Defining eradication units to control invasive pests



    Corresponding author
    1. School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch, New Zealand
      Bruce C. Robertson, School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch, New Zealand (fax + 64 3364 2024; e-mail
    Search for more papers by this author

    1. School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch, New Zealand
    Search for more papers by this author

Bruce C. Robertson, School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch, New Zealand (fax + 64 3364 2024; e-mail


  • 1Pest eradication is an important facet of conservation and ecological restoration and has been applied successfully to invasive rat species on offshore and oceanic islands. Successful eradication requires the definition of a target population that is of manageable size, with low recolonization risk. We applied a molecular genetic approach to the identification of populations suitable for eradication (eradication units) to provide a new tool to assist the management of brown rats Rattus norvegicus on South Georgia (Southern Ocean).
  • 2A single eradication attempt on South Georgia (4000 km2) would be an order of magnitude larger than any previously successful rat eradication programme (110 km2). However, rats are demarcated into glacially isolated populations, which could allow sequential eradication. We examined genetic variation at 18 nuclear microsatellite loci to identify gene flow between two glacially isolated rat populations. One population, Greene Peninsula (30 km2), was earmarked for an eradication trial.
  • 3Genetic diversity in 40 rats sampled from each population showed a pronounced level of genetic population differentiation, allowing individuals to be assigned to the correct population of origin.
  • 4Our study suggests limited or negligible gene flow between the populations and that glaciers, permanent ice and icy waters restrict rat dispersal on South Georgia. Such barriers define eradication units that, with due care, could be eradicated with low risk of recolonization, hence facilitating the removal of brown rats from South Georgia.
  • 5Synthesis and applications. We propose that the molecular definition of eradication units is a valuable approach to management as it (i) provides a temporal perspective to gene flow, which is important if dispersal events are rare; (ii) allows an eradication failure (i.e. surviving individuals) to be distinguished from a recolonization event, opening the way for adaptive management in the face of failure; and (iii) can aid the management of pest species in habitat continua by resolving meta-population dynamics, so guiding pest eradication/control strategies. This study further illustrates the developing array of applied ecological issues in which molecular techniques can help guide management.


Eradication is a powerful strategy in conservation and ecological restoration (Myers et al. 2000). It provides a means to alleviate and/or remove the detrimental effects that exotic species typically have on their host ecosystems (Jackson 2001; Courchamp, Chapius & Pascal 2003; O'Dowd, Green & Lake 2003). To be applied successfully, however, the target population earmarked for eradication (the eradication unit) must be clearly defined. Small island populations or populations limited to ‘habitat islands’ are intrinsically well-defined eradication units (e.g. mammalian eradications on New Zealand offshore islands; Towns & Broome 2003; Courchamp, Chapius & Pascal 2003). Populations on larger islands or those that display no distinct structure, however, are more problematic (Carey 1991; Hampton et al. 2004). While large-scale eradications are feasible (Taylor, Kaiser & Drever 2000; Towns & Broome 2003), they are logistically difficult (Courchamp, Chapius & Pascal 2003). Success requires considerable planning (Myers et al. 2000), not least a strategy to define units that are both of manageable size and low recolonization risk (Parkes 1990; Bomford & O’Brien 1995). Attempting to eradicate a fraction of a population, or a sink population within an unidentified source–sink dynamic (Pulliam 1988; Hanski 1999), would inevitably result in rapid recolonization and a waste of resources (see examples in Myers et al. 2000).

Molecular genetics has provided a valuable means of identifying population structure, particularly with regards to defining units of conservation, management and evolutionary significance (Moritz et al. 1996). The degree of genetic structure within a population is indicative of connectivity; negligible genetic differentiation between spatially isolated populations is indicative of significant gene flow, while significant differentiation between adjacent populations indicates limited dispersal. With as few as one migrant per generation required to abate genetic differentiation (Wright 1969; but see Mills & Allendorf 1996), the partitioning of genetic variation can allow identification of distinct population units with negligible immigration. With appropriate care, these population units could be eradicated with little chance of recolonization.

