Introduction

Distributions of many species are currently shifting, triggered by climate change (Chen et al. 2011) and human-assisted introductions of non-native species (García-de-Lomas and Vilà 2015). Shifts in the distribution of species often go along with differences in phenotypic traits (reviewed by Chuang and Peterson 2016), whereby individuals in newly colonized areas typically show, for example, greater dispersal ability, reproductive investment, or levels of foraging activity, which lead to a faster pace of life, compared with their counterparts in long-established populations (e.g., Hill et al. 2011; Pintor et al. 2008; Therry et al. 2014a, b, c, 2015). Two main non-exclusive mechanisms can explain this phenotypic change. First, dispersers are a non-random subset of individuals from the source populations that are characterized by a set of traits that favor a better colonization capacity, which includes, for example, high risk-taking behavior and metabolic rate (Cote et al. 2010; Carere and Gherardi 2013). Second, range limits are shaped by a complex interplay of abiotic and biotic factors (reviewed in Gaston 2009), with conditions at species’ range edges being often more stressful than those in core populations (Hardie and Hutchings 2010), which may shape organism phenotype through plasticity or selection. For instance, during northward range expansions, organisms face strong shifts in climatic conditions. Ectotherms at higher latitudes are often darker in color (i.e., a higher level of melanism), which is advantageous for thermoregulation at lower temperatures (Clusella Trullas et al. 2007).

While a growing number of studies documented changes in morphological, physiological, and behavioral traits at the range expansion front (Zeuss et al. 2014; Chuang and Peterson 2016; and references therein), none of these studies, according to our knowledge, focused on changes in covariations among phenotypic traits (i.e., phenotypic integration; Kim and Velando 2015). Determining changes in phenotypic integration during range expansion is required for understanding the evolutionary constraints and the proximate basis of changes in single traits during range expansion. The impact of a phenotypic trait on fitness indeed often depends on the value of another trait, resulting in correlational selection, which can lead to adaptive trait covariation (Roff and Fairbairn 2012). Therefore, the covariation among traits will vary with local environmental conditions that shape the selection pressures on correlated traits (Endler 1995; Bell and Sih 2007; Dingemanse et al. 2007; Raffard et al. 2019). Since environmental conditions, and therefore, selection pressures typically differ between core and expanding edge populations (Hardie and Hutchings 2010; Phillips et al. 2010), trait covariations are likely to vary between core and marginal populations.

In particular, ectotherms have evolved morphological and behavioral traits that allow them to cope with different conditions at higher latitudes, such as climatic conditions or season length. For example, species and individuals are often darker in color and larger in colder environments compared to warmer environments (Zeuss et al. 2014) or have faster growth rates at higher latitudes due to time constraints (Therry et al. 2014a). An investment in body size or a fast growth rate generates higher energetic needs and forces individuals to have higher foraging activity, increasing predation risk (growth–predation trade-off; Werner and Anholt 1993; Stamps 2007). Therefore, predation can lead to correlational selection between body size, exploratory or activity behaviors, and anti-predator defense traits (Dingemanse et al. 2007; Raffard et al. 2019). For example, risk-taking juvenile three-spined sticklebacks (Gasterosteus aculeatus) are better camouflaged through darker body coloration compared to risk-adverse juveniles (Kim and Velando 2015), and this covariation may be the result of selection pressures induced by predators in the environment (Dingemanse et al. 2007). Trait covariations can also result from pleiotropic effects, linkage disequilibrium between genes, and similar plastic responses to conditions among traits. For example, body darkness in invertebrates often correlates with behavioral traits and immune function, whereby darker individuals are more active, explorative, and bold or possess a higher immune defense (Verhoog et al. 1998; Armitage and Siva-Jothy 2005; Mafli et al. 2011). Cuticular melanization indeed occurs via the phenoloxidase cascade, which plays a key role in immune responses (González-Santoyo and Córdoba-Aguilar 2012), and dopamine is a neurotransmitter produced as an intermediate of the melanization pathway that influences activity and explorative behavior (van Oers and Mueller 2010, Hodgetts and O’Keefe 2006).

