Winters restrict a climate change–driven butterfly range expansion despite rapid evolution of seasonal timing traits

Significance

Species ranges are changing rapidly due to climate change, but it is unclear how evolution during range expansion affects these range shifts. Our field experiments within and beyond the range of a native butterfly reveal that although two traits crucial for the timing of winter dormancy have evolved in expanding range margin populations, selection on them is weak at most times during most years. Instead, the range is constrained by cold winters, to which the butterflies have not adapted, possibly because historical natural selection at the range margin has depleted genetic variation for this trait. Thus, to understand the role of evolution in range expansion, we must identify the traits that limit the range and their potential for contemporary evolution.

Abstract

Climate change pushes species toward higher latitudes and altitudes, but the proximate drivers of range expansions vary, and it is unclear whether evolution facilitates climate change–induced range changes. In a temporally replicated field experiment, we translocated wall brown butterflies (Lasiommata megera) descending from range interior and range margin populations to sites at 1) the range interior, 2) the range margin, and 3) beyond the current northern range edge. Thereby, we tested for local adaptation in seasonal timing and winter survival and evaluated to what extent local adaptation influences the ongoing, climate-driven range expansion. Almost all individuals from all populations entered diapause at an appropriate time, despite previously identified among-population variation in diapause induction thresholds. Caterpillars of northern descent, however, grew faster than those from southern populations at all field sites. This may be a countergradient adaptation to compensate for the short, northern growing seasons, but we found no evidence for prewinter body mass affecting winter survival. In fact, winter survival was low overall—extremely so at the beyond range site—regardless of population origin, indicating that the primary constraint to range expansion is an inability to adapt to winter conditions. Hence, although range-expanding wall browns show clear local evolution of two traits related to seasonal timing, these putative local adaptations likely do not contribute to range expansion, which is instead limited by winter survival. To predict future range changes, it will be important to distinguish between the traits that evolve during range expansion and those that set the range limit.

Climate change redistributes species, pushing them away from areas that are no longer habitable and enabling them to colonize newly habitable areas (1–4). However, even when species expand poleward, upward, or into deeper waters—thereby tracking favorable temperatures—other environmental features may require new adaptations (5–8). For example, poleward-expanding species will face both more pronounced seasons than in their previous environments and unfamiliar daily and annual light cycles. Moreover, simply tracking warming during part of the year can fail because different seasons do not warm equally (9) and warming alters season length as well (10). To persist, species may therefore need to both move and evolve, and understanding how range-expanding populations adapt to novel environments is crucial for predicting where and how fast species ranges will shift (11–13).

Range margin populations serve as pioneers into uncharted territory. Because range margin conditions often resemble those beyond the range, local adaptation to range margins could facilitate expansion (14, 15). However, even if adaptive evolution does occur at the range margin [which can be hindered by genetic and demographic constraints (16–20)], it may not facilitate range expansion, if it does not affect the specific traits that limit the species’ range. Whether or not the traits that evolve at the range margin also are the ones that limit range expansion could greatly influence the pace of climate change–driven range expansions.

The wall brown butterfly, Lasiommata megera (Linnaeus 1767), has—despite declining dramatically in parts of Europe (21, 22)—expanded northward in Sweden since approximately 2000 (23). This recent and rapid expansion makes an excellent system for assessing the role of local adaptation in range expansion of a native species. L. megera is a diet and habitat generalist (24, 25) and hypothesized to depend heavily on warm temperatures (26, 27), so its range is likely to be climate-limited. Specifically, long and cold winters could directly reduce survival, or northern growing seasons may be too short for the complete life cycle. The yearly life cycle of L. megera always includes at least two adult generations (bivoltinism), after which offspring of the last adult generation overwinter as caterpillars in diapause (seasonal dormancy), which is triggered by short days well before winter. Such photoperiod-induced diapause can complicate range expansion across latitudes because winters arrive earlier and late-summer days are longer at higher latitudes (28, 29), yet local adaptation could facilitate correct diapause timing and prevent facing winter in an inappropriate life stage (a “lost generation”; ref. 30). Evolution of daylength thresholds for diapause induction has indeed been described in many species and can be rapid (31–38), but the role of local adaptation to novel photoperiods in facilitating climate-driven range expansions is not well understood.

