Studied species

The larvae of P. icarus feed on small herbs in Fabaceae, and the species spends the winter in a late larval instar. It is commonly found in grasslands and other open areas over the whole of the European continent, except for some islands on the northern tip of Scandinavia at latitudes > 70° N (Dal 1978). Lycaenids are small butterflies generally considered to be relatively sedentary (Hughes 2000). Therefore we assumed that P. icarus would have the opportunity to be locally adapted and show local genetic life history adaptations to climatic gradients, as was earlier found in some other lycaenids (Nylin and Svärd 1991).

P. icarus is known to have two generations per year in Southern Sweden, e.g. on the Baltic island of Öland (locality L in Figure 2; Professor O. Leimar, Stockholm University, personal communication). We initially assumed that the species is univoltine in Northern Sweden, with a transition zone in Central Sweden, but the details regarding voltinism were revealed in a study of the museum collections of P. icarus (see below). P. icarus has been shown to respond to photoperiodic cues by plastically adjusting its size and development time according to perceived date in the season (Leimar 1996).

Figure 2.

Map of Sweden depicting the source populations for the common-garden experiment: L = low latitude (Öland), I = intermediate latitude (Stockholm), H = high latitude (Borgsjö). One longitude (degrees East, vertical line) and three latitudes (degrees North, horizontal lines) given.

Museum collection

The Swedish Natural History Museum has a collection of some 500 individuals of P. icarus collected from different latitudes in Sweden. The collection contains butterflies sampled between the years 1880 to 2001. Latitude and date of collection were determined from labels. Wing lengths (base to tip) of all individuals were measured under a standard laboratory microscope (6x). The average of right and left wings was used.

For purposes of illustrations and statistical analyses of size patterns, latitudes for collection sites were rounded off to nearest integer degree, whereas for purposes of studying the shift in generation number they were rounded off to nearest half degree. Rounding was done to increase sample sizes in latitudinal categories. We excluded the extreme latitudes of 55° and 69° due to very low sample sizes.

Common-garden experiment

Females were collected in the wild from three populations for rearing of their offspring in the laboratory (Figure 2). Sixteen females were collected from the low-latitude population (two sites on Öland, 56.5° N), four from the intermediate population (Stockholm, 59.5° N) and two from the high-latitude population (Borgsjö, 62,5° N). In Britain P. icarus shows a shift from a predominantly bivoltine to a univoltine phenology approximately at latitude 54°, but with variation among years (Asher et al. 2001). In Sweden the transition zone seems to be even more ambiguous and individuals following a bivoltine pathway can regularly be found at higher latitudes than in Britain (Professor O. Leimar, personal communication, and see Results from the museum collection). However, the butterfly is clearly normally univoltine at the high-latitude site and generally bivoltine at the low-latitude site. In the intermediate area a second generation is not uncommon but neither is it a dominant strategy.

Length of the favourable season follows a strong and clear north-south pattern within Sweden (Nylin and Svärd 1991) due to a combination of factors: the effect of latitude itself combines with generally higher altitudes and more inland climates in the north. The low-latitude and the intermediate sites are situated close to the Baltic Sea. The high latitude sampling area is more continental, about 80 km inland and for instance has a visibly different flora compared to the coastal regions at the same latitude (Nygren, personal observation). This implies that despite the uniform latitudinal distances between the tree sampling areas, the high latitude sample is actually climatically more distant to the intermediate area than suggested by the geographical distance alone. The location of the low-latitude sites on an island should even further enhance the climatic distances between sites, in terms of day-degrees available in the favourable season.

To control for phenological changes in host plant quality, collected butterflies from all three sites were reared simultaneously. Wild-caught females and later eggs from low latitudes were stored for a short time period in a cold room to delay larval hatching and achieve synchronization. The experiment was initiated over a ten-day period in the second half of July. Newly hatched larvae were individually placed in plastic jars containing fresh bird's foot trefoil, Lotus corniculatus L. (Fabales: Fabaceae) and reared in a climate chamber at a photoperiod of 22: 2 LD at 22°C (low latitude n = 320, intermediate n = 52 and high n = 34). The plants were checked daily and changed when signs of wilting were found. Larvae of different latitudinal origin were mixed in the climate room to prevent position effects, and also moved in the room at least every second day.

