Reinvasion by ship rats (Rattus rattus) of forest fragments after eradication

Transcription

1 Biol Invasions (2011) 13: DOI /s ORIGINAL PAPER Reinvasion by ship rats (Rattus rattus) of forest fragments after eradication Carolyn M. King John G. Innes Dianne Gleeson Neil Fitzgerald Tom Winstanley Barry O Brien Lucy Bridgman Neil Cox Received: 16 December 2010 / Accepted: 25 June 2011 / Published online: 9 August 2011 Ó Springer Science+Business Media B.V Abstract Reinvasions provide prime examples of source-sink population dynamics, and are a major reason for failure of eradications of invasive rats from protected areas. Yet little is known about the origins and population structure of the replacement population compared with the original one. We eradicated eight populations of ship rats from separate podocarp-broadleaved forest fragments surrounded by open grassland (averaging 5.3 ha, scattered across 20,000 ha) in rural landscapes of Waikato, New Zealand, and monitored the- re-establishment of new populations. Rats were kill-trapped to extinction during January to April 2008, and then again after reinvasion in April May (total n = 517). Rats carrying Rhodamine B dye (n = 94), available only in baits C. M. King (&) B. O Brien L. Bridgman Department of Biological Sciences, University of Waikato, Private Bag 3105, Hamilton 3240, New Zealand c.king@waikato.ac.nz B. O Brien b.obrien@waikato.ac.nz L. Bridgman bridgmanl@landcareresearch.co.nz J. G. Innes N. Fitzgerald L. Bridgman Landcare Research, Private Bag 3127, Hamilton 3240, New Zealand innesj@landcareresearch.co.nz N. Fitzgerald fitzgeraldn@landcareresearch.co.nz placed 1 2 months in advance in adjacent source areas located m (average 228 m edge to edge) away, appeared in 7 of the 8 fragments from the first day of the first eradication. The distribution of age groups, genders and proportions of reproductively mature adults (more immature juvenile males and fewer fully mature old females) was different among marked rats compared with all other rats (P = 0.001, n = 509); in all rats caught on days 7? of the first eradication compared with on days 1 6 (P = 0.000); and in the total sample collected in fragments by trapping to and after local extinction compared with in brief, fixed-schedule sampling of populations in continuous forests (P = 0.000). Genotyping of 493 carcases found no significant population-level differentiation among the 8 fragments, confirming that the D. Gleeson T. Winstanley Ecological Genetics Laboratory, Landcare Research, Private Bag 92170, Auckland 1142, New Zealand gleesond@landcareresearch.co.nz T. Winstanley tom.winstanley@gmail.com N. Cox Invermay Statistics Group, AgResearch, Invermay Research Centre, Puddle Alley, Private Bag 50034, Mosgiel, New Zealand neil.cox@agresearch.co.nz

2 2392 C. M. King et al. rats in all fragments belonged to a single dynamic metapopulation. Marked rats of both genders travelled up to 600 m in a few days. Conservation of forest fragments is compromised by the problem that ship rats cannot be prevented from rapidly reinvading any cleared area after eradication. Keywords Invasive predators Reinvasion Roof rat Black rat Forest fragment Genetic differentiation Eradication units Introduction In a fragmented landscape, the survival of functional populations of native species in the remaining patches of preferred habitat, and the rate at which invasive species colonise those patches, depends on the species-specific dispersal abilities and inter-patch movements of individuals (Bowne and Bowers 2004). Relatively little is known about how dispersal affects the important early stages of either colonisation or reinvasion (Puth and Post 2005), when population structure and behaviour might be different from those of established populations (Russell et al. 2009a). Therefore, understanding how the spatial configuration of habitats influences dispersal distances and population dynamics is essential for wildlife management, both to protect native species affected by forest fragmentation and to control introduced species (With 2002; Hastings et al. 2005). Eradication programmes on islands are now clearing more and larger areas of invasive species, with huge conservation benefits (Brooke et al. 2007; Towns 2009). Forest fragments share certain characters of islands (Watling and Donnelly 2006), and some aspects of conservation management developed on islands could potentially be applied to managing mainland fragments. The main limitation for managers of forest fragments is that the primary condition required to match the successes of island pest eradication programmes (isolation from reinvasion) is hard to meet on the mainland. Lowland rural landscapes in most settled countries