In line with adaptive predictions for populations at distinct latitudes, L. megera populations from the species’ northern range margin in Sweden have evolved to enter diapause in longer days than populations from the range interior in southernmost Sweden (an area where the species has existed for as long as there are entomological records and that thus can be considered part of the species’ core range) (23). Conversely, Ittonen et al. (39) found no evidence for evolution of cold tolerance among some of the same populations, despite low survival of simulated cold winters in the laboratory. Growth rate during larval development toward diapause was not explored in the two previous studies, but is another potentially important life history trait, as its evolution might help northward-expanding populations maintain a consistent life cycle despite a shorter growing season (40, 41). To test for both among-population differences in these three traits and their fitness effects under natural conditions, we reciprocally transplanted offspring from two populations in the range interior in southern Sweden and two populations at the northern range margin 300 to 400 km further north (hereafter southern and northern populations) to field sites in the range interior, at the range margin, and north of the species’ current range (hereafter range interior, range margin, and beyond range sites; Fig. 1A). Our transplant design allowed us to evaluate first, whether short growing seasons or cold winters limit performance beyond the range margin—indicated by decreased overall performance beyond the range—second, whether demonstrated evolution at the range margin is adaptive in the field—indicated by northern populations performing better than southern populations at the range margin—and third, whether local adaptation indeed facilitates range expansion—indicated by northern populations performing better than southern populations beyond the range.

Fig. 1.

The field experiment setups. (A) A map of the sampled populations (colored circles), experimental sites (black diamonds), and the current range of the wall brown butterfly (gray shading). (B) Our experimental sites during the autumn experiment (each site had 20 cages, five for each population). (C) A drawing of an autumn experiment cage. Two bricks kept each cage in place, on top of the bricks laid a plastic pot of grass on a drainage plate, and the front of had a zipper opening (not depicted). (D) Pictures of our winter experiment setup. Map adapted from ref. 23, originally published under a CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) license.

In our experiments (summarized in Table 1), we studied three potentially range-limiting traits. First, in the autumn experiment, we placed eggs from wild-caught butterflies from all populations in field cages at the three experimental sites, beginning in August (Fig. 1 A–C). At the end of the growing season, we assessed whether the individuals had adequately interpreted seasonal cues and entered diapause, and we weighed all diapausing caterpillars to compare growth rates and prewinter weights. Second, in the laboratory experiment, we assessed among-population differences in growth rate using common garden laboratory treatments, allowing separation of the effects of photoperiod and temperature, both of which can affect growth rates (42, 43). Finally, in the winter experiment (Fig. 1D), we assessed (mal)adaptation to winter conditions by having caterpillars from both the autumn experiment and the laboratory experiment overwinter at our field sites and scoring their survival.

Table 1.

Overview of the performed experiments and their timing

Experiment	Years (temporal replicates)	Months	Studied traits	Life stages	Origin of animals
Autumn*	2021, 2022	August–October	Diapause or direct development and prewinter mass, both assessed in late October.	Egg–pupa	F1 offspring of wild females collected in August the same year
Laboratory	2022	August–October	Growth rate during development into diapause	Caterpillar	F1 offspring of wild females collected in August the same year
Winter	2021 to 2022 and 2022 to 2023	October–March/April	Survival (scored in late March or early April 2022 and 2023)	Caterpillar (in diapause)	Diapausing caterpillars from the autumn experiments; in 2022 to 2023 also (separately) diapausing individuals from the laboratory experiment

^*See SI Appendix, Appendix 1 for the pilot study done in 2020.

1. Results

1.1. Diapause Induction.

Almost all individuals that survived the autumn experiment entered diapause: all 482 individuals in 2021 and all but seven out of 678 individuals (99%) in 2022. The nondiapause developers were all from the two southern populations (five individuals from Hässleholm and two from Vejbystrand) and reared at the range margin site, where there was a significant difference among populations (Fisher’s exact test, P = 0.009). In 2020, we performed a pilot experiment (SI Appendix, Appendix 1), which was started about 2 wk earlier than the 2021 and 2022 experiments and yielded remarkably lower diapause incidence: Only 30% (72 out of 242 surviving individuals) entered diapause.

1.2. Growth Rate in the Field.

By the end of the autumn experiment, caterpillars from the northern populations had grown larger than those from the southern populations (Fig. 2; F_3,94.5 = 26.1, P < 0.0001), at all sites (population–site interaction: F_6,94.4 = 1.49, P = 0.19). All pairwise comparisons except the one between the two southern populations (P = 0.14) showed significant differences (SI Appendix, Appendix 2, Table S1). As expected given warmer conditions (see weather data in Table 2 and SI Appendix, Appendix 3), caterpillars generally grew larger the further south they were reared (Fig. 2; F_2,98.8 = 289, P < 0.0001), and, despite a year–site interaction (F_2,98.7 = 23.7, P < 0.0001), all sites differed from each other in both years (P < 0.0001 for all comparisons; SI Appendix, Table S2). Overall, caterpillars grew larger in the warmer year, 2022, than in 2021 (Fig. 2; F_1,98.9 = 55.1, P < 0.0001), except at the range interior site (SI Appendix, Table S3).

