Habitat destruction and fragmentation result in decreased habitat patch size and increased habitat patch isolation. Consequences of this common and now well-studied phenomenon include, amongst others, modified community composition and structure, smaller population sizes and decreased population connectivity 1. Both population size and connectivity significantly affect population genetic diversity, which can subsequently affect fitness. Indeed, as population size decreases, genetic stochasticity increases, resulting in increased allele fixation with each generation due to higher genetic drift. Homozygosity also increases due to a higher frequency of mating among relatives. Decreased connectivity augments the effects of small population sizes because of a reduction in successful dispersal events. Consequently, no or few alleles arrive to enhance the existing gene pool. Accordingly, smaller and more isolated populations are generally characterized by low genetic diversity (e.g. 2). Theory predicts that loss of genetic diversity generates adverse consequences for fitness, and subsequently for population demography, in small isolated populations 3, which is indeed confirmed by empirical evidence 2456. Hence, the major goal of conservation managers is to maintain minimum viable population sizes not only by increasing population size but also by enhancing connectivity between populations (e.g. 7).

In this study, we analysed the relationship between genetic diversity and fitness in a butterfly metapopulation within a highly fragmented landscape. Contrary to most studies relating genetic diversity to fitness, and despite the high amount of habitat loss and fragmentation within the study area, the chosen model species still has local populations of relatively large sizes and high connectivity. Both demographic and genetic techniques were used to estimate genetic diversity and fitness in five populations of the chalk-hill blue butterfly Polyommatus coridon (Poda) (Lycaenidae, Lepidoptera) situated in southern Belgium, at the northern edge of the species distribution range (Figure 1). These five populations were the largest populations in the area. Several other suitable sites were visited in the neighbourhood, but no or very few P. coridon individuals were observed. Polyommatus coridon is a thermophilic univoltine habitat specialist of calcareous grasslands 89, where caterpillars feed exclusively (T. Schmitt, personal communication) on the horseshoe vetch, Hippocrepis comosa. It is one of the six most characteristic butterfly species of European calcareous grasslands 10. Within the Walloon region (i.e. southern Belgium), P. coridon is classified as rare and in decline, and its distribution range has regressed by 70% since the 1950s 11. Consequently, it is a species of high conservation priority.

Calcareous grasslands are biodiversity hotspots not only because they are amongst the richest plant and insect communities in Europe 12, but also because they harbour numerous species from diverse regions, including Southern Mediterranean and Eastern Asia. They are also a central issue in nature conservation management in Europe because most of the calcareous grasslands are semi-natural habitats, i.e. created by human activity. Since the beginning of the 20th century, calcareous grasslands and their associated communities have declined severely following the progressive abandonment of agropastoral practices due to a lack of economical interest, land use intensification, and urbanisation 13. Indeed, most former calcareous grasslands have been afforested or have developed into woodlands by natural succession 14. Our study area reflects this general trend 15. In 2004 only 4.76% of the calcareous grassland surface of 1905 remained. This drastic decrease in calcareous grassland area is accompanied by a sharp decrease in average patch area (from 7.8 ha in 1905 to 0.8 ha in 2004), in patch number (22 large patches in 1905, 45 small patches in 1965, and 10 small patches in 2004), and in average patch connectivity (connectivity in 2004 is only 6% of that in 1905). Habitat loss and fragmentation in this region has had a profound effect on the butterfly community, including the extinction of specialist species and a subsequent increase in the over-representation of generalist species. Butterfly abundance, for both specialist and generalist species (based on common versus rare status), has also decreased significantly during this time period within the study region 15. These modifications have also been observed elsewhere (i.e. on western German calcareous grasslands 16). Consequently, calcareous grasslands are of high conservation priority, not only because they are biodiversity hotspots but also because of their historical (and cultural) heritage. The fact that the study system is found at the edge of P. coridon's distribution range does not make it less important for conservation management, especially when the main factors driving extinction are external (e.g. 17).

The primary aim of this study was to analyse the relationship between genetic diversity and fitness in a species of high conservation value within a landscape that has deteriorated significantly over the last century. Demographic data were used to estimate population size, effective population connectivity (number of successful dispersal events) and a fitness component, lifetime expectancy. We then investigated how genetic diversity affected lifetime expectancy. Lifetime expectancy is an important fitness component in insects because it directly affects lifetime reproductive success of many species 18, particularly polygamous species mating throughout their life, as P. coridon. Indeed, a longer life span will allow a greater number of matings (both sexes of P. coridon mate more than once). Furthermore, P. coridon females with a higher lifetime expectancy will lay more eggs since they lay eggs singly and continually until they die, but need to attain a certain maturity after hatching before being able to lay eggs, and lay fewer eggs in early life compared with later on 9.