include many scattered remnants of original mainland ecosystems, most in private ownership, supporting a disproportionately large repository of threatened native ecosystems and species (Walker et al. 2006). In New Zealand, as elsewhere, it is especially important to protect such remnants and their biota from invasive predators, because forest fragments on private land represent the principal opportunity to protect biodiversity in lowland environments (Department of Conservation and Ministry for the Environment 2000; Green and Clarkson 2006). Ship rats (Rattus rattus, also called black or roof rats) are widespread and abundant pests world-wide. Their depredations on the fauna of native forests on the New Zealand mainland and offshore islands were recently reviewed by Innes et al. (2010a). As arboreal, nocturnal omnivores, ship rats are key predators of small forest passerines (Innes et al. 1999; Armstrong et al. 2006a, b; Brown et al. 1998), lizards, invertebrates (Gibbs 2009; Towns et al. 2006) and seeds. They can search the canopy for vulnerable native fauna with great thoroughness and agility, especially in forests that offer a complex, three-dimensional habitat (Foster et al. 2011). Within continuous mainland forests of the North Island, ship rats are very common, and vitally necessary control operations are undone within months by rapid repopulation of cleared areas (Innes et al. 1995). Small and isolated forest fragments are much more numerous than large ones, and yet the times required to achieve a specified conservation outcome in them, and conditions favouring or preventing it, have seldom been analysed. The commonest and simplest form of protection is stock fencing, but, unexpectedly, we have found that invasive rats may be significantly more abundant in fenced compared with unfenced fragments (Innes et al. 2010b). The demographic characters of invaders are almost unknown, largely because they cannot be determined from the results of control programmes (the majority) using toxic baits, and reinvasion is rarely monitored by carcase analysis. In short stature habitats of various kinds, such as alpine tussock, very young pine forests and grassland, ship rats are much less common than in forest (Innes 2005). One implication of this is that they might be expected to avoid the open, short-grassed pasture that mostly surrounds these fragments, where they would be more vulnerable to attack from predators. Hence, one might expect that ship rats could be slow to disperse to forest fragments surrounded by grazed pasture, much as populations of the Australian native rat R. fuscipes eradicated from forest patches could be

3 Eradication and reinvasion by Rattus rattus in forest fragments 2393 slow to recolonise across surrounding cleared regions (Peakall and Lindenmayer 2006). Confirmation of any factor inhibiting reinvasion would greatly encourage the implementation of pest control operations in privately-owned forest fragments, with potentially valuable consequences for biodiversity conservation in rural areas. Reinvasion is a serious threat to all pest control programs, because it is the most likely cause of failure to achieve protection of core values (Pascal et al. 2004; Abdelkrim et al. 2007; Russell et al. 2010). The sink effect, induced by eradication of local populations connected by gene flow to a wider meta-population (Russell et al. 2009b), makes reinvasion inevitable. Eradication operations are bound to fail unless confined to eradication units isolated from reinvasion, which may be much larger than their obvious geographical components. For example, off the coast of France, some eradication units for Norway rats comprise whole archipelagos, not individual islands (Abdelkrim et al. 2005). The limited data available for ship rats suggests that effective eradication units are likely to comprise archipelagos of forest fragments. For example, Russell et al. (2009b) found no genetic distinction between samples of ship rats from two mainland fragments separated by 400 m of grazed pasture. Likewise, Abdelkrim et al. (2010) found no significant genetic isolation between ship rats living in Puketi Forest, Northland, and those in two adjacent fragments each 3.5 km away. Therefore, documenting the extent of potential gene flow between local pest populations is vital information for pest managers, for two reasons. (1) Local populations that are not complete eradication units must be eradicated simultaneously (Abdelkrim et al. 2010) and (2) Genetic mixing across broad geographical areas reduces the chances that posteradication samples could distinguish between invaders and survivors at the population level (Russell et al. 2009b, 2010). In this study we aimed to integrate