The influence of ecological and life history factors on ectothermic temperature–size responses: Analysis of three Lycaenidae butterflies (Lepidoptera)

Abstract Body size has been shown to decrease with increasing temperature in many species, prompting the suggestion that it is a universal ecological response. However, species with complex life cycles, such as holometabolous insects, may have correspondingly complicated temperature–size responses. Recent research suggests that life history and ecological traits may be important for determining the direction and strength of temperature–size responses. Yet, these factors are rarely included in analyses. Here, we aim to determine whether the size of the bivoltine butterfly, Polyommatus bellargus, and the univoltine butterflies, Plebejus argus and Polyommatus coridon, change in response to temperature and whether these responses differ between the sexes, and for P. bellargus, between generations. Forewing length was measured using digital specimens from the Natural History Museum, London (NHM), from one locality in the UK per species. The data were initially compared to annual and seasonal temperature values, without consideration of life history factors. Sex and generation of the individuals and mean monthly temperatures, which cover the growing period for each species, were then included in analyses. When compared to annual or seasonal temperatures only, size was not related to temperature for P. bellargus and P. argus, but there was a negative relationship between size and temperature for P. coridon. When sex, generation, and monthly temperatures were included, male adult size decreased as temperature increased in the early larval stages, and increased as temperature increased during the late larval stages. Results were similar but less consistent for females, while second generation P. bellargus showed no temperature–size response. In P. coridon, size decreased as temperature increased during the pupal stage. These results highlight the importance of including life history factors, sex, and monthly temperature data when studying temperature–size responses for species with complex life cycles.