Additionally, since this study was conducted within the framework of a nature conservation project with concrete management guidelines as one of the main objectives, we relate our demographic and genetic results to surrogate indices often used in applied conservation studies: habitat patch area and habitat quality descriptors as surrogates for population size, and structural connectivity as a surrogate for the intensity of dispersal and associated gene flow. Central to the metapopulation theory, the basic idea is that increased patch size and/or connectivity lead to increased population sizes (e.g. 19). Populations of larger size are expected to be more viable because they are subjected less to demographic, genetic, and environmental stochasticity 20. The long-term viability of populations is also expected to decrease with increased geographic isolation. Indeed, due to a decrease in successful dispersal events, locally isolated populations will not benefit from population replenishment by immigrants, which is necessary for population subsistence when the birth rate is exceptionally low or the death rate high 21. The supply and exchange of genes (alleles) also decreases in isolated populations, resulting in lower genetic diversity and increased inter-population differentiation 22 (but see 23).

The causal link between genetic diversity and fitness cannot be demonstrated without large experimental breeding and/or translocation manipulations at the population level, which is almost impossible to put into practice, especially with a threatened species. In such study systems it is therefore often necessary to correlate the variation observed in several traits, and we used this approach in our study; this correlation approach often generates rich conclusions (e.g. 24). Obtaining high quality estimates of genetic, demographic, and dispersal data was a clear prerequisite to be able to assess existing relations between these three kinds of trait. This required a huge investment in data collection and analysis, forcing a trade-off with the number of populations analysed in this study. The relatively small number of populations (five) studied here placed certain limits on the power of statistical tests. As a consequence, we did not interpret the very statistical level associated with the test of any single particular correlation coefficient, but focused on patterns emerging from sets of correlations involving alternative measures of genetic diversity.

Demography of the five populations was studied with the Capture-Mark-Recapture (CMR) methodology. The large CMR data set (7228 (re)captures of 2789 different individuals) was analysed using high-performance statistical procedures to obtain estimates of demographic parameters (survival and birth rates). Lifetime expectancy at birth (LTE), i.e. the number of days a newborn butterfly is expected to live, was computed for each of the five populations from these values of survival and birth rates. Genetic diversity was evaluated using three alternative and partially redundant measures based on 30 ISSR (intersimple sequence repeat) loci: He (expected heterozygosity, calculated from the null allele frequency and based on the Hardy-Weinberg equilibrium, with allele frequencies calculated from null homozygote frequencies assuming panmixia and corrected for dominance), Div (average gene diversity, i.e. mean number of pairwise haplotype differences), and PPL (percentage polymorphic loci). From here on, we refer to He, PPL and Div collectively when using the term 'genetic diversity'.

Population sizes (total over the flight period) varied between 2301 and 171 individuals (sexes pooled), and lifetime expectancy LTE varied from 8.46 to 10.86 days (Table 1). More detailed demographic results are out of the scope of the present paper; they are therefore not displayed here.

The average gene diversity Div over all ISSR loci and for all populations was 0.347 ± 0.176 (Table 1), indicating that about 35% of the loci differed between all pairs of ISSR genotypes. The expected heterozygosity He was 0.321 and the percentage of polymorphic loci PPL was 96.7%. As expected, the three measures of genetic diversity were all positively correlated to each other, although strictly speaking only the He-PPL correlation was statistically significant at the 0.05 level (He-Div: r = 0.700, p = 0.188; He-PPL: r = 0.949, p = 0.014; Div-PPL: r = 0.632, p = 0.252). Population differentiation was low but significant: 0.0268 and 0.0225 (p = 0.001) for F_st and θ_st, respectively. Fondry des Chiens and Abannets were the most genetically differentiated populations (Table 2). No significant (geographic) isolation by distance was detected (r = -0.040, p = 0.495).

Patterns of correlations were clearly visible among all the genetic, demographic, and dispersal measures: whereas correlations presented in Table 3 (genetic diversity, LTE, population size and dispersal rate) are high (10/11 are higher or equal to 0.63), those presented in Table 4 and Table 5 (genetic diversity and LTE with surrogates for population size and dispersal rate) are substantially lower (only 1/20 is higher than 0.63).