In this study, we investigated the phenotypic variation and covariation across southern and more recently colonized areas in northern Europe of a range-expanding dragonfly (Crocothemis erythraea). We measured multiple phenotypic traits in larval and adult stages (activity, body size, and coloration), to fulfill three specific objectives. First, we tested differences in the mean value, variation among individuals, and variation within individuals (i.e., repeatability and plasticity between life stages and thermal conditions) of phenotype between southern (i.e., core) and more northern (i.e., edge) populations. We expect individuals from populations in northern Europe to exhibit higher activity, darker coloration, and larger body size than individuals from southern populations due to higher dispersal and thermoregulation capacity. Furthermore, we expect higher temperature-induced behavioral plasticity in younger populations at higher latitudes, since phenotypic plasticity has been associated with the invasive potential of populations and evolves upwards during range expansion (Aubret and Shine 2009; Chevin et al. 2010; Valiente et al. 2010). Second, we tested whether phenotypic covariations differ between southern and northern populations. Notably, activity, body size, and coloration should be correlated if correlations lead to higher adaptive values. For example, active individuals may display darker colorations to be better camouflaged for predators while foraging. Also, we expect that the enhanced growth–predation trade-off in the northern populations translates into a stronger covariation among traits associated with food acquirement (activity) and anti-predator defense (coloration). Finally, we tested a potential carry-over of phenotypic traits and phenotypic correlations from the larval to the adult stage to predict whether potentially observed differences among populations would translate into fitness differences in adults.

Materials and methods

To achieve our objectives, we collected larvae from four South-European and five north-European populations, and we scored larval behavioral traits (activity) at cold (19 °C) and warm (25 °C) temperatures. It allowed us to test the differentiation of both behavioral types and behavioral reaction norms in response to a major environmental condition that varies along the expansion gradient (we expected colder temperatures in northern populations). After behavioral tests, we quantified larval darkness and size. Then, the larvae were reared in a common garden until adulthood to quantify similar traits (adult darkness, adult size, and relative thorax length) at the reproductive life stage. We focused on the risky behaviors associated with food acquirement (activity) and anti-predator defense (camouflage through higher body darkness, True 2003), since it has been shown that patterns of phenotypic integration depend on predation regimes (Dingemanse et al. 2007).

Study system, collection, and housing

Crocothemis erythraea (Brullé 1832) is a dragonfly species with a predominantly African distribution, with its European native breeding range confined to the Mediterranean (Dijkstra 2006). In the 1960s, a climate-driven northward range expansion through Europe started from southern populations (Ott 2007). The southern populations belong to the central European distribution area colonized in the 80s. In the 1980s, the species had colonized the entire France, although with a patchy distribution and only low-density populations in regions north of the Loire River (Dommanget 1987). The species further colonized northern Europe in the next decades (Ott 2007), with the northernmost populations currently located in Northern Germany (Brockaus et al. 2015).

The species breeds in a wide range of stagnant water habitats and seeks out warmer microclimates in the north, predominantly shallow waters with dense aquatic vegetation (Dijkstra 2006). We studied four southern populations from Southern France (S1: Bram, S2: Camargue, S3: Albi, and S4: Brenne) and five northern populations from North-West Europe (N1: Mainz, Germany; N2: Ghent, Belgium; N3: Soerendonk, the Netherlands; N4: Dortmund, Germany; and N5: Braunschweig, Germany). Populations S4, N1, N2, and N4 consist of two subpopulations located < 25 km from each other [Electronic Supplementary Material (ESM) A; Table S1 and Fig. S1], and were pooled in the analyses. The first reproductions in the region of Mainz and Ghent were observed in the early 1990s (Deknijf 1995; Ott 2007), near Dortmund in 1995 (Bauhus 1996), and up to latitudes of Braunschweig in 2000 (Lohr 2003).