demographic and genetic data, gathered during the course of a closely monitored eradication programme, that could document the processes of recovery of eight ship rat populations eradicated from separate forest fragments scattered over a large area (20,000 km 2 ), and to produce results of general significance for the management of rodent pests on mainland New Zealand. Objectives This paper describes part of a wider study of the interactive effects of routine management (stock fencing and pest control) on key ecological processes on pastoral land around rural Waikato, New Zealand. Elsewhere we have described the vegetation (Burns et al. 2010) and abundance of invertebrates (Didham et al. 2009) in 53 forest fragments managed under different combinations of strategies (Dodd et al. 2011), and the abundance of ship rats in eight (Innes et al. 2010b). Our primary hypotheses were as follows. (1) After an eradication, immigrant rats might be expected to differ from the previous residents in their age structure and breeding potential. To test this hypothesis, we identified rats known to have visited or arrived from outside as those marked with vital dye obtainable only in potential source areas, and searched for detectable differences in age, gender and breeding status between marked and unmarked rats. (2) The processes of eradication and replacement of rat populations in all eight fragments could be influenced by the nature of the fragments (isolated or connected to source areas, fenced or grazed by livestock); by precise timing (fragments 1 4 and 5 8 differed in trapping dates); and by longer-term seasonal effects. To test this hypothesis we applied a generalised linear mixed model with binomial distribution and logit link to the complete dataset, searching for possible inter-relationships between the proportion of marked rats collected at each site and for correlations suggesting how the above factors might influence it. (3) The replacement populations appearing in the eradicated fragments could represent a slightly different gene pool from the original population. To test this hypothesis, we genotyped all rats collected throughout the study, and checked for genetic differentiation between fragments and between original and replacement populations. Study areas The eight fragments chosen are scattered along a range of hills km SE of Hamilton City in the central

4 2394 C. M. King et al. Waikato Region, North Island (Fig. 1), as described in Table 1 and in further detail by Innes et al. (2010b). All the fragments were small (\10 ha) cutover remnants of previously continuous conifer/broadleaved evergreen native forest invaded by ship rats after about 1860 (Atkinson 1973). Forest clearance began centuries ago after Maori settlement, but these fragments probably achieved their current form by about 1900 (Clarkson et al. 2002). All are now surrounded by grazed pasture. The northern-most five fragments surrounding Te Miro, a 400 ha forested reserve, are all within 5 km of each other and km from the southern-most three in the Whitehall area, which are all within 2 km of each other. The fragments averaged 5.3 ha in size (range ha); four were isolated (separated from potential source areas in adjacent forests by 55, 100, 105, and 250 m, measured edge to edge) and four connected by woody vegetation (e.g. strips of forest or hedgerows) to source areas. Fenced fragments protected from grazing by livestock supported much higher densities of rats before (though not after) eradication, but connectedness to an adjacent source of immigrants did not influence either original density or the chances of reinvasion (Innes et al. 2010b). The forest canopies of all the fragments were dominated by tawa Beilschmiedia tawa, mangeao Litsea calicaris, mahoemelicytus ramiflorus, pukatea Laurelia novae-zelandiae, and rewarewa Knightia excelsa and treeferns. Data on regeneration of key canopy and understorey vegetation, on litter mass and decomposition rates, and on the composition of the detritivore fauna, in relation to management (fencing, pest control etc.) regimes, are reported elsewhere. None of the fragments had been subject to effective pest control programmes (trapping, poisoning) targeting brush-tail possums and ship rats in the 2 years before our field sampling began. For comparison between the populations of ship rats in the fragments trapped to extinction, and a population sampled only lightly at 1 3 month intervals, we used data from Pureora Forest Park, a large forest in the central North Island supporting comparable forest vegetation (Innes et al. 2001). Methods We used a combination of data collection methods, each with different advantages and all linked to precisely defined locations