Intersexual differences in density‐dependent dispersal and their evolutionary drivers

Abstract

Dispersal is well recognized as a major driver of evolutionary processes in local populations. Nevertheless, dispersal abilities should also be perceived as a life‐history trait, being subject to evolutionary changes in response to various drivers. Empirical studies investigating these drivers rarely consider that they may influence male and female dispersal differently. The purpose of our study was to document intersexual differences in density‐dependent emigration from local habitat patches. As a model system, we used a metapopulation of Maculinea (Phengaris) teleius butterfly, in which densities of both sexes vary greatly throughout the flying season. Following intensive mark–release–recapture surveys, the parameters and predictors of dispersal were analysed with the Virtual Migration model and the multi‐state recapture model. The emigration rate in males was substantially higher in the early season, especially at smaller habitat patches. With the proportion of females increasing with the season progression, males became reluctant to emigrate from their natal patches. In turn, higher female emigration in the later part of the season was most strongly associated with female tendency to reduce intraspecific competition experienced by their offspring. Our findings provide evidence for the impact of reproductive strategies on dispersal in both sexes. The difference in reproductive strategies of males and females explains sex‐biased dispersal in different parts of the season, which carries important implications for metapopulation functioning.

Abstract

Our findings give a perspective on the impact of reproductive strategies on dispersal in both sexes. With the proportion of females increasing with the season progression, males became reluctant to emigrate from their natal patches. In turn, higher female emigration in the later part of the season was most likely triggered by female tendency to reduce intraspecific competition experienced by their offspring.

Open in new tabDownload slide

INTRODUCTION

Many species nowadays live in increasingly fragmented landscapes, forming metapopulation systems with discrete habitat patches inhabited by spatially separated local populations (Hanski & Gaggiotti, 2004). In such a scenario, dispersal represents the crucial process that drives gene flow, thus constituting one of the major drivers of evolutionary processes in local populations such as allele frequency changes or founder effects (Clobert, Baguette, Benton, & Bullock, 2012). At the same time, dispersal abilities should also be perceived as life‐history traits of organisms, affected by genetic changes as well as by phenotypic plasticity (Clobert et al., 2012; Hanski, 1999; Hanski & Gaggiotti, 2004).

Individuals may be inclined to move from one habitat patch to another in order to avoid predation or competition for resources, as well as to increase opportunities to find more (or better quality) mating partners. Despite the potential benefits, dispersal also induces substantial costs that dispersing individuals may have to pay. Bonte et al. (2012) suggested different possible sources of such costs. Direct costs are represented by time investment and energy expenditure to move between habitat fragments, as well as by the risk of dying or suffering an injury during dispersal. Additionally, indirect costs of decreasing individual fitness through ending up in a worse habitat may be expected.

Among the empirical studies that investigated individual variation in dispersal in response to various drivers, few considered that male and female dispersal behaviour may be differently influenced (but see Casacci et al., 2015; Goff, Yerke, Keyghobadi, & Matter, 2018; Trochet et al., 2016). They also tended to neglect the evolutionary implications of such intersexual differences, despite the fact that the existence of strongly sex‐biased dispersal is predicted by theoretical works on dispersal evolution (Gros, Hovestadt, & Poethke, 2008; Li & Kokko, 2019; Perrin & Mazalov, 2000).

Several evolutionary reasons for the occurrence of intersexual differences in dispersal behaviour have been suggested (Li & Kokko, 2019; Matthysen, 2012). The avoidance of inbreeding plays a primary role in this respect (Matthysen, 2012; Moore & Ali, 1984), followed by differences in sex‐specific requirements which drive intraspecific competition (Perrin & Mazalov, 2000; Waser, 1985). Such a competition can be expressed as male–male conflict (for access to a sexual partner), female–female competition (for food and convenient habitats for laying eggs) or male–female competition (where the fitness of one sex increases at the expense of the opposite sex; Trochet et al., 2013).

In order to maximize their reproductive success, both sexes adopt different behavioural strategies. Li and Kokko, in their recent review (2019), point out that an important albeit understudied aspect of dispersal is the extent to which animal movements are interpretable as a direct consequence of mate‐searching strategies. The classic literature (Greenwood, 1980) associates different patterns of sex‐biased dispersal with mating systems, with female‐biased dispersal typically occurring in monogamous species and male‐biased dispersal in polygynous ones. One sex is predicted to carry all the cost load of dispersal, since in order to avoid the risk of inbreeding it is sufficient that individuals of one sex emigrate from the natal sites. On the other hand, if mating opportunity was the only factor triggering dispersal of a particular sex, then philopatry would be a poor choice for the other sex individuals. In fact, if one sex is selected for dispersal, the natal patches will have a surplus of the less dispersing sex and this would create a reactive selection on them to disperse more as well (Shaw & Kokko, 2014). Moreover, the relation between polygyny and male‐biased dispersal is not straightforward in the models where local resources limit female reproduction (Li & Kokko, 2019). The above rationale highlights that it is important to consider the mobility of two sexes separately in the context of variable conspecific densities and individual decision‐making rules.