Recent studies have provided insights into the many ways ecological and life history factors are likely to affect the direction and strength of temperature-size responses (but a full synthesis is needed). These factors include, but are not limited to, sex, voltinism, trophic level, immature stage, and habitat type. For example, aquatic ectotherms exhibit a decrease in size with warming temperatures but the strength of response may vary depending on trophic level (Forster, Hirst, & Atkinson, 2012;Wilson-Brodie, MacLean, & Fenberg, 2017). In contrast, some terrestrial species increase in size with warming temperatures, but response can vary based on sex and voltinism (Fenberg et al., 2016;Horne et al., 2015). In some holometabolous insects, univoltine species increase in size with temperature while multivoltine species decrease in size, whereas bivoltine species appear to show no change . While these are important findings, approaches thus far have not generally considered whether individual generations and/or sexes have varying responses. This can be significant because individuals from each generation will experience different temperatures. Fenberg et al. (2016) showed this is particularly important during growth of the final larval instar. Sex can also be an important factor especially if there is sexual size dimorphism (SSD), as males and females do not always respond in the same way to temperature (e.g., Fenberg et al., 2016;Høye et al., 2009). Thus, omission of such ecological and life history factors when conducting analyses may oversimplify the study system and/or cause some temperature-size responses to be masked.
Climate warming potentially provides a longer growing season for some insect species, particularly for temperate species emerging early in the season (Forrest, 2016). This is especially significant for obligate univoltine species (restricted to one generation per year) which often attain a larger size at maturation with increasing temperature . Some species, however, are able to increase voltinism and produce an extra generation per year at higher temperatures, resulting in a decrease in adult size due to a trade-off between increased single generation growth and producing more generations with smaller individuals (Forrest, 2016;Horne et al., 2015;Van Dyck, Bonte, Puls, Gotthard, & Maes, 2015;Zeuss, Brunzel, & Brandl, 2016). Thus far, it is unclear how each generation of multivoltine species (more than two generations in one year) or obligate bivoltine species, which are restricted to two generations each year, will respond to increasing temperatures.
A recent study using museum collections and monthly temperature records over a 100-year period of a British population of the univoltine butterfly species, Hesperia comma, found that adult size increased with increasing temperature; this, however, was only the case for males and not females, which showed no significant change in size (Fenberg et al., 2016). The females of this species are larger than males and therefore, these results suggest SSD decreases with increasing temperature. This differs from field research on the Arctic spider Pardosa glacialis, which found that females (but not males) increased in size with earlier snowmelt and the species became more sexually dimorphic in size (Høye et al., 2009), and from field research on two Arctic butterflies, Boloria chariclea and Colias hecla, which found that size was negatively correlated with temperature in both males and females despite the difference in size between the sexes (Bowden et al., 2015). These studies highlight both the importance of studying males and females separately, and that species may have different responses to climate change, reflecting different life history strategies, rather than following a particular rule.
Museum collections are a useful resource for studying historical size changes and can be used to create a time series which can be compared to climate variables over the same period . Furthermore, these collections often cover a wide range of taxa over large spatial areas and temporal periods . There are, however, some potential problems in using museum collections as metadata are often incomplete or missing altogether, which reduces the number of useable specimens Lister et al., 2011), and specimens are usually collected opportunistically rather than systematically as part of long-term projects (Kharouba, Lewthwaite, Guralnick, Kerr, & Vellend, 2018).
Additionally, museum specimens often do not provide an ecological context, such as other species present or abiotic environmental factors, and therefore, the drivers of change may not be apparent (MacLean, Nielsen, Kingsolver, & Buckley, 2018). When species are well represented in collections, however, they provide a unique opportunity to study changes in species body size over a long time period (decades to centuries; Kharouba et al., 2018). As long as specimens are used critically (i.e., by disregarding those without sufficient metadata or that are in poor condition) and paired with appropriate temperature records, these collections can provide important insights into recent historical size changes, and may prove useful in predicting future changes.
Butterfly collections provide a useful resource for studying the effect of various factors on temperature-size responses. Good study species for this type of research are ones in which males and females, and in some cases separate generations, can easily be identified and climate data are readily available for analyses. This approach was used in previous research on H. comma (Fenberg et al., 2016) in which monthly temperature records were compared to adult size, rather than a single annual or seasonal temperature value. This has the advantage of assessing which monthly temperature has the most impact on adult size and therefore, in which life stage growth is most affected by temperature. Laboratory experiments often use a constant, single temperature per replicate (for example, studies used for meta-analysis in Horne et al., 2015), but in the natural environment, temperature fluctuates and individuals may experience a wide range of temperatures during the growth period and from one year to the next as shown in field research by Bowden et al. (2015) and in historical data used by Fenberg et al. (2016).
In contrast to previous studies, which have not considered generational differences in temperature-size responses, we have chosen to examine how temperature affects the adult size of each generation in a bivoltine species, the Adonis Blue butterfly (Polyommatus bellargus), compared to the response of two univoltine species, one from the same habitat (the Chaklhill Blue butterfly Polyommatus coridon) and one from a different habitat type as P. bellargus (the Silver-studded Blue butterfly Plebejus argus). Polyommatus bellargus has two temporally distinct generations (separated by approximately two weeks) and for all three species, the sexes are easily identifiable, and are present in large numbers in the collections of the Natural History Museum (NHM; London), making them ideal study species to investigate temperature-size responses over many decades.
Based on the meta-analysis by Horne et al. (2015), we hypothesize that the univoltine species, P. coridon and P. argus, will increase in size with temperature and, as a bivoltine species, the size of P. bellargus will not change in response to temperature warming. Horne et al. (2015) did not include the effect of generation or sex in their analysis; however, life history traits, such as timing of the final larval stage and sex, can influence whether there is a response to temperature (Fenberg et al., 2016). Therefore, when sexes and generations are analyzed separately, we may find that a subset of the population is responding differently to temperature changes. If this is the case, we predict that for P. bellargus, generation one will show a greater size response to temperature than generation two due to the extended growing period available to generation one in warm years and the less favorable growing conditions during cool springs (Thomas, 1983;Thomas & Lewington, 2014). In particular, we would expect May temperatures to be important for generation one as this is when the species is in the final larval stage (Thomas & Lewington, 2014).
For the two univoltine species, we would expect a similar response as Hesperia comma (Fenberg et al., 2016); therefore, we predict that the size P. argus and P. coridon will increase in size with increasing temperature for the month corresponding to the final larval stage (June for both species [Thomas & Lewington, 2014]), and there will be a greater size change for the smaller sex.

