four can be sometimes very different. For the regressions variables. SEASONAL FORM and THERMAL the final estimate used was the mean of the two regres- FORM are the first two components of a principal sions (Lam & Calow, 1989; Holloway etal., 1993b). component analysis, so their overall phenotypic corre- Jackknifing is very sensitive to outliers, but it also lation must be zero. Correlations within temperatures provides a good opportunity to detect them (Devlin et and sexes can, however, be different from zero, and the al., 1975; Hinkley, 1978). Jackknifing usually produces same applies to rgs. The expression of a character in a normal distribution of values after appropriate trans- different environments had to be measured on separate formation. Here a family was considered to be an out- individuals, and therefore the usual statistical method her and omitted from the calculation if it deviated by for calculation of the genetical correlations (e.g. more than three standard deviations from the mean of Falconer, 1989) is not applicable. They can, however, all the estimates and if such a family was very small be calculated by correlation of the family means in two (only one or two members in one of the temperatures), environments of families split over the two environ- i.e. when the estimation of the mean was probably ments(Via, 1984,1991): corrupted by the small sample size. This led to the removal of only one or two families from each calcula- COVmi,m2 tion (Table 1). rg Though rgs between two characters within environ- ( V1)(Vm2) ments are normally estimated with an ANCOVA-like whereCO Vmj,m2 is the covariance between the family procedure (Falconer, 1989), here the same jackknife means in the two environments, and Vmi, Vm2 are the procedure was used as for the rgs across temperatures. variances of the family means in the two environments These will give similar results to the ANCOVA-like proce- respectively. Such estimates will underestimate the rg, dures if the sample sizes are large (Via, 1984). especially for small family sizes, because of the All rgs in this study are estimated from full-sibs. So presence in the (co)variances of error (or within-family) they contain not only additive genetic effects but also variance, due to sampling error, in addition to the nonadditive genetic, maternal and common environ- REACTION NORMS AND GENETIC CORRELATIONS 463

mental effects. These nonadditive effects are probably not important for the wing pattern characters Results SEASONAL FORM and THERMAL FORM (cf. SEASONALFORM was more or less constant Kingsolver & Wiernasz, 1991). Common environ- between 17°C and 20°C (Fig. 2; reaction norms for mental effects might have somewhat more influence on SEASONAL FORM vs. DEVELOPMENT also DEVELOPMENT and WEIGHT, since these charac- reflect the reaction norm of SEASONAL FORM on ters are more likely to be influenced by amounts and temperature, since DEVELOPMENT is strongly quality of food. These common environmental effects correlated to temperature). In the range 20—28°C there will be somewhat reduced because large families were was a large increase in SEASONAL FORM. The TgS raised in two or more cages. Common environmental within seasonal form were neither strongly positive nor effects will cause a higher proportion of the phenotypic negative (Table 2). This corresponded to the form of correlations to be included in the rgs within tempera- the bundle of reaction norms with neither parallel reac- tures. The rgs across temperatures can be decreased by tion norms (for positive rgs), nor reaction norms cross- common environmental effects because the common ing at a single point (negative rgs). In the reaction norms environmental effects across temperatures are for DEVELOPMENT vs. temperature (inset Fig. 2) a uncorrelated; if the mean of a character for one family point around which reaction norms cross is very cJearly is increased in one temperature (e.g. by a relatively present just above 20°C. Corresponding negative rgs better food quality in its cage) it does not have to be were found between the higher and lower tempera- increased in another temperature. So the rgs within tures. temperatures, especially those with WEIGHT and The slopes of the regression lines of SEASONAL DEVELOPMENT, will have to be interpreted with FORM on DEVELOPMENT seemed to decrease caution. with increasing development time: positive at 28°, flat