The reproductive success of an adult female should primarily depend on its ability to secure resource for the offspring, whereas for a male such success mostly reflects its ability to find a fertile female and successfully mate. If males try to maximize the number of encounters with females and to minimize male–male competition for females, then their emigration from local habitat patches should rise with increasing male density and decreasing female density.

In turn, female tendency to reduce intraspecific competition experienced by their offspring should promote emigration in case of high female densities and low resource availability. Moreover, female propensity to emigrate may also be influenced by male density. A positive effect of density on emigration rate has been demonstrated in already‐mated females fleeing from harassment by overabundant males (Baguette, Vansteenwegen, Convi, & Neve, 1998; Odendaal, Turchin, & Stermitz, 1989; Shapiro, 1970). In such a case, male‐biased sex ratio or simply high abundance of males may trigger female dispersal as an escape behaviour performed in order to avoid the risk of potential physical injury caused by unwanted male courtship or the loss of energy invested in deterring the courtship. However, sexual harassment‐driven dispersal of females appears to be a relatively rare phenomenon when compared with dispersal to reduce intraspecific competition for resources experienced by offspring (Baguette, Clobert, & Schtickzelle, 2011).

Consequently, it is clear that the link between mating strategies and sex‐biased dispersal is complex and may depend on various factors exerting simultaneous impacts. Relying on the above rationale, the main purpose of the present study was to analyse intersexual differences in dispersal in response to temporal changes in conspecific density. As a model species, we used a specialist butterfly Maculinea (= Phengaris) teleius (Bergsträsser, 1779), which is known to experience strong fluctuations in the abundance of adults throughout the flight period (Nowicki, Witek, Skórka, Settele, & Woyciechowski, 2005; Timus, Czekes, Rákosy, & Nowicki, 2017). Furthermore, as in many temperate zone butterflies, the within‐season changes in local densities of both sexes in this species are asynchronous and predictable due to the phenomenon of protandry (Wiklund & Fagerström, 1977). Adult males emerge earlier and fly in high numbers in the first half of the flight period but become less abundant afterwards, whereas females prevail later in season. Consequently, we tested the following hypotheses: (1) male dispersal decreases with the flight period progression, and it is predominantly influenced by the availability of mating partners; (2) female dispersal increases with the flight period progression, and it is primarily triggered by the diminishing resource availability. Alternatively to the latter hypothesis, we also considered the possibility that female dispersal is mainly caused by male harassment, and if so it should be positively related to male density and generally decrease with the flight period progression.

MATERIALS AND METHODS

Study species

Maculinea (Phengaris) teleius is a univoltine species, with adult flying season in Europe normally extending from early July to mid‐August (Nowicki, Bonelli, Barbero, & Balletto, 2009; Nowicki & Vrabec, 2011). Being classified as a hygrophilous species, it occurs in wet meadows where both the Sanguisorba officinalis foodplants and ants of the genus Myrmica, required for the development of the Maculinea larvae, are present (Thomas, 1995). Females prefer greenish, not fully developed flowerheads of the great burnet plant (S. officinalis) for oviposition, almost always laying only one egg in a given flowerhead (Figurny & Woyciechowski, 1998). After a short period of feeding on foodplants, Maculinea larvae drop to the ground and are adopted by ants, representing specific species of the genus Myrmica, and they complete their development in ant colonies, acting as social parasites (Elmes & Thomas, 1992; Thomas, 1995).

Specific foodplants, also serving as primary nectar source for adults, and specific Myrmica ant hosts are thus the two essential resources required by M. teleius. Due to the patchy distribution of these resources, the focal species typically occurs in classic metapopulation systems, with discrete habitat patches easy to define (Dierks & Fischer, 2009; Nowicki et al., 2007, 2009). Local populations of M. teleius are usually small, but in appropriate conditions the species can reach very high densities of several thousand adult individuals per hectare in the season (Nowicki, 2017; Nowicki et al., 2007). The adults are characterized by relatively low mobility when compared with other butterfly species (see Bink, 1992, categorizing the mobility of 145 butterfly species) with an average inter‐patch movement distances typically in the range of several hundred metres (Nowicki & Vrabec, 2011; Nowicki et al., 2014).

Study area

The study was conducted within a large meadow complex located in the Vistula River valley on the south‐west outskirts of Kraków city, southern Poland (282.9 ha; 50º01'N, 19º54'E). The complex consists mostly of Molinion wet meadows and lowland hay meadows representing the Arrhenatherion elatioris community, with occasional Festuco‐Brometea xerothermic grasslands occupying elevated land fragments.