Date of pupation was noted, and two days after pupation the pupal mass was measured and individuals were sexed. Wing length of adults was not taken for the laboratory-reared individuals, but pupal mass typically correlates well with adult weight and adult wing length among individuals within sexes in butterflies (Nylin 1992; Kemp 2000; Fric and Konvicka 2002) and moreover only differences among categories that were large enough to make the body measure chosen more or less irrelevant were used.

Growth rate was not measured independently of pupal mass, but for illustrative purposes a measure of growth rate was calculated as the natural logarithm of pupal mass divided by larval development time (in days).

All statistical analyses were performed with STATISTICA 1999 Edition, Kernel release 5.5.

Museum collection

Above latitude 60° N, few P. icarus butterflies from a potential second generation, i.e. with a late collection date, could be found in the museum collection. At these latitudes, collecting dates were distributed approximately normally with a peak at mid-July. At latitudes below 60° N, the distributions of collecting dates had two peaks (late June and early August). Consequently, the butterfly appears to have shifted from a univoltine to a bivoltine life cycle at this latitude in Sweden. However, below latitude 60° the butterfly probably is only partly bivoltine as not all individuals develop directly, and this generation may not occur in all years at all latitudes (the collection was not large enough to permit further analysis of phenological details). It is unclear where an obligate bivoltine lifestyle takes over, if ever within Sweden, but below lat. 57.5° the second peak was consistently large, and hence bivoltinism appears to be the dominant strategy. These results confirmed our initial assumptions regarding where the bivoltine and univoltine populations would be found in the field.

In the “bivoltine area”, below latitude 60° N, the butterflies were classified as belonging to either the first or the second generation. There were some individuals collected between the two peaks that may belong to either of the tails of the two overlapping generations or in some cases represent a single generation in a cold year. These butterflies were excluded from relevant analyses. Since the generations differed in size in bivoltine areas, we show results of depicting latitudinal patterns with these more southern areas represented by either all individuals (Figure 3a), only the first generation from over-wintering larvae (Figure 3b) or only the second, directly developing, generation (Figure 3c). In the field collection female size was consistently smaller (Figure 3a—c; GLM ANOVA, n = 522, p<0.001); for this reason the sexes are shown separately.

Figure 3.

Wing length of Polyommatus icarus collected at different latitudes in Sweden. Arrows show the approximate shift from univoltine to bivoltine phenology, a) Average for both generations in the bivoltine area, b) only the over-wintering (first) generation in all areas, c) only the directly developing (second) generation in the bivoltine area. Bars show standard errors.

When all individuals from both generations were included there was no clear size pattern according to latitude (Figure 3a). When only insects with a history of hibernation were included there was a pattern faintly reminiscent of the predicted saw-tooth pattern (Figure 1), albeit with more smooth peaks and troughs (Figure 3b), and there was a significant effect of latitude on size (GLM ANOVA, n = 275 individuals, 10 latitudes, p<0.001; sex included as a factor). The butterflies were fairly large in the southernmost areas of Sweden, with size decreasing to the north. As predicted from the saw-tooth model, sizes increased again to a local maximum at around latitude 60° N, where the predominantly univoltine area begins, followed by a new decrease further north. Finally, size again increased at the very highest latitudes (Figures 3a—c). When only the second generation was chosen to represent the bivoltine areas (Figure 3c) the pattern was similar, but the maximum at latitude 60° N was even more evident. This is because wing length was always smaller for the directly developing generation than for the over-wintering generation in the bivoltine area below 60° N (compare Figures 3b—c; GLM ANOVA, n = 303 individuals, 4 latitudes, p<0.001; sex included as a factor).

Common-garden experiment

Development time did not differ between butterflies reared from individuals collected from the two low latitude sampling areas, i.e. those on the island of Öland (t-test, p = 0.84). Pupal mass did differ significantly (t-test, p<0.05), but variation among latitudes was much larger. For simplicity of analysis and illustration, these two southern sites were therefore pooled in all statistical analyses and illustrations.

Female butterflies reared from individuals collected from southern and intermediate areas had lower pupal mass than males, but this was not true for the strain from the high latitudinal areas (Figure 4a); this interaction was however not statistically significant (GLM ANOVA, two-way interaction between sex and latitude as categorical predictors, n = 304, p=0.095). When the interaction term was removed, the effect of sex was significant (p<0.001). If latitude was instead treated as a continuous predictor, the interaction was significant (p<0.05), as well as the effects of sex (p<0.001) and latitude (p<0.001). The heaviest pupae were produced in the southern strain and the lightest in the northern strain (Figure 4a).