1.7

0.85

0 z C

-0.85

-1.7 3.2 3.6 4 4.4 4.8 DEVELOPMENT Fig.2 Reaction norms for SEASONAL FORM—DEVELOPMENT time,for females. Symbols indicate means of families (hatched first experiment, solid black second). The means of the same family in different temperatures are connected, and thus form a reaction norm. Inset: reaction norms for DEVELOPMENT on temperature for females in the second experiment. for the and the and with were were in slope to 0.091 0.485* 0.252 0.221 0.029 0.178 middle 0.094 0.532* 0.533* 0.159 0.270 average reaction changes, Exp.2 —0.133 —0.176 —0.078 —0.143 —0.510 —0.211 —0.179 —0.278 —0.061 —0.069 Females, —0.072 large lines regression norms the experiment slopes families SEASONAL in similar SEASONAL sign females the whole

experiment 2 longer regression for low 0.243 the been 0.679* 0.258 0.698* 0.416 0.367 0.323 0.030 0.041 0.201 relatively 0.342 0.214 0.523* 0.013 0.006 0.037 0.479 second a from reaction the Males, Exp. —0.366 —0,015 —0.167 —0.255 —0.133 —0.237 positive regression outlying showed 28°, between FORM but the second have the 17° at when were in two somewhere rgs the outliers temperatures. 1 At by somewhat are in slope as removed 0.119 0.046 20° would The 0.414 0.987* 0.150 0,551* 0.439* where outliers —0.085 —0.221 —0,284 —0.072 a Exp. —0.292 —0.383 —0.067 —0.256 —0.134 —0.003 —0.049 —0.454 —0.437 experiment families Females, —0,136 —0.262 at they within than cross. been 28°. at caused not WEIGHT first point is at negative 1 are These removed they

DEVELOPMENT 20° correlation FORM—WEIGHT the FORM-DEVELOPMENT FORM-THERMAL negative the FORM-WEiGHT 0.051 FORM-DEVELOPMENT 0.413 0.772* 0.241 0.117 DEVELOPMENT 0.826* 0.498* 0.639* 0.123 0,733* not characters, Males, Exp. —0.650 —0.180 —0.205 —0.405 —0.207 —0.098 —0.314 —0.289 —0.002 —0.126 In single Had at They considered; and and bundle. 2). is estimates above were The experiment, genetic 23° average the 17° 17° 20° 23° 28° 17° 20° 23° 28° 17° 20° 23° 28° Between 17° 20° 23° 28° 17° 20° 23° 28° 20° 23° 28° FORM. norm thus analysis. other FORM DEVELOPMENT (Fig. DEVELOPMENT DEVELOPMENT flat. first an of and THERMAL at found SEASONAL just (b) SEASONAL SEASONAL ThERMAL

C Q Ca- > . Ca- Ca C) I Ca.-COCa -?a) a) CO 00 a) Ca 'C CaCa Ca Caa) Ca Ca Ca Ca C)Ca Ca, CO Ca Caa) Ca0 CaC) Ca' C)Ca Ca .0 a) '.-dcl-a) ci, a)a) 'C CO E 00 .0 E a) zo

C m '.0 C N m r C * — CN C '1' . — * * —1- N'O' CC CC CC OON NC CC Cf)CfCS CC —5 .C,S,CC -i't: — C' CS — clW CC CC N C C— N C S C S — C mI NN QCN N '.0 C N N N * - C CN C * * C°". * * cc,c. '.000 00 —C .-' C°'NC d© — CC S C'.0 '.000 '.r- CC N,CC * C- C — * CN 1Nj CC * S' CC r". C C * '.05 '.0' *C * - * N N — C C C I * N * C — * * C C '.0C * CC Cr'.O0S CC '.05 55 SN, — SNCS '.ON CC '.0Cc V C C'&CC N°t0000 CC C C * N C * or'•'. CN S CC CN 't•Q CC * C C I * *S C q CN N 00 re'. C C C I N N '.0N '.0C N C S — C N C * S C C t * * — -- ciciJ C'S C C NCIfl qCC I * SC,CC SCN N—' CC — —C NC C'.0 NS S'.C CC C * C C •rC NC1 C C C - rS N C N * * * S