Fig. 2.

Larval mass at the end of the autumn experiment in October 2021 and 2022. All data points are shown with black circles, the black lines represent means, the violins show the estimated kernel probability density (violin width indicating the proportion of the data at the y-axis values), and numbers below each violin indicate sample sizes. See legend for population colors; S = southern population, N = northern population.

Table 2.

Climate variables for the autumn and winter experiments

Site	Year	T first 30 d	Autumn GDD	T Dec–Feb	Days below −16	Winter onset	Spring onset
Range interior	2021	15.1	237	2.0	0	NA†	NA†
Range interior	2022	17.1	286	1.8	2	Dec 7	Dec 24
Range margin	2021	13.2	169	–1.0	1	Nov 26	Feb 12
Range margin	2022	15.3	218	–0.4	2	Dec 3	Mar 14
Beyond range	2021	12.6	131	–2.6	1	Nov 26	Mar 18
Beyond range	2022	13.5	153	–3.9	8	Dec 3	NA†

T first 30 d = mean temperature (°C) for the first 30 d of the autumn experiment inside cages (values were averaged across loggers); Autumn GDD = accumulated growing degree days^* during the autumn experiment; T Dec–Feb: mean temperature (°C) inside overwintering jars (averaged across loggers) for the months December, January, and February; Days below –16 = the number of days with daily minimum temperature inside overwintering jars reaching –16 °C, a temperature that induced mortal freezing in roughly half of individuals in ref. 39; Winter onset = the first date out of the first series of seven consecutive days with a mean temperature below 0 °C; Spring onset = the first date out of the first series of seven consecutive days with a mean temperature above 0 °C that came after the last sequence of seven days with a mean temperature below 0 °C.

^*The base temperature for calculating GDDs was 10 °C, and the cutoff temperature was 30 °C. The sum was calculated from 96 temperature measurements per day, with each contributing their fraction to each day’s GDD value. GDD sums from different loggers within a site were averaged.

^†With our definitions, winter did not arrive at the range interior site in 2021 to 2022, and it did not end before the end of the experiment at the beyond range site in 2023.

1.3. Growth Rate in the Laboratory.

In the laboratory experiment, we kept caterpillars individually in controlled temperatures and daylengths (a full factorial design with a warmer and a colder daily temperature cycle and two photoperiods) and weighed each individual repeatedly. On the last (49th) day of the experiment, caterpillars in the warm treatments were heavier than those in the cold treatments (F_1,284 = 66.4, P < 0.0001), but there were no significant differences among either photoperiod treatments (F_1,282 = 1.87, P = 0.17) or populations (F_3,25.7 = 1.99, P = 0.14). Nevertheless, between day 0 and day 31 (the chosen knot for our spline regression model; see Materials and Methods, Section 3.3), northern caterpillars grew faster than southern ones (Fig. 3), as indicated by a significant interaction between population and the number of days passed since starting the experiment (see Table 3 for all model statistics). After 31 d, the growth of northern caterpillars generally decelerated, and southern caterpillars caught up, leading to the similar end weights among populations.

Fig. 3.

Caterpillar growth during the laboratory experiment. The four treatments are shown in separate panels. Colored lines show growth curves (obtained by connecting data points from approximately weekly weighings) of individual caterpillars, and the black solid, dotted, and dashed curves show backtransformed model predictions from spline regression models fitted separately for each population–treatment combination. Three obviously directly developing individuals were excluded from the data. Sample sizes per treatment (excluding dead individuals) were 17 to 21 for Hässleholm, 21 to 25 for Vejbystrand, 24 to 28 for Katrineholm, and 7 to 10 for Rindö. See legend for population colors; S = southern population, N = northern population.

Table 3.