by GIS, in order to build an integrated picture of the physical, demographic and genetic characteristics of the original and replacement populations. Footprint tunnels detected the presence of rats in a fragment, and the arrival of replacements, without disturbing them; bait markers detected the previous presence of some individuals in a source area; snap trapping implemented the eradication programme and provided samples for necropsy and genetic analysis; hair tubes supplemented the other detection methods and also provided information on movements of those individuals that entered both hair tubes and traps; and genotyping identified the extent of the wider population sampled in the eight fragments. This combination of ecological and molecular techniques applied over such a large geographic area is unusual (Peakall and Lindenmayer 2006) but very informative. Eradications Tracking tunnels containing Black Trakka TM tracking cards were set on a ± 8 m grid throughout each fragment, operated for one night immediately before the first eradication and at monthly intervals thereafter. The resulting tracking indices, calculated according to standard methods, were reported previously (Innes et al. 2010b). Victor snap traps (passed by NAWAC standards as humane kill traps for rats) were set in tunnels at and in the spaces between tracking stations (i.e., on a m grid) in numbers and on dates shown in Table 1. The traps were set inside tunnels just large enough to accommodate the trap, to discourage nontarget species, guide target rats across the treadle, and to provide public safety. They were baited with peanut butter, checked and re-baited daily until no rats were caught. Then, the tracking stations were reset, and both trapping and tracking continued together. Our aim was first to eradicate the entire population of rats in each fragment, and then to monitor how long the fragment would remain clear of replacements. For the first eradication, we set a rigorous end point for any given fragment, defined as the day on which no rats had been trapped or tracked for the three previous consecutive nights. At monthly intervals thereafter, we set tracking tunnels, checked daily, to test for reinvasion. If rats had reappeared,

5 Eradication and reinvasion by Rattus rattus in forest fragments 2395 Fig. 1 Map of the study area in the Waikato province east of Hamilton, showing the relative positions of the eight forest fragments. Lake Karapiro occupies part of the course of the Waikato River southeast of Cambridge Te Miro 2 8 Whitehall S 170 E Lake Karapiro 0 Study fragment Other forest 5km Lake Karapiro we immediately began snap-trapping again. For the second eradication, we defined our end point in all fragments as 10 days since the start of trapping. We divided the eight fragments into two groups, numbers 1 4 and 5 8, arranged to enable the paired comparisons of fencing and isolation that we expected to have primary influence on the results. Because it was logistically impossible to treat all the fragments simultaneously, both eradications in fragments 1 4 were done earlier than in fragments 5 8 (Table 1). Although the difference between the two groups of fragments was analysed under the heading Location, the distinction between them was primarily a matter of timing. We did not expect any geographical effect, so did not attempt to control for it: the northern group included Fragments 1, 2, 4, 7

6 2396 C. M. King et al. Table 1 Study areas, all located km SE of Hamilton, central Waikato, where rats were trapped between December 2007 and June 2008 Fragment Area (ha) Fenced or Grazed G F G F G F G F Connected to source, or isolated C C I I C I I C Distance from source (m) Number of traps RhB baits placed or renewed 11 Dec, 10 Jan, 21 Feb, 28 Mar, 10 Ap 17 Dec, 10 Jan, 27 Feb, 27 Mar, 10 Ap 10 Dec, 10 Jan, 27 Feb, 28 Mar, 28 Ap 11 Dec, 10 Jan, 26 Feb, 31 Mar, 28 Ap 10 Dec, 17 Jan, 26 Feb, 26 Mar, 23 Ap 17 Dec, 17 Jan, 27 Feb, 27 Mar, 28 Ap 12 Dec, 17 Jan, 26 Feb, 28 Mar, 23 Ap 13 Dec, 17 Jan, 26 Feb, 27 Mar, 24 Ap Distance bait station to nearest fragment edge (m) Traps set for 1st eradication 10 Jan 10 Jan 10 Jan 10 Jan 14 Feb 14 Feb 14 Feb 14 Feb Traps set for 2nd eradication 11 Ap 11 Ap 29 Ap 29 Ap 21 May 8 May 21 May 21 May Number of hair tubes a 10/6 21/10 33/10 19/9 22/6 20/10 10/10 24/10 Dates hair tubes set 9 Jan 21 Feb 17 Mar 10 Ap 9 Jan 11 Mar 10 Ap 9 Jan 11 Mar 28 Ap 9 Jan 17 Mar 28 Ap 13 Feb 20 Mar 23 Ap 20 May 13 Feb 7 May 13 Feb 23 Ap 20 May 13 Feb 20 May Genotypes/matches b 0 12/ /1 12/3 During the first eradication, trapping ended when no rats had been trapped or tracked for 3 consecutive nights, and during the second, after 10 days. For density indices derived from