Larvae of C. erythraea in the penultimate and last larval instars were collected in the field using an aquatic kick net between 12 May and 22 June 2015 and transported to the laboratory (see ESM A for sample dates per population). Larvae were housed individually in 300 ml opaque vials filled with dechlorinated tap water and daily fed with ad libitum chironomid larvae. The vials were kept in an incubator at 22 °C and a 15.5:8.5 L:D photoperiod. Larvae were housed on average for 18.7 ± 8.3 (SD) days in the incubators under controlled laboratory conditions before the start of the behavioral trials.

Experimental setup

Behaviors of 220 larvae (southern: 104; northern: 116) in the last larval instar were tested at both 19 °C (low-temperature treatment) and 25 °C (high-temperature treatment). Tests were performed at 24-h intervals between the tested temperatures (see ESM A for sample size per population). Half of the larvae per population were first tested at low temperature, while the other half was first analyzed at high temperature (testing order). To standardize conditions among individuals, larvae were deprived of food and acclimatized to testing temperatures for 20 h before the start of each behavioral assay. At the initiation of the test, each larva was introduced in the center of a white plastic tray (30 cm × 20 cm) filled with 2 cm dechlorinated tap water. The activity of each larva in the experimental arena was recorded for 200 min, with webcams connected to a computer with the iSpy® software (version 6.3). Twenty larvae were tested at the same time using 20 arenas (arena ID), and larvae of different range locations (southern vs. northern) were simultaneously tested, and their positions among the 20 arenas were randomized. We extracted the total distance moved (activity) from the video recordings using the Ethovision ® software (version XT8). Due to the poor quality of some videos, behavioral scores of 76 larvae (southern: 35; northern: 41—cold treatment: 33; warm treatment: 43) were obtained in only one of the two temperatures analyzed (i.e., 17% of trials).

After the second behavioral trial, larvae were photographed (dorsal side) in the laboratory using a Canon Powershot G16 ®. To perform this, larvae were gently blotted dry with absorbent tissue and placed on a white background with standardized light conditions (i.e., in the same place and with the same light intensity). Larval head width and body darkness were quantified using the ImageJ ® (vs. 1.51) software. Head width was used as a proxy for larval size (Benke 1970), and body darkness was used as a proxy for cuticle melanization (Fedorka et al. 2013). We measured the darkness of the dorsal areas of thorax and abdomen by quantifying (using ImageJ) the level of grey value presents in the selected areas. The ImageJ software produces a mean grey score for selected pixels (from 0 = totally black to 255 = totally white). We removed the grey value from a control blank to correct for a potential session effect (i.e., the grey value of control blank minus the grey value of the sample). The control value was the average of the mean grey values of three pictures of a white piece of paper (for each session) placed behind the larvae. Higher levels of grey value correspond to higher darkness (from 92.2 to 156.2, mean = 125.0, variance = 140.5). Then, larvae were placed back in their original breeding vials and reared until the adults emerged. Forty-three percent of the larvae survived until the adult stage (N = 96; see ESM A for sample size per population). When the adult exoskeleton had hardened, 24 h after emerging at 22 °C, a picture of the dorsal side of the adult was taken using a Canon Lide 210 scanner ® and a picture of the lateral side of the adult was taken using a Canon Powershot G16 ®. The darkness of the dorsal area of thorax and abdomen was measured using the dorsal picture with the method described for larval darkness quantification. The length of the thorax (from the junction of pronotum and thorax to the caudal attachment of the hind wing with the thorax) and the length of the abdomen (from the caudal attachment of the hind wing with the thorax to the caudal end of the abdomen) were measured on the lateral picture. The total length was calculated as the sum of thorax and abdomen length, and the relative thorax length (RTL) was calculated as the ratio between the thorax length and the total length. The relative thorax length is correlated with flight muscle mass in odonates (Therry et al. 2015), and greater investment in flight muscle mass has been associated with greater flight performance in insects (Schilder and Marden 2004; Therry et al. 2014c).