N CN N S N N CN S 00 N N CN S N

C-Ca Ca Ca Ca Ca a) Ca Ca- Ca a) a) Ca C-Ca Ca C-I) E a)) a) a) 'Ca 'Ca H WINDIG J. Ca C00 Ca C I) CMa) Ca,E Ea) 'C a) UCa C- CO C CO J. -Ca - Ca CM a) CMa) C- Ca a) a) CMCM C C- C) Ca CMC- a) C.) CaC- Ca U HE 464 Cd, Cd, a)Ca- a) Cd,C- a) C.)Ca C- Ca C)C a) a) a) E - C C', C Ca a) C0 a) Ca Cd,a) Cd,Ca a)I- Ca 'C -C 'C . REACTION NORMS AND GENETIC CORRELATIONS 465 in the range from 28° to 200,correspondingto the reac- 28° to negative at 20° (Fig. 3). At 17° near-vertical lines tion norms (Table 2). seemed present. Slopes of regression lines, however, Many reaction norms ran parallel for THERMAL were not significantly different from zero, but this is FORM (Fig. 3). Some families had consistently lower expected for verticallines. The Tg5 between values than average and others consistently higher over THERMAL FORM and SEASONAL FORM showed the whole environmental range. The r0s within a corresponding sign change from positive at 28° to THERMAL FORM, between temperatures, were negative at 20°. There was no clear pattern in the reac- nearly all strongly positive (Table 2). Only for males, in tion norms for THERMAL FORM on DEVELOP- experiment 2, did negative correlations occur. The MENT and none of the rgs between THERMAL corresponding bundle of reaction norms indeed shows FORM and DEVELOPMENT were significant. crossing reaction norms between the high and low The bundle of reaction norms for WEIGHT was a parallel reaction norms. horizontal band with some crossing over of reaction In the graph for SEASONAL FORM-THERMAL norms. Nearly all TgS were positive, but only some, FORM (Fig. 3)themales and females were clearly mostly between neighbouring temperatures, were separated, females having a higher value for significant. There was a trend for the r5s to decrease THERMAL FORM (thus being paler and larger). when the temperatures were further apart (Table 2). Within sexes the butterflies of the first experiment were The rgs for WEIGHT with the other characters generally above those of the second experiment. were similar in different temperatures (Table 2): Regression lines in all groups, however, followed a negativefor WEIGHT-DEVELOPMENT and similar pattern. The slopes changed from positive at WEIGHT-SEASONAL FORM and positive for

1.9 V

1.0

U-0 -J 0.0 ILU I—

-tO

-1.9 -1.7 0.85 0 0.85 1 .7 SEASONAL FORM Fig. 3 Reaction norms for SEASONAL FORM-THERMAL FORM. Symbols as Fig. 2. 466 J. J. WIN DIG