The brown rat Rattus norvegicus (Berkenhout) is one of three rat species implicated in the majority of recent extinctions of birds on islands (Atkinson 1985; Courchamp, Chapius & Pascal 2003). The brown rat was unintentionally introduced to South Georgia when commercial sealing started there in c. 1786. As elsewhere (Imber, Harrison & Harrison 2000; Stapp 2002), brown rats have negatively impacted on the island's avifauna, reducing populations of the Antarctic pipit Anthus antarcticus (Cabanis), the South Georgia pintail Anas georgica (Gmelin) and various burrowing petrel species (Pye & Bonner 1980). The exact impact of rats on this ecosystem is unknown, but it is clear from comparisons of areas with and without rats that rat-free areas contain unique assemblages of plants and animals that appear to be unable to sustain populations in the presence of rats (McIntosh 1999). These findings have prompted a call for the total eradication of rats from South Georgia.

Although brown rats are distributed over the entire north-east coastline of South Georgia (4000 km2), their eradication may be aided by the island's extensive glaciation (56% of the island area; Smith 1960). A total reliance on coastally distributed tussac Parodiochloa flabellata (Lam. Poaceae) for winter survival (Pye & Bonner 1980) results in a discontinuous string of coastal populations delimited by glacial activity, permanent snow and ice, and icy waters (Pye & Bonner 1980; McIntosh 1999). If such barriers preclude dispersal, then each discrete population could be considered as an eradication unit. Eradication of rats from South Georgia could then proceed sequentially through space and time with low risk of natural recolonization.

In this study, we assessed the effectiveness of barriers to rat dispersal by using genetic variation at 18 nuclear microsatellite markers to identify gene flow between putatively isolated populations. Isolated rodent populations undergo rapid microevolution (Pergams & Ashley 2001), so barriers to dispersal should be readily identifiable by genetic differentiation at the population level. We compared the glacially isolated population of brown rats on Greene Peninsula, South Georgia, with the adjacent population at Hestesletten, South Georgia, the most likely source of immigrants. Greene Peninsula, the focus of a potential rat eradication trial, is an ecologically important area, as it is one of the few places on South Georgia that has not been grazed by introduced reindeer Rangifer tarandus (McIntosh 1999).



We obtained liver samples from 40 brown rats from each of two populations on South Georgia (54°S, 37°W). Samples were immediately stored in 70% ethanol at 4 °C. Whole genomic DNA was extracted from all 80 rat liver samples using c. 1 mm3 of liver tissue and a 5% Chelex (BioRad, Hercules, CA, USA) protocol (adapted from Walsh, Metzger & Higuchi 1991).

Rats on the Grytviken Peninsula (70 km2) were kill trapped between 15 and 18 February 2001 on the coastline of Moraine Fjord, within an area located between Discovery Point and a point 2 km to the south-west (hereafter referred to as the Hestesletten population). The Greene Peninsula (30 km2) rat population was also sampled using kill trapping. Rats were trapped between 17 and 23 February 2001 on the western coast of the peninsula (Fig. 1) within a 2-km section of coastal tussac located 1–3 km from Dartmouth Point (hereafter referred to as the Greene Peninsula population).

Figure 1.

A map of South Georgia (54°S, 37°W) showing the two regions (shaded) from which brown rats were sampled: 1, Hestesletten; 2, Greene Peninsula. Crosses in Moraine Fjord represent kelp beds. Other shading represents glaciation, permanent ice and snow. Numbers refer to peak heights in feet.

The vast Nordenskjöld Glacier (> 3 km wide, terminating over deep water), the Hamberg and Harker Glaciers (both terminating in deep water), permanent snow and ice, and the icy subantarctic waters of Moraine Fjord demarcate Greene Peninsula (Fig. 1). In this study, we made the assumption that the Nordenskjöld Glacier is an insurmountable hurdle to rat dispersal and that natural gene flow, if any, into Greene Peninsula would occur from the adjacent Hestesletten rat population.

dna analyses

Amplification of nuclear microsatellite loci (dinucleotide) was achieved by polymerase chain reaction (PCR) using 18 rat-specific primer pairs (Whitehead Institute/MIT Center for Genome Research, Rat Genomic Mapping Project, Data Release 7, January 2000, R. G. Steen et al., personal communication) and direct incorporation of αP33-dCTP. Loci were selected on the basis of reported levels of polymorphism, and only one locus per chromosome pair was screened. The PCR was carried out in 10-µL reactions containing c. 50 ng of template DNA, 0·5 pmol of each primer, 5 nmol each of dATP, dGTP and dTTP, 0·5 nmol of dCTP, 0·1 µCi of αP33-dCTP, 50 mm KCl, 10 mm Tris–HCl, pH 9·0, 1·5 mm MgCl2 and 0·1 unit Taq polymerase (ROCHE, Mannheim, Germany). The thermal cycling parameters were an initial 3-min denaturation at 94 °C, followed by 30 cycles at 94 °C/30 s, annealing temperature (Table 1)/30 s and 72 °C/40 s and then a final extension step at 72 °C/10 min. Following amplification, PCR products were size-fractionated on 6% denaturing PAGE gels and exposed to Biomax MR autoradiography film (Kodak). The size of amplified fragments was scored manually against a sequencing reaction (forward M13 primer and pBSMB plasmid).

