Latitudinal insect body size clines revisited: a critical evaluation of the saw-tooth model

Authors


Correspondence author. E-mail: sami.kivela@oulu.fi

Summary

1. Insect body size is predicted to increase with decreasing latitude because time available for growth increases. In insects with changing voltinism (i.e. number of generations per season), sharp decreases in development time and body size are expected at season lengths where new generations are added to the phenology of a species, giving rise to saw-tooth clines in these traits across latitudes. Growth rate variation may affect the magnitude of variation in body size or even reverse the saw-tooth cline.

2. In this study, we analyse latitudinal body size clines in four geometrid moths with changing voltinism in a common laboratory environment. In addition to body size, we measured larval development time and growth rate and genetic correlations among the three traits.

3. The patterns of clinal variation in body size were diverse, and the theory was not supported even when saw-tooth body size clines were found. Larval development time increased and growth rate decreased consistently with increasing season length, the clines in these traits being uniform.

4. The consistencies of development time and growth rate clines suggest a common mechanism underlying the observations. Such a mechanism is discussed in relation to the complex interdependencies among the traits.

Introduction

Understanding large-scale geographical variation in life-history traits within a species lies at the core of modern evolutionary biology. Body size is one of the key life-history traits (Roff 1992; Stearns 1992), and it often varies spatially as latitudinal or altitudinal clines (reviewed by Mousseau 1997; Blackburn, Gaston & Loder 1999; Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006). Bergmann’s rule states that endotherm species are larger in cold than in warm environments (Blackburn, Gaston & Loder 1999), but the rule is also applicable to the intraspecific level (Blackburn, Gaston & Loder 1999) and to many ectotherms (Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006). Bergmann’s rule translates to clinal variation where body size increases with increasing latitude or altitude (but see Geist 1987). The temperature-size rule states that ectotherms grow larger in cold than in warm environments because of phenotypic plasticity (von Bertalanffy 1960; Atkinson 1994; Atkinson & Sibly 1997), which could generate a Bergmann cline in body size. Bergmann clines in body size arise, at least partially, as a response to temperature (von Bertalanffy 1960; Atkinson & Sibly 1997; Chown & Gaston 1999, 2010), but the underlying mechanisms are controversial (Geist 1987; Van Voorhies 1996, 1997; Atkinson & Sibly 1997; Mousseau 1997; Partridge & Coyne 1997; Angilletta & Dunham 2003).

The pattern of variation in body size is opposite to the Bergmann cline (i.e. a converse Bergmann cline; body size decreases with increasing latitude or altitude) in many ectotherms (reviewed by Mousseau 1997; Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006). Converse Bergmann clines arise as a result of variation in season length (Masaki 1967; Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004). In temperate seasonal environments, only a part of the year is favourable for ectotherm growth and reproduction, this season becoming shorter with increasing latitude or altitude. If generation time is less than a year, time available for growth easily becomes a limiting factor for ectotherm size. This may constrain body size because larger size may be attained only by prolonging growth (Roff 1992; Stearns 1992). Hence, a converse Bergmann cline emerges in body size, and cogradient variation (see Conover & Schultz 1995), in development time. An additional constraint for life-history evolution arises because often a particular developmental stage has to be reached in due time to survive the adverse season (Tauber, Tauber & Masaki 1986).

Insect generation times are predominantly short compared with the seasonal cycle. Accordingly, converse Bergmann clines are common in insects (reviewed by Mousseau 1997; Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006), but Bergmann clines exist as well (reviewed by Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006). Blanckenhorn & Demont (2004) showed that relatively large insects with long development times tend to express converse Bergmann clines, whereas relatively small insects with short development times tend to express Bergmann clines. The clines are often genetic (Masaki 1967; Blanckenhorn & Fairbairn 1995; Telfer & Hassall 1999; Blanckenhorn & Demont 2004; Schutze & Clarke 2008), which suggests an adaptive value for body size variation, but nonadaptive explanations have also been suggested (Van Voorhies 1996, 1997).

Variation in number of generations per season (i.e. voltinism) affects time constraints. The traditional view that how latitudinal or altitudinal gradient in season length affects the evolution of development time and body size in potentially multivoltine insects was developed by Masaki (1967, 1972) and formalized by Roff (1980, 1983) and Iwasa, Ezoe & Yamauchi (1994). They assumed that development time and body size are positively correlated among genotypes, and insects use all available time for growth to maximize adult body size and thus fecundity (see Honěk 1993). Consequently, both development time and body size are expected to increase along a gradient of increasing season length (i.e. with decreasing latitude or altitude) as long as voltinism does not change, giving rise to a converse Bergmann cline. However, sharp decreases in both traits are expected at the season lengths that facilitate the emergence of a new generation within the season, because time available per generation decreases at these transitions. Thus, the traditional saw-tooth cline (Fig. 1a) arises across phenologies because of genetic adaptation to local conditions (Masaki 1967; Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994).

Figure 1.