WEIGHT-THERMAL FORM. Only some of these Many studies, mainly in Drosophila, report a sign were significant. change in rgs across environments (Murphy et a!., 1983; Service & Rose, 1985; Mukai, 1988; Newman, 1988; Gebhardt & Stearns, 1988, 1992). This study Discussion found two out of six correlations with sign changes Thisstudy had the objective to analyse a system which across environments. Reaction norms for wing pattern is adaptively plastic, with the help of reaction norms characters in B. anynana generally do not run parallel, and genetic correlations (rgS). It found that each and genetic variation in slopes of the reaction norms character had its own specific form for its bundle of for SEASONAL FORM is also present (Windig, reaction norms (question 1 of the Introduction). 1994) SO sign changes were not unexpected from this Consequently the rgs within characters, across point of view. temperatures (question 2) differed for the four charac- To clarify the sign change in the rg for SEASONAL ters and ranged from negative for DEVELOPMENT, FORM-DEVELOPMENT a simple graphical model around zero for SEASONAL FORM, to positive for with reaction norms for a limited number of genotypes WEIGHT and THERMAL FORM. Most rgs between was made (Fig. 4). In the model the negative rgs within characters, within temperatures, were not significant DEVELOPMENT can be seen because the genotypes (question 3), except for DEVELOPMENT-SEA- that have relatively fast development times at high SONAL FORM and THERMAL FORM-SEA- temperatures are relatively slow at low temperatures. SONAL FORM which changed sign from positive at Because there is no differentiation in SEASONAL 28° to negative at 200. Reaction norms and rgs were FORM at 23° and 17° the rgs around zero with these generally consistent with each other (question 4). temperatures are also explained. The instability of the From a theoretical point of view absence of strong rgs at 20° can be explained because the reaction norms positive correlations within a character together with cross around this temperature and at the same time changes of rgs between two characters are generally they flatten off. So although the reaction norms are not expected in plastic characters (Stearns et a!., 1991). linear, crossing of reaction norms leads to a sign Only when their reaction norms run parallel over the change. whole environmental range will this not be the case. A similar model was made for THERMAL For nonpiastic characters the same may apply. The FORM-SEASONAL FORM correlations (Fig. 5). only difference from plastic characters is that on The positive rgs for THERMAL FORM are reflected average, over all genotypes, the phenotype remains the by the fact that the rank of the genotypes is the same same over the whole environmental range. Individual for each temperature. At intermediate values for genotypes may show (limited) plasticity over (part of) SEASONAL FORM the values for THERMAL their environmental range in such a way that the rank FORM are identical for all genotypes in this model, order of the genotypes changes. explaining the horizontal regression lines and insignifi- Most studies that have analysed reaction norms have cant correlations at 23°. Only a slight modification, found bands in which crossing-over occurs, but not letting the reaction norms cross instead of touch each at one specific point (e.g. Mazer & Schick, 1991; other at intermediate temperatures, is needed to get Rawson & Hilbish, 1991; Gebhardt & Stearns, 1992). negative rgs. In this model a sign change in rgs can be Consequently nearly all TgS across environments explained with more or less parallel reaction norms in reported in the literature are positive (e.g. Via, 1984; one of the characters. Wade, 1990; Platenkamp & Shaw, 1992; Ebert et