Effects of temperature and drought on early life stages in three species of butterflies: Mortality of early life stages as a key determinant of vulnerability to climate change?

Abstract Anthropogenic climate change poses substantial challenges to biodiversity conservation. Well‐documented responses include phenological and range shifts, and declines in cold but increases in warm‐adapted species. Thus, some species will suffer while others will benefit from ongoing change, although the biological features determining the prospects of a given species under climate change are largely unknown. By comparing three related butterfly species of different vulnerability to climate change, we show that stress tolerance during early development may be of key importance. The arguably most vulnerable species showed the strongest decline in egg hatching success under heat and desiccation stress, and similar pattern also for hatchling mortality. Research, especially on insects, is often focussed on the adult stage only. Thus, collating more data on stress tolerance in different life stages will be of crucial importance for enhancing our abilities to predict the fate of particular species and populations under ongoing climate change.


| INTRODUCTION
The Earth's mean surface temperature and the frequency of extreme weather events such as heat waves have already increased as a result of anthropogenic climate change (Hansen, Sato, & Ruedy, 2012;McKechnie & Wolf, 2010;Meehl et al., 2007). These changes have in turn resulted in phenological and range shifts as well as abundance changes in a plethora of species (Chown et al., 2010;Parmesan & Yohe, 2003;Sunday, Bates, & Dulvy, 2012;Thomas, 2010). In this context, the extreme temperatures associated with heat waves seem to be more important than changes in mean temperatures because they typically exert a much stronger selection pressure (Anderson, Collinge, Hoffmann, Kellett, & McKechnie, 2003;Kellermann et al., 2012;Zimmermann et al., 2009). However, in addition to increasing temperature stress, terrestrial organisms will likely experience higher levels of desiccation and food stress, due to detrimental effects of drought periods on water supplies and food plant quality and availability (Clusella-Trullas, Blackburn, & Chown, 2011;Hoffmann, Chown, & Clusella-Trullas, 2013).
The changes outlined above are considered to be a major threat to biodiversity (Pimm et al., 2014;Thomas et al., 2004). However, while some species will suffer, others may benefit from ongoing climate change (e.g., many warm-adapted species). Thus, responses to climate change are likely species specific, probably depending on a given species ability to cope with extreme temperatures, desiccation, and associated food stress (Anderson et al., 2003;Coumou & Rahmstorf, 2012). To identify which species are most at risk from climate change is of prime importance to predict the future consequences of ongoing climate change Chown et al., 2010;Beaumont & Hughes 2002;Hoffmann & Sgrò 2011;Rosset & Oertli 2011). Unfortunately, the specific biological features determining whether a given species is becoming a "winner" or "loser" of climate change are largely unknown (Brook et al., 2009;Williams, Shoo, Isaac, Hoffmann, & Langham, 2008). In this context, it should be noted that research, especially on insects, is often focussed on the adult stage only, which may bias predictions regarding a species' survival under climate change (Kingsolver, 2009;Kingsolver et al., 2011;Klockmann, Günter, & Fischer, 2017;Radchuk, Turlure, & Schtickzelle, 2013). Typically, stress tolerance varies throughout ontogeny in insects (Bowler & Terblanche, 2008;Kingsolver et al., 2011), such that it is necessary to identify the most vulnerable life stage (Bowler & Terblanche, 2008;Klockmann, Günter, et al., 2017;Radchuk et al., 2013). Here, early life stages, often facing high mortality, may be particularly crucial although temperature stress perceived early in life may not necessarily affect later life (Klockmann, Günter, et al., 2017;Potter, Davidowitz, & Arthur Woods, 2011).
However, matters are complicated as other factors may also play an important role in the mortality of early life stages, for instance, the specific microclimatic conditions provided by the host plants (Potter, Davidowitz, & Woods, 2009;Smith, 1978;Woods, 2013 Figure 1). Currently, L. tityrus shows positive, L. dispar largely stable, and L. helle negative population trends (Brunzel, Bussmann, & Obergruber, 2008;Settele et al., 2008;Habel et al. 2011;Lindman et al., 2015). These differences seem to be associated with different distribution areas and habitat requirements, with L. tityrus inhabiting different types of habitat including hot and dry stands, L. dispar mainly wetlands, and L. helle cool and moist habitats (Ebert & Rennwald, 1991;Settele et al., 2008;Habel et al. 2011). Consequently, these species may also differ in (heat) stress tolerance and concomitantly in their vulnerability to climate change, ranging from low to high risk (Ebert & Rennwald, 1991;Settele et al., 2008;Habel et al. 2011; see further below and Table 1). We focus on early developmental stages because earlier studies showed that differences in vulnerability are unlikely to be caused by differential responses to thermal stress during larval and pupal development (Klockmann, Karajoli, Reimer, Kuczyk, & Fischer, 2016;Klockmann, Schröder, Karajoli, & Fischer, 2016) as well as adult stress resistance (Klockmann, Wallmeyer, & Fischer, 2017). We hypothesize that (i) mortality rates increase at higher temperature and additionally with reduced humidity in all species, (ii) and that L. helle will suffer most strongly from simulated heat and drought stress. Such species differences in the sensitivity to stress are statistically indicated by species × treatment interactions, for which we explicitly test here.