The entire study system includes ca. 50 S. officinalis patches, of variable area and isolation, inhabited by local populations of M. teleius. Our extensive study in 2012 involved a large fraction of the system, with 20 foodplant patches surveyed, of which two relatively isolated ones were also selected for a highly intensive survey in the following year. The foodplant patches form a classic metapopulation, with local M. teleius populations acting as independent demographic units (Nowicki, 2017; Nowicki et al., 2007). The surveyed metapopulation also fulfils further requirements of the study, namely the investigated patches were free from direct human impacts, either negative (e.g. habitat destruction) or positive ones (e.g. targeted conservation management). All the habitat patches were mapped with high‐precision GPS, and their spatial parameters were subsequently derived in GIS software ArcGIS and Idrisi for Windows (see the map in Figure S1).

Field sampling

The field surveys, carried out in two consecutive seasons (2012 and 2013), were based on intensive mark–release–recapture (MRR) sampling. Butterflies were captured using a butterfly net, and each individual was marked with a waterproof marker (Staedtler Lumocolor 313) by writing a unique number on the underside of its back wing. After marking, the specimens were immediately released at the place of their capture. For each individual, the assigned number along with the sex, catch time and code of the habitat patch was recorded. The MRR sampling was performed between 9:00 and 17:00 on a daily basis with few gaps due to unfavourable weather conditions. To ensure adequate sampling effort, the time spent in each patch was adjusted to its area and the abundance of butterflies flying (based on the experience from earlier years), varying from 1.5 to 4 person‐hours per day, which assured daily capture probabilities of ca. 25% individuals per patch. The order in which patches were visited each day was randomly changed throughout the season in order to avoid sampling the same patch at similar hours every day. This procedure allowed to avert potential biases in the frequencies of butterfly movements recorded between particular patches due to the fact that butterflies’ activity may be influenced by daily weather patterns, especially temperature (cf. Cerrato, Lai, Balletto, & Bonelli, 2016).

During the first campaign aimed at documenting the differences in dispersal parameters of both sexes between early and late season, the sampling was conducted at 20 habitat patches (size range: 0.02–4.54 ha, inter‐patch distance range: 0.18–4.73 km) from 10‐Jul‐2012 to 10‐Aug‐2012, thus comprising almost the entire flight period of the focal species. Although the very beginning and the very end of the flight period were not surveyed because of logistic constraints, this had very little effect on the sample size collected since very few individuals flew in those periods. For the second objective of our research, that is the detailed analyses of within‐season dynamics of emigration rate in males and females, a more intensive sampling was concentrated on two habitat patches (patch A: 4.54 ha, patch B: 1.07 ha) in the following year. The field activities lasted from 08‐Jul‐2013 to 14‐Aug‐2013 to cover the whole flight period of the butterfly. The two patches were selected so as to meet the assumptions of the planned analysis, that is they were located close to each other with the separation distance of ca. 23 m but fairly isolated from all other patches in the investigated system. As indicated by the frequencies of butterfly movements recorded in 2012, such a location of the two patches ensured relatively frequent movements between them, but hardly any movements occurring to (and from) other patches in the system.

Estimation of dispersal parameters

The parameters of dispersal within the investigated metapopulation were estimated separately for males and females in both early and late season based on the MRR data collected in 2012. The early season was defined as the period from 10‐Jul‐2012 to 28‐Jul‐2012 and the late season as the period from 31‐Jul‐2012 to 10‐Aug‐2012. Although such a division was arbitrary to some extent, it roughly corresponded to the two halves of the flight period and, more importantly, it reflected the natural split of the flight period by a prolonged spell of heavy rain between 29 and 30‐Jul‐2012 that only a small number of adults managed to survive. The few butterflies captured in both parts of the season (13 males and 7 females) were excluded from the analysis.

The estimates of dispersal parameters were derived with the help of the Virtual Migration (VM) model, which represents a well‐established standard for analysing dispersal using MRR data (Hanski, Alho, & Moilanen, 2000). The model requires the following: (a) individuals inhabit a network of discrete habitat patches with different size and connectivity, (b) at least 7–10 patches have been sampled with MRR, (c) the sampling has involved 10 or more capture occasions, and (d) spatial information is available for both the sampled patches as well as those that were not sampled, because their presence affects the connectivity of the former group (Hanski et al., 2000; Schtickzelle, Mennechez, & Baguette, 2006). All these underlying assumptions were clearly met in our study. The basic parameters estimated by the model include the following: (μ) mortality in habitat patches; (η) emigration propensity, defined as daily emigration rate scaled to 1‐ha patch; (ξ_em) emigration scaling with natal patch area; (ζ_im) immigration scaling with target patch area; (λ) scaling of mortality during dispersal with natal patch connectivity; and (α) distance dependence of dispersal.