a!., 1993; Etges, 1993; Thomas & Bazzaz, 1993; Geneticcorrelations, developmental and functional Andersson & Shaw, 1994) and negative rgs are excep- tional (e.g. Via, 1991). This indicates that genes explanations influence a character in different environments in a Geneticcorrelations between characters will be strong similar way, and that selection in one environment has if they are developmentally or functionally related. a similar effect in a different environment. For this Thus developmental and/or adaptive processes might study the same applies to THERMAL FORM and explain the observed pattern of rgs. On a develop- WEIGHT. For SEASONAL FORM rgs were low and mental level characters that share part of their develop- this suggests that selection in one season has little effect mental pathway will show stronger rgs. At a functional on the expression of the phenotype in the other season. level stronger rgs will evolve if an increase in one For DEVELOPMENT the negative rgs imply that character is only adaptive when another character is selection in one season will have the opposite effect in increased (positive correlations) or decreased (negative the other season. correlations) simultaneously. When characters are considered in different environments their rgs will REACTION NORMS AND GENETIC CORRELATIONS 467

E C specific regression line • 0 0 genotypic values C C Fig. 4 Graphical model explaining the observed pattern for genetic correla- tions between DEVELOPMENT and SEASONAL FORM. The two lines indicate the two most extreme geno- types. Intermediate reaction norms are left out, except for one, for clarity. Points indicate DEVELOPMENT and SEASONAL FORM within one environment (e.g. temperature). One extreme genotype reacts more on tem- perature differences in the lower tem- peratures, the other on differences in the higher temperatures. development time

specific regression line • 0 0 genotypic values

0 LL —I ,cj

- Iwcc I, I- .0' S. I, Fig. 5 Graphical model explaining the observed pattern for genetic correla- tions between THERMAL FORM and SEASONAL FORM. Explanation as Figure 4. SEASONAL FORM change over environments if (variation in) their that there is a polymorphism for pigments which are developmental pathways and/or their functions differ produced independently of larval temperature, explain- in the environments. In plastic characters this will ing the genotypes that are darkest or lightest through- generally be the case. out the environmental range. In DEVELOPMENT r5s If functions of a character change completely over are positive between neighbouring temperatures but environments low TgS within a character can be negative between the more distant temperatures. expected. This allows the different expressions of such Genotypes with enzymes that have their kinetic optima a character to evolve relatively independently. This is at different temperatures (e.g. Watt, 1977; Watt et al., exactly what can be seen in SEASONAL FORM. The 1983) might explain such a pattern. observed low r5s are also in agreement with the The Tg between two characters will change over observed substantive amounts of additive variation for environments if a corresponding change in function or plasticity in this character (Windig, 1993). development occurs. A developmental or functional A developmental relationship between the expres- relationship between DEVELOPMENT and sion of a character in two environments is obvious; THERMAL form is difficult to envisage. The absence after all it is the same character. Crucial for the rgs is of significant rgs is then possibly the result of the whether flexibility exists in the developmental pathway. absence of any functional or developmental rela- It may well be, for example, in THERMAL FORM tionship. 468 J. J. WINDIG