footprint tracking, taken immediately before each eradication started, see the matching Table 1 in Innes et al. (2010b) a Number of hair tubes set in eradication and reinvasion source areas respectively b Number of hair samples genotyped, and the number of genotypes matching other hair or ear samples and 8, and the southern group numbers 3, 5 and 6 (Fig. 1). Rat population data All rats caught were returned to the lab for dissection. We removed all the whiskers and half an ear from one side, and sent them for further analysis. We recorded gender, colour morph, whole body weight, head and body length, tail length and reproductive condition. Sexual maturity Sexual maturity was classified as a simple binary category, yes or no, regardless of whether the individuals so classified were in active reproductive condition at the time of capture. Extensive breeding data for ship rats analysed by Innes et al. (2001) from various North Island forest types at Pureora Forest Park have previously shown that no indicator of reproductive activity varied significantly with habitat, including % females pregnant; mean number of live embryos; % with uterine scars; mean number of uterine scars; % lactating; % pregnant females with resorbing embryos; % males with scrotal testes (Innes et al. 2001: tables 6 and 7). For females, these characters also did not vary between summer and autumn, which were the only two seasons sampled here. The proportion of fertile males did not vary with habitat, but declined slightly in autumn. For the present analysis, we therefore defined sexually mature females as those with active uteri ( string or cord condition, as opposed to the thread condition of juveniles) or pregnant or with uterine scars. Sexually mature and fertile males were

7 Eradication and reinvasion by Rattus rattus in forest fragments 2397 defined as those with scrotal, active testes containing visible tubules in the epididymes. The lack of any correlation between these classifications and habitat in the Pureora sample (n = 547, summer and autumn captures only) showed that the same classifications could be applied with confidence to the entire sample of rats collected from our forest fragments that were in good enough condition to be classified by age, gender and reproductive condition (n = 509). Age groups Innes et al. (2001) also showed that body weight controlled for season, year and gender increases predictably with age estimated from tooth wear, regardless of habitat. For this study we divided the samples into three age-related groups by weight. The lower boundary of the middle group, the young adults, was defined as the upper limit of body weights of all individuals still not sexually mature, which was found by inspection of the ranked data from the current (forest fragments) samples (109 g in males and 79 g in females). The upper boundary of this group was based on the average whole body weight of rats from Pureora (males 141 g, females 121 g). We confirmed that these groups represented real categories distinctly different from each other in age by applying the same weight-class definitions retrospectively to the Pureora data set (summer-autumn samples only) (Table 2). Table 2 Definitions of weight groups used as a proxy for age classifications of 509 Fragments rats, compared with the actual toothwear classes of 547 Pureora rats allocated to the same weight groups Age group Fragments Body weight range (g) Pureora n Mean toothwear class Males 1 \ [ Females 1 \ [ Totals SD Identifying the sources of re-established populations Rhodamine B marking In forest adjacent to each fragment, located m (average 228 m edge to edge) from each fragment, we set out up to ten possum-proof bait stations at 40 m spacing a month before trapping was due to begin. Each initially contained 200 g of bait pellets (RS5 cereal prefeed pellets, made by Animal Control Products, Wanganui) surface-coated with Rhodamine B (RhB) biomarker at 0.1% concentration. RhB dye can be detected in external keratinous tissue such as hair and claws (Fisher 1999). Baits were replaced with 100 g of fresh stock at monthly intervals, on the dates shown in Table 1. During necropsy, all rats were checked visually for RhB staining on the fur of the paws, face and belly. We then selected 6 7 medium to long whiskers from each carcase, preferably those with an intact follicle and a fine tip, for examination under flourescence microscopy. Whiskers were washed and drymounted, at least 6 to a glass microscope slide, under a 60 mm #2 coverslip, and examined at 1009 under a compound microscope fitted with a 50 W mercury vapour arc lamp and a Leitz Ploemopak epifluorescence unit. Green excitation was produced by an M2 filter set (excitation BP