The model assumes that an individual survives in any habitat patch with a probability of ϕ until the end of the unit of time (in the case of butterflies the unit is usually scaled to one day) or until emigrating from the patch. The assumption of equal survival probability in all the patches is reasonable for butterfly metapopulations, including the investigated one (cf. Nowicki et al., 2005) where all patches experience the same weather, which is the predominant factor affecting the survival of adult individuals. The dispersal‐independent mortality in each habitat patch (μ) is therefore 1–ϕ. This parameter is not a dispersal parameter per se, and as such it was not of prime interest for our analysis. Consequently, as an independent evaluation of this parameter with the Cormack–Jolly–Seber model (Schwarz & Arnason, 1996) revealed absolutely no intersexual difference or variation in time, we assumed its uniform value for males and females in both parts of the season in order to increase the precision of further parameters derived with the VM model, following the approach by Schtickzelle et al. (2006). The VM models with nonfixed within‐patch mortality, which we tried for comparison, yielded highly consistent values of this parameter for both sexes and season parts (Table S1). At the same time, they brought virtually no change in the estimates of all the other model parameters, except for their wider confidence intervals.

The emigration rate ε_j from a specific patch j is modelled as a function of patch area:

where η defines emigration propensity (here expressed as a daily emigration rate from a 1‐ha patch) and the emigration scaling parameter (ξ_em) reflects the steepness of the negative power relationship of actual emigration rate with patch area (A_j). For instance, the emigration scaling value of 0 means that emigration rate is independent of patch area, the value of 1 implies emigration rate being inversely proportional to patch area, and the value of 2 indicates 100‐fold decrease in emigration rate with 10‐fold increase in patch area. The rationale for this negative relationship is that with decreasing patch area, the individuals are more likely to leave their natal patches (Hambäck & Englund, 2005). The negative power dependence of emigration on patch size is well‐grounded in the theory and supported by numerous empirical studies (Englund & Hambäck, 2007; Hambäck & Englund, 2005; Turchin, 1986).

Whereas survival within habitat patches is assumed to be constant across all the patches, survival during dispersal is assumed to be affected by the natal patch connectivity, because an individual leaving an isolated patch is less likely to reach any other patch than an individual leaving a well‐connected patch. These assumptions allow to disentangle mortality within habitat patches from mortality during dispersal, based on the effect of habitat patch connectivity on the latter. Specifically, the VM model estimates the parameters of a function that relates the probability of successful dispersal to a measure of patch connectivity. In order to do so, the model considers the following sequence of events: (a) an individual either survives or perishes within the natal patch with a constant probability; (b) if it survives, it may emigrate and, (c) if so, it may either survive the emigration process or die before reaching the target habitat patch.

The survival of dispersing emigrants from a patch j (φ_mj) is assumed to increase with the patch connectivity S_j (i.e. the inverse index of its spatial isolation), with a relation described by a sigmoid function:

where the scaling parameter λ represents the connectivity value at which half of the dispersers successfully reach other patches. The connectivity index is calculated as:

where d_jk is the Euclidean distance between patches and A_k refers to the target patch area with which the immigration rate is linked through the scaling parameter ζ_im. Finally, α represents the scaling of distance dependence of dispersal, that is the coefficient of the kernel describing the probability of dispersal at a given distance. Successful dispersers are thus distributed among target patches proportionally to their contributions to the natal patch connectivity.

The dispersal‐related mortality scaling (λ) and distance dependence of dispersal (α) can be converted into the mean level of mortality experienced by dispersers and the mean distance they cover, respectively. With the negative exponential function (NEF) adopted as the dispersal kernel, the mean dispersal distance is calculated as 1/α. The NEF kernel fit our MRR data much better than the inverse power function (IPF) considered as an alternative as revealed by the VM model goodness‐of‐fit tests (Hanski et al., 2000). Likewise, the goodness‐of‐fit tests indicated that our data set conformed well to the assumptions of uniform within‐patch survival across the investigated patches as well as to the negative relationship of emigration with natal patch area and positive relationship of immigration with target patch area.

The maximum likehood (ML) values and the 95% confidence intervals (95% CIs) for the VM model parameters were obtained using the VM2 and VMSIM programs, respectively. We tested the significance of the differences between the parameters obtained for both sexes in early and late season by comparing their 95% CIs. The difference between a pair of parameters should be considered significant at p < .05 if the ML value of one parameter falls outside the 95% CIs of the other one and vice versa (Matter, Roland, Moilanen, & Hanski, 2004).