In life history theory, growth rate and size are traits Phenotypic correlations between SEASONAL involved in an energy trade-off (Stearns, 1992). Such FORM and DEVELOPMENT are strongly negative trade-offs are believed to lead to the evolution of nega- within temperatures (Windig, 1992, 1994). A simple tive rgs (Lande, 1982; Rose, 1984). Since length of physiological model can explain this: the amount of a DEVELOPMENT time is opposite to growth rate, a certain chemical at the end of the larval stage deter- positivegeneticcorrelationis expectedfor mines the 'wetness' of the wing pattern of a butterfly, WEIGHT-DEVELOPMENT. Negative rgs were, and a constant fraction of this chemical is synthesized however, found in all temperatures. Though rgs con- or denatured each day in the larval stage. If there are sistent with the operation of trade-offs have sometimes differences in the rate of production (denaturation) of been found in other studies (e.g. Holloway et al., 1990; the chemical or in the initial amount this will lead to Rose & Charlesworth, 1981; Soliman, 1982), this is variation in SEASONAL FORM. A lower rate of pro- not always so (e.g. Stearns, 1983; Murphy eta!., 11983; duction (denaturation) that is genetically linked to a Bell,1984a,b; Holloway eta!., 1993b). Several authors faster development can explain the observed significant have suggested why unexpected rgs are sometimes rs. A better understanding of the physiology is needed found (Service & Rose, 1985; van Noordwijk & de to explain fully the relation between SEASONAL Jong, 1986; Holloway et a!., 1990; Charlesworth, FORM and DEVELOPMENT, and possibly the 1990). No definite conclusions can be drawn from this observed rgs. study since many of the negative rgs were not signifi- At 28° a clear negative relationship exists between cant, and the possibility of common environmental and SEASONAL FORM and THERMAL FORM. This maternal effects having influenced the rgs exists. An might reflect a trade-off: butterflies can either develop experimental design better suited to evaluate life his- into a form that is suited to good fliers: small with dark tory traits is needed to demonstrate whether the nega- wings (optimal pattern in males), or a form that suits tive rgs found here for WEIGHT-DEVELOPMENT better motionless (e.g. egg-laying) butterflies (optimal in really exist. females). In dark wings a pattern with a white band and WEIGHT and THERMAL FORM are characters large eyespots etc. might interfere with the thermal that are developmentally related. Both are, at least requirements, whereas in a large, pale wing large eye- partly, measurements of body size (Windig, 1993). spots might be more effective and more important in Apart from size THERMAL FORM is also deter- motionless butterflies. In the dry season, butterflies mined by the darkness of the wings. Nevertheless, might be either optimized for survival in the dry period THERMAL FORM and WEIGHT are clearly (large and dry forms), or optimized for reproduction in developmentally related and positive correlations are the wet period (small and wet forms). This might to be expected between the two characters. Indeed explain the negative rgs between SEASONAL and nearly all correlations found between them are posi- THERMAL FORM in the lower temperatures as well tive. The fact that only a few values are significantly as the negativergs between WEIGHT and different from 0 is probably more the result of the SEASONAL FORM. weakness of power of the test, than any reflection of a real absence of correlation. A trade-off between SEASONAL FORM and Concludingremarks DEVELOPMENT might explain the positive rgs Theargument of Stearns et a!. (1991), that TgS them- found at the higher temperatures. In the wet season selves are not interesting but that they might indicate directionalselection for DEVELOPMENT is the underlying developmental and functional rela- expected; faster developing individuals can probably tionships between characters (see also Cheverud, foster more offspring. If the production of a wet wing 1984), is emphasized by this study. Physiological pattern is costly, for example in energetic terms, and a (DEVELOPMENT-SEASONAL FORM) or func- dry wing pattern is not, a trade-off with growth rate tional (SEASONAL FORM-THERMAL FORM) might result in positive correlations between processes can explain the observed rgs and their sign SEASONAL FORM and DEVELOPMENT in the changes across environments. Evaluation of the fit- wet season, and an absence of trade-offs in the dry nesses of different genotypes in the field is needed, on season. It is, however, doubtful if the production of a the one hand to determine whether selection could wet season wing pattern is costly. French & Brakefield have led to the observed rgs, and on the other to deter- (1992) have demonstrated that the production of eye- mine how correlations across as well as within environ- spots in B. anynana can be induced or inhibited by ments influence the outcome of selection. Better simple cautery of some cells on the wing epidermis in knowledge of physiological processes is also needed to the pupal stage. understand some rgs, including those between DEVE- REACTION NORMS AND GENETIC CORRELATIONS 469