Interestingly, the latitudinal variation in mass did not correspond to variation in larval development time in the simple way assumed by Roff's theory on latitudinal patterns. Time spent in the larval stage differed significantly according to the latitudinal origin of strain (Figure 4b; GLM ANOVA, p<0.001). However, larval development time was in fact shortest in the southern strain and longest in the northern, with the intermediate strain showing intermediate values (Figure 4b). There was no significant effect of sex on larval development time (p = 0.96). Pupal development time differed significantly according to latitudinal origin (GLM ANOVA, p<0.05) but showing a pattern opposite to that for larval time (not shown). However, variation in pupal time was much smaller, so that the pattern for total development time was similar to that for larval time (effect of latitude p<0.001, not shown). The pattern of shorter development time yet larger mass in the southern areas corresponds to faster growth in southern strain, as illustrated in Figure 4c (GLM ANOVA, effect of latitude p<0.001). The interaction between sex and latitude can also be seen for growth rate, but it is not significant (p = 0.46). Effect of sex was not significant with the interaction included in the model (p = 0.76), but when the interaction was removed a significant difference could be seen, with males growing faster than females (p<0.01). If latitude was instead treated as a continuous predictor, the interaction was significant (p<0.05), as well as the effects of sex (p<0.01) and latitude (p<0.001).

There was a significant effect of family (p<0.001 in all cases) in the analyses described above, if family was added as a random factor nested in latitude, strongly suggesting a genetic component behind the observed variation. However, other results did not change; this is an indication that the geographical patterns may be robust even though there was a low number of families from two of the populations.

In the F1 generation no correlation between development time and mass among individuals could be found in the low (n = 253, p = 0.9) and intermediate (n = 41, p = 0.17) latitudinal sample, but in the high latitude strain they were positively correlated (n = 15, r² = 0.36, p<0.05).

Museum collection

The geographical position of the shift from a univoltine to a bivoltine life cycle by P. icarus in Sweden is somewhat ambiguous (see also Asher et al. 2001 regarding a similar situation for populations in Britain). This can, at least partly, be explained by gene flow, preventing exact local adaptation. Climatic variation among years is another likely cause; early instar larvae will respond in a plastic manner to cues signalling how much time is left in the season, with different outcomes in different years. In warm years this will result in two generations and in cold years in a single generation, so that the “transition zone” (as seen in the combined museum material) will extend over large geographical areas. This is not simply an artefact of pooling years, but in a real sense represents the temporally fluctuating environment to which local populations must adapt. In other words, most genetic differences between areas must reflect the probability of a second generation in a given area, rather than a fixed life cycle pattern.

The size of the P. icarus decreased towards higher latitudes with shorter seasons in the bivoltine area, but then increased when they shifted to a univoltine lifestyle, according to predictions. Interestingly, this pattern was not evident when the generations were pooled in the bivoltine area (as is the usual procedure when phenology is not clear, e.g. Nylin and Svärd 1991), because the generations differed in size (see also Fric et al. 2006). This size difference in itself strongly suggests that adult size is, in part, determined by plasticity, since the major developmental pathway - diapause or direct development - followed by an individual is an outcome of high-level plasticity according to seasonal cues such as photoperiod (Nylin and Gotthard 1998).

Observations of this kind also suggest that size trends in insects may have been underestimated in the scientific literature because of the difficulties introduced by generation differences, and furthermore introduce a problem regarding which pattern should be used to test Roff's hypotheses (1980, 1983). The pattern seen with only the first generation of adults (which has experienced larval hibernation) representing the bivoltine area (< 60° in Figure 3b) could be argued to be the most correct test. This is because the univoltine individuals further north (≥ 60°) always over-winter and they are thus more comparable to the first generation of adults in the south. Hibernation is associated with special demands on an insect and could incur metabolic costs visible as reductions in adult size or, conversely, larger size if winter survival is positively size-dependent (Kemp and Jones 2001).

On the other hand, it is the last generation of the favourable season that has the most direct information on how much time remains before winter in the bivoltine areas, and if size is partly plastically determined this generation may show the clearest size responses as a result of end-of-season cues from the environment (Roff 1983). Plasticity theory (Leimar et al. 2006) suggests that in such ambiguous cases, information from the environment should affect what phenotype will result from an individual's genotype. In a sense, the genotype of an individual organism provides it with information on what was the fittest life cycle in past environments, but only in a probabilistic way, and in some situations the current environment provides more precise information that should be exploited if possible.