Analysis of emigration dynamics

The detailed analysis of within‐season dynamics of emigration in both sexes and their predictors was conducted using the multi‐state recapture models of Brownie, Hines, Nichols, Pollock, and Hestbeck (1993). With different habitat patches adopted as “states,” the model makes it possible to estimate survival (ϕ_j) and capture probability (p_j) within each patch as well as the probability of transition (= movement) from one patch to another for each pair of patches (ψ_jk). Since the number of model parameters grows factorially with increasing number of patches, the model can be effectively applied to investigate dynamics of the parameters only for a limited system of patches (optimally n ≤ 4) and it requires high‐quality data with a relatively high number of recaptures. A further requirement is that there are few movements into and out of the system (Brownie et al., 1993). Finally, negligible mortality during dispersal is highly desirable because in such a situation the probability of successful transition from a particular patch reflects the probability of emigration from this patch. Otherwise, the emigration probability is negatively biased because the emigrants that do not succeed in getting to any another patches are not accounted for. All the above conditions were met in our MRR campaign in 2013 when highly intensive sampling was performed in a small and relatively isolated metapopulation fragment consisting of two habitat patches. The dispersal mortality was considered negligible, as in the previous year dispersal mortality estimated for the entire metapopulation was next to zero (see Results). Hence, it can be assumed that, in our two‐patch system, the transition probability is tantamount to emigration probability.

The multi‐state recapture model parameter estimates were derived with MARK 8.0 software (White & Burnham, 1999). Following the outcome of the model selection procedure based on the Akaike information criterion corrected for small sample size (AICc) (Burnham & Anderson, 2001), we relied on the model variant ϕ_A(.)ϕ_B(.)p_A(s*t)p_B(s*t)ψ_A→B(s*t)ψ_B→A(s*t), that is the one assuming constant and equal survival for both sexes as well as intersexual differences and time variation in both capture probabilities and emigration probabilities. The lack of any significant intersexual differences in survival was also evident in the estimates of the model variants that assumed sex‐specific survival (Table S2 in the Electronic Supplement). The mean adult life span was estimated from the survival rate as (1 − ϕ_j)⁻¹ − 0.5 (Nowicki et al., 2005).

The obtained estimates of male and female emigration rates for the intervals between consecutive capture days constituted the dependent variables in the analyses of factors affecting within‐season emigration dynamics. As the predictor variables, we used (a) day in season, (b–c) densities of co‐occurring males and females, (d) the resulting female‐to‐male ratio and (e) the total number of previously flying females adopted as proxy of egg load laid prior to the date. The first of the above predictors, although alone not particularly informative, was tested for conformity with the earlier coarse analysis of the VM model estimates. The relative abundances of conspecifics of both sexes, as well as their sex ratio, determine the availability of breeding partners and the extent of potential competition for them. In turn, the progressive increase in the number of previously oviposing females reflects the decreasing number of available sites for laying eggs (i.e. flowerheads of S. officinalis foodplants). Although its effect was expected to influence only female emigration, it was tested as a potential predictor of male emigration for the sake of consistency. It should be stressed that assessing the dynamics of foodplant availability directly by counts of M. teleius eggs laid (or foodplants without eggs) repeated throughout the season was not possible because such a method is too laborious to be feasibly applied at habitat patch scale and destructive for the surveyed eggs. We thus used the number of previously flying females as a proxy of egg load laid as in Timus et al. (2017; see this reference for details). The rationale of this approach is based on the assumptions that (a) only a certain (fairly constant) fraction of flowerheads developed by S. officinalis is in proper phenological state for oviposition by M. teleius (Figurny & Woyciechowski, 1998); (b) M. teleius females typically lay a single egg per flowerhead and avoid flowerheads with conspecific eggs present (Figurny & Woyciechowski, 1998) and (c) the lifetime female fecundity in Maculinea is known and shows relatively little variability (Thomas, Clarke, Elmes, & Hochberg, 1998).

With the exception of the day in season (which did not differ between the two patches as they were surveyed on exactly the same days), the aforementioned predictor parameters were calculated separately for each patch as explained below. The numbers of males and females present on a particular day were derived on the basis of the number of captured individuals (n_i) and the capture probability (p_i) as $Ni=ni/pi$ and subsequently converted into densities (through dividing them by patch areas), as well as used to assess the sex ratio. Based on the estimated daily number of butterflies (N_i), as well as their survival rate (ϕ; constant value) and emigration probability (ψ_i), we could also assess the recruitment (B_i), that is the number of new individuals entering the population within the interval between consecutive capture days, using the formula:

and likewise in the case of B_B,i, where d_i is the length of the interval (in days) between capture days i and i + 1. The power transformation of survival rate and emigration probability was necessary so as to account for the interval length whenever it exceeded one day since the estimates of these parameters produced in MARK 8.0 were scaled to daily periods. The number of previously flying females was then obtained as the sum of female recruitments prior to a given date. For comparative purposes, we also estimated the absolute number of eggs laid by multiplying the number of previously flying females by the average fecundity of 75 eggs per female (Timus et al., 2017). The egg load dynamics assessed in this way was compared with the availability of S. officinalis flowerheads. The flowerhead density was surveyed for each habitat patch at 20 randomly distributed sampling quadrates of 2 × 2 m. The overall flowerhead availability was subsequently calculated as the product of patch area, flowerhead density and the proportion of flowerheads in suitable phenological state for oviposition, assumed as one third after Figurny and Woyciechowski (1998).