nm, dichroic mirror 580 nm, barrier filter LP 580 nm) which passes all wavelengths in the yellow, orange and red range of the spectrum. Autofluorescence, a natural property of some biological materials, can be distinguished from a true RhB signal by adding an I2 block. The resultant blue excitation gives a stronger signal from the autofluorescent sources and a substantially reduced rhodamine excitation. We scored whiskers showing this relative change of intensity as negative for RhB. We examined how the proportions of RhB-marked rats collected at each site and in each trapping session might vary with isolation (connected to a source or not), habitat disturbance (grazed or not), trapping history (first eradication or second), timing of the first eradication (earlier in fragment locations 1 4 than in 5 8) and day of operation (days 1 6 or day 7?). Connectedness was included because, although we have already shown it did not affect the chances that some rats would reappear after the first eradication (Innes et al. 2010b), it

8 2398 C. M. King et al. might affect the age, gender, reproductive or genetic distributions of which rats arrived. We checked for possible inter-relationships between all these factors by applying a generalised linear mixed model with binomial distribution and logit link to the complete dataset. The model included random terms (hence extra-binomial variation) for the individual fragments and repeats within these. It adjusted the estimated percentage frequencies for each variable by controlling for imbalances in the others. To check for a possible seasonal variation in the supply of immigrants, we divided the total sample into four periods of time (roughly a month each) defined from the dates on which the Rhodamine B baits were renewed in each area (dates given in Table 1). Period 2 started on the date of the first renewal of marked baits in mid January and ran until the second renewal at the end of February, and so on to period 5 running from the last bait renewal in late April until the end of the study in late May. We assumed that any rats we caught within a fragment that were clearly marked with RhB (but not only these) could be classified as definite replacements. We do not call them reinvaders because some could have been wandering residents. Hair tubes Before and during the eradications, the rat populations in the fragments, and in adjacent reinvasion source areas, were monitored with hair tubes (Gleeson et al. 2010) baited with peanut butter and set at trap sites (on a m grid) 2 4 times through the study as shown in Table 1. GPS positions of all hair tubes were recorded to ±8 m. Hair samples were collected after 1 day, and stored dry pending DNA extraction. We searched for exact matches between genotypes derived from hair samples collected from more than one hair tube, and between hair samples and ear tissue from carcases, assuming that an exact match represents the same animal sampled on separate occasions. Distances between matching sample sites indicate minimum movements by individuals between or within fragments. DNA extraction and genotyping We extracted genomic DNA from ear samples stored in 95% ethanol, and from dry hair samples including at least five clearly visible follicles, using the Corbett X-tractor Gene automated DNA solid tissue extraction system following the manufacturer s recommendation with external digest (Corbett Robotics, Brisbane, Australia). We used a total of nine microsatellite markers, previously developed from R. norvegicus genome mapping (Jacob et al. 1995), to genotype each sample: D2Rat234, D5Rat83, D7Rat13, D10Rat20, D11Mgh5, D15Rat77, D16Rat81, D18Rat96, D19Mit2. We performed PCR amplification in 10ul volumes, following Abdelkrim et al. (2010), and ran the PCR products as two multiplex runs on an ABI prism 3130 capillary electrophoresis system (Applied Biosystems). We derived genotypes using GeneMapper software (Applied Biosystems). From the tissue genotypes, we calculated an overall P sib score using GenAlEx v 6.2 (Peakall and Smouse 2006) in order to determine the probability that a sibling would possess an identical genotype. From the hair samples, we made three replicates per sample, and did not accept a genotype as complete unless there was consensus from at least two of the three replicates. We assessed evidence for allelic drop-out, scoring error due to stutter, and the presence and frequency of any null alleles with MICRO-CHECKER (Oosterhout et al. 2004) using a Bonferroni adjusted 95% confidence interval and 10,000 repetitions. We calculated genetic diversity indices and pairwise genetic differences, and conducted Principal Components Analysis (PCA) using GENALEX v 6.2 (Peakall and Smouse 2006), and