Which generation should for these purposes be considered the “last of the year”, in the case of a bivoltine insect with late larval winter diapause such as P. icarus in Southern Sweden? This is not entirely clear either, but we suggest it is the second adult peak (Figure 3c). The first peak consists of individuals that spent the very last part of the previous summer as half-grown larvae and thus had a growth period at a later date than is ever the case for the second peak. However, the second peak consists of individuals that pupated later in the summer. Since growth of insects typically follows exponential or power functions, large size differences can occur due to small differences in the number of days spent in the last larval stage (Nylin 1994). This suggests that plastic size variation should mostly accumulate just before pupation, by the “decision” to continue growing or not as suggested by Leimar (1996) for P. icarus. The same conclusion can be reached based on the physiological model of body size variation in Lepidoptera presented by Davidowitz and Nijhout (2004), as they suggest that body size variation is the product of growth rate variation, coupled with variation in critical mass for pupation, and in the interval to cessation of growth, once the critical mass has been reached. The fact that the second generation is consistently smaller in size supports this interpretation, since it may suggest that this generation has to “pay the cost” of a bivoltine pathway by pupating early to ensure that enough time in the season remains for its offspring to reach the hibernating stage.

An unexpected pattern was found within the univoltine area. As predicted, sizes decreased going north within each of the areas of voltinism. However, further north in the univoltine area, P. icarus again increased its size, corresponding to field observations (K. Fiedler, University of Vienna, personal communication). If this pattern is to be explained by the saw-tooth theory, it would suggest that above latitude 65° P. icarus has adopted a two-year life-cycle (Johansson 2003). However, a clear possibility remains that environmental factors and biological responses other than those included in this theory come into play at such extreme latitudes.

Common-garden experiment

Regarding the laboratory experiments, note that the low number of individuals used to create the strains for the intermediate and high latitude populations means that we cannot claim to have a good picture of the genetic make-up of these populations; more extensive sampling will be needed. Nevertheless, we believe that the differences between geographical localities may prove robust, for two reasons. First, there was little difference between two sites at the more extensively sampled low latitude locality. Second, adding family as a factor to statistical investigations did not change the overall results.

The highly significant effects of latitudinal origin of strain demonstrate that there is a strong genetic component to the observed life history variation within Sweden, not only in size/mass but also in development time and growth rate. The potential for local adaptation is also suggested by the apparent presence of genetic variation for these life history traits, visible as significant variation among families. The presence of genetic latitudinal patterns in a common garden environment lead us to propose that the life history observed in a given strain reflects adaptation to the probability that the source population (averaged over years) will have enough time to complete two generations. This probability may well increase to lower latitudes in a near-linear fashion over the area covered by the three populations, and we propose that the decreasing development times and increased growth rates in the lower latitudes (Figure 4b—c) reflect this higher probability of a second generation. At the high latitude site a second generation is very rare or non-existent, hence these butterflies probably always have more than enough time to complete the single generation per year (at even higher latitudes the situation might be different, see below). Accordingly, it is likely that at this site the butterflies have been selected for slow growth, since fast growth could be costly (Gotthard et al. 1994, Gotthard 2000). Conversely, at the site from the lowest latitude it is likely that P. icarus has been selected for fast growth, because there is generally enough time to complete two generations, but only barely.

The strain developed from individuals collected at the intermediate latitudes, were situated in a zone where development is not predictably uni- or bivoltine, and showed intermediate growth rates. This may be a result of selection from an intermediate probability of a second generation; however, gene flow from areas with more predictable life cycles probably also contributes. The most adaptive life history in this situation could be very fast growth under direct development (not the observed intermediate rates, because they have even less time than the southern population to complete two generations) and very slow growth under development to diapause. Although differences between developmental pathways in this general direction are commonly found in butterflies and other insects (Nylin and Gotthard 1998) gene flow and other constraints on reaching the optimal reaction norm (such as genetic correlations among developmental pathways; Via and Lande 1985) may cause the intermediate growth rate and development time that we observed.