| Study system
The three study species are European Lepidoptera in the Lycaenidae with northern range edges in the UK. These species have different life history and ecological traits, which are summarized in Table 1.
Further details about the species are given below.
In Britain, P. bellargus and P. coridon are restricted to calcareous grasslands on the south-facing slopes of chalkhills, and populations are discrete (Brereton, Warren, Roy, & Stewart, 2008;Harper, Maclean, & Goulson, 2006;Thomas, 1983). Plebejus argus occurs on lowland heaths, mosses, calcareous grasslands, and sand dunes; this species has rapidly declined across the UK in the last century, and its stronghold is now on the New Forest heaths in Hampshire (Thomas, 1985;Thomas & Lewington, 2014). Polyommatus bellargus is obligatorily bivoltine, with the first generation emerging in May and June, and the second in August and September (Thomas, 1983). This makes P. bellargus an interesting study species as it will not respond to temperature by attempting to fit in an extra generation. In contrast to this, P. coridon and P. argus are both univoltine and adults are on the wing from mid-July to early September and late June to August, respectively (Thomas, 1985;Thomas & Lewington, 2014).
Despite laying their eggs on the same plant species, there is limited TA B L E 1 Some traits and characteristics of the three lycaenid study species (Brereton et al., 2008;Thomas, 1985Thomas, , 1983Thomas & Lewington, 2014)  as there is little overlap between the timings of the larval stages (Thomas, 1983). Like many European Lycaenidae species, P. bellargus, P. coridon, and P. argus are all associated with certain ant species, including Lasius alumius, L. niger, and Myrmica sabuleti (Fiedler, 1989;Kitching & Luke, 1985). The butterfly larvae produce secretions and are "milked" by ants. The ants nurture the larvae and bury them in earth cells when they are inactive or molting (Thomas, 1983;Thomas & Lewington, 2014).

| Climate variables
Temperature records were obtained from the UK Meteorological Office (http://www.metof fice.gov.uk/clima te/uk/summa ries/datasets). Regional records of annual, seasonal (spring-summer average), and monthly mean temperatures were used from the South East and South-central England (for P. bellargus and P. argus) and the East Anglia regions (for P. coridon), as these areas included our study sites.
These data cover the years from 1910 to the present day and provided appropriate coverage for the majority of our butterfly data set.

| Image analysis
The NHM has recently digitized their entire British butterfly and moth collections (Paterson et al., 2016). A large number of specimens of P. bellargus with accompanying collection date and locality metadata were present in the collections from Folkestone (51°4′53.05″N, 1°10′10.20″E; 1,396 out of 4,814 specimens for that species). The most common site in the collections for P. coridon was Therfield Heath (52°2′24.00″N, 0°1′12.00″W; 2,300 out of 3,097 specimens).
For P. argus, many specimens were collected from the New Forest in Hampshire (centered around 50°52′12.00″N, 1°37′48.00″W; 1,049 out of 2,121 specimens). No other single area had as high a density of specimens for those species, and therefore, we focused on those locations to remove any potential effects of locality on the results.
Images of these specimens were checked for usability and separated into year groups and by generation. The generations of P. bellargus were separated following the methods used in Brooks et al. (2017). To be usable, the images needed to be sharp (i.e., not blurred) and in dorsal view with undamaged forewings. For P. coridon, some years had an excessive number of specimens so 50 usable specimens were selected at random for those years. The number of specimens used in analyses is given in Table 2, along with the years covered by the specimens. It should be noted that for P. argus, 48 specimens were not included as they were collected prior to 1910 (when local temperature records began). For P. coridon, males in 1919 were all considerably smaller than males in other years and therefore were excluded from analyses as outliers.
For each image, the sex of the specimen was noted and the length of the forewings were measured using ImageJ software.
First, the scale was set in ImageJ for each individual image using the scale bar imbedded in each image to measure out a known distance, for example 10 mm. Each forewing was measured from the point at which the wing meets the thorax to the apex of the wing (not including the scales at the very edge of the wing as these were often absent), and an average forewing length was calculated for each individual. Forewing length of museum specimens has been found to correlate strongly with wing surface area (Fenberg et al., 2016) and has been used in previous studies as a proxy for overall body size (Bowden et al., 2015;Fenberg et al., 2016). Temperature data were added to the data sets for the corresponding years of collection.