 Predicted clinal variation in relation to season length in age and size at maturity under different patterns of growth rate variation (left) and the predicted genetic correlations among these three traits (right; − = negative correlation, 0 = no correlation, + = positive correlation). Insects are assumed to use all available time for growth – hence, the predicted cline for age at maturity is the same in all of the cases, that is, decreasing season length (i.e. increasing latitude or altitude) within a given phenology results in a shortened time for growth. In (a), growth rate does not vary, and as a consequence, age and size at maturity covary positively giving rise to traditional saw-tooth clines in both of the traits. In (b), growth rate increases with decreasing season length within a particular phenology and compensates perfectly for the change in age at maturity, which results in constant size at maturity irrespective of variation in season length and phenology. In (c), growth rate overcompensates the change in age at maturity as season length decreases within a given phenology, reversing the body size cline predicted in a).

Blanckenhorn & Demont (2004) introduced countergradient variation (Conover & Schultz 1995) of growth rate into the analysis, albeit they ignored variation in voltinism. In the traditional view, growth rate was constant over the gradient of season length, whereas a higher (genetic) growth rate evolves as a response to increasing time constraints (i.e. shortening generation time) under countergradient variation. If increasing growth rate perfectly compensates for the decrease in generation time, body size remains constant over the whole gradient of season length (Fig. 1b). If a decrease in generation time increases growth rate more than is needed to maintain constant body size (i.e. overcompensation), the traditional saw-tooth cline will be reversed, so that body size decreases with increasing season length within a particular phenology (Fig. 1c). Countergradient variation in growth rate may also give rise to a traditional-like saw-tooth cline with a smaller magnitude of body size variation (i.e. undercompensation) than predicted by the traditional model.

The saw-tooth clines arise because both season length and voltinism set constraints for development time. The saw-teeth arise as a result of variation in voltinism, because the time available per generation sharply decreases when an additional generation emerges, but the effect of changing voltinism on time constraints decreases with increasing number of generations. Seasonal time constraints for growth probably decrease as the number of generations increases (Chown & Gaston 1999), and the aseasonal selection regime is approached (Nylin & Gotthard 1998; Chown & Gaston 2010). Consequently, saw-tooth clines are expected in species with long development times in relation to season length (i.e. few generations per season) (Chown & Gaston 1999), that is, in species where converse Bergmann clines in body size are expected (Blanckenhorn & Demont 2004). These kinds of species have provided empirical support for the traditional saw-tooth model and its genetic basis (Masaki 1972; Mousseau & Roff 1989; Burke et al. 2005). The traditional saw-tooth model fails to explain body size variation in species having numerous generations per season, like many dipterans, that generally show Bergmann clines in body size (reviewed by Chown & Gaston 1999, 2010; Blanckenhorn & Demont 2004; Dillon, Frazier & Dudley 2006).

The traditional saw-tooth model or its extensions do not seem to provide a general explanation for body size variation even in species with long development times. Despite a shift between uni- and bivoltine phenologies, a uniform converse Bergmann cline has been found both in the laboratory (Blanckenhorn & Fairbairn 1995) and in the field (Blanckenhorn & Demont 2004; data from Nylin & Svärd 1991). Moreover, two species of butterflies follow Bergmann clines in the field (Blanckenhorn & Demont 2004; data from Nylin & Svärd 1991), suggesting overcompensation of growth rate. Variation in growth rate should therefore be analysed rigorously to assess the potential role of countergradient variation in generating body size clines.

We studied latitudinal variation in body size in four species of geometrid moths in a common laboratory environment. The studied latitudinal gradient includes a shift between uni- and bivoltine phenologies in each study species. We analysed whether body size variation follows either the traditional saw-tooth cline or the countergradient variation scheme. To assess the proposed patterns of growth rate variation (constant growth rate vs. countergradient variation, Fig. 1), we measured larval growth rates in each species. Finally, we estimated genetic correlations between body size, development time and growth rate as the theory predicts a genetic basis for clinal variation, and the alternative hypotheses predict different genetic correlation structures among the studied traits (see Fig. 1). As body size, development time and growth rate are developmentally and physiologically tightly connected (Nijhout, Roff & Davidowitz 2010), the signs of genetic correlations between these traits should be insensitive to environmental factors (Stearns, de Jong & Newman 1991). Thus, the genetic correlation structure estimated in the common laboratory environment reflects the genetic correlations in the field conditions, justifying the common garden approach.

Materials and methods

Study species

Four widespread geometrid moths (Lepidoptera: Geometridae), Cabera exanthemata (Scop. 1763), Cabera pusaria (L. 1758), Chiasmia clathrata (L. 1758) and Lomaspilis marginata (L. 1758), were studied. The species are bivoltine in southern Finland and univoltine in northern Finland (Mikkola, Jalas & Peltonen 1989; see below for more details), which strongly suggests that the species have long development times in relation to season length under natural conditions. The species’ diapause generations fly in early summer (June–early July), and nondiapause generations, in late summer (late July – August). Only the pupal stage is able to overwinter (Mikkola, Jalas & Peltonen 1989). The species can be classified as capital breeders (see Tammaru & Javoiš 2000 for C. pusaria and C. clathrata) because adult feeding on nectar is rare (Mikkola, Jalas & Peltonen 1989; own observations), and females have a large number of eggs ready to be laid at adult eclosion (own observations). Larvae of C. exanthemata and L. marginata feed mainly on Salix spp., while Alnus and Betula species are the main hosts of C. pusaria (Mikkola, Jalas & Peltonen 1989). C. clathrata feeds on several leguminous plants (Fabaceae) and on Galium (Mikkola, Jalas & Peltonen 1989).