Statistical analyses

Phenotypic differentiation between southern and northern individuals

We analyzed the effect of range location (southern vs. northern, assuming that northern areas correspond to edge habitats, whereas southern areas could be seen as core habitats) on larval activity, larval darkness, larval size, adult darkness, adult size, and relative thorax length using linear mixed models. We ran these analyses with range location instead of latitudes for the sake of comparison with the covariation analyses (i.e., structural equation modeling, SEM), in which we aimed at comparing covariations among traits between core and edge populations. Hence, SEM can only be performed with a categorical co-variable (i.e., southern vs. northern) and not with a continuous co-variable (i.e., latitude). Nonetheless, we also ran supplementary analyses (i.e., linear mixed models) with latitude as a continuous variable, and the results were similar (ESM B). Overall, we ran six models to test whether the environment (range location and/or temperature) can affect phenotypic traits. For larval darkness, the mean grey value corrected for the control blank was included as a dependent variable, range location, larval size, and temperature of the second behavioral trial (i.e., to control for an effect of temperature on coloration since pictures were taken after the second behavioral trial at either 19 °C or 25 °C; Garcia et al. 2003; Fedorka et al. 2013) as fixed factors, and the population ID as random intercept. For larval activity, the dependent variable was the activity level of individuals for both temperatures (i.e., two values per individual) and full models included range location, larval size, testing temperature (19 °C vs. 25 °C), and its interaction with range location as fixed effects to test the variation of behavioral plasticity with range location. Larval ID, population ID, arena ID, and testing order (first vs. second trial) were added as random effects. Larval size variation was tested with range location as a fixed effect and population as a random effect. Models that tested variation in adult darkness and relative thorax length included range location, adult size, sex, and its interaction with range location as fixed effects, and populations as a random effect. Finally, for adult size, the model included range location, sex, and its interaction with range location as fixed effects and populations as a random effect. The interaction between range location and sex was included, since in dragonflies females and males can differ in their life history traits and responses to environmental conditions, including temperature and parasitism (De Block and Stoks 2003), and therefore, in their responses to range expansion (Hughes et al. 2007). Larval or adults sizes were added as a covariate when analyzing larval and adult traits, since body size often influences morphological, physiological, and behavioral traits (Bonner 2006).

The repeatability (i.e., variance explained by inter-individual differences) of behaviors across testing temperatures was assessed using linear mixed models with larval ID as a random intercept. Intraclass correlation coefficients (ICC = Varianceinter/Variancetotal) were calculated, and likelihood ratio tests were used to assess the significance of larval ID. Repeatability was scored in the overall dataset and in the separate subsets of southern and northern larvae, to verify the consistency of repeatability across range locations. Standard errors in trait values for each population were quantified as measures of within-population variability, and Mann–Whitney U tests were used to evaluate differences in variability between the southern and northern populations. Activity (scored as total distance moved) was log-transformed, and all continuous variables were standardized (mean: 0, SD: 1) before the start of the statistical analyses. These analyses were performed using SAS© v.9.4.

Phenotypic covariation in southern and northern individuals

Structural equation modeling (SEM) was used to test: (1) the presence of phenotypic covariation among larval traits, (2) the presence of phenotypic covariation among adult traits, (3) the presence of carry-over effects from larval to adult traits, and (4) the differences in these covariations and carry-over effects between southern and northern populations.