| Statistical analysis
Within each species (or generation where applicable), specimens were only included in the statistical analyses if there were three or more individuals for each sex per year. For each species, average forewing length was compared to mean annual temperature; all data were used in this model without taking into account generation or sex, which is consistent with methods used in previous studies and meta-analyses of temperature-size responses (e.g., Baar et al., 2018;Horne et al., 2015). This was repeated using mean spring-summer temperatures, which were an average of March to August temperatures to include the main growing periods for all species.
Data were then analyzed using R statistical packages MASS and June to August for generation two; for P. coridon and P. argus, each month from March to July). These months were selected to cover the entire postwinter larval growth period and included the pupal phase as well (Thomas & Lewington, 2014). The data were analyzed separately for males and females, and for generations one and two in P. bellargus.
Additionally, where there were significant correlations between size and monthly temperatures, a percentage change in size per °C change in temperature was calculated for the most important months for predicting adult size so that change could be compared between species and sexes. Percentage wing size change was calculated using the formula (exp (slope) −1) × 100; slopes were calculated using the natual log of wing lengths to account for any scaling effects that may have resulted from the differences in size between species and sexes (Fenberg et al., 2016;Forster et al., 2012).

| Sexual size dimorphism
For P. bellargus, generation one adults were significantly larger than generation two adults (F 1, 718 = 25.278, p < .001) but there was no overall difference in size between sexes (F 1, 718 = 1.286, p = .257). The males of generation one were larger than females in generation one Therefore, the SSD was limited to generation one, as a result of males being larger in this generation. For P. coridon, males were significantly larger than females (t = −3.7492, p < .001; male average size = 15.414 mm, female average size = 14.951 mm; Figure 1).

| Annual and seasonal temperature
Average forewing length was first compared to annual mean temperature using a linear model, irrespective of sex and generation, for each species. The model was not significant for P. bellargus (adjusted R 2 = .00224, F = 2.617, df = 1 and 720, p = .106) or P. argus

| Polyommatus bellargus
Multiple linear regression of the P. bellargus size data using mean monthly temperatures for males and females in both generations showed that the model was significant for generation one males (  Figure 3). In addition, for males, there was also a 3.03% increase in size with April temperatures and a 2.02% decrease with March temperatures. The models were not significant for generation two males or females (Table 3).

| Polyommatus coridon
For P. coridon, multiple linear regression models with mean monthly temperatures were significant for both males and females. For males, mean June temperature was the only significant variable, but all variables were included in the final model after stepwise regression (  (Figure 2d).

| Plebejus argus
The multiple linear regression was significant for male, but not female P. argus (Table 3)   When data were analyzed with no consideration of sex or generation and only using one temperature measure for each year (annual or spring-summer averages), there was no significant relationship between size and temperature for the bivoltine species, P. bellargus. This is consistent with results shown in Horne et al. (2015). In contrast to those results, however, the univoltine species studied here did not increase in size when compared to a single temperature value; P. argus showed no response to temperature and P. coridon decreased in size with increasing temperature. Often these annual and seasonal values may be the only climate data available but they are TA B L E 3 Results of the linear models for predicting average forewing length of Polyommatus bellargus (both generations), Polyommatus cordion, and Plebejus argus using mean monthly temperatures as variables; values are adjusted R 2 (AR 2 ), the F statistic (F), degrees of freedom (df), and the p-value (p) biologically less meaningful than using the growth period monthly averages for short-lived species (such as those studied here). As we show, the temperature-size relationship of an organism can also vary between and within life stages, which would not be detected unless temperature records for the appropriate months are included in analyses. This is important, since any significant effect obtained with annual or seasonal temperatures could result from a correlation with temperatures during one or more months during the life cycle, in which case the full complexity of the response to temperature will be missed. Furthermore, for species with more than one generation in a year, each generation may experience different environmental conditions (e.g., climate and food quality and quantity), which can differentially affect size between generations (Horne, Hirst, & Atkinson, 2017). If generational differences are not taken into account (e.g., by averaging wing length of all individuals regardless of generation), or if there is a bias toward a particular generation in the data set, especially if that generation is not responsive to temperature, any effects of temperature on size may be masked.