Sampling procedure

Moths were sampled from 22 locations (judged as populations) along a latitudinal gradient of changing season length in Finland (Fig. S1; Table S1, Supporting information). The species are bivoltine with a relatively abundant nondiapause generation in southern Finland and univoltine in northern Finland (Hyönteistietokanta 2009; own observations). A small nondiapause generation regularly arises in central Finland (Hyönteistietokanta 2009; Matti Ahola, personal communication), indicating a partially bivoltine phenology there. Altitudinal variation was slight among the study populations (all within 60–250 m above sea level). Females were caught from the field with a net. Only diapause generations were sampled to avoid any variation arising from different developmental pathways (e.g. Spence 1989; Wiklund, Nylin & Forsberg 1991; Blanckenhorn 1994). Sampling was conducted in June and July. The generation individuals belonged to was subjectively determined based on the date of capture.

Estimating season length

Season length was estimated as the length of the season (days) when average daily temperature is above 10 °C. A generalized additive model [function gam (Wood 2006) in r 2.10.1 (R Development Core Team 2009)] was fitted to the data (years 2001–2008) on average daily temperatures in each sampling location. The raw data were interpolated daily temperatures in 10-km grid (mean daily temperature, Finnish Meteorological Institute 2001–2008). The beginning and termination dates of the season were determined by the intersection of the fitted curve and the 10 °C threshold temperature. Season length was defined then by the number of days between these dates (the endpoint dates are included).

Common garden experiment

Wild-caught females were placed into 0·1-l plastic containers and provided with host plant for oviposition. Salix phylicifolia L. was used as a host for C. exanthemata and L. marginata, Alnus incana (L.) for C. pusaria and Lathyrus pratensis L. for C. clathrata. The females were allowed to oviposit ad libitum. The eggs were monitored daily for hatching, and five offspring of each female were taken for rearing. The female parent formed a family-factor used in the analyses. The numbers of families per population were affected by the numbers of females we were able to capture and the hatching success of the eggs (Table S1, Supporting information). In total, we got 125, 103, 114 and 99 families for C. exanthemata, C. pusaria, C. clathrata and L. marginata, respectively.

Neonate larvae were placed individually in 0·2-l plastic containers provided with moist garden peat at the bottom and a fresh host plant. Containers were distributed in spatially randomized design within a species in a laboratory and monitored daily. A fresh host plant was substituted and water was added to maintain humidity when necessary. The larvae were reared until pupation at a constant temperature of 20 ± 1 °C and under a photoperiod of 8 h : 16 h (light : dark). The short day length was applied to invariably induce diapause in genotypes of different latitudinal origins (see Masaki 1972; Tauber, Tauber & Masaki 1986; Burke et al. 2005). The larval period was considered to end on the day when a larva burrowed into the peat to pupate. Five days later, the pupa was weighed to the nearest 0·01 mg with a precision balance (Mettler Toledo MT 5) to get a surrogate of adult body mass. Then, the pupae were individually placed in 0·2-l plastic containers with moist moss (Sphagnum spp.) and transferred to a dark refrigerator (5 °C) for overwintering. After 6–9·5 months of hibernation, the pupae were exposed to continuous light at 20 °C. Adult moths were frozen on the day they emerged. The sex of individuals that died during the pupal stage was determined on the grounds of genital scars on the pupal cuticle (Scoble 1992). The midfemur lengths of eclosed individuals were measured to get a linear measure of adult body size. The left midleg of each individual was detached (if the left midleg was damaged, the right midleg was detached instead) and photographed through a microscope on a millimetre paper for scaling. The femur length was measured from the photographs using ImageJ 1.41o (Rasband 2009). Each individual was measured twice to control for measurement error. All measurements of a particular species were taken by a single person to avoid error at the measurer level.

Determination of larval growth rate

Insect growth trajectory data indicate that growth rate increases towards the end of the larval period (e.g. von Bertalanffy 1960; Tammaru & Esperk 2007; Gotthard 2008). This kind of growth has been described with the exponential model

image(eqn 1)

where m is mass of the larva at time t, m0 is mass of a neonate larva and k is a growth parameter affecting growth rate (growth rate increases with increasing k). Under this model, an index of growth rate for the larval stage is the relative growth rate, k, calculated by

image(eqn 2)

where mpupa is pupal mass and tlarva is larval development time. Tammaru & Esperk (2007) argued that larval growth in Lepidoptera follows a power function and can be described as

image(eqn 3)

where m is mass of the larva at time t, c is a parameter affecting growth rate (growth rate increases with increasing c) and = 1/(1 − b), b being an allometric exponent relating anabolism to body mass. In eqn 3, c has a comparable role to k in eqn 1, so c can be used as an index of growth rate under this model. With a realistic estimate of the allometric exponent b, the index is calculated by

image(eqn 4)