Statistics from the linear mixed model on the laboratory growth data

Variable	F	df	Residual df	P
Population	2.32	3	161	0.077
Temperature	10.5	1	1,040	0.001
Photoperiod	0.29	1	1,087	0.59
Days	949	1	1,251	<0.0001
Late days	54.0	1	1,252	<0.0001
Population × Temperature	5.89	3	412	0.0006
Population × Photoperiod	1.53	3	1,176	0.20
Population × Days	11.9	3	1,254	<0.0001
Population × Late days	35.0	3	1,247	<0.0001
Temperature × Days	0.001	1	1,265	0.97
Temperature × Late days	0.009	1	1,253	0.93
Photoperiod × Days	0.39	1	1,261	0.53
Population × Photoperiod × Days	4.15	3	1,270	0.006
Population × Temperature × Late days	3.93	3	1,242	0.008

P values below 0.05 are highlighted in bold typeface. The variable “days” is the number of days since the start of the experiment, and “late days” is a modification of days to account for the changing growth patterns after the spline model’s knot (31 d).

Caterpillars grew faster in the warm than in the cold temperature treatments, and the difference between northern and southern populations was more pronounced in the warm treatments (population–temperature interaction). There was no clear effect of photoperiod, and the significant three-way interaction between population, photoperiod, and days (the number of days since starting the experiment) might primarily be an effect of the short-day treatments being slightly warmer than the long-day treatments (Materials and Methods, section 3.3), which could explain the faster growth in short-day treatments.

1.4. Winter Survival.

Northern winters were colder than southern winters, and the 2022 to 2023 winter was colder than the previous one, particularly at the beyond range site (Table 2; see SI Appendix, Appendix 3 for additional temperature data).

Winter survival after the autumn experiment was lower in the 2022 to 2023 winter than in the 2021 to 2022 winter (Fig. 4A; χ²₁ = 48.9, P < 0.001). Survival also varied among sites (Fig. 4A; χ²₂ = 51.8, P < 0.001), being much lower at the beyond range site (P < 0.0001 for both pairwise comparisons) but not differing between the range interior and range margin sites (P > 0.99; see SI Appendix, Table S6 for full Tukey test results). Populations survived in similar proportions, regardless of site (χ²₃ = 0.50, P = 0.92 for the main effect of population; χ²₆ = 2.25, P = 0.90 for the population–site interaction).

Fig. 4.

Winter survival of (A) individuals overwintering after the 2021 and 2022 field autumn experiments and (B) individuals moved to field sites after the 2022 laboratory experiment. Error bars show 95% CIs. Sample sizes per population, site, and year were between 25 and 67 in (A) and between 17 and 54 in (B); all sample sizes are found in SI Appendix, Appendix 2, Tables S4 and S5. See legend for population colors; S = southern population, N = northern population.

Of the caterpillars overwintering at the range margin and beyond range sites after the laboratory experiment, many more (all) individuals died at the beyond range site than at the range margin site (Fig. 4B: χ²₁ = 71.4, P < 0.001), and, again, survival did not differ among populations (χ²₃ = 4.59, P = 0.20). Of the laboratory experiment treatments, photoperiod did not significantly affect winter survival (χ²₁ = 1.00, P = 0.32), but caterpillars that were reared in the colder laboratory treatments tended to survive better than those reared in the warmer temperatures (χ²₁ = 3.83, P = 0.050). Prewinter mass did not affect winter survival (χ²₁ = 0.00, P = 0.98).

2. Discussion

We reciprocally transplanted L. megera between their range interior in southern Sweden and their northern range margin to assess among-population differences in winter survival as well as in two seasonal timing traits that determine whether butterflies overwinter in an appropriate life stage. Simultaneously, we transplanted individuals beyond the current range of L. megera to reveal what limits the current range and test whether local adaptation to range margin conditions can facilitate further expansion. At all sites in both years, almost all individuals from all populations entered diapause. Hence, at least in these two years, a southern-type response to daylength cues would be unlikely to impede northward range expansion. Nevertheless, the slight among-population differences that we did find were in the same direction as previous laboratory results (23) and expectations following local adaptation to daylength conditions. Similarly, during development into diapause, caterpillars of northern descent consistently grew faster than southern caterpillars—indicating adaptation to short, northern growing seasons—yet we did not find clear fitness consequences of these differences. This indicates that the demonstrated genetic differences in diapause induction thresholds and larval growth rate (ref. 23 and this study) do not facilitate range expansion. Instead, overwinter survival beyond the range margin was universally poor, providing solid evidence for cold winters restricting the northern distribution of L. megera.