| D ISCUSS I ON
When growth period monthly temperatures, sex, and generation are included separately in the models for the three species, there is a significant relationship between size and temperature, apart from female P. argus and generation two P. bellargus, which showed no TA B L E 4 (a) Outputs for variables included in linear models for Polyommatus bellargus generation one males and females using mean monthly temperatures. (b) Outputs for variables included in linear models for Polyommatus coridon males and females using mean monthly temperatures. (c) Outputs for variables included in linear models for Plebejus argus males using mean monthly temperatures temperature during early larval stages may be due to a trade-off between using energy for growth and energy for producing secretions to attract ants, which are more active during warmer conditions (Thomas & Lewington, 2014). Another possible explanation is that ecdysis (moulting) takes place more slowly and at a higher metabolic cost at lower temperatures (which is more likely earlier in the season), and therefore, early stage larvae in cool conditions take in more food over a longer period at a higher efficiency, and are able to grow to a larger size during early development (Karl & Fischer, 2008  advantage for access to females. At cooler temperatures, there is a trade-off between reaching a large size and emerging earlier. Therefore, males tend to be smaller during years when cooler conditions coincide with the late larval stages . Males must also actively compete with other males for females (in P. bellargus males swarm around freshly emerged females; Thomas & Lewington, 2014), so large males are at an advantage.
However, emerging in time to compete for females may be more important for males than growing to a larger size, hence, they tend to be smaller in years with lower temperatures during late larval stages (e.g., Figure 3). For females, only P. bellargus generation one showed an increase in size following an increase in temperature during late larval stages. In bivoltine (and multivoltine) species, females in the first generation may be under time pressure to emerge and lay eggs. Therefore, in cooler years, when growth is slower, females emerge at a smaller size.
Conversely, in univoltine species, females are under less time pressure to develop so may prioritize growth, to increase fecundity, and therefore female size may be less affected by temperature.
Nevertheless, for female P. coridon, our results showed a decrease in size with increasing May temperatures, although it was smaller than the decrease in size shown by the males. May is towards the end of the period occupied by early larval stages in this species. A possible explanation is that the relationship with ants starts in May and, therefore, the larvae use some energy for producing the secretions that are attractive to ants, and this activity increases under warmer conditions (Thomas & Lewington, 2014). This may also be the reason for the decrease in adult size when higher temperatures occur during the pupal stage, because the pupa of this species also produces secretions and sounds to attract ants (Thomas & Lewington, 2014). This contrasts with results in a study of the univoltine butterfly Anthocharis cardamines, which is not associated with ants in the larval or pupal stages, which found that an increase in adult body size was correlated with temperature increases during the pupal stage of development, and that both sexes responded in the same way (Davies, 2019).
Our findings suggest that a reversal of the TSR occurs only during late larval instars for males in univoltine species and in both sexes in the first generation of bivoltine species. It is likely that the second generation of bivoltine species (and any further generations in multivoltine species) do not respond to temperature by changing size due to the limited time available for growth and conditions being more favorable for growth later in the season. For example, in second generation P. bellargus, the larval foodplant covers a larger area, the plants are taller, and the larvae are not restricted to the warmest short vetches (Thomas & Lewington, 2014). This is reflected in the greater abundance of individuals in generation two in the museum collections and in population monitoring data (Thomas, 1983;Thomas & Lewington, 2014;UKBMS, 2018). A phenological study of British butterfly species found that in warmer years, both generations one and two of P. bellargus emerge earlier than in cooler years (Brooks et al., 2017). Therefore, it is possible that any increase F I G U R E 3 Plot of average forewing length (mm) versus May temperature for generation one Polyommatus bellargus males. Circles represent individual specimens, the solid line is the predicted values given by the linear model, and the dashed lines represent two standard errors above and below the predicted values. Note, the increasing trend during years covering cool to moderate May temperatures (10-11.5°C)  TA B L E 5 Stage-specific temperaturesize trends in three lycaenid spcies, and Hesperia comma (Hesperiidae; results from Fenberg et al., 2016) in growth rate due to increased temperature is counteracted by an earlier emergence time leading to no overall change in average size between years for generation two.
In contrast to our results, a meta-analysis of a large number of arthropod species found that, in general, there was no difference in the strength and direction of the temperature-size responses in males and females of the same species, and therefore, SSD did not change with temperature . Yet in agreement with other previous studies, our results show that temperature can have variable effects on the sizes of each sex, resulting in a shift away or towards SSD (Fenberg et al., 2016;Høye et al., 2009). Unlike most butterfly species, in which females are the larger sex and have a longer development time (Teder, 2014;Wiklund & Kaitala, 1995), the males in all three of our study species are larger, despite males emerging earlier than females (Thomas, 1985). Therefore, an increase in female size (as the smaller sex) with increasing temperature was predicted. Yet, for P. argus and P. coridon, males increased in size with increasing temperatures during the late larval stages, while the females did not (  (Ravenscroft, 1990). An additional factor to consider is the life cycle of the species. Polyommatus bellargus, for example, overwinters as a larvae whereas P. argus and P. coridon overwinter as eggs (Thomas & Lewington, 2014). Yet, they are all direct developers (i.e., there is no pupal diapause) and therefore may be subject to more seasonal time constraints and variation than species which diapause in the pupal phase, such as A. cardamines (Davies, 2019). This may explain why all three lycaenid species we have studied, and H. comma, respond to temperature in the larval stage ( Advances in phenology may result in certain life stages (particularly early larval stages) being exposed to shorter days, so that, although day-time temperature is higher, these stages will be exposed to cooler night-time temperatures for longer. This is particularly relevant for P. coridon and P. argus, which are nocturnally active (Thomas & Lewington, 2014). Yet, it is likely that ground temperatures, especially around the larval host plants, will increase even during the night if day-time temperatures are higher, which is probably why these two species show the same general temperature-size trends as P. bellargus, which has diurnal larvae.