According to Tammaru & Esperk (2007), = 0·8 seems a realistic estimate for our study species and was thus applied. The correlation between c (eqn 4) and k (eqn 2) is strong and positive in C. exanthemata, where we have data on m0, when b equals 0·8 (Spearman correlation rs = 0·995, n = 489, P < 0·0001) or 0·67 (rs = 0·945, n = 489, P < 0·0001), indicating that the two indices similarly describe differences in growth rate among individuals. The strong correlation between c and k is not surprising as both the exponential model and the power function model are related so that when = 1, growth is exponential. The models describe different growth trajectories only when b ≠ 1. Because the power function (index c) deviates more from the exponential model (index k) with decreasing b, the analysis of clinal variation in growth rate was repeated by using = 0·67 to evaluate how robust the results are in relation to the value of b.

Data analyses

Support for a particular hypothesis presented in Fig. 1 would require that clinal variation in body size and development time qualitatively fit the predictions under that particular hypothesis and that the signs of the genetic correlations between body size, development time and growth rate match the predictions. Lack of fit between the data and even a single specific prediction would refute a hypothesized model.

Statistical analyses were performed with r 2.10.1 (R Development Core Team 2009). The measurement error in midfemur length was calculated as a percentage of total variance of the data that was because of variance within individuals. The variance components needed to calculate the measurement error were extracted with a linear mixed-effects model [function lme (Pinheiro et al. 2009)]. The model was fitted with the restricted maximum-likelihood method, with midfemur length as a response variable, round of measurement (factor with two levels) as a fixed factor and individual as a random factor.

The geographical variation in midfemur length, pupal mass, larval development time and larval growth rate was analysed with linear mixed-effects models (function lme) fitted with the maximum-likelihood method. Random effects were initially defined so that random families were nested within random populations. In many cases (see Table 1), the standard deviation at the population level was negligible, indicating no need to include population in random effects (Pinheiro & Bates 2000). In these cases, population was excluded from the random effects. Season length (as a quantitative variable) and sex were set as fixed factors. Initially, interactions between sex and each of season length, square of season length, cube of season length and the fourth power of season length were included in the model, and then, the fixed effects were reduced according to the principle of hierarchy to find the definitive model. In the analysis of larval growth rate in C. clathrata, the two highest order terms of season length had to be omitted as the algorithm did not converge if they were included in the fixed effects. Polynomial regression was selected to objectively test (the exact location of the change in phenology is vague) whether the data would indicate existence of the predicted saw-tooth cline. The traditional saw-tooth cline (Fig. 1a) would generate a significant positive cubic term for season length, whereas the reversed saw-tooth cline owing to overcompensating growth rate (Fig. 1c) would generate a significant negative cubic term for season length.

Table 1.   Parameter estimates for linear mixed-effects models explaining variation in the measured variables in relation to season length in the four geometrid moths studied. Season stands for season length, for simplicity
SpeciesVariableParameterEstimateSEd.f.tP
  1. aFamily as a random factor.

  2. bHeteroscedasticity was modelled with the varIdent variance function (Pinheiro et al. 2009), so that different variances were allowed for females and males (more variance in females).

  3. cRandom families nested within random populations.

C. exanthemataMidfemur lengthaIntercept−2·293·00365−0·7640·45
Season0·1580·08651191·820·071
Season2−0·001550·000820119−1·890·062
Season34·98 × 10−62·5 × 10−61191·9560·053
Sex (male)0·03670·009273653·960·0001
Pupal massa,bIntercept−356185379−1·920·056
Season12·85·341202·400·018
Season2−0·1260·0505120−2·480·014
Season30·0004030·000161202·570·012
Sex (male)−7·370·629379−11·7<0·0001
Development timeaIntercept13·80·68036820·2<0·0001
Season0·05600·006621188·47<0·0001
Sex (male)−0·6130·110368−5·55<0·0001
Growth rateaIntercept6·61 × 10−54·06 × 10−636816·3<0·0001
Season−3·65 × 10−73·95 × 10−8118−9·24<0·0001
Sex (male)1·84 × 10−67·35 × 10−73682·500·013
C. pusariaMidfemur lengthaIntercept−10022·8313−4·38<0·0001
Season3·940·855984·61<0·0001
Season2−0·05600·011998−4·70<0·0001
Season30·0003517·3 × 10−5984·80<0·0001
Season4−8·17 × 10−71·67 × 10−798−4·88<0·0001
Pupal massa,bIntercept−47961592335−3·010·0028
Season18859·6983·160·0021
Season2−2·710·83098−3·270·0015
Season30·01720·0051983·370·0011
Season4−4·05 × 10−51·17 × 10−5983·480·0008
Sex (male)−10·30·656335−15·7<0·0001
Development timecIntercept14·61·3133511·1<0·0001
Season0·05700·0121114·710·0006
Sex (male)−0·7650·136335−5·61<0·0001
Growth ratecIntercept5·17 × 10−57·45 × 10−63366·94<0·0001
Season−2·60 × 10−76·87 × 10−811−3·780·0031
C. clathrataMidfemur lengthaIntercept2·570·00846287304<0·0001
Sex (male)0·1440·0085228716·9<0·0001
Pupal massaIntercept45·90·384294120<0·0001
Sex (male)−0·8110·398294−2·030·043
Development timecIntercept0·05635·112910·01100·99
Season0·1610·049373·270·014
Sex (male)−0·3290·121291−2·730·0067
Growth ratecIntercept2·08 × 10−44·80 × 10−52914·33<0·0001
Season−1·62 × 10−64·63 × 10−77−3·510·0099
Sex (male)3·27 × 10−61·38 × 10−62912·370·019
L. marginataMidfemur lengtha,bIntercept2·270·073414030·9<0·0001
Season0·002140·000691913·100·0026
Sex (male)0·2150·017514012·3<0·0001
Pupal massaIntercept42·10·336230125<0·0001
Sex (male)−1·620·376230−4·31<0·0001
Development timecIntercept15·11·3322911·4<0·0001
Season0·02670·0126132·120·054
Sex (male)−0·8080·139229−5·82<0·0001
Growth rateaIntercept5·33 × 10−55·61 × 10−62299·50<0·0001
Season−2·60 × 10−75·19 × 10−895−5·02<0·0001
Sex (male)5·65 × 10−61·05 × 10−62295·39<0·0001