Winter survival did not differ between the range interior and range margin but was far lower (zero in the second winter) at the beyond range site, less than 100 km from the northernmost recorded L. megera occurrences. This holds for individuals previously reared either on site (Fig. 4A) or in common-garden laboratory environments (Fig. 4B), so larval conditions during the northern autumn do not explain the poor winter survival. Further, survival did not differ among populations, so there is no evidence for better winter tolerance having evolved in northern range-margin populations. Both the low winter survival beyond the current range and the lack of local adaptation to winter conditions match with earlier laboratory results (39). The variation in mortality among field sites (in the current study) and among laboratory treatments (in ref. 39) were also both due to differences in winter temperature, not winter length—in the current study, caterpillars overwintered for similar times at all field sites. Our temperature data (Table 2 and SI Appendix, Appendix 3) support this view; the beyond range site had the coldest and most frequent low-temperature extremes, and the much lower survival in 2022 to 2023 than in the previous winter is also associated with colder low-temperature extremes. The lack of local adaptation in response to cold winters contrasts with rapid cold tolerance evolution in some invasive insect populations (44–47). Perhaps evolution of cold tolerance at native cold range margins is constrained by trade-offs or physiological limits, whereas the founders of alien populations may come from native range interiors, where populations have not yet evolved to their cold tolerance limits.

Comparing the same populations used in our experiments, a laboratory common garden study (23) showed that northern caterpillars enter diapause in longer photoperiods than southern caterpillars, and this trait has evolved quickly in other insects as well (33–35). However, our reciprocal transplants show that, at the geographical scale of our experiment, late-summer temperatures and the timing of egg laying are likely much more important than local adaptation to daylength cues in determining whether caterpillars enter diapause. The autumn experiment was started when most wild eggs would be laid at the northern range margin (SI Appendix, Appendix 2, Fig. S1), and very few individuals developed directly without diapause. In stark contrast, most surviving individuals developed directly in the 2020 pilot experiment (SI Appendix, Appendix 1). As this pilot experiment was started approximately two weeks earlier in the season, it is likely to reflect the fate of eggs laid by the earliest females of the generation. Our combined results thus suggest that diapause thresholds can evolve rapidly even when selection acts only on this limited part of the generation or in especially warm years, when adults lay eggs earlier and these offspring develop faster than usual.

Unnoticed in previous studies of the system (23, 39), northern caterpillars grew faster than southern ones during development toward diapause, in the laboratory as well as at all field sites in both years of the field experiment. This is a clear example of countergradient variation, suggesting that northern caterpillars compensate for their short and cool growing seasons by growing fast. Countergradient variation (48) is found across many different ectotherms (49) and can explain latitudinal clines in larval growth rate in insects (50). Our results demonstrate that such a pattern can evolve quickly in range-expanding populations. The slower growth of southern individuals could be due to a cost of growing fast—such as physiological costs and higher predation risk (51)—or due to relaxed selection, if growing past a certain overwintering size offers few benefits. Higher prewinter mass often enhances winter survival (52–56), but, in agreement with ref. 39, we found no evidence for such an effect in L. megera. It may be that the tested individuals were all large enough to successfully overwinter and that fast growth has evolved through natural selection only in very cold years or on late-emerging individuals that have little time to grow before winter. Hence, diapause induction thresholds and growth rate may evolve through analogous selection on opposite extremes: either the early or late part of the generation and in either warm or cold years.

We have shown that although two important prewinter timing traits—thresholds for diapause induction and growth rate during development into diapause—have evolved rapidly during range expansion of a native butterfly, developmental timing seems to have limited importance, at least during the early stages of range expansion (dispersal and establishment; see refs. 57, 58). Instead, we found that these early stages are strongly limited by cold northern winters. Nevertheless, proper seasonal timing may benefit the last stage of range expansion—long-term persistence—by maximizing the number of individuals that enter diapause, even if the proportion that survives winter is low (59, 60). This way, the local evolution of seasonal timing seen within the current range may be important; winter mortality was high even in the more favorable winters within the species’ actual range, as seen in some other species as well (60, 61).

Our field study allowed us to capture both the full natural variation in photoperiod and close to the natural variation in temperature and precipitation. Compared to conditions in the wild, however, caterpillars experienced somewhat less-variable environments. The field cages used for the autumn experiments slightly buffered temperature relative to the surrounding environment (SI Appendix, Appendix 3), and we did not let caterpillars choose what plants to feed on or where to overwinter. The winter experiment may have been somewhat harsher than natural conditions, as we did not offer caterpillars fresh grass during winter [diapausing L. megera can feed (62)], but our among-site comparisons are conservative because we ended the winter experiment at almost the same time at all sites despite continued winter at the beyond range site. Thus, the evidence that winter conditions ultimately limit the range is compelling. Nevertheless, winter and summer performance interact (59–61, 63, 64), and summer conditions may be important for determining the timing of egg-laying, a factor whose effects we controlled for but did not investigate. Future research would benefit from considering the role of the full seasonal cycle in range expansions.

