#### description of study site and field work

*C. mercuriale* was sampled from Beaulieu Heath (50°47·8′N, 01°29·9′W) in southern England, UK. This site (4·6 × 3·7 km), isolated from other colonies by more than 4 km of woodland and heath, is a matrix of a large central ‘population’ and four satellite colonies (Fig. 1). For convenience during sampling, the central site was divided into seven areas, which are not separate (but appear semi-isolated where the habitat narrows) except where they are bisected by a road. Therefore, for this analysis, the central area was divided into an east and west population (Fig. 1). The area of suitable *C. mercuriale* habitat at each site was taken from recent surveys (Daguet 2006).

*C. mercuriale* emerges from May until the end of July, with a peak during June (Purse & Thompson 2003). To estimate population size, we undertook capture–mark–recapture (CMR) between 11 June and 14 July 2002. Adults were searched for every day (09.30–16.00 h) except during poor weather when they are not active. All unmarked, mature damselflies were caught, marked and re-observed using standard methods described previously (Rouquette & Thompson 2007a), with the position of every encounter geo-referenced using a differential GPS.

#### demographic estimates of adult population, *N*_{e} and rates of migration

Daily population sizes were calculated using a full Jolly–Seber model (Jolly 1965; Seber 1973) that makes a number of assumptions, including that animals are unaffected by being marked, marks are not lost, marked animals become mixed within the population and there is an equal probability of catching classes of animals (i.e. different sexes, age classes, marked and unmarked animals). Deviations from these assumptions are minimal, for example marks are not lost as they were written in indelible ink and removal of a single leg for genetic analyses (see below) does not affect fitness in damselflies (Fincke & Hadrys 2001). However, female damselflies were encountered less often than males at breeding sites despite an even sex ratio (Rouquette & Thompson 2007b), so we calculated population size using male CMR data only and then doubled this estimate to account for the more cryptic females. Our sampling did not take place over the full flight season of *C. mercuriale*. Estimates of the daily numbers of adult damselflies present when CMR was not undertaken were made from the trend in logistic growth (or decline) based on the available increasing (or declining) daily population estimates and assuming zero adults on the first (or last) date that *C. mercuriale* were sighted in England (6 May and 25 September in 2002, D.K. Jenkins pers. comm.). Censuses on 23 June, 4 July and 6 July were lower than expected from the trend in daily population size, probably because poor weather reduced capture efficiency, and were replaced with expected sizes from the trend in logistic growth or decline. Dividing the sum of all daily censuses by the average life span (mean time between first and last captures in days) provided a total adult population estimate (*N*).

Ecological estimates of *N*_{e} were calculated using the approximation *N*_{e} ≈ 8*N*/(*V*_{kf} + *V*_{km} + 4), where *V*_{kf} and *V*_{km} are the respective VMS for females and males (Falconer & Mackay 1996) and the population meets the assumptions of a Wright–Fisher model, which are given in the Introduction; using lifetime mating success data (Purse & Thompson 2005) we estimated *V*_{kf} and *V*_{km} of *C. mercuriale* to be 7·4 and 13·5 (Watts *et al*. 2007). Immigration rates (*m*) are the proportions of animals recorded moving into a site different to that in which they were first captured (Table 1).

Table 1. Numbers of adult damselflies, *C. mercuriale*, caught, recaptured and observed moving between separate study sites during a CMR study on Beaulieu Heath, UK during 2002. See Fig. 1 for locations of sample sites | No. caught | No. recap'd | No. of observed immigrant damselflies |
---|

ROU | HAT | GRE | BAG | BHW | BHE |
---|

Source site |

ROU | 1 744 | 878 | – | | | | | |

HAT | 241 | 66 | | – | | | | |

GRE | 152 | 82 | | | – | | 1 | |

BAG | 104 | 39 | | | | – | | 1 |

BHW | 2 934 | 893 | | | 1 | | – | 1 |

BHE | 5 084 | 2200 | | | | 1 | | – |

Total | 10 259 | 4158 | | | 1 | 1 | 1 | 2 |

#### genotyping and genetic data analysis

Full details of DNA extraction and genotyping are described elsewhere (Watts *et al*. 2004a). DNA was extracted from single legs that were removed from between 47 and 192 damselflies per sample during 2002 and 2004 (Fig. 1), which represents a single generation interval as *C. mercuriale* is semivoltine in the UK. We observed no significant effect of our sampling upon recapture rate (D.J. Thompson unpublished). Every individual was genotyped at 14 unlinked microsatellite loci: LIST4-002, LIST4-024, LIST4-034, LIST4-037, LIST4-062, LIST4-063, LIST4-023, LIST4-030, LIST4-031, LIST4-035, LIST4-042, LIST4-060, LIST4-066 and LIST4-067 (Watts *et al*. 2004c; Watts, Thompson & Kemp 2004b).