In many cases, Akaike Information Criterion (AIC) suggested that adding a variance function to model heteroscedasticity would significantly increase model goodness-of-fit. The inclusion of a variance function into the models tended to result in problems in estimating the within-group standard deviation. Thus, the results are presented for a model without any variance functions, whenever the problem arose. This is justified because both visual evaluation of the residual plots indicated that heteroscedasticity would not be a serious problem and the parameter estimates were insensitive to the inclusion of a variance function. Details for the cases where a variance function was included in the model are presented in Table 1. Excluding the cases where AIC supported the inclusion of a variance function, the definitive model was always the one with the lowest AIC value.

Finally, broad-sense genetic correlations between pupal mass, larval development time and larval growth rate were estimated. Bayesian approach in the animal model framework (Kruuk 2004; Wilson et al. 2010) was used by using Markov chain Monte Carlo methods as implemented in the R function MCMCglmm (Hadfield 2010). A trivariate animal model was fitted to the data in each species. Pupal mass, larval development time and larval growth rate were set as response variables. Sex-specific trait means were set as fixed effects [i.e. genetic variances and covariances were estimated after conditioning for sex (Wilson 2008)] and individual breeding value and population as random effects. The whole data (all the populations along the cline) were used in this analysis in each species, so it was important to include population in random effects to control for genetic differentiation among populations from different regions [See Appendix S1 (Supporting information) for full details].

Methodological aspects

In the beginning of the experiment, when the larvae from the southernmost populations hatched, C. clathrata was provided with Trifolium pratense L. as a host. Neonate larvae appeared unable to feed on this species, and thus, we changed T. pratense to L. pratensis immediately. Moreover, our technique to keep the herbaceous host fresh improved during the course of the experiment. To avoid any biases because of the aforementioned mentioned reasons, the two southernmost populations were omitted from all analyses concerning C. clathrata.

We stress that potential seasonal variation in the quality of tree leaves (see Schroeder 1986; Ayres & MacLean 1987) does not bias our conclusions concerning species feeding on trees (S. phylicifolia and A. incana), although all larvae could not be reared simultaneously for practical reasons. First, young leaves of the host trees were available for almost the whole summer on growing shoots. Second, many of the populations from regions where a nondiapause generation may emerge were sampled twice to generate temporal overlap in development among larvae from different populations. Within the populations sampled twice, late larval hatching date had no effect on the measured life-history traits in C. exanthemata and C. pusaria but resulted in decreased development time and increased growth rate in L. marginata (Appendix S2, Supporting information). However, the inferred clinal variation in larval development time and growth rate in L. marginata seems quite robust (Appendix S2, Supporting information), but some caution is needed in evaluating the genetic correlations in this species.

The possibility of nongenetic maternal effects on life-history traits (Mousseau & Dingle 1991; Mousseau & Fox 1998) was not explicitly controlled for in this study (see Conover & Schultz 1995; Bernardo 1996), but the results are likely to reflect the genetic basis of variation in the analysed traits. In C. exanthemata, no confounding maternal effects were found in an additional experiment (Appendix S3, Supporting information). In the other species, we expect the potential maternal effects on offspring phenotype to generate normally distributed noise around the genetic effects, not to systematically bias the phenotypic mean from the mean of genetic effects. Therefore, establishment of numerous families from each population should reveal the average genetic component of the variation along the cline. Secondly, the potential maternal effects were implicitly taken into account in the statistical analyses of clinal variation, as the random effects for families include both genetic and nongenetic variation among families. As the distribution of random effects is normal with a mean of zero (Pinheiro & Bates 2000), the random effects for families would approximate the individual breeding values halved (i.e. additive genetic effects) in the absence of maternal effects and at Hardy–Weinberg equilibrium (Falconer & Mackay 1996). We expect a strong genetic component in the phenotypic variation at the family level because additive genetic variation in life-history traits is ubiquitous (e.g. Roff 1992), especially in geographically widespread species (see Kellermann et al. 2009).