| CON CLUS IONS
Temperature-size responses can vary according to life cycle factors (e.g., voltinism and life history stage) and sex. If these factors are not included in analyses, adult size of a species may not appear responsive, especially when compared to annual or seasonal temperatures.
However, a temperature-size response can emerge if studies separately analyze sex, generation, and life history stage with monthly temperatures during growth. While the species studied here all responded in some way to temperatures during immature stage development, the responses varied in strength and direction. Generation one male and female P. bellargus and males of P. coridon and P. argus responded in a similar way to previously studied univoltine species (Fenberg et al., 2016;Horne et al., 2015) and became larger in years with warmer temperatures during late larval stages and, therefore, showed the reverse of the TSR. Conversely, adult body size decreased in size with increasing temperature during early larval instar development and, in P. coridon, during pupal development, in line with the TSR. This suggests that, within a species, the temperaturesize response can change direction as the immature stages progress (Table 5). Furthermore, these results suggest that for species which follow this trend, there will likely be an increase in SSD due to climate warming, if increases in size during late larval instars are greater than decreases in size during early larval instars. This not only highlights the importance of integrating life history factors into temperaturesize response analyses, but also emphasizes that size declines in response to climate warming are not "universal" and that the TSR is too simplistic, especially for species with complex life cycles.
While not all aspects of ecology can be included directly in a study using museum specimens, these collections can provide useful insights into the responses of organisms to temperature change in the recent historical past. These data, when used in combination with monthly temperature records, can unravel the complex interactions between individuals and some of the factors that control body size. Furthermore, museum specimens provided the best opportunity to study the size response of these species to a wide range of temperatures as field studies would require continuous monitoring over many years and laboratory experiments would be hard to conduct due to the strong relationship between the juvenile stages of these species and ants.
Future studies examining temperature-size responses for species with complex life histories should be aware that the strength and direction of temperature effects may vary according to growth stage, sex, and generation. Where possible, studies using natural history collections can be used in combination with field studies and laboratory experiments to give a more holistic approach to studying temperature-size responses and predicting how these responses may be affected by climate change.

ACK N OWLED G M ENTS
We would like to thank the anonymous reviewers for their detailed and helpful comments, which have greatly improved the paper. We would like to thank Ben Price and the NHM digitization team for providing us with the high-resolution specimen images. This work was supported by the Natural Environmental Research Council (grant number NE/L002531/1).

CO N FLI C T O F I NTE R E S T S
There are no competing interests to declare.

AUTH O R S ' CO NTR I B UTI O N S
All authors contributed in conceiving the ideas, designing methodology, and writing, and agreed to the final manuscript; RJW collected and analyzed the data, and led the writing of the manuscript.