Table 1.  Genetic diversity at 18 dinucleotide, microsatellite loci in two brown rat populations from South Georgia
Locus°CHestesletten (n = 40)Greene Peninsula (n = 40) FST
  • °C, annealing temperature; AN, allele number; HO, observed heterozygosity; FIS, level of inbreeding; FST, level of genetic differentiation between populations.

  • *

    P < 0·05, after sequential Bonferroni correction.

D2Rat1856030·25 0·2830·5−0·25 0·03
D3Rat1836340·65 0·1230·6 0·06 0·04*
D4Rat246040·35 0·2920·18−0·08 0·09*
D5Rat336030·59 0·1250·4 0·32 0·05
D6Rat1056340·7 0·0350·53 0·23 0·14*
D7Rat976020·34 0·230·61−0·09 0·02*
D8Rat1236020·4−0·1340·53 0·08 0·07*
D9Rat1106030·48 0·0320·38−0·03 0·003
D10Rat516030·35* 0·4130·53 0·21 0·01
D11Rat506220·36−0·0310·00 0·26*
D12Rat386250·3 0·4730·28−0·11 0·12*
D13Rat886040·4 0·0730·43 0·03 0·33*
D15Rat776030·26 0·1530·54 0·03 0·06*
D16Rat576350·55 0·1930·75−0·25 0·06
D17Rat1156260·63 0·1150·62 0·16 0·06
D18Rat966030·43* 0·2220·03−0·00 0·55*
D19Rat596020·35 0·2120·36 0·20−0·02
D20Rat466030·78−0·1630·43 0·08 0·19*
Average (SE) 3·39 (1·14)0·45 (0·16) 0·143·06 (1·11)0·43 (0·2) 0·06 0·12

statistical analyses

Genetic diversity was summarized for each population as the average number of alleles per loci and average observed heterozygosity. Allele frequency, observed heterozygosity and FIS were calculated for each locus in both populations using genepop 3.1 (Raymond & Rousset 1995). The FIS value provides an indication of population inbreeding when it is consistently high across multiple loci within a colony, and an indication of null alleles (alleles that do not amplify due to mutations in the PCR primer sites) when high values are recorded at a single locus (Pemberton et al. 1995). We tested for deficiency of heterozygotes at all loci, under the assumption of Hardy–Weinberg (H-W) equilibrium, using randomization tests implemented in genepop 3·1. Departures from H-W would be expected if a population substructure was present in the form of localized heterozygote deficiencies, which is suggestive of localized inbreeding. We estimated effective population size (Ne) for both populations to gauge the potential effect of genetic drift on allele frequencies, using a coalescent approach in the program migrate 1·7·3 (Beerli & Felsenstein 1999) and a mutation rate (µ = 1·0 × 10−3) estimated in the middle of the cited range for dinucleotide microsatellites (Hedrick 1999).

We investigated genetic differentiation between the Hestesletten and Greene Peninsula rat populations using four approaches. First, we compared allele frequency distributions in both populations using exact probability tests in genepop 3·1, where the explicit assumption was that significant differences in allele frequency distributions are indicative of reproductively isolated populations. Secondly, we calculated a pairwise estimate of FST (θ; Weir & Cockerham 1984) as a measure of genetic differentiation over subpopulations (Hedrick 1999) and tested this value for significant departure from zero using permutation procedures in fstat (Goudet 2001). Thirdly, we assessed gene flow between the two populations using the private alleles method of Slatkin (1985a), which uses the linear relationship between the average frequency of alleles found in only a single population (i.e. private alleles) and Nm, the number of migrants per generation. FST was not used to estimate Nm because recently isolated populations are unlikely to be at equilibrium, a key assumption of this approach (Whitlock & McCauley 1999). Finally, we tested the reliability of assignment of individuals to their population of origin based on allele frequencies using two different assignment tests: a Bayesian-based approach (Rannala & Mountain 1997), which combines prior beliefs about the probability of a hypothesis with the likelihood of the hypothesis; and assignment based on reference population allele frequencies (Paetkau et al. 1995). Assignment testing was performed with geneclass v.1·0·02 using observed allele frequencies and the ‘leave one out’ option (Cornuet et al. 1999).