We used F1 offspring of wild-caught females, which enabled us to follow the natural phenology of L. megera, but this also introduced the possibility of transgenerational effects (65) influencing our results. However, a genetic basis for differences in diapause induction thresholds among these populations of L. megera was previously demonstrated with a common garden experiment on F2 offspring of wild butterflies (23). Further, while environmental differences experienced by the field-collected females could have influenced the growth rates of their offspring, summer temperatures differ little among our study populations (SI Appendix, Table S7). Our observations over several years of field work also indicate no major differences in other environmental factors (such as available host plants) known to induce transgenerational effects in Lepidoptera (65). Finally, the consistency of our results across both spatial (two populations per region) and temporal (two years) replication speaks against any major role of transgenerational effects in explaining our results.

Identifying what limits the range and inferring the importance of local adaptations for range expansion was possible because we transplanted both range interior and range margin individuals to the range interior, the range margin, and beyond the range—a study design that, despite its benefits, has seldom been used (66). Our confidence in our results is strengthened by the combination of laboratory and field studies, which led to more accurate conclusions than either would have alone. Genetic differentiation among populations, as shown for diapause induction thresholds in ref. 23, can easily lead to assumptions about its general adaptive importance, which our field study lends limited support for. Yet this differentiation clearly exists; it just would not have been detected with our field study alone. Similarly, our laboratory experiment—at the end of which southern individuals caught up with the initially faster-growing northern individuals (Fig. 3)—would have missed the field experiment’s clear and consistent among-population differences in prewinter mass. Future studies should try to both identify range-limiting factors and determine how local adaptations in various traits influence range expansions, and combined field transplant and laboratory common garden experiments will be powerful tools for this task.

Our results show that the growing season climate just north of the present range margin would allow L. megera to time its life cycle appropriately, but cold winters induce a level of mortality that prevents establishment. Because L. megera shows no evidence of evolution in response to winter cold, the further expansion of its northern range margin will likely continue at a rate dictated by climate change, specifically the rate of winter warming. In this regard, it is perhaps not surprising that range expansion is mainly constrained not by the most evolvable traits but instead by traits with limited evolutionary potential. A similar pattern of constraint-by-unevolvable-traits could characterize range expansions caused not just by climate change, but by a variety of anthropogenic and nonanthropogenic environmental changes. To understand and predict range expansions, it will be crucial to know both what traits limit range expansions and what traits may evolve during them—and to realize that these are not always the same traits.

3. Materials and Methods

3.1. Study Species, Populations, and Experimental Sites.

L. megera is a common grassland butterfly in large parts of the Western Palearctic (67), and its caterpillars feed on several common grasses (24, 25). The caterpillars are sensitive to daylength and temperature cues for diapause induction in their second and third larval instars and typically enter diapause in their third instar (23, 62, 68).

We collected mated females from four areas (populations) in Sweden: two (Hässleholm, 56.2 °N, 13.7 °E, and Vejbystrand, 56.3 °N, 12.8 °E) in the species’ southern-Swedish range and two (Katrineholm, 59.0 °N, 16.3 °E and Rindö, 59.4 °N, 18.4 °E) near the northern range margin (Fig. 1A). All butterflies were collected from elevations below 100 m above sea level. We transplanted F1 offspring of butterflies from each population to three field sites: Stensoffa Ecological Field Station, Lund (the range interior site; 55.70 °N, 13.45 °E; 20 m above sea level); Tovetorp Research Station, Nyköping (the range margin site; 58.95 °N, 17.15 °E; 35 m above sea level); and the village of Vassbo, Falun (the beyond range site), where we used two nearby sites (60.53 °N, 15.53 °E in 2021 to 2022 and 60.52 °N, 15.53 °E in 2022 to 2023; 125 to 130 m above sea level).

3.2. Autumn Experiment.

At each field site, we reared L. megera from egg on potted grass (Dactylis glomerata “Donata” grown from commercially available seeds) inside white mesh cages (Fig. 1C). If the original grass wilted or was mostly consumed, we added another pot of D. glomerata, sometimes obtained from the wild. The cages (which were randomly assigned among the four populations) stood approximately 2 m apart in 4 × 5 cage grids in flat areas of grassland (Fig. 1B), except at the beyond range site in 2021, where we had two grids (4 × 2 and 2 × 6 cages) with an about 10-m distance and 3-m elevation difference between grids. The cages were mostly unshaded, and the plastic windows faced north. We watered the plants using automatic, solar-powered dripping irrigation systems (SOL-C24 and SOL-C24L, Irrigatia).