Results

Body size

Midfemur length and pupal mass were strongly and positively correlated in females and males in each species (Table S2, Supporting information). This indicates that, within a sex, both of these measures may be used as surrogates of adult body size with reasonable accuracy.

Measurement error was negligible in midfemur length (range: 0·043–0·73%). Midfemur length varied in relation to season length in each species except in C. clathrata (Table 1; Fig. 2). Constant midfemur length across latitudes in C. clathrata fits the expectation under perfectly compensating countergradient variation in growth rate (cf. Fig. 1b). Midfemur length variation followed the traditional saw-tooth cline (cf. Fig. 1a) both in C. exanthemata and in C. pusaria (Fig. 2), which is indicated by a significant positive cubic term for season length (Table 1). In C. pusaria, there was also a significant negative fourth power term for season length as midfemur length decreased with increasing season length in the three southernmost populations (Table 1; Fig. 2). In L. marginata, midfemur length increased linearly with increasing season length (positive linear term for season length; Table 1; Fig. 2), indicating a converse Bergmann cline across phenologies, and no support for the predictions that are presented in Fig. 1. Males had longer midfemora than females in each species except in C. pusaria, in which there was no difference between the sexes (Table 1; Fig. 2).

Figure 2.

 Midfemur length (top row), pupal mass (2nd row), larval development time (3rd row) and larval growth rate (bottom row) in relation to season length in C. exanthemata (1st column), C. pusaria (2nd column), C. clathrata (3rd column) and L. marginata (4th column). Dots indicate population means, and whiskers, their 95% CIs. Females are marked with filled circles and males with open circles. The regression curves [when the sexes differ, continuous lines indicate females and dashed lines males; see Table S4 (Supporting information) for regression equations] are shown for the cases where clinal variation was found. The populations left from the vertical dotted line are univoltine, whereas a nondiapause generation may emerge in populations on the right side of that line.

Pupal mass varied in relation to season length in C. exanthemata and C. pusaria (Table 1; Fig. 2), again as predicted by the traditional saw-tooth model (cf. Fig. 1a; see the significant and positive cubic term for season length in Table 1). Also this time, there was a significant negative fourth power term for season length in C. pusaria, as pupal mass decreased with increasing season length in the three southernmost populations (Table 1; Fig. 2). Pupal mass did not vary in relation to season length in C. clathrata and L. marginata (Table 1; Fig. 2), which is expected under perfectly compensating countergradient variation in growth rate (cf. Fig. 1b). Female pupae were, on average, heavier than male pupae in each species (Table 1; Fig. 2), although the difference was small in both C. clathrata and L. marginata (Table 1).

Larval development time

Larval development time linearly increased with increasing season length in each species as there was a positive linear term for season length (Table 1; Fig. 2), which was significant in each species except L. marginata (Table 1). In L. marginata, the linear term was marginally insignificant (Table 1), but when an exponential variance function [function varExp (Pinheiro et al. 2009)] was included in the model that was slightly heteroscedastic, AIC value decreased 7·8 units and the linear term for season length became significant (estimate = 0·0293, SE = 0·0111, t13 = 2·64, P = 0·020). However, the inclusion of variance function resulted in inaccuracy in the estimation of within-group standard deviation, but as the two slope estimates differ only by 9%, it is safe to conclude that larval development time increased linearly with increasing season length also in L. marginata. Hence, shift in phenology did not influence larval development time (no significant cubic or higher-order terms for season length) in any of the species, which contradicts the prediction (see Fig. 1). Males had a shorter larval period than females in each species (Table 1; Fig. 2).

Larval growth rate

Growth rate linearly decreased with increasing season length as there was a significant negative linear term for season length in each species (Table 1; Fig. 2). Contrary to the prediction, the shift in phenology was not reflected in growth rate variation (no significant cubic or higher-order terms for season length). Males had a higher growth rate than females in all species except C. pusaria, in which growth rate did not differ between the sexes (Table 1; Fig. 2).

The results concerning clinal variation in growth rate are robust for the value of b used in this study as the results remained qualitatively unchanged when = 0·67 was used instead of = 0·8. However, the difference between the sexes disappeared in C. exanthemata, C. clathrata and L. marginata when = 0·67 was used instead of = 0·8.

Genetic correlations

There was a strong negative broad-sense genetic correlation between larval development time and larval growth rate in each study species (Table 2), which is expected under countergradient variation in growth rate (see Fig. 1b,c). This correlation is not just a consequence of the opposite clinal variation in the two traits, because the same correlation was significant at the phenotypic level within different subsets of the cline in each species (Table S3, Supporting information). In addition, there was a positive broad-sense genetic correlation between pupal mass and larval development time in C. exanthemata (Table 2) as expected under the traditional saw-tooth model (see Fig. 1a). In L. marginata, there was a positive broad-sense genetic correlation between pupal mass and larval growth rate (Table 2), which is expected under overcompensating countergradient variation in growth rate (see Fig. 1c). All other correlations were insignificant in the sense that the interval including the correlation with 95% probability (Bayesian inference) included zero (Table 2). In C. pusaria and C. clathrata, the signs of the three estimated genetic correlations matched the expectation under perfectly compensating countergradient variation in growth rate (Fig. 1b). In C. exanthemata and L. marginata, the estimated correlation structures deviated from all predictions that are presented in Fig. 1.