As predicted from our selection of variable microsatellite loci, the 18 measured loci were all polymorphic (Table 1). When examining each locus per population and correcting for multiple tests, only two loci did not meet expected H-W proportions (Table 1); D10Rat51 and D18Rat96 were deficient of heterozygotes in the Hestesletten population only. The lack of concordance in heterozygosity deficiency between populations and the observation that FIS values for these loci were not consistent or significantly different from zero (Table 1), suggested that heterozygote deficiency was unlikely to be due to the presence of null alleles. We found no evidence of linkage disequilibrium, which was consistent with each of the 18 loci being on a different chromosome.

The Greene Peninsula population displayed slightly less genetic diversity than the Hestesletten population, with both allele number and observed heterozygosity lower in the Greene Peninsula population (Table 1). The Greene Peninsula population was also estimated to have a smaller effective population size (Ne = 72) than the Hestesletten population (Ne = 118).

Multiple lines of evidence suggested that negligible gene flow occurred between the rat populations of Greene Peninsula and Hestesletten. First, the two populations displayed significantly different allele frequency distributions at the locus and population levels (P < 0·05; Table 1). Secondly, the overall FST value (0·12, P < 0·05 after correction for multiple tests) indicated a pronounced level of genetic differentiation between the two populations (Wright 1978), which was comparable to mainland rat populations separated by 40 km (Kohn, Pelz & Wayne 2003). Thirdly, using the linear relationship between the average frequency of private alleles and migration (Slatkin 1985a), we found that migration was a rare occurrence between the two populations (Nm = 0·85 migrants per generation). Finally, the observed genic differentiation was sufficient to allow the correct assignment of most individuals to their populations of origin, regardless of method employed (Table 2).

Table 2.  Percentage of rats correctly assigned to their population of origin using Bayesian and frequency assignment tests. Likelihood-based assignment (Likelihood) and probability-based assignment (Probability) were done following Cornuet et al. (1999)
Reference populationnBayesianFrequency
Hestesletten40 98100 98100
Greene Peninsula40100 93100 85


eradication of rats from south georgia

Our study shows pronounced genetic differentiation between brown rats from the glacially isolated Greene Peninsula population and those from the adjacent population in the Hestesletten tussac. Increased quarantine measures at South Georgia (McIntosh 1999) and minimal human activity in the area suggests that this differentiation has probably occurred within the last 40 years since cessation of whaling at Grytviken (Headland 1984). We conclude that the Greene Peninsula rat population is a discrete eradication unit that could be eliminated with low natural recolonization risk.

Natural dispersal routes are important considerations when defining eradication units. Elsewhere, brown rats are known to use sublittoral dispersal (Calmet, Pascal & Samadi 2001) and surveys of near-shore islands around South Georgia indicate that kelp beds exposed at low tides (e.g. a ‘kelp bridge’ at the head of Moraine Fjord; Fig. 1) present a dispersal route to rats (S. Poncet, unpublished data). Brown rats are also strong swimmers (Whishaw & Tomie 1997), but in the subantarctic indirect observation of rat dispersal suggests that swimming distance drops from 600 m (Atkinson 1986) to less than 50 m (Taylor 1986). The genetic differentiation we noted indicates that rodent dispersal on South Georgia is limited to land bridges and impeded by glaciers, permanent ice and water barriers, which raises the possibility of the eradication of rats from the whole of South Georgia using glacially defined eradication units.

An attempt to eradicate rats from South Georgia (c. 4000 km2) in a single effort would be an order of magnitude larger than any previously successful eradication (e.g. Campbell Island, c. 110 km2; Towns & Broome 2003) and is almost certainly logistically impracticable. The island's extensive glaciation, however, subdivides the island's rat population into eradication units of manageable size (i.e. < 110 km2) that could be sequentially eradicated using currently available techniques (Towns & Broome 2003). Sequential eradication would allow the considerable expense of rat eradication (Towns & Broome 2003) to be spread over a number of financial years, thus making the programme(s) more attractive to funding agencies. Reliance on the island's glaciation necessarily requires an ongoing commitment to eradication as glacial barriers may retreat with climatic warming (Gordon & Timmis 1992), allowing the introduction of rats into previously rat-free areas (McIntosh 1999).

defining eradication units and monitoring eradication with molecular genetics

The use of molecular genetics to define population structure is a well-established, practical alternative to direct methods, such as mark and recapture, and provides the added advantage of a temporal perspective to gene flow (Slatkin 1985b; Neigel 1997). Such an approach is especially useful in the evaluation of recolonization risk in eradication programmes, as dispersal events may occur infrequently (e.g. in association with rare neap tides and sublittoral rodent dispersal; Calmet, Pascal & Samadi 2001).