Analysis of path models was carried out using the software package lavaan.survey in R (Oberski 2014; R Core Team 2013) with parameters estimated by robust maximum likelihood, and population included as a hierarchical effect. Due to mortality in the larval stage, sample sizes for adult traits were smaller than for larval traits. Traits were scaled to the mean within each range location before running the models. We ran a model that included all covariations (i.e., bidirectional paths) among the three larval traits, but no paths from each larval trait to each adult trait. The default model was the model with equal covariances between the southern and northern populations for all paths. First, we compared this model with a model with all unconstrained covariances to investigate whether matrices of covariances were, on average, different between southern and northern populations. Second, we compared the default model to each alternative model with one covariance unconstrained to be equal between southern and northern populations. Alternative models for which ΔAIC > 2 (where ΔAIC = AICdefault model – AICalternative model) have substantial support compared to the default model. The goodness of fit of the alternative and default models was compared with likelihood ratio tests. We obtained a final model by selecting unconstrained covariances that improved the AIC compared to the default model. The adequate fit of the final model was tested using χ² tests, and estimates were derived from this model. The same approach was used to test covariation among adult traits (size, darkness, and relative thorax length). Third, we added unidirectional paths from each larval morphological trait to each corresponding adult morphological trait (larval size and darkness to adult size, relative thorax length, and darkness) to the final model, while covariances between larval and adult traits were fixed based on the final model. The steps for unidirectional paths from larval to adult traits were similar to those of larval covariations.

Results

Phenotypic differentiation between southern and northern populations

We found that northern larvae were darker than southern larvae, while larvae from southern and northern populations were similar in body size (Table 1a, Fig. 1). The activity was not different between southern and northern larvae (Table 1a, Fig. 1). However, when measuring the alternative measure of risk-taking behavior, we found that northern individuals were bolder than southern individuals (see ESM C for details on the alternative measure of risk-taking behavior). The activity level of individuals was repeatable across testing temperatures (ICC = 0.40, χ2 = 26.17, P < 0.001), even if larvae were more active at warmer testing temperature. However, both the repeatability and the temperature-depedent plasticity did not differ between southern and northern populations (repeatability: ICC = 0.39, χ2 = 12.54, P < 0.001 and ICC = 0.40, χ2 = 12.54, P < 0.001 for southern and northern populations, respectively, plasticity: Table 1a, Fig. 1, no significant interaction term). None of the traits quantified in adults (adult darkness, adult size, and relative thorax length) differed significantly between southern and northern larvae (Table 1b, Fig. 1). Overall, adult males were darker, larger, and had a greater relative thorax length than adult females (Table 1b, Fig. 1). The between-individual variances for the larval and adult traits did not differ significantly between southern and northern populations (ESM A).

Table 1 a Linear mixed models testing for the effects of range location (RL: southern vs. northern) and testing temperature (Temp: 19 °C vs. 25 °C) on activity; and testing for the effect of range location on darkness and size in last instar Crocothemis erythraea larvae. b Linear mixed models testing for the effects of range location and sex (female vs. male) on darkness, size and relative thorax length in C. erythraea adults
Full size table
Fig. 1
figure 1

Activity (a), body darkness (b), and size (c) of Crocothemis erythraea larvae from southern and northern populations. Activity is plotted for both testing temperatures. Darkness (d), size (e) and relative thorax length (f) of females and males of C. erythraea adults from southern and northern populations. Given are means ± 1 SE on scaled variables

Full size image

Covariation in larval traits

The comparison of path models revealed that covariations between darkness and size and between darkness and activity were different between southern and northern populations (Table 2, Fig. 2a and b). Specifically, larval darkness and activity were positively correlated in the North, while this covariation was not significantly different from zero in southern populations (Fig. 2, ESM D). Larval darkness covaried negatively with larval size in northern populations, while this covariation was not significant in southern populations (Fig. 2, ESM D). However, the difference in covariation only tended to be significant, even if it decreased the AIC of the model (Table 2). The final model included differences in covariations between southern and northern populations for larval darkness and activity, and for larval darkness and larval size (Fig. 3). This final model better fitted the data than the default model with all covariances equal among locations (Table 2), and did not differ from the observed covariance matrix (χ2 = 1.05,  df = 1, P = 0.31).