| Study organisms and egg sampling
To investigate vulnerability to climate change, we used three species of Copper butterflies (Lycaena spp.; cf. . The Sooty Copper L. tityrus (Poda, 1761) is a widespread temperate-zone butterfly, ranging from western Europe to central Asia (Ebert & Rennwald, 1991). The species has 1-3 generations per year and inhabits a variety of biotopes like grassland, sandy heathland, bogs, and open woodland (Brunzel et al., 2008;Settele et al., 2008). The principal larval host plant is Rumex acetosa L., but some congeneric plant species (e.g., R. acetosella L., R. scutatus L.) are utilized as well (Ebert & Rennwald, 1991;Settele et al., 2008;Tolman & Lewington, 2008). This species is not listed in the EU Habitat Directive and is considered least concern in the Red List of Germany (Settele et al., 2008;Binot-Hafke et al., 2011; Table 1). Lycaena tityrus has recently colonized previously unoccupied mountain ranges in central Europe and is expanding its range northward in northeastern Europe (Brunzel et al., 2008;Settele et al., 2008). Because of these range expansions and its ability to inhabit even dry and hot habitats, the species is expected to benefit from climate change and its according vulnerability is consequently considered to be low. Mated females were caught in two bivoltine Bad Nenndorf, Germany) under naturally fluctuating temperatures to improve ecological realism (i.e., control conditions: mean 19.4°C, 75% relative humidity, and L17:D7 photoperiod; Figure 2). The second species, the Large Copper L. dispar (Haworth, 1802), is a transpalaearctic butterfly, ranging from western Europe across temperate Asia to the Amur region and Korea (Ebert & Rennwald, 1991;Settele et al., 2008). The species has 1-2 generations per year (Lindman et al., 2015;Settele et al., 2008) and typically occurs in wetland habitats including lakeside and riverside areas (Lindman et al., 2015;Settele et al., 2008). Eggs are laid on nonacidic sorrels (e.g.,  T A B L E 1 Summary of key ecological and conservation attributes for Lycaena tityrus, L. dispar, and L. helle F I G U R E 2 Graphical illustration of the conditions used to investigate stress tolerance of Copper butterfly eggs. Treatments lasted for 2 days and started 2 days after egg-laying. For the desiccation treatment, control conditions were used except that the glass vials containing the eggs were transferred on two consecutive days to a box with silica gel to reduce the relative humidity hygrophilous butterfly colonizing mires, swampy grassland, and moist meadows (Ebert & Rennwald, 1991;Fischer, Beinlich, & Plachter, 1999;Settele et al., 2008). The only larval food plant in central Europe is B. officinalis (Fischer et al., 1999). The species is declining in large parts of its range (Van Swaay & Warren, 1999), is listed in the EU Habitat Directive, and is considered endangered in the Red List of Germany (Binot-Hafke et al., 2011;Settele et al., 2008). Because of the strong population declines and its specialized habitat requirements, being confined to moist and cool stands, we consider the species' vulnerability to climate change as being high (Habel et al. 2011).

| Experimental design
We investigated egg and hatchling mortality with the eggs obtained using a split-brood design for L. tityrus and L. dispar, while for L. helle eggs were randomly divided into groups. In the first experiment, we investigated the effects of heat and desiccation stress on egg mortality. Both stresses seem to be ecologically relevant in the egg stage. For testing, eggs were placed 2 days after laying into glass vials in groups of 10, using 9-20 replicates per treatment and population. Replicates  (Limberg & Fischer, 2014). Thus, we used a temperature cycle typical of average conditions in June (i.e., within larval period) as control and the data of a particularly hot June day to simulate a heat wave (Limberg & Fischer, 2014). Although the humidity chosen is very low, note the short exposure time mimicking conditions that may occur, for instance, during spells of direct sun exposure. Our field data obtained from larval habitats confirm that such low humidities are ecologically relevant.
Except from exposure to heat waves or low humidity, all eggs were kept under control conditions. Egg hatching success per glass vial (10 eggs) was scored under control conditions as percent.
In the second experiment, we investigated effects of heat and food stress on hatchling mortality. We used a food stress rather than a desiccation treatment here for ecological reasons. While eggs certainly suffer from low humidity, larvae may not as they are able to obtain water from plant material. We therefore decided to manipulate plant quality rather than humidity. Hatchlings were placed 2 days after hatching, separated by female for L. tityrus and L. dispar, in groups of 10 into translucent plastic boxes (250 ml (ii) heat (provided with fresh cuttings and exposed for 2 days to a simulated heat wave), and (iii) food stress (provided with wilted leaves to mimic the results of a period of drought for 2 days under control conditions). Control and heat conditions were identical to the first experiment (cf. Figure 2). To produce wilted leaves, leaves were cut off the plant and stored for 24 hr at 20°C and 50% r.h. in a climate cabinet without water supply. This handling resulted in levels of host plant wilting frequently experienced in the natural habitats of the species. All animals remained under control conditions before and after the treatments. The mortality rate per box was scored as percent on day 6 of larval development.