conducted tests for Hardy Weinberg Equilibrium (HWE) and linkage disequilibrium using GENEPOP v 4.0 (Raymond and Rousset 1995). The fixation index (F ST ) is calculated from the correlation of randomly chosen alleles within one sub-population relative to that found in the entire population. Population differentiation can then be estimated from a series of pairwise comparisons. Values for F ST between are consistent with little genetic differentiation (near panmixis); indicates moderate, and high differentiation between populations. We calculated pair-wise F ST estimates using FSTAT (version ) (Goudet 1995), and determined their significance by Fisher s method (Manly 1985: ; Raymond and Rousset 1995). We used the Bayesian clustering method implemented in STRUCTURE (Hubisz et al. 2009)to estimate the best-supported number, K, of genetic

9 Eradication and reinvasion by Rattus rattus in forest fragments 2399 clusters. We chose the admixture model with correlated allele frequencies, and conducted 10 replicates for each value of K varying from 1 8 for each run, consisting of a burn-in period of 100,000 MCMC (Markov Chain Monte Carlo) steps followed by 106 iterations. Results Altogether, 517 ship rats were collected from the 8 fragments over two eradication attempts (mean 46, range each; Table 3), of which 509 were classified by age and reproductive condition, and 465 by RhB status. Sample and sub-sample sizes are given in context as appropriate in Tables 2, 3, 4, 5, 6, 7, 8, 9, 10. We also caught 11 Norway rats (in two fragments), one weasel and two mice. Summer eradication In January-April 2008, the mean time needed to meet our definition of eradication in each of the eight fragments was 33 nights (range 3 67). Summer rat density estimated from the first six nights of trapping was significantly higher in fenced (6.5 rats ha 1,SE 1.4) than in stock-grazed fragments (mean 0.5 rats ha 1, SE 0.4; P = 0.02, and the time to complete the first eradication averaged 47 nights in fenced and 19 nights in grazed fragments (Table 3). Ship rats replaced the eradicated populations in all fragments within a month, regardless of fencing or isolation (Innes et al. 2010b). Replacement (autumn) rat populations The mean time between initial eradication and the second tracking index (Table 1) was 63 nights, range nights. The relative abundances of the reestablished rat populations were often different from those measured in the same fragments 4 months previously (Table 3). Rats were no longer significantly more abundant in the fenced fragments, at least over the 10 days we observed them (Innes et al. 2010b). The autumn populations were not pursued to local extinction. Distribution of rats marked with RhB Rhodamine B dye had been available since December only in adjacent areas outside the fragments (Table 1). From January onwards, only one of the eight fragments did not produce at least one marked rat from among the previously untrapped populations. Of the 517 rats examined, the RhB status was ambiguous or unknown in 52, and definite in 465, of which 94 were marked (Table 3) The overall mean minimum distance they had moved, from the centre Table 3 Proportions of marked rats caught in eight forest fragments (total 94 of 465 rats with definite RhB status) Fragment First eradication (summer) Days to complete Rats/ha (days 1 6) Rats with clear RhB status Rats marked n (total 79) Rats marked % of Total rats trapped Second eradication (autumn) Rats/ha (days 1 6) Rats with clear RhB status Rats marked n (total 15) Rats marked % of Total rats trapped Pre-eradication tracking indices are given by Innes et al. (2010b). Density estimates are given only for the first 6 days of trapping (Watkins et al. 2010), after which immigration is expected to increase the total catch to many more rats than could live in each fragment at one time. The second eradication attempt was stopped after 10 days

10 2400 C. M. King et al. Table 4 Distribution of 465 rats with determined Rhodamine B status Factor SD Significance Isolation Connected Isolated % marked 13.7% 18.1% 6.9% P = n Eradication 1st 2nd % marked 17.9% 13.9% 4.9% P = n Location % marked 8.4% 27.8% 6.8% P = n Disturbance Fenced Grazed % marked 15.1% 16.5% 7.1% P = n Day of trapping 1 6 7? % marked 10.9% 22.3% 3.8% P = n Period since renewal of RhB Jan Feb Feb Mar Mar Ap Ap May % marked 16.3% 39.9% 5.8% 12.0% 9.2% P = n For each factor listed, the model calculates the percentage of rats marked in each of two conditions plus the standard error and significance of the differences attributable to that factor, after allowing for all the other factors. The two groups comprising Fragments 1 4 and 5 8 (in sub-heading Locations ) are distinguished by the slightly