Every population and the entire sample was tested for departure from expected Hardy–Weinberg equilibrium (HWE) conditions using the randomization procedure (5000 randomizations) implemented by fstat ver. 2·9·3 (Goudet 1995). This software was used to calculate expected heterozygosities (*H*_{e}) and the level of genetic differentiation, *F*_{ST} (Weir & Cockerham 1984), throughout the study area, with 95% confidence intervals (95% CI) for *F*_{ST} made by bootstrapping over loci.

Temporal methods to estimate *N*_{e} are based on the premise that the magnitude of temporal fluctuations in allele frequencies is directly related to *N*_{e}. Recent statistical improvements on the original temporal methods (e.g. Berthier *et al*. 2002; Wang & Whitlock 2003) present a variety of techniques with which *N*_{e} may be estimated. It is beyond the scope of this paper to examine all of these, so we compared (1) Waples's (1989) original method with (2) moment and (3) maximum-likelihood (ML) estimators derived by Wang & Whitlock (2003) as the former is still widely used and the latter two methods were developed to estimate *N*_{e} and immigration rate (*m*) jointly, but may be applied to calculate *N*_{e} of a closed population (note that Waples’ method assumes that the population is isolated, a condition which is not met in this study; Table 1). Wang & Whitlock's methods estimate *N*_{e} and *m* of a focal population under the assumption that immigrants are provided by an infinitely large source, but are robust to deviations from this model and may be applied to a source comprising one or more finite subpopulations. Here, source populations consisted of the pooled Beaulieu Heath genotypes after those of the focal population had been removed. NeEstimator ver. 1·3 (Peel, Ovenden & Peel 2004) was used to calculate *N*_{e} according to Waples (1989) and MNe ver. 2·3 (Wang & Whitlock 2003) was used to calculate *N*_{e} according to the moment and maximum-likelihood estimators of Wang & Whitlock. Ninety-five per cent confidence intervals are calculated for the moment and ML methods of Waples (1989) and Wang & Whitlock (2003), respectively.

Because many studies are limited to a single sampling season, estimates of *N*_{e} were made using Hill's (1981) LD method, implemented by NeEstimator (Peel *et al*. 2004), to provide some guide to its reliability. Nonrandom associations between alleles at different loci (linkage disequilibrium, LD) may arise through selection, migration, assortative mating and genetic drift. At neutral loci the degree of LD within an isolated Wright–Fisher population is determined by genetic drift, such that the extent of association between alleles provides an estimate of *N*_{e} (Hill 1981).

Under a stepping-stone model of dispersal, which is appropriate for *C. mercuriale* (Watts *et al*. 2004a, 2007), several models of gene flow and selection (Slatkin 1973; Nagylaki & Lucier 1980; and others, reviewed by Slatkin 1985; Adkison 1995) can be used to assess the potential for local population adaptation. Briefly, adaptation is expected when two conditions are fulfilled. First, there is a ‘characteristic length’ of spatial variation in allele frequencies due to gene flow and selection (*l*_{c}) that is defined by *l*_{c} = σ/*√s*, where σ is the standard deviation of dispersal distances and *s* is the strength of selection, and local adaptation cannot occur if the direction of selection varies at distances less than *l*_{c} (Slatkin 1973, 1985). Second, Nagylaki & Lucier (1980) demonstrated that the relative importance of selection and genetic drift can be captured by the single parameter where *m* is the migration rate, *s* is the strength of selection and *N*_{e} is the effective population size. If the first condition is met, then local adaptation occurs when β >> 1, but not if β << 1 as selection is overwhelmed by the effects of genetic drift. This approach is discussed in detail by Adkison (1995). We estimated *l*_{c} using demographic and genetic estimates of dispersal distances (σ = 118 and 176 m, respectively, Watts *et al*. 2004a) and β, again using demographic- and genetic-based estimates of *N*_{e} and *m *[calculated using Wang & Whitlock's (2003) maximum-likelihood method]. Both *l*_{c} and β were calculated for a range of selection coefficients (*s*) that varied from 0·1 to 0·0001 to describe the influence of strong to weak selective pressure.