| Statistical analyses
We analyzed mortality rates (percentage of dead individuals per box; starting with 10 individuals per box) for eggs and hatchlings using hierarchical general linear mixed models (GLMMs) with treatment and species as fixed categorical effects, and population and group (either family or random group in L. helle) as random categorical effects. Population was nested within species, and group was nested within species and population. Note that we used replicated populations for each species. The effect of replicate population was modeled as random effect in order to account for the variance explained by differences among populations rather than species. Group was modeled as random effect to account for the nonindependency of siblings, as a split-brood design was used for two of the three species (see above) such that per female one group of offspring was allocated to each treatment. Separate analyses were run for eggs and hatchlings due to differences in the treatments used. Data were analyzed using Statistica 8.0 (StatSoft, Tulsa, OK, USA). Pair-wise comparisons after GLMMs were performed employing Tukey's HSD for unequal sample sizes. Throughout the text, means are given ±1 SE.

| RESULTS
Average mortality rates of eggs varied significantly across treatments, being lowest under control conditions (14.2 ± 1.9%) followed by the heat (28.8 ± 1.9%) and finally the desiccation treatment (38.8 ± 1.9%; control < heat < desiccation, Tukey's HSD; and 31.0 percentage points for the heat and desiccation treatment, respectively), but weaker in L. dispar (11.6 and 25.9 percentage points) and especially in L. tityrus (7.8 and 15.6 percentage points).
Mortality rates differed significantly among populations and groups.