We put eggs into the cages on August 20, 2021, and August 18 to 19, 2022 (SI Appendix, Appendix 2, Table S8). The eggs were laboratory-laid offspring of adult females collected from our study populations in early August (see SI Appendix, Table S9 for the numbers of wild-caught females). For each population in each year, we first mixed all eggs and then divided them into 15 groups of 20 eggs (five groups per site), yielding 300 eggs per population. We had fewer eggs from the northern population Rindö (234 in 2021 and 243 in 2022), so we divided these as evenly as possible among the 15 groups (SI Appendix, Table S9). At the field sites, we placed each group of eggs near the middle of each cage’s grass, within an approximately 2 × 2 × 2 cm mesh basket from which hatchlings could easily crawl onto the grass.

On October 15 to 27, 2021, and October 26 to 28, 2022 (SI Appendix, Table S8), we scored pupae and adults as directly developing and caterpillars as diapausing, except for one caterpillar that had clearly entered a direct development path (being about twice as heavy as the largest diapausing individuals). Dead caterpillars and missing individuals were excluded from analyses. To analyze prewinter body mass, we weighed all live caterpillars zero to two days after surveying the cages. Per site, all individuals were weighed on the same day, and those not immediately weighed were kept outside in Stockholm in 1L jars with fresh grass.

We used Fisher’s exact test to analyze among-population differences for the only year–site combination with variation in the binary response variable (diapause or nondiapause development). For the mass data, we used linear mixed models with mass as response variable. By backward elimination, we excluded those nonsignificant interactions for which we did not have hypotheses to test. The final model then included the explanatory variables year, site, population, the interaction between site and population, and the interaction between year and site, along with cage as a random effect. We generated P values using F tests with Kenward–Roger degrees of freedom approximation (69, 70) and type III sums of squares and used Tukey tests for pairwise comparisons.

3.3. Laboratory Experiment.

In autumn 2022, we reared siblings of the eggs used in that year’s autumn experiment in a laboratory experiment using four climate cabinets (KB8400-L, Termaks) with controlled photoperiods and temperature cycles. With one climate cabinet per treatment, we used two photoperiods (short: 12 h of light, 12 h of darkness; long: 15 h of light, 9 h of darkness) and two fluctuating temperature treatments that each followed a 24 h cycle (warm: 8.0, 11.5, 15.0, 18.5, 22.0, 18.5, 15.0, and 11.5 °C; cold: 5.3, 8.6, 12.0, 15.3, 18.6, 15.3, 12.0, and 8.6 °C; temperatures changed every 3 h). We based the treatments loosely on temperatures logged in September 2021 at the range interior site (mimicked in the warm treatments) and the beyond range site (mimicked in the cold treatments). In photoperiod, the treatments differed more than our field sites do, but both photoperiods should induce diapause in almost all individuals (23). Due to variation among climate cabinets, measured temperatures were slightly warmer in the short-day treatments than in the long-day treatments. Mean temperatures ± SD were 16.1 ± 4.1 °C in the warm, short-day treatment; 15.5 ± 4.5 °C in the warm, long-day treatment; 13.1 ± 3.8 °C in the cold, short-day treatment; and 12.5 ± 3.9 °C in the cold, long day treatment.

After the eggs hatched (in 17 °C and a 12 L:12 D photoperiod), we placed each first-instar caterpillar into a 0.5-L jar with Poa annua grass, whose roots reached a solution of water and liquid fertilizer (nitrogen, phosphorus, and potassium; “Växtnäring universal”, Plantagen Sverige AB) through a hole in the bottom of the jar. The grass was changed whenever it started wilting. We weighed each caterpillar weekly until it had been in the experiment for 49 d. For analyses, we excluded all individuals that died during the experiment, as well as three whose early growth indicated nondiapause development. The final sample sizes were 7 to 28 individuals per population and treatment (SI Appendix, Table S10).