The effects of the aforementioned predictor factors were analysed with the logistic regression since the distribution of the dependent variable was clearly zero‐inflated, that is the emigration probabilities from both surveyed patches were estimated at zero for most intervals during the season. All the predictor factors considered are clearly nonindependent (e.g. the sex ratio derives from male and female densities) and thus obviously correlated (positively or negatively) with one another. Consequently, their effects were tested separately in alternative logistic regression models run in the Statistica 10 software (StatSoft 2010). The main aim of this testing was to evaluate the relative strength of the predictors in explaining the temporal dynamics of emigration rather than merely detect the significant relationships.

RESULTS

Dispersal patterns within metapopulation

In the metapopulation‐scale study conducted in 2012, we captured and marked a total of 1,317 butterflies, including 321 males and 204 females in the early season (sex ratio of 1.57 males per female), and 326 males and 446 females in the late season (sex ratio of 0.73 males per female). The sex ratio of captured individuals thus significantly differed between the two parts of the flight period (Chi‐square test: $χ12$ = 44.3203, p < .0001), confirming protandry. Altogether 405 butterflies (31%) were recaptured at least once, among them 76 individuals (19%) changed their habitat patch.

The estimated mortality rate of adult individuals within their habitat patches (μ) was 0.3538 (with a 95% CIs of 0.3238–0.3846), which can be interpreted as ca. 35% of butterflies dying per day. This corresponds to the average adult life span estimated at 2.34 days (95% CIs: 2.10–2.59 days). Female emigration propensity (η) was significantly higher in the late season than in the early season (Figure 1). In contrast, in males the emigration propensity was slightly (but not significantly) elevated in the early season as compared with the late season estimate (Figure 1). In addition, the negative relationship of male emigration rate with natal patch area was clearly steeper in the first part of the flight period when the emigration scaling parameter (ζ_em) reached a significant higher value (Figure 2). Taking into account the values of both above parameters, the actual rate of male emigration calculated using Equation (1) turned out to be significantly higher in the early season at all the smaller habitat patches (below 0.5 ha; i.e. 14 out of 20 surveyed patches), where female abundance was relatively low (Figure 3). In females, the emigration scaling with natal patch area was almost identical in both parts of the season (Figure 2), but the difference in emigration propensity was so high that the estimated emigration rate in the late season reached higher levels at all the patches regardless of their size.

No significant differences related to season part or sex were found for any other dispersal parameter (Table S3 in the Electronic Supplement). The dispersal mortality scaling estimates (λ) were invariably very close to zero, which indicates very low (if any) disperser mortality of at most a few per cent of individuals leaving their habitat patches. The dispersal kernel parameter (α) reached higher values in females, which translates into their slightly shorter mean dispersal distances (ca. 90–120 m versus 170–180 m in males). However, the precision of these estimates was relatively low (Table S3).

Emigration dynamics and its predictors

During the intensive MRR campaign of 2013, altogether 513 individuals (280 males and 233 females) were captured in patch A and 316 individuals (159 males and 157 females) in patch B. Out of 261 (32%) recaptured butterflies, 27 (10%) were found to move between the two patches. Apart from the very beginning of the flight period when few butterflies were present (<10 individuals of each sex captured per day in patch A, and < 5 in patch B), their daily densities were estimated at 8–64 males per ha and 6–36 females per ha in the first half of the season. In the second half of the season, the respective values were 8–55 per ha for males and 12–62 per ha for females. The within‐season dynamics of estimated sex ratio followed a typical protandrous pattern, with a relative predominance of males at the beginning of the season followed by a slow increase in the relative proportion of females with the flight period progression (Figure 4). The daily survival rate of butterflies was estimated at a slightly higher level than in the previous year, reaching 0.7576 (±SE = 0.0051) at patch A and 0.7327 (±SE = 0.0225) at patch B. This corresponds to the average adult life span estimated at 3.63 (95% CIs: 3.46– 3.80) for patch A and 3.31 (95% CIs: 2.69–3.93) for patch B.