Table 2.   Estimated broad-sense genetic correlations between pupal mass, larval development time and larval growth rate in the four geometrid moths studied. Correlation coefficients that are different from zero with at least 95% probability (Bayesian inference) are in boldface. The 95% highest posterior density intervals are in parentheses. See Fig. 1 for the predicted signs of the genetic correlations
SpeciesPupal mass vs. larval timePupal mass vs. growth rateLarval time vs. growth rate
C. exanthemata0·530 (0·111, 0·665)−0·348 (−0·559, 0·101)0·912 (−0·953, −0·850)
C. pusaria0·131 (−0·336, 0·436)0·127 (−0·317, 0·429)0·877 (−0·937, −0·765)
C. clathrata−0·148 (−0·373, 0·339)0·211 (−0·0305, 0·561)0·903 (−0·940, −0·792)
L. marginata−0·384 (−0·704, 0·179)0·586 (0·0796, 0·826)0·895 (−0·950, −0·740)

Discussion

Clinal variation in body size and assessment of the hypotheses

We found adult body size to vary in a saw-tooth pattern across a latitudinal gradient of season length in C. exanthemata and C. pusaria in the common laboratory environment (Table 3). These saw-tooth clines fit the traditional saw-tooth model (Masaki 1967, 1972; Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994) as the transition between univoltine and bivoltine phenologies was reflected in body size as predicted. Body size of C. pusaria clearly decreased towards increasing season length also at the southern end of the studied gradient, which would imply a transition to trivoltine phenology. This is, however, unlikely as trivoltinism has not been recorded in this species in southern Finland (Hyönteistietokanta 2009; own observations). A more plausible, yet unexamined, explanation is that bivoltine phenology changes from a facultative to obligatory strategy in southernmost Finland, the major change in body size occurring at this point and not at the point where a partially bivoltine phenology becomes possible. In C. clathrata and L. marginata, a transition between the phenologies was not reflected in adult body size. There was no latitudinal variation in adult body size in C. clathrata, but a converse Bergmann cline (i.e. body size increased with increasing season length) continued across the entire seasonal gradient in midfemur length in L. marginata (Table 3).

Table 3.   Summary of the empirical results concerning clinal variation in the measured traits in the four geometrid moths studied
TraitObserved clinal variation
C. exanthemataC. pusariaC. clathrataL. marginata
  1. aThe variable of interest is positively correlated with season length.

  2. bThe variable of interest is negatively correlated with season length.

Midfemur lengthSaw-toothSaw-toothNo clineConverse Bergmanna
Pupal massSaw-toothSaw-toothNo clineNo cline
Development timeCogradientaCogradientaCogradientaCogradienta
Growth rateCountergradientbCountergradientbCountergradientbCountergradientb

Clinal variation in body size was estimated qualitatively consistently with both midfemur length and pupal mass in each species except L. marginata (Table 3), where midfemur length followed a converse Bergmann cline but pupal mass did not show clinal variation. This inconsistency between the different surrogates of body size implies that body shape of L. marginata does not remain constant across the gradient of changing season length, even though pupal mass and midfemur length are strongly and positively correlated. A slight variation in relative abdomen mass may explain the deviation between the patterns of body size variation when body size was measured either as a pupal mass or as a linear dimension of the body. In capital breeders, female fecundity is directly proportional to body mass (Honěk 1993), which may explain why females were heavier than males in each species, although midfemur length showed the opposite sexual size dimorphism in three species and no dimorphism in C. pusaria.

The observed saw-tooth clines in body size in C. exanthemata and C. pusaria fit qualitatively into the traditional saw-tooth model (Masaki 1967, 1972; Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994), but both species violate the model as the development time clines do not show the expected saw-tooth pattern and the genetic correlation structures do not fit the prediction. There was a negative genetic correlation between development time and growth rate in each species (Table 3), which indicates countergradient variation in growth rate (see Fig. 1b,c). The observed combination of genetic correlations in C. exanthemata does not fit any of the predictions presented in Fig. 1, but it loosely fits the expectation under undercompensating countergradient variation in growth rate. In C. pusaria and C. clathrata, the genetic correlations fit the expectation under perfectly compensating countergradient variation in growth rate (Fig. 1b). In L. marginata, the observed genetic correlation structure deviated from the predictions that are presented in Fig. 1, yet it loosely fits the one expected under overcompensating countergradient variation in growth rate (Fig. 1c).

The hypotheses presented in Fig. 1 implicitly assumed that the genetic correlation structure among body size, development time and growth rate is constant across the entire cline. However, the phenotypic correlations (Table S3, Supporting information) suggest that the correlation structure may vary along the cline.