| DISCUSSION
Our experiments show that mortality rates were higher when simulating heat waves compared with control conditions. This result was expected (Andrew, Hart, Jung, Hemmings, & Terblanche, 2013;Rukke, Aak, & Edgar, 2015; and may be caused by, e.g., denaturation of proteins, disruption of membrane structure and function, interactions with oxygen supply, and dehydration impairing evaporative cooling (Chown & Terblanche, 2006;Klose & Robertson, 2004;Potter et al., 2009). Mortality rates of eggs further increased after exposure to low humidity, likely caused by dehydration facilitated by the low egg mass and a concomitantly high volume-surface ratio (Addo-Bediako, Chown, & Gaston, 2001;Chown, Sørensen, & Terblanche, 2011). The level of humidity used in our treatments was very low (10%), but exposure time was quite short (2 × 2 hr).
Therefore, we do think that our treatment conditions were reasonable to mimic effects of low humidity during drought periods, although it is evidently difficult to extrapolate such laboratory results to field conditions. For instance, eggs are in nature attached to host plants and may benefit from buffering microclimatic conditions (Potter et al., 2009;Smith, 1978). Furthermore, wilted as opposed to fresh leaves and the associated food stress increased hatchling mortality rates as expected.
All above results are in agreement with our first hypothesis, although T A B L E 2 Results of general linear mixed models (GLMMs) for the effects of treatment (control, heat, desiccation; fixed), species (fixed), population (nested within species; random), and group (nested within species and population; random) on egg (a) and hatchling (b) mortality rates in three Copper butterfly species  it is interesting to note that effects of desiccation and food stress were even more pronounced than those of heat stress. This suggests that exclusively focusing on upper thermal limits while neglecting the impact of drought periods is insufficient for predicting the fate of species and populations under changing climate (Cooper, Tharp, Jernberg, & Angilletta, 2012;Rezende, Tejedo, & Santos, 2011). Mosquito distribution, for instance, may be limited by egg desiccation resistance (Kearney, Porter, Williams, Ritchie, & Hoffmann, 2009), and fruit flies from the Australian wet tropics are known to have a low desiccation resistance and concomitant heritability, which is likely a crucial factor limiting their distribution (Hoffmann, Sørensen, & Loeschcke, 2003;Kellermann et al., 2012).
The most interesting result though is that indeed the three species investigated here differed in their responses to environmental stress, as evidenced by the significant species by treatment interaction for egg mortality, in agreement with our second hypothesis. Specifically, egg mortality increased most strongly in the arguably most vulnerable species L. helle, followed by L. dispar and finally L. tityrus (Figure 3a).
Note that these results fit very well with our predictions based on the general ecology of the three species concerned (Ebert & Rennwald, 1991;Fischer et al., 1999;Brunzel et al., 2008;Settele et al., 2008;Lindman et al., 2015;Hampe & Jump, 2011; Table 1). While the glacial relict species L. helle is associated with cool and moist habitats (Fischer et al., 1999) and should therefore suffer from both, heat and desiccation (Habel et al. 2011), L. dispar is associated with moist but (at least partly) warm habitats (Lindman et al., 2015), such that the species should in first place suffer from desiccation rather than heat stress.
Concomitantly, L. dispar responded less strongly to heat stress than L. helle but (nearly) as strongly to desiccation, while L. tityrus, being a habitat generalist with respect to moisture and temperature (Brunzel et al., 2008;Ebert & Rennwald, 1991), showed moderate responses only to both stressors. Note that the above conclusions rest mainly on comparisons relative to controls. This might be problematic given that L. tityrus showed the highest baseline mortality, which may drive the significant interaction for egg mortality (Figure 3). However, this pattern for L. tityrus is unusual as egg hatching success in this species is typically around 90% or higher under control conditions as found in both other species (K. Fischer, personal observation). The most likely explanation for the reduced hatching success in L. tityrus under control conditions seems random variation in female quality, for instance, caused by differences in female age (i.e., we assume that field-caught L. tityrus females were on average a bit older than in both other species). We exclusively used field-caught females here, as these species do not mate in captivity. To account for such effects, we are convinced that the most straightforward way to interpret our data is comparisons relative to control levels. Anyway, note in addition that L. helle showed the highest mortality rates under stress also in absolute terms.
In contrast to the above results on egg mortality, the respective interaction between species and treatment was not significant for hatchling mortality, although the pattern obtained was very similar (see Figure 2). These results may suggest that vulnerability to stress decreases during development, which is in line with an earlier study on the same species as investigated here, in which we found that differences in vulnerability to climate change are unlikely to be caused by differential responses to thermal and desiccation stress during (later) larval and pupal development  or in the adult stage (Klockmann, Wallmeyer, et al., 2017). Thus, in the species considered here, the early developmental stage, especially the egg stage, seems to be the most critical life stage determining vulnerability to climate change.
The significant group (family) and population effects in the first experiment indicate that, besides the differences among species, heat and desiccation resistance may additionally differ among populations and families. This may again be explained by random effects including variation in the quality or condition of field-caught females (see above).
Alternatively, such variation may suggest a heritable component that can be exploited by natural selection. However, heritability in such traits is typically very low, such that evolutionary rescue appears to be unlikely given the pace of current climate change (Blackburn, van Heerwaarden, Kellermann, & Sgrò, 2014;Hoffmann, Shirriffs, & Scott, 2005;Kellermann et al., 2012).
An increased frequency of extreme weather events such as heat waves and periods of drought are important consequences of ongoing climate change (Battisti & Naylor, 2009;Coumou & Rahmstorf, 2012).
Our results indicate that this may have important consequences for extant biodiversity, as simulated heat waves and drought stress generally increased mortality rates during early development, as would be expected. Our data also stress the importance of considering detrimental effects of drought when trying to forecast species responses.
Importantly, we found that closely related species, arguably differing in their vulnerability to climate change, seem to differ in their responses to environmental stress. However, such variation was restricted to early developmental stages, while different levels of stress seem to have little effect on fitness during further development (i.e., in older larvae, pupae, and adults; Potter et al., 2011;Klockmann, Günter, et al., 2017;Klockmann, Schröder, et al., 2016;Klockmann, Wallmeyer, et al., 2017). We suggest that, in the three species investigated here, stress tolerance during early development might be a major determinant of vulnerability to climate change and may explain recent population declines in L. helle along with habitat deterioration (Bauerfeind, Theisen, & Fischer, 2008;Fischer et al., 1999). These findings, if similar patterns were found in a larger array of species, may have important implications for enhancing our abilities to predict the fate of particular species and populations under ongoing climate change. For instance, it might be worth comparing models that do or do not consider differences in thermal tolerance across life stages, or models could be improved by integrating results from the arguably most sensitive stage. Recent studies on a tropical butterfly also indicated that the egg stage comprises the most vulnerable developmental stage, that body mass may be a crucial constraint on stress tolerance, and that stress experienced early in life could affect later life stages (Klockmann, Kleinschmidt, & Fischer, 2017;Klockmann, Günter, et al., 2017). Further progress regarding specific traits underlying vulnerability to climate change will likely be achieved by collating more data on stress tolerance throughout development from a broader range of taxa.