To compare growth curves, we fitted a spline regression model with a knot separating early and late growth at 31 d after the start of the experiment. Thirty-one days was the breakpoint estimated with the R package segmented 1.6.2 (71) for the Rindö population in the warm, short-day treatment—the treatment–population combination in which caterpillars grew the fastest. This choice allowed accurate comparison of initial growth rates. The response variable was square root–transformed mass, and the explanatory variables were days (the number of days since the start of the experiment), late days (only the days that came after the 31st day, representing the change in slope after the knot), temperature, photoperiod, population, and a set of two- and three-way interaction terms (listed in Table 3). We chose the interaction terms based on backward elimination but retained the nonsignificant interactions temperature × days and population × photoperiod × days, which we specifically wanted to test. Further, we included the random effects family and individual (nested within family). We generated P values with type III Wald F tests with Kenward–Roger degrees of freedom approximation. Because this analysis focused on early growth, we also analyzed differences in final mass. Mass on the last (49th) day of the experiment was the response variable; temperature treatment, photoperiod, and population were the explanatory variables; and family was the random effect in our linear mixed model. All interactions (for which we did not have hypotheses to test) were excluded through backward elimination. We used type II Wald F tests with Kenward–Roger degrees of freedom approximation to generate P values.

3.4. Winter Experiment.

3.4.1. Winter survival after the autumn experiment.

After the autumn experiment, we placed all diapausing caterpillars in 1-L plastic jars (covered with mesh) with fresh Dactylis glomerata; the grass reached water in a below jar through a hole in the bottom. All caterpillars from one autumn cage were pooled into one jar. We kept the jars outside (covered from direct precipitation) in Stockholm until transporting them to their respective field sites on November 11 to 24 in 2021 and October 26 to 28 in 2022 (SI Appendix, Table S8). At each field site, we kept the jars on trays with high sides (which kept the jars upright; Fig. 1D) in the middle of the area used in the autumn experiment. To avoid precipitation filling the jars, we covered them with a plywood board, leaving about 5 cm between the jars and the board (Fig. 1D). We scored survival the following spring: on March 24 to 28, 2022, and March 30–April 6, 2023 (SI Appendix, Table S8). Spring arrived later in 2023 than in 2022 (Table 2), hence the later scoring. We scored survival based on appearance and immediate response to touch stimulus. In a few unclear cases, the caterpillar’s position was marked, and the individual was scored as dead if it had not moved anywhere after being left at room temperature overnight.

To analyze the binary survival data, we first fitted a full binomial generalized linear mixed model with year, site, population, and all possible interaction effects as explanatory variables, and then, by backward elimination, removed those nonsignificant interactions for which we did not have hypotheses to test. We kept the nonsignificant interaction between site and population, which tests for local adaptation to different winters. The overwintering jars were included as a random effect. For the beyond range site in 2022 to 2023, which had zero survival, we added one “alive” observation to each population to improve model performance. Our estimates of differences between the beyond range site and the other sites are thus slightly conservative. We used likelihood ratio tests with type II (because the interaction was not significant) sums of squares to generate P values and Tukey tests for pairwise comparisons.

3.4.2. Winter survival after the laboratory experiment.

For the winter 2022 to 2023, we also placed diapausing caterpillars from the laboratory experiment (Section 3.3) at the range margin and beyond range sites. Unlike for the field-reared caterpillars, we had individual prewinter masses of these caterpillars to test whether prewinter mass affects winter survival. After weighing, the caterpillars were placed individually in 0.5-L plastic jars with fresh Poa annua grass. The jars were covered with mesh and kept outside (covered from direct precipitation) in Stockholm for 5 to 11 d before transport to the field sites. On October 26 to 28, 2022, we placed 152 caterpillars (in their jars) at the range margin site and 151 caterpillars at the beyond range site, all covered like described above (Section 3.4.1). We assigned caterpillars to each site randomly but divided those reared in different temperature and photoperiod treatments evenly among sites. The relative positions of jars within field sites were also randomized. We scored survival on the same days as for the individuals overwintering after the 2022 field autumn experiment.

Our initial generalized linear mixed models estimated the variance of the random effect (“family”, the wild-caught mother) at or very near zero, regardless of what fixed effects were included. Thus, we left out the random effect and analyzed the binary survival data using a binomial generalized linear model with site, population, prewinter mass, temperature, and photoperiod as explanatory variables (temperature and photoperiod were the treatments in the laboratory experiment). We dropped the nonsignificant interactions for which we did not have hypotheses to test through backward elimination and generated P values using likelihood ratio tests.

3.5. Software.

We used R 4.2.2 (72) for analyses and the package lme4 1.1.32 (73) for fitting models. For statistical tests, we used the package car 3.1.1 (74), except for likelihood ratio tests—which we did with anova in base R and the package afex 1.2.1 (75)—and post hoc tests, which we did with emmeans 1.8.5 (76).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Experimental data, original temperature measurements, and the R script used for analysis are available in Zenodo at https://zenodo.org/records/15496741 (77).