Table 2 Comparison of the model fit of alternative models (with a single covariance/regression difference between the southern and northern subset) and the default model (all covariances/regressions equal between the southern and northern subset)
Full size table
Fig. 2
figure 2

Distributions and relationships among larval traits from southern (a) and northern (b) populations, and among adult traits from southern (c) and northern (d) populations. Estimates are obtained using path analysis. Bold values represent significant covariances (α = 0.05). RTL relative thorax length

Full size image
Fig. 3
figure 3

Results from path analysis. Black lines are covariations and regressions that were significant (α = 0.05). Non-significant covariations (grey lines) are shown as indications. A double arrow represents a covariation and a single arrow represents a directional path. Symbols indicate the sign of the path coefficient. See ESM D for details on parameter estimates

Full size image

Covariations among adult traits

The covariations among adult traits were also different among range locations (Table 2, Fig. 2c, d). Adult size and darkness were positively correlated in northern populations, while this covariation was not significant in southern populations (Fig. 3, ESM D). On the contrary, adult darkness positively covaried with relative thorax length only in southern populations (Fig. 3, ESM D). The relationship between adult size and relative thorax length did not differ among range locations (Table 2), and was not significant in both southern and northern populations (ESM D). The final model included differences in covariations between southern and northern populations for adult size and darkness, and for adult darkness and relative thorax length (Fig. 3). This model better fitted the data than the default model and did not differ from the observed covariance matrix (χ2 = 0.002, df = 1, P = 0.97).

Carry-over from larval to adult traits

We identified two carry-over effects from larval to adult traits (Fig. 3). Notably, larger larvae emerged into larger adults with a greater relative thorax length (Fig. 3, ESM D). Nonetheless, the larval darkness was not carried over to adults (Fig. 3). The carry-over effects were not different between southern and northern populations (i.e., none of the directional paths from larval to adult traits differed; Table 2). The default model was not significantly different from the data (χ2 = 7.45, df = 18, P = 0.99).

Discussion

Our main results show that some phenotypic traits and phenotypic integration (i.e., patterns of trait covariations) were different between southern populations and northern populations of the northward expanding dragonfly C. erythraea. Furthermore, the observed carry-overs of morphological traits from larvae to adults support the phenotypic coupling across life stages previously observed in organisms with a complex life cycle (Stoks and Cόrdoba-Aguilar 2012).

Our first expectation was that individuals from populations at the expanding edge were phenotypically different from individuals from core populations. It could be due to phenotype-biased founding events or differences in environmental conditions, and the resulting selective pressures, across latitudes. More specifically, we were expecting individuals in northern populations to be more active, darker, larger, and more plastic than individuals in southern populations. Contrary to our expectations, we detected only a difference in larval darkness between southern and northern populations, while there were no differences in body size, activity, and temperature-dependent activity between northern and southern populations. Specifically, larvae from northern populations were, on average, darker than their southern counterparts, which may be explained by several non-exclusive mechanisms. Terrestrial ectotherms often show a gradient of increasing dark coloration with latitude, since darker bodies absorb more solar radiation, improving thermoregulation in colder climates (Clusella Trullas et al. 2007; Pinkert et al. 2016). However, caution should be exercised since darker coloration does not always improve thermoregulation in aquatic organisms as water has a much higher specific heat capacity than air (Garcia et al. 2003), and since larval differences in melanism were not carried over to adults. Alternatively, darker bodies may result from a covariation with unmeasured traits under spatially varying selection. For example, a higher investment in immune function has been observed at higher latitudes (De Block et al. 2008) and at the expansion front in odonates (Therry et al. 2014b), while melanism and immune functions are tightly linked through pleiotropic effects and shared responses to environmental conditions (Côte et al. 2018). Further investigations are needed to investigate these explanations, particularly by measuring the links between darkness, other phenotypic traits, and individual fitness with laboratory-raised individuals (i.e., F1 and F2) from different range locations and assessing the relative contribution of plasticity and selection in darkness differentiation.