LOPMENT and SEASONAL FORM at different tem- EBERT, D., YAMPOLSKY L. AND VAN NOORDWIJK, A. J. 1993. peratures. It will only be possible to understand fully Genetics of life history in Daphnia magna. II. Phenotypic evolutionary processes in the field if we understand the plasticity. Heredity, 70, 344—352. development and function of the characters under ETGES, w. .i. 1993. Genetics of host—cactus response and life- history evolution among ancestral and derived popula- study. tions of cactophilic Drosophila mojavensis. Evolution, 47, 750—767. Acknowledgements FALCONER, iis.1989. Introduction to Quantitative Genetics, Iam grateful to Paul Brakefield and Graham Holloway 3rd end. Longman, London. for many suggestions and helpful comments during the FRENCH, V. AND BRAKEFIELD, P. M. 1992. The development of experiments and analysis. Graham was especially help- eyespot patterns on butterfly wings: morphogen sources or ful with the calculations of the genetic correlations. sinks? Development, 116, 103—109. GEBHARDT,M. D. AND STEARNS, S.c. 1988. Reaction norms for Their comments on the manuscript together with those developmental time and weight at eclosion in Drosophila of Gerdien de Jong were also very useful. Bert de mercatorum. J. Evol. Biol., 1,335—354. Winter, Diana Lockhorst and Marianne van Genderen GEBHARDT, M. D. AND STEARNS, S. C. 1992. Phenotypic plasticity helped raise all the very hungry larvae, for which Els for life history traits in Drosophila melanogaster. II. Effect Schlatmann, Wim Kanters and Bert de Winter raised of the environment on genetic parameters. Gen. Res., 60, the food plants. 87—101. GIESEL, J. T., MURPHY, P.A. AND MANLOVE, M. N. 1982.The References influence of temperature on genetic interrelationships of life history traits in a population of Drosophila melano- ANDERSSON,S. ANDSHAW,R.G. 1994.Phenotypic plasticity in gaster: what tangled data sets we weave. Am. Nat., 119, Crepis tectorum (Asteraceae): genetic correlation across 464—479. light regimens. Heredity, 72, 113—125. GROETERS, F. R.AND DINGLE, H. 1987.Genetic and maternal ARVESEN, J. M. AND SCHMITZ, T. H. 1970. Robust procedures for influences on life history plasticity in response to photo- variance components problems using the jackknife. period by milkweed bugs (Oncopeltusfasciatus). Am. Nat., Biometrica, 26, 677—686. 129,332—346. BELL, G. 1984a. Measuring the cost of reproduction. 1. The HINKLEY,D. v.1978. Improving the jackknife with special correlation structure of the life table of a plankton rotifer. reference to correlation estimation. Biometrika, 65, Evolution, 38, 300—3 13. 13—21. BELL, G. 1984b. Measuring the cost of reproduction. II. The HOLLOWAY, G. J., POVEY, S. R.AND SIBLY,R.M. 1990.Theeffectof correlation structures of the life tables of five freshwater newenvironment on adapted genetic architecture. invertebrates. Evolution, 38, 314—326. Heredity, 64, 323—330. BRAKEFIELD, P. M. AND LARSEN, T B. 1984. The evolutionary HOLLOWAY, 0. J., BRAKEHELD,P. M. AND KOFMAN, s.1993a.The significance of dry and wet season forms in some tropical geneticsof wing pattern elements in the polyphenic butter- butterflies. Biol. J. Linn. Soc., 22, 1—12. fly Bicyclus anynana. Heredity, 70, 179—186. BRAKEFIELD,P. M. AND REITSMA, N. 1991.Phenotypic plasticity, HOLLOWAY, 0. J., DE JONG, P. W. AND OTFENHEIM, M. 1993b. The seasonal climate and the population biology of Bicyclus quantitative genetics and cost of defence in the 2-spot butterflies in Malawi. Ecol. Entomol., 10,291—303. ladybird (Adalia bipunctata) and the effect of sex on the CHARLESWORTH, B. 1990. Optimization models, quantitative additive genetic variance/covariance matrix. Evolution 47, genetics, and mutation. Evolution, 44, 520—538. 1229—1239. CHEVERUD, i. M. 1984. Quantitative genetics and develop- HUGHES, D.. 1992.Genotype—environment interactions and mental constraints on evolution by selection. J. Theor. relative clonal fitness in a marine bryozoan. J. Anim. Biol., 110, 155—171. Ecol., 61,291—306. CONDAMIN, M. 1973. Monographie du genre Bicyclus de JONG, G. 1989. Pheriotypically plastic characters in isolated (Lepidoptera Satyridae). Mem. Inst. Fond. Afr. Noire, 88, populations. In: Fontdevila, A. (ed.) Evolutionary Biology 1—324. of Transient Unstable Populations, pp. 3—18. Springer, COWLEY, 0.F.ANDATCHLEY,W. R. 1990. Development and Heidelberg. quantitative genetics of correlation structure among body de lONG, G. 1990a. Quantitative genetics of reaction norms. J. parts of Drosophila melanogaster. Am. Nat., 135, Evol. BioL, 3, 447—468. 242—268. de JONG, G. 1 990b. Genotype by environment interaction and DEVLIN,S.J., GNADESIKAN, R. ANDKETFERING,J. R.1975.Robust the genetic covariance between environments: multilocus estimation and outlier detection with correlation coeffi- genetics. Genetica, 81, 171—177. cients. Biometrika, 62, 531—545. KINGSOLVER, J. G. AND WIERNASZ, D. c. 1991. Development, DENNIS, R. L. H. 1993. Butterflies and Climate Change. function and the quantitative genetics of wing melanin Manchester University Press, Manchester, U.K. pattern in Pieris butterflies. Evolution, 45, 1480—1492. DINGLE, H., EVANS, K. E. AND PALMER, .i. o. 1988. Responses to LAM, P. K. S. AND CALOW, P. 1989. Intraspecific life—history selection among life history traits. Evolution, 42, 79—92. variation in Lymnea peregra (Gastropoda: Pulmonata). II. 470 J. J. WINDIG