New directions for the study of clinal variation

The predicted (Masaki 1967, 1972; Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994; Fig. 1) saw-tooth clines in development times or growth rates were not found, indicating that the shift in phenology did not clearly reflect to these traits. Interestingly, the clines in development time and growth rate were consistent across species (Table 3). Development time increased and growth rate decreased with increasing season length in each species. The consistent clinal variation in development time and growth rate and the strong genetic correlation between these traits suggest that there is a common mechanism underlying the observations.

A potential explanation for why the shift in phenology did not clearly reflect in the development time and growth rate clines is that the life histories of individuals entering the alternative developmental pathways are different. Individuals entering the diapause pathway, which was studied here, may only experience the time constraints set by season length and not at all the time constraints because of a shift in phenology that takes place at the population level. Although the time available per generation is halved at the transition from univoltine to bivoltine phenologies, this change in time constraints may become evident only in the life history associated with the alternative developmental pathway, that is, in individuals that develop directly into adults within the same season. Accordingly, directly developing individuals have shorter development times and higher growth rates than individuals entering diapause (Wiklund, Nylin & Forsberg 1991). In many insects, including our study species, change in phenology is not abrupt as predicted in theory (Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994), but gradual with a transition region where a partial nondiapause generation emerges (e.g. Masaki 1972; Mousseau & Roff 1989; Spence 1989; Blanckenhorn & Fairbairn 1995; Välimäki et al. 2008). In the transition region, offspring of the diapause generation individuals enter both developmental pathways, because diapause induction is phenotypically plastic (e.g. Masaki 1972; Tauber, Tauber & Masaki 1986; Burke et al. 2005).

The species-level consistency in development time and growth rate clines, and the inconsistency in body size clines, may be explained by the mechanistic determination of body size and development time. In Manduca sexta (Lepidoptera: Sphingidae), both body size and development time are determined by the growth rate, the critical weight after which metamorphosis is possible without further nutrition and the time interval between the attainment of the critical weight and the cessation of growth (Nijhout, Roff & Davidowitz 2010). The phenotypic landscapes for both body size and development time are largely almost orthogonal, and thus, there is no strong correlation between these traits (Nijhout, Roff & Davidowitz 2010). Because the predicted evolutionary trajectories for body size and development time are almost orthogonal, there will be a conflict when both traits are simultaneously under directional selection (Nijhout, Roff & Davidowitz 2010). We assume the aforementioned mechanism to be applicable to all lepidopterans. Then, the observed consistency of development time and growth rate clines suggests that the evolutionary trajectories of these traits have been parallel in each study species. The ultimate reason for this would be that development time is the trait under the strongest selection in species with relatively long generation times. The intensity of selection for short development time increases when season length decreases and the seasonal time constraints become more intense, resulting in cogradient variation (Conover & Schultz 1995). Selection for short development time is likely to generate a correlated selection towards increasing growth rate, as suggested both by the observed strong negative genetic correlation between these traits (this study) and by the observed response to selection in M. sexta (Nijhout, Roff & Davidowitz 2010). Thus, the emergence of countergradient variation in growth rate would be only a proximate consequence of selection on development time. Simultaneous fecundity selection for large body size would conflict with the selection for short development time, so the response of body size depends on the initial location of the population within the phenotypic landscape as well as the amount of additive genetic variation in each trait. Therefore, the resulting body size clines may vary among species, which is the case in Lepidoptera (Blanckenhorn & Demont 2004; data from Nylin & Svärd 1991; Burke et al. 2005; this study).

Conclusions

The latitudinal body size clines were diverse and, even when saw-tooth clines were found, the theory (Roff 1980, 1983; Iwasa, Ezoe & Yamauchi 1994) was not supported because the patterns were apparently produced by different mechanisms. In particular, development times consistently decreased northwards in each species, resulting in cogradient variation between development time and season length. Probably as a correlated response to selection on development time, there was consistent countergradient variation in growth rate across species. The results suggest that development times of individuals entering diapause are mainly affected by season length, not voltinism, and that there is a common mechanism underlying the development time and growth rate clines. The complex developmental and physiological interdependencies between body size, development time and growth rate (Nijhout, Roff & Davidowitz 2010) offer a potential explanation. These three key life-history traits need to be evaluated simultaneously to make realistic inferences on the evolution of body size.

Acknowledgements

We thank H. Pöykkö for help both in the field and in the laboratory. We thank him, M. Angilletta, J. Aspi, W. Blanckenhorn, M. Blows, J. Forsman, H. Kokko, T. Tammaru, J. Tuomi and an anonymous reviewer for helpful comments on an earlier draft of the manuscript. We are also grateful to H. Huiskonen for help in rearing the larvae and M. Mutanen for providing us with some moths. The study was financed by the Ella and Georg Ehrnrooth foundation (grants to S.M.K and P.V.), Societas pro Fauna et Flora Fennica (S.M.K.), Societas Biologica Fennica Vanamo (S.M.K.), the Jenny and Antti Wihuri foundation (S.M.K.), the Finnish Cultural Foundation (P.V.) and the Oskar Öflund foundation (P.V.). All Finland’s guidelines and legal requirements for the use of animals in research were followed.

Ancillary

Advertisement