In both patches, the male emigration noticeably declined and female emigration increased with season progression (Figure 4). The logistic regression analyses yielded significant effects for most of the predictors tested (Table 1). The clear exception was male density, for which we detected no significant effect on the emigration of same‐sex individuals. Male emigration decreased with increasing abundance of females, both flying on a given day or previously present in the population, but as expected its temporal pattern within the season was best explained by the changes in female‐to‐male sex ratio (R² > 0.80 for the emigration from patch A and > 0.65 from patch B). In turn, female emigration was negatively influenced by male density and positively by female density, as well as female‐to‐male sex ratio (Table 1). Nevertheless, again according to our predictions, the number of previously flying females used as a proxy of eggs laid proved to have the strongest explanatory power, accounting for > 95% of temporal variation in female emigration from patch A and > 75% from patch B.

It is also worth pointing out that the recorded increase in female emigration from both investigated patches coincided very well with the period when the carrying capacity of the foodplant flowerheads was reached. The density of S. officinalis flowerheads was assessed at 1.43 (±SE = 0.21) per m² at patch A, and 3.10 (±SE = 0.54) per m² at patch B, which translates to the overall numbers of the flowerheads available for oviposition at ca. 21.5 thousand and ca. 11 thousand, respectively. These levels were reached by the estimated numbers of eggs laid at both patches in the early days of August (Figure S2), which corresponds well with the onset of female emigration (Figure 4).

DISCUSSION

There are strong theoretical arguments that the emigration process should be highly dependent on the abundance of conspecific individuals (Poethke & Hovestadt, 2002), and many empirical studies demonstrated cases of both positive and negative density‐dependent emigration (see reviews in Clobert et al., 2012; Lambin, Aars, & Piertney, 2001; Matthysen, 2005). An increasing number of co‐occurring individuals influences the tendency to emigrate due to an increasing competition for resources (Waser, 1985), although conspecific density is likely to trigger emigration only above certain threshold values (Nowicki & Vrabec, 2011).

In agreement with this scenario, the results of our study make it clear that females become more prone to emigrate from the native habitat patch as the season progresses. One could be tempted to interpret this higher emigration of females in the second part of the flying season as a quest for males, which become rare at the end of the season (Calabrese & Fagan, 2004). Nevertheless, it must be stressed that although male abundance in the late season was indeed considerably lower than in the early season, it was still relatively high as implied by the numbers of captured individuals in 2012 or the estimated densities in 2013, which were in the typical range of values recorded for other populations of the investigated species (cf. Nowicki et al., 2005, 2009; Timus et al., 2017). Consequently, increased female emigration in the late season appears better explained by the attempt of females to avoid intraspecific competition for the diminishing resources, namely, foodplant flowerheads in proper phenological condition and free from conspecific eggs already laid. This explanation is in line with the existing theory that considers breeding sites to be the main limiting resource influencing reproductive success of females (Li & Kokko, 2019). Further evidence confirming the adopted hypothesis is the fact that the number of previously flying females, reflecting the overall number of conspecific eggs already laid, turned out to be the best predictor of female emigration. Moreover, the timing of the onset of female emigration from both investigated patches matched very well the period when the carrying capacities of the foodplant flowerheads available for oviposition were likely to be reached (Figure S2).

Based on this rationale, it is reasonable to interpret our results in the light of evolutionary cost/benefit balance influencing the emigration decision in females. In the second part of the season, direct investment costs of dispersal as well as its indirect costs, related to the uncertainty of finding better resources in a new habitat patch, are probably lower than the costs of searching for the diminishing resources in the natal habitat patch. Female strategy to emigrate in such conditions thus helps to reduce the competition among their larval offspring.

In turn, our results do not support the hypothesis of male harassment as a driver of female emigration, as we found a negative effect of male density on female emigration probability. This result, even though in contrast with the theory on sexual conflict as well as the outcomes of several studies on other butterfly species (Baguette et al., 1998; Odendaal et al., 1989; Shapiro, 1970; Trochet et al., 2013), is in line with an earlier research on Maculinea (Nowicki & Vrabec, 2011), suggesting that female emigration triggered by sexual harassment appears unlikely in these butterflies.

In the case of males, their prevalence at the beginning of the flight period is likely to make their emigration more beneficial than intra‐sexual competition in early season. In our study, a higher emigration rate was found in the first part of the flight period but only for small habitat patches, where opportunities to find mating partners are low. Our result is concordant with an earlier study of Petit, Moilanen, Hanski, and Baguette (2001) on Boloria (= Proclossiana) eunomia, which demonstrated that males are more likely to emigrate from small patches than females, but at the same time they are more prone to be sedentary in large habitat patches.