different timings of the operations in each (Table 1), not their geographic locations (Fig. 1). The exact significance values are given for all comparisons, in bold type if P \ of a minimum convex polygon enclosing the locations of the set of bait stations nearest the fragment to the point of capture within it, was 387 m (range m, n = 88); means for marked rats collected from individual fragments varied from 282 to 547 m. The distribution of marked rats through the samples was uneven, but they continued to appear in all fragments throughout the eradication process. Of the 94 marked rats collected, 79 were caught during the first eradication (Table 3), and 10 of these were caught on the first day of trapping (12 January), whereas only 15 marked rats were caught during the second eradication. After accounting for other factors influencing the samples, our GLM model estimated these proportions as 19.9 and 13.9% respectively (P = 0.413, Table 4). The proportion of marked rats was significantly higher (P = 0.004) in fragment numbers 5 8 (Table 4), where the first eradication was done a month later than in fragments 1 4 (Table 1). The proportion of marked rats was also higher after the first 6 days of trapping (P = 0.003), as would be expected if they represent the immigrant group, but it was not higher in connected versus isolated fragments, or in fenced versus grazed fragments, or during the first versus the second eradications. Visual inspection of the plotted capture positions of the marked rats collected within each fragment showed no perceptible clustering along the edge habitats closest to the sources of RhB-marked bait. We compared the proportion of marked rats caught during four successive monthly periods, defined in Table 4, to check for seasonal effects whilst controlling for isolation, location and other factors as before. This comparison confirmed a significant variation in proportion of marked rats through the trial, but almost all of that effect was due to the large influx of marked rats (40%) observed from mid January to the end of February. The total number of rats caught increased slightly in May, but only 12% of these were marked, implying no increase in the supply of immigrants after the end of the breeding season in autumn.

11 Eradication and reinvasion by Rattus rattus in forest fragments 2401 Table 5 Distribution of weight groups and sexual maturity, total fragments sample Body weight range (g) n % sexually mature Difference between weight groups in % mature by gender Males 1 Juveniles \ Young P \ Old [ Females 1 Juveniles \ Young P \ Old [ Totals Table 6 Distribution of age/gender classes among rats with and without RhB marker dye, which they could have acquired only outside the study fragments Total classified by age/gender Marked rats (% marked of total age group) Unmarked, Eradication 1 only Unmarked, Eradication 2 only Males 1 Juveniles (28.3) Young 47 4 (8.5) Old (20.8) Females 1 Juveniles (20.3) Young (23.8) Old (8.1) Totals * 129* The age structure of marked rats was significantly different from that of all unmarked rats (P = 0.001), and also from that of the replacement population caught during Eradication 2 (P = ) * Excluding rats of uncertain RhB status Population structure (age and gender) In the 509 rats collected in good enough condition to be classified, the six groups defined from gender and weight were distinctly different in age by, on average, one toothwear class (Table 2). This difference in age is consistent with the significant increase in the proportions of rats in the top compared with the middle weight classes of each gender that were sexually mature (Table 5). There was no overall difference in the proportions of males and females that were mature (P = 0.144). The age and gender ratios among the 94 marked rats differed significantly (P = 0.001) from those of unmarked rats (Table 6). The largest single group of marked rats comprised juvenile males (28%), and the smallest, old breeding females (8%). Impact of removal trapping on age and gender ratios The reference samples from Pureora were collected according to a fixed, intermittent schedule in continuous forest, in which traps were never set for more than 12 days at a time (Innes et al. 2001), whereas the first trapping programme continued until a robust predetermined criterion of eradication had been met in each fragment, which took up to 67 consecutive

12 2402 C. M. King et al. Table 7 Distribution of age/gender groups in samples of rats collected by sampling to and after local extinction (Fragments programme) and by fixed-period short-term sampling (Pureora), summer/autumn data only Group Fragments Pureora n % n % Males 1 Juveniles Young Old Mean male Body weight (g