Variations between populations in genetic diversity were positively correlated to variations in LTE (Table 3; Figure 2). This correlation was highly significant for He. While not significant at the 0.05 level, Div and PPL also showed largely positive covariation with LTE. The example of the He-LTE correlation is used to show that this correlation remained largely unaltered when the uncertainties in the estimates of He and LTE were fully taken into account, and is robust to a large decrease of the number of loci on which genetic diversity was computed (Figure 3). This positive relation implies that butterflies from populations with the highest genetic diversity had a mean lifetime expectancy up to c. 25% higher than butterflies from populations characterised by a low genetic diversity.

Genetic diversity was positively correlated with population size (Table 3), but not with its surrogates (Table 4). Likewise, genetic diversity was positively correlated with number of immigration dispersal events (Table 3) but not with its surrogate (Table 5). No correlation was observed between LTE and population size or its surrogates (Tables 3 and 4); LTE was positively correlated with number of immigration events, but not with its surrogate (Tables 3 and 5); LTE was also positively correlated with the number of emigration events (r = 0.900, p = 0.037), but not with the net number of immigrants (immigration minus emigration: r = 0.300, p = 0.624).

Using both demographic and genetic techniques on a metapopulation of P. coridon consisting of large and well connected populations, three conclusions were reached: (1) the most important conclusion is that fitness of butterflies positively correlated with genetic diversity; (2) immigration rate and population size were also positively correlated with genetic diversity; and (3) CMR-based estimates of population size and dispersal rates were better indicators of genetic diversity than were some of their more conventionally used surrogates: habitat patch area, various measures of habitat quality, and geographic patch isolation.

Lower genetic diversity was coupled to a sharp decrease in adult lifetime expectancy (LTE): P. coridon butterflies of some populations lived for up to 25% (2.4 days) longer on average than their counterparts of other populations, therefore significantly increasing their individual fitness. LTE is definitely a major component of individual fitness via lifetime reproductive success in this species. Indeed, both sexes mate more than once and females with a higher LTE will lay more eggs, because females lay eggs singly and continually until they die but need to attain a certain maturity after hatching before being able to lay eggs 9.

Despite the limits inherent to the correlational approach and the restricted number of populations sampled in this study, the validity of our conclusions is supported by three main elements. First, the accuracy of estimates for genetic, demographic and dispersal parameters was very high compared with the standard level achieved in other studies. Genetic diversity measures were based on a relatively large number of loci. Demographic and dispersal measures were based on a very large CMR dataset. LTE not only integrates survival over the entire adults' life-span (daily survival changed with time) but also incorporates the variation in birth rate over time (which followed a parabolic distribution), which results in one robust and comprehensive value of average adult lifetime expectancy per population. Second, the He-LTE correlation remained highly significant even when the errors associated with demographic and genetic estimates were taken into account, indicating that the restricted number of populations was sufficient for detection of trends, despite the estimation errors. And third, clear trends of similar and consistent correlations were observed with alternative measures of genetic diversity.

Furthermore, although artefacts could potentially have influenced the observed correlation between fitness and genetic diversity, they do not alter the validity of our conclusions. First, the issue of population structure (i.e. significant population differentiation) has been raised as one of the most likely causes of association between genetic diversity and fitness related traits. Indeed, sampling individuals from different geographic origins could confound associations because of environmental heterogeneity at the sites 25. We are convinced that this explanation can be ruled out in the present case for several reasons. First of all, the studied populations were all located within a very restricted geographic area and were readily identifiable because of the very specific ecological requirements of P. coridon's host plant. Additionally, CMR data confirmed that the five populations were in fact independent but interconnected units: dispersal within populations (intra-patch movements) was more than 10 times as large as between population dispersal, and none of the populations functioned as source or sink. Genetic analyses confirmed these results: population differentiation was significant but extremely low. Low population differentiation at local and regional scales is typical for this species 26. Second, a positive correlation between genetic diversity and LTE could also be due to the recognition by P. coridon adults of differential habitat quality among patches. Indeed, prime habitat would facilitate long life spans in butterflies living there and attract, and keep, many individuals from outside the patch, leading to increased genetic diversity. As a consequence, populations in prime habitat will show an increased LTE and an increased genetic diversity, both caused by habitat quality, and the link between genetic diversity and LTE might not express an effect of genetic diversity on fitness. If this scenario were the case in our study, we would expect to find both a positive relation between LTE and some of the usual habitat quality descriptors (habitat area, microclimate, adult nectar resources, host plant abundance, structural connectivity), and between LTE and the net number of immigrants (immigrants minus emigrants), which was not the case. Some of the harshest conditions were even found on the site characterised by the highest genetic diversity and LTE. Indeed, vegetation in this site had completely dried up long before the end of the P. coridon flight season, including its host plant (SV, personal observation). Third, LTE might be biased by emigration rate, because CMR analyses consider as identical the death or the emigration of an individual, both being disappearance events that decrease the estimate of survival rate (and then LTE). In this case, a negative trend between LTE and the number of emigration events is expected. However, emigration rate from the five populations was relatively low compared with the amount of demographic data (3% of captures), and a positive correlation was observed between LTE and the number of emigration events, i.e. in the opposite direction to the expected bias. The bias of LTE due to emigration decreased differences between populations in LTE instead of exaggerating them, and the observed positive correlations between LTE and genetic diversity were therefore conservative.