Interestingly, our analysis of predictors of dispersal revealed that male emigration is not influenced directly by male density. Even though Maculinea butterflies are known to establish home ranges within their habitat patches (Hovestadt & Nowicki, 2008), they cannot be considered territorial per se, as they do not present the typical territorial behaviour performed by males of some other butterfly species, which show direct male to male rivalry (e.g. Fischer & Fiedler, 2001; Kemp, 2002). Female‐to‐male ratio instead proved to be the best predictor for male emigration, which suggests that male dispersal is primarily driven by their attempt to maximize the number of encounters with females. It is worth noting that the mating opportunities depend not only on the number of occurring females, but also the number of males competing for them (Perrin & Mazalov, 2000). Therefore, it is not surprising that the changes in sex ratio determined male tendency to emigrate most accurately in our study.

All concerned, our study provides solid empirical evidence for sex‐biased density dependence of dispersal, showing that differences in optimizing mating success between males and females explain differences in their propensity to emigrate from natal habitat patches in particular periods of the flight season. It remains unclear whether the recorded differences in dispersal propensity between butterflies of both sexes occurring in different parts of the flight period have a genetic background, that is are due to genetic differences between such individuals, or they merely reflect phenotypic plasticity of a common genotype, the reaction of which is differently shaped by the environment in males and females due to intersexual differences in emigration response to external drivers. In univoltine species, individuals flying early and late in season may represent distinct developmental strategies. Since adult life span in Maculinea is much shorter than their flight period and thus individuals occurring in different part of the season constitute separate cohorts, one may expect a certain degree of genetic differentiation also between early and late flying butterflies of the same (meta‐)population, especially that they are likely to represent different developmental strategies, originating from two‐year and one‐year developing larvae, respectively (Witek et al., 2006).

Moreover, butterflies are characterized by chromosomal sex determination, with males being homogametic (ZZ) and females heterogametic (ZW) (Traut, Sahara, & Marec, 2007). According to Brom, Massot, and Laloi (2018), the heterogametic sex is more prone to disperse than the homogametic sex in order to avoid inbreeding, since the Z chromosome inherited from a heterogametic parent is identical between siblings, and higher identity can be expected by descent for W‐linked genes. The interaction between genetically determined abilities and stimuli from the external environment should shape specific behaviours which are activated when needed (i.e. when oviposition sites are scarce), thus being modulated according to a dynamic cost/benefit balance. Consequently, strong within‐season differences in dispersal behaviour are expected especially in female butterflies.

On the other hand, males should be more prone to patrol a specific area in search of mates, moving around and avoiding clearly directional flights, and to keep to their habitat patches, especially large ones (Petit et al., 2001). However, in a small habitat patch at the beginning of the season the chances of meeting a female are scarce due to protandry. Under such circumstances, males are more likely to cease their unsuccessful patrolling activity and emigrate from a habitat patch more often than in the late season, where more females are available. The aforementioned differences in the evolutionary context of male and female behaviours thus shape intersexual and within‐season differences in dispersal.

The results we obtained also offer a novel perspective for our understanding of metapopulation functioning, due to the different roles played by males and females in this respect. It should be underlined that whereas inter‐patch movements by both sexes can contribute to gene transfer (Piaggio, Navo, & Stihler, 2009; Solmsen, Johannesen, & Schradin, 2011), only female (post‐mating) dispersal enables effective (re)colonization of vacant habitat patches (Bergman & Landin, 2002). In the case of butterfly metapopulations, unfavourable conditions (such as bad weather) reducing life expectancy of adult individuals in the second part of the flight period when the tendency of females to emigrate is the highest may thus drastically limit the colonization potential, which determines metapopulation persistence (Hanski & Gaggiotti, 2004). In contrast, the occurrence of an analogous situation in the first part of season, when male dispersal prevails, should have less serious consequences, because male dispersal does not affect directly the colonization dynamics.

ACKNOWLEDGEMENTS

The study was funded by the Polish National Science Centre grant DEC‐2013/11/B/NZ8/00912, whereas the field surveys of Maculinea butterflies were supported by the Jagiellonian University through its DS/WBINOZ/INOŚ/761 funds, and conducted with a proper permission from the Polish General Directorate for Environmental Protection. We are grateful to Marta Kopeć, Monika Makowska and Jacek Marczyk for their assistance in the fieldwork, as well as to David S. La Mantia for improving the English of the manuscript. Julia Schroeder and two anonymous reviewers provided valuable comments on an earlier version of the manuscript.

AUTHOR CONTRIBUTIONS

PN and EP originally formulated the idea, TM and PN conducted fieldwork, EP and TM analysed the data, and EP and PN wrote the manuscript. The authors declare that they have no conflict of interest.

Peer Review

The peer review history for this article is available at https://publons.com/publon/10.1111/jeb.13688.

DATA AVAILABILITY STATEMENT

The dryad link to the article JEB 13688 https://doi.org/10.5061/dryad.nk98sf7rb.

REFERENCES