Genetic definition of eradication units can be an important component of any eradication strategy, regardless of taxonomy, especially when dispersal pathways are not known or the effectiveness of putative barriers is unquantified. Importantly, information gained in a pre-eradication survey of genetic variation provides a valuable means of predicting eradication success and identifying the cause of an eradication failure (e.g. eradication attempts on Californian Medfly populations; Davies, Villablanca & Roderick 1999). Individuals remaining following eradication can be examined genetically using assignment tests (Cornuet et al. 1999) and identified as either eradication survivors (indicating a failure of protocol) or as re-invaders (highlighting either an unappreciated dispersal route or a quarantine breach). Such assignment testing is a powerful tool that can be readily applied to populations with low, but significant, genetic population differentiation (Gaggiotti et al. 2002). From an adaptive management perspective (sensuLessard 1998), the element of retrospective pathology that a pre-eradication survey affords managers may turn past eradication failures into important experiments that can guide future eradication strategy.

A potential challenge to the molecular definition of eradication units lies in accounting for situations in which dispersal does not translate into gene flow (i.e. effective dispersal is not equal to total dispersal). Under this scenario, dispersal is undetectable by molecular means unless fortuitously sampled. Available evidence suggests that the probability of acceptance of an immigrant into a population may be related to population demographics and the behaviour of a species (Scribner & Chesser 1993). In rodents, population density affects territorial behaviour and the mode of mating system (Moore 1999), with aggressive exclusion of immigrants observed in high-density populations (e.g. Granjon & Cheylan 1989, cited in Calmet, Pascal & Samadi 2001). In such situations genetic variation displays fine-scale population substructuring at the kilometre scale (Calmet, Pascal & Samadi 2001; Peakall, Ruibal & Lindenmayer 2003). In the present study, we found no evidence of population substructuring (i.e. systematic heterozygosity deficiency; Cramer et al. 1988) in either low-density population (F. Carpenter, personal communication), suggesting that behaviour was unlikely to be a barrier to dispersal. However, behavioural and demographic barriers to dispersal highlight the importance of knowledge of the dispersal capability of target species in pest control and eradication programmes (Myers et al. 2000; Byrom 2002; Courchamp, Chapius & Pascal 2003).

With the growing number of eradication successes on islands (Courchamp, Chapius & Pascal 2003; Towns & Broome 2003), attention has begun to re-emphasize management of pest species in habitat continua (Ji, Clout & Sarre 2000). One such example is the management of mammalian predators on New Zealand mainland ‘islands’ (Saunders & Norton 2001). Here, the boundaries of an eradication unit are unclear and control is probably more achievable than eradication. Removal of pests from the core management area is an on-going, expensive exercise due to continual re-invasion pressure from neighbouring unmanaged habitat (Saunders & Norton 2001). Such an ecosystem-based management approach to biodiversity conservation would benefit from an understanding of the landscape genetics of pest species (i.e. the marriage of genetic patterns with habitat patterns; Manel et al. 2003). Under such a paradigm, control of pest species could be aided by the identification of source populations for eradication, thereby potentially alleviating some of the re-invasion pressure on mainland managed areas. Understanding the source–sink dynamic (Pulliam 1988; Hanski 1999) in a seemingly homogeneous habitat could also refine strategic policy in pest eradication campaigns.


We thank the Government of the South Georgia and South Sandwich Islands for permission to sample the rat populations, and Sally Poncet, Mark Carpenter and Fraser Carpenter for sample collection. Fred Allendorf, Fraser Carpenter, Mark Carpenter Andy Cox, Raphael Didham, Tania King, Ed Minot, Steve Ormerod, John Parkes, Sally Poncet and two anonymous referees read the manuscript critically. We thank the University of Canterbury (B. C. Robertson) and New Zealand Department of Conservation for financial support.