## Introduction

Effective population sizes (*N*_{e}) and dispersal rates are fundamental evolutionary parameters that determine the spatial variation in allele frequencies as well as the efficacy of selection in shaping genomic architecture (Slatkin 1973, 1985; Nagylaki & Lucier 1980; Adkison 1995). Migration permits gene flow and re-colonization following local extinction and, accordingly, quantifying dispersal rates has received widespread attention (Slatkin 1985; Clobert *et al*. 2001; Hanski 2003), but field estimates of *N*_{e}, the focus of this study, have been largely neglected. The concept of *N*_{e} was introduced by Wright (1931) to predict the genetic properties of a finite population that meets the assumptions of random mating, constant population size and nonoverlapping generations (the Wright–Fisher model) (Fisher 1930; Wright 1931). In the absence of significant migration, selection or mutation, the allele frequencies of this ‘idealized’ population will vary among generations due to random sampling of gametes during reproduction; the amount of this genetic drift being inversely proportional to population size. Natural populations seldom conform to the Wright–Fisher model and various factors reduce the number of reproductively successful individuals (i.e. *N*_{e}) below that of the total adult census (*N*) (Frankham 1995). This has widely recognized consequences for species’ conservation because it is *N*_{e} and not *N* that determines levels of inbreeding and the rate of loss of genetic diversity, which are important correlates of population persistence (Saccheri *et al*. 1998; Spielman, Brook & Frankham 2004) and future evolutionary potential (Franklin 1980).

Numerous demographic- and genetic-based methods have been developed to calculate *N*_{e} (reviewed by Nunney & Elam 1994; Wang & Caballero 1999; Wang 2005) yet it remains challenging to quantify. For example, there are manifest difficulties associated with acquiring accurate estimates of population fluctuations, sex ratio and variance in mating success (VMS), which are the principal factors that determine *N*_{e}. The corollary is that most estimates of *N*_{e} are based on the frequencies of genetic markers in populations, as these are a retrospective measure of all processes that determine the successful breeders (but not the relative significance of each demographic factor). Of the genetic methods formulated to estimate *N*_{e}, such as the extent of temporal variation in allele frequencies (Nei & Tajima 1981; Waples 1989; and others), linkage disequilibrium (Hill 1981) or heterozygote excess (Pudovkin, Zaykin & Hedgecock 1996), the first is believed to have the greatest precision and is the most frequently employed.

For species existing as a dynamic network of partially connected patches (captured by the metapopulation paradigm, see Hanski 2003 for review) several issues may complicate estimation of *N*_{e}. First, the original genetic methods developed to calculate *N*_{e} assume that populations are closed and this is often invalid; migration limits genetic divergence and this biases estimates of *N*_{e} if not taken into account. For this reason, new techniques have been derived that are able to jointly estimate *N*_{e} and immigration rates (*m*) (Vitalis & Couvet 2001; Wang & Whitlock 2003) and there are now several examples of the application of these techniques to characterize vertebrate populations (Wilson, Hutchings & Ferguson 2004; Jehle *et al*. 2005). Second, the effect of spatial genetic structure should be examined, as the *N*_{e} of a structured population is a function of the amount of subdivision. For instance, under an island model (populations exchange equal numbers of migrants) *N*_{e} = *nN*_{eS}/(1 – *F*_{ST}), where *n* is the number of subpopulations, *N*_{eS} is the effective population size of the subpopulations and *F*_{ST} is the standardized variance in gene frequencies among populations (Wright 1943; Wang & Caballero 1999). Accordingly, with increasing spatial structure it is possible for the *N*_{e} of a metapopulation to be greater than that of a random mating population of equivalent total size (Nichols, Bruford & Groombridge 2001). The alternative view that subdivision can reduce *N*_{e} (Whitlock & Barton 1997) highlights the point that the overall *N*_{e} of a structured population depends on the particular metapopulation dynamics.

With increasing population fragmentation through anthropogenic habitat modification, predicting the long-term population viability of poorly known species from their life history and other ecological characteristics is clearly important. The parameters *N*_{e} and *m* certainly impact on population survival *per se*; however, they also inform on the relative influence of random genetic drift, gene flow and selection (Slatkin 1973, 1985; Nagylaki & Lucier 1980; Adkison 1995). Hence, their quantification can be indicative of the potential for adaptive divergence among populations, an understanding of which affords a better practice of species’ conservation that is able to recreate contemporary evolutionary patterns and processes that is preferable to static preservation (Crandall *et al*. 2000; Stockwell, Hendry & Kinnison 2003). Invertebrates in particular are poorly studied, and this is particularly true for analyses of *N*_{e} in insect populations (see also Thompson, Watts & Saccheri 2007). This deficiency is alarming as they form a dominant functional group in many ecosystems, and some 800 species are in the IUCN (2006) red list data book (excluding the least concern category).

In this paper we (1) quantify adult population sizes (*N*), *N*_{e} and *m* in a fragmented population matrix of the endangered damselfly *Coenagrion mercuriale* (Charpentier, 1840) (Odonata: Zygoptera). We (2) contrast genetic and demographic methods of estimating the latter two parameters, (3) examine whether there are simple correlates of *N*_{e}, such as habitat size or the numbers of adults *per se*, and (4) use estimates of *N*_{e} and *m* to assess whether local *C. mercuriale* populations are more likely to be influenced by the action of selection or genetic drift. We find that *N*_{e} is always less than the population census (*N*), likely because of variance in family size, and that the ratio *N*_{e}/*N* varies considerably with population size. Making accurate estimates of *m* in weakly differentiated systems remains challenging. These results are discussed with respect to conservation of *C. mercuriale*.