So far, the interaction between genetic diversity and fitness (i.e. survival and/or reproductive success) has only been revealed when effective population size becomes low and/or populations are extremely isolated (e.g. 2728). Small population sizes favour inbreeding behaviour and allow genetic drift to erode genetic diversity, while increased isolation limits gene flow, i.e. arrival of new alleles and non-related individuals. Whether the positive genetic diversity-lifetime expectancy correlation is due to heterosis, where heterozygous genotypes are superior to any of the homozygous genotypes, or due to partial dominance, where decreased allelic diversity leads to the expression of recessive or partially recessive deleterious alleles, cannot be determined with the present results but certainly deserves further research.

In the present study, demographic analyses indicated that the populations were neither small nor isolated. Although estimating effective population size (Ne) remains difficult, detecting inbreeding effects in population sizes of several hundreds to several thousands of individuals may suggest that Nes were only a tiny fraction of census population sizes (N) in this study area. The same pattern of decrease in genetic diversity in small and isolated local populations was recently detected within a metapopulation of lesser kestrel (Falco naumanni) (Fleischer) (Aves, Falconidae), a vagile and migratory raptor with high dispersal potential 29. Altogether, these pieces of evidence support the idea that there is a considerable difference between the total and the effective population size in metapopulations.

The second conclusion of our study is that number of dispersal events positively correlated with genetic diversity. That increased functional connectivity leads to greater genetic diversity is expected and was also detected in a large scale study of P. coridon 30. However, it is quite surprising to detect this relationship at such a small spatial scale and with a relatively high connectivity between populations. Likewise, genetic diversity was positively correlated to population size. Larger population sizes are indeed generally more resistant to genetic erosion.

The third conclusion concerns the widespread use of surrogate variables. Building on the metapopulation theory, many conservation studies use patch area and structural connectivity measures as indicators of population size and isolation, respectively, two components with major demographic and genetic effects on the short- and long-term viability of species systems (e.g. 313233). Our study hints that these measures may not always be very reliable surrogates (see also 34 for patch area). Patch area and a series of well chosen habitat quality descriptors failed to reveal the links observed when a high quality CMR-based estimate of population size was used; similarly, structural connectivity failed to reveal the link observed when numbers of immigrants were used. This latter point connects to the fact that structural connectivity estimates do not capture the intra-specific variation in individual behaviour related to mobility, which has been shown to be strongly affected by both landscape fragmentation and metapopulation history (e.g. 35363738). These results therefore suggest that demographic (such as CMR) studies provide more realistic and reliable indicators necessary for conservation programmes to be effective in the long-term. Unfortunately, they also necessitate much more investment.

A positive correlation between genetic diversity and mean lifetime expectancy was detected at a very local scale in a metapopulation consisting of large high-density populations interconnected by considerable dispersal rates. Even in such a system, genetic diversity can be decreased and directly affect the fitness of individuals. From a biodiversity conservation perspective, this study clearly demonstrates the benefits of both in-depth demographic and genetic analyses and highlights that in order to ensure the long-term survival of the populations conservation actions should not be solely based on patch area and structural isolation. This result may be especially pertinent for species at their range margins, particularly in this era of rapid environmental change.

SV and MB designed the research; SV performed the research; SV and NS analysed the data; SV, NS, and MB wrote the paper. All authors contributed equally to this work.