Our results show that larval traits covaried in northern populations, but this did not occur in southern populations. In northern populations, both a positive (correlation between larval darkness and activity and a negative correlation between larval darkness and larval size were detected. Variances of trait values did not differ across range locations. Therefore, the lack of covariations among traits in southern populations was not driven by a smaller variability in phenotypic traits. Nonetheless, unmeasured traits, which can underpin trait covariations (i.e., a common causal parent), can vary differently across the range expansion gradient and, in turn, modify the covariations among traits. The adaptive value of covarying traits depends on the selective pressures acting on a population. For example, behavioral correlations between activity, exploration, and aggressiveness were present in three-spined stickleback populations that live sympatrically with predators but were absent in relaxed predator-free populations (Dingemanse et al. 2007). In our current study system, although we have no evidence that predator communities consistently differ between southern and northern habitats, enhanced predation risk is expected in northern populations given the shorter growth season and the associated faster life history in high latitude populations. In those populations with high predation risk, active animals are likely to gain a large fitness benefit by investing in anti-predator strategies. In contrast, less active individuals can benefit from an ‘energy-saving’ strategy, whereby investment in costly traits is avoided, explaining the positive covariation between activity and darkness in northern populations. Melanin is indeed costly to synthesize and may be included in a trade-off with other costly traits (Roff and Fairbairn 2013; Côte et al. 2018), which could explain the negative covariation between larval darkness and larval size in the northern larvae. Altogether, the observed covariations, between darkness and activity and between darkness and body size, may reflect different strategies, where some individuals invested their energy in growth rate to reach larger body size and reproduce faster, while others may invest energy in melanism to avoid predators.

Contrary to covariations of larval traits, ontogenetic covariation from larval to adult traits did not differ between southern and northern individuals. Odonata shows a striking habitat shift through metamorphosis, from an aquatic larval stage to a terrestrial adult stage, so it is unlikely that environmental stressors experienced in the larval stage predict stressors in the adult stage. The lack of common environmental stressors that cover both life stages probably eliminates the environmental impact of correlational selection on ontogeny. Larger larvae emerged into larger adults with a greater relative thorax length, which is in line with other studies (Stoks and Cόrdoba-Aguilar 2012). The size of adults largely determines adult fitness. For example, larger animals have higher survival (Marden and Rowan 2000), and the size in females is closely related to fecundity (Cordero 1991). Thorax size is fixed in adults and determines the maximum space available for flight muscles. Therefore, it determines flight power output (Schilder and Marden 2004) and impacts sexual selection, predator avoidance, and dispersal (De Block and Stoks 2007; Cόrdoba-Aguilar 2008). Interestingly, the covariation between body size and darkness was reversed in adults with larger individuals being darker only in northern populations (see Figs. 2, 3). Phenotypic differences between adult males and females from different range locations could have primarily explained the covariation between size and darkness. Males were larger and darker than females, and these differences were more pronounced in northern populations (Fig. 2). However, the size–darkness covariation was not canceled by the addition of a sex effect in a linear model (P = 0.04). Another explanation could be that adult darkness, which was not carried over from the larval to the adult stage, reflects different color hues produced by different pigments (Futahashi 2016), with different physiological constraints (e.g., oxidative stress) and different functions (e.g., reproduction instead of camouflage). To sum up, further investigations are required to better understand the carry-over of traits in dragonflies and other taxa with metamorphosis and their adaptation to changing environments.

Phenotypic integration is increasingly being studied to understand the evolutionary and ecological implications of trait variation among populations living in heterogeneous environments (Beckerman et al. 2010; Therry et al. 2019). Here, we showed the differentiation between core and marginal populations in larval darkness, the covariation in larval behavior and morphology, and the ontogenetic carry-over from larval size to investment in flight morphology. The association of larval size with adult flight morphology and size suggests that constraints at the larval stage only (e.g., season length) also influences phenotypic traits and potentially associated fitness traits at the adult stage. Our results highlight the complex interplay among phenotypic traits of individuals from core and margin populations and emphasize the need for studies to understand the drivers of species’ responses to climate change.