Environmental or genetic variance? J. Anim. Ecol., 58, STEARNS, s. c. 1983. The genetic basis of differences in life 589—602. history traits among six populations of mosquito fish LANDE, R. 1982. A quantitative genetic theory of life history (Gambusia affinis) that shared common ancestors in 1905. evolution. Ecology, 63,607—615. Evolution, 37, 6 18—627. L0F5vOLD, D. 1986. Quantitative genetics of morphological sTEARNS, s. c. 1992. The Evolution of Life History. Oxford differentiation in Peromyscus. 1. Tests of homogeneity of University Press, New York. genetic covariance structure among species and sub- STEARNS, S. C., de JONG, G. AND NEWMAN, R. A. 1991. The effects species. Evolution, 20, 5 59—573. of phenotypic plasticity on genetic correlations. Trends MAZER, S. J. AND SCHICK, C. T. 1991. Constancy of population Ecol. Evol., 6, 122—126. parameters for life history and floral traits in Raphanus ThOMAS, S. C. AND BAZZAZ, F. A. 1993. The genetic component sativus L. 1. Norms of reaction and the nature of genotype in plant size hierarchies: norms of reaction to density in a by environment interactions. Heredity, 67, 143—156. Polygonum species. Ecol. Monogr., 63,231—249. MILLER, R. o. 1974. The jackknife—a review. Biometrika, 61, THOMPSON, .D.1991. Phenotypic plasticity as a component of 1—15. evolutionary change. Trends Ecol. Evol., 6, 246—249. MuKAI, T. 1988 Genotype—environment interaction in relation VIA, S. 1984. The quantitative genetics of polyphagy in an to the maintenance of genetic variability in populations of herbivore.II. Genetic correlations in larval Drosophila melanogaster. In: Weir, B. S., Eisen, E. J., performance within and among host plants. Evolution, 38, Goodman, M. J. and Namkoong, G. (eds) Proceedings of 896—905. the Second International Conference on Quantitative VIA, S. 1991. The genetic structure of hostplant adaptation in Genetics, pp.2l—3l. Sinauer, Sunderland, MA. a spatial patchwork: demographic variability among MURPHY, P. A., DIESEL, T. A. AND MANLOVE, M. N. 1983. Tempera- reciprocally transplanted clones. Evolution 45, 827—852. ture effects on life history variation in Drosophila simu- VIA, S. AND LANDE, R. 1985. Genotype environment interaction Ions. Evolution, 37, 1181—1192. and the evolution of phenotypic plasticity. Evolution, 39, NEWMAN, R. A. 1988. Adaptive plasticity in development of 505—522. Scaphiopus couchii tadpoles in desert ponds. Evolution, WADE, M. .i. 1990. Genotype—environment interaction for 42,671—678. climate and competition in a natural population of flour vanNOORDWIJK,A. J. AND de JONG, G. 1986. Acquisition and allo- beetles Tribolium castaneum. Evolution, 44, 2004—2011. cation of resources: their influence on life history tactics. WA'IlT, W. B. 1977. Adaptation at specific loci. I. Natural selec- Am. Nat., 128, 137—142. tion of phosphoglucose isomerase of Colias butterflies: PLATENKAMP, G. A. J. AND SHAW, R. 0. 1 992. Environmental and biochemical and population aspects. Genetics, 87, genetic constraints on adaptive population differentiation 177—194. in Anthoxanthum odoratum. Evolution, 46,341—352. WATT, W. B., CASSIN, R. C. AND SWAN, M. s. 1983. Adaptation at RAWSON, .D.AND HILBISH, T. H. 1991. Genotype—environment specific loci. III. Field behaviour and survivorship differ- interaction for juvenile growth in the hard clam ences among Colias PGI genotypes are predictable from Mercenaria mercenaria (L.). Evolution, 45, 1924—1935. in vitro biochemistry. Genetics, 103,725—739. ROSE, M. R. 1984. Genetic covariance in Drosophila life WINDIG, i. i. 1991. Quantification of butterfly wing patterns history: untangling the data. Am. Nat., 123, 565—569. using an image analyser. J. Res. Lepid., 30, 82—94. ROSE, M. R. AND CHARLESWORTH, n. 1981. Genetics of life history WINDIG, j.j. 1992.Seasonal polyphenism in : a in Drosophila melanogaster. I. Sib analysis of adult continuous reaction norm. N. J. Zool. 42, 583—594. females. Genetics, 97, 187—196. WINDIG, ..i.1993. The Genetic Background of Plasticity in SCHEINER, S. M. AND LYMAN, R. F. 1989. The genetics of pheno- Wing I'attern of Bicyclus Butteif lies. Ph.D. Thesis, typic plasticity. I. Heritability. J. Evol. Biol., 2, 95—107. University of Leiden. SCHLICHTING, C. D. 1976. The evolution of phenotypic plasti- WINDIG, j.j. 1994.Reaction norms and the genetic basis of city in plants. Ann. Rev. Ecol. Syst., 17, 667—693. phenotypic plasticity in wing pattern of the butterfly SERVICE, P. M. AND ROSE, M. R. 1985. Genetic covariation among Bicyclus anynana. J. Evol. Biol., (in press). life history components: the effect of novel environments. WINDIG, J. J., BRAKEFIELD, P. M., REITSMA N., AND WILSON J. G. M. Evolution, 39, 943—945. 1994. Seasonal polyphenism in the wild: survey of wing SOLIMAN, M. n. 1982. Directional and stabilizing selection for patterns in five species of Bicyclus butterflies in Malawi. developmental time and correlated response in reproduc- Ecol. Entomol., (in press). tive fitness in Tribolium castaneum. Theor. Appl. Gen., ZUBERI, M. I. AND GALE, j.s.1976. Variation in wild populations 63,111—116. of Papa ver dubium. X. Genotype—environment interaction associated with differences in soil. Heredity, 36,359—368.