It is interesting to note that although patterns of pupal mass according to latitudinal origin in the common-garden experiment followed a geographical pattern, the highest mass occurred at the lowest latitudes (Figure 4a). The field data show no such size pattern in either the first or the second generation, for these latitudes (compare Figure 3b—c with Figure 4a). Moreover, this pattern of higher mass at lower latitudes is not reflected by longer development times, the basic trade-off assumed by Roff (1980, 1983), but rather the opposite. Instead the more time-stressed butterflies at lower latitude grew faster (Figure 4c), achieving high mass in a short time. This can be seen as an example of counter-gradient variation (Conover and Schultz 1995), but not (as is typical) in relation to field temperatures in the source populations, as this would have lead to the opposite growth rate cline. This is noteworthy because counter-gradient variation in response to temperature and time stress often co-vary and cannot readily be decoupled; this is the situation in, for instance, the common frog Rana temporaria where frogs from Northern Sweden grow faster (Laugen et al. 2003). It should be noted that the three sites for the common-garden experiment did not include such a time-stressed high latitude population, where there is just enough time for a single generation. At such sites we would expect laboratory growth rates to increase again, as in Alaskan Papilio canadensis butterflies when compared to Michigan populations at the same temperature (Ayres and Scriber 1994).

Combining the evidence

We suggest that the discrepancy between field and laboratory patterns in P. icarus can be ultimately explained by the fact that body size is the complex product of growth rate and development time (Wiklund et al. 1991; Abrams et al. 1996; Nylin and Gotthard 1998; Davidowitz and Nijhout 2004) and proximately by the constant laboratory environment. In the field, seasonal cues are available to signal the progress of the season and sometimes the need to cut development short at the expense of final size. In the common-garden experiment such cues were absent (other than a constant long photoperiod and high temperature) suggesting that such cues of high summer induced direct development in all individuals. Pupal mass was then free to reach some value that was, at least in part, a function of larval growth rates. Shorter photoperiods were not included within the scope of the present study, because the expected variation in diapause induction within and among sites - with concurrent life history variation as noted above — would have added further to the complexity of interpretation. Short photoperiods can be expected to speed up growth in individuals destined for direct development, but not do so to the same extent in those destined for diapause (Wiklund et al. 1991; Nylin 1992; Leimar 1996).

Similarly, the observed interaction between sex and site of origin for pupal mass (not seen in the field material) and growth rate demonstrates how observations of sexual size dimorphism in the laboratory may in part be the result of variation in growth rate, rather than reflecting selection on mass per se. This suggests that the lack of sexual differences in growth rate and mass - or even reversed sexual difference - in the rearing of the northern population (Figure 4a and c) reflects release from selection for protandry under direct development in this univoltine population (Wiklund et al. 1991). Here, males can achieve protandry by simply breaking diapause early; hence in obligate univoltine populations there is no need for high larval growth rate in males. It has previously been shown for P. icarus (Fiedler and Hölldobler 1992; Leimar 1996) that the degree of sexual size dimorphism in this species is plastic. It has been demonstrated here that such laboratory results may not translate directly into size variation under natural conditions, although undoubtedly they demonstrate the presence of sexual differences in reaction norms for life history traits, and can illuminate how life history decisions are taken according to sex.

Despite evidence of growth rate variation in the laboratory, offsetting the correlation between final size and development time assumed by Roff (1980, 1983), in the field we could still see a saw-tooth pattern fully consistent with his predictions. P. icarus is a relatively large insect restricted by season length and, hence, is predicted to follow the converse to Bergmann's rule, and its extension, the saw-tooth theory (Chown and Gaston 1999; Blanckenhorn and Demont 2004). Evidently it does so in the field albeit not in a common environment. The grasshopper Chorthippus brunneus provides an interesting comparison. In this species, time-constrained northern populations grow faster but have also evolved a lower critical mass (fewer instars) and thus shorter development time and lower adult mass (Telfer and Hassall 1999), in line with both Roff's mechanistic assumptions and with his theoretical predictions. In another grasshopper, Omocestus viridulus, populations from different altitudes grow at the same rate but again insects from time-stressed populations have both a shorter development time, fewer instars and lower final mass in the laboratory (Berner and Blanckenhorn 2006). Future studies may reveal whether this stronger role for genetic variation in development time, as measured in the laboratory, is a general pattern explaining size variation in grasshoppers. An indication that this may perhaps not be the case is the fact that a review of intraspecific variation in the number of larval instars in insects (Esperk et al. 2007) found that such variation only rarely functions to couple development time positively with adult mass, as envisaged by Roff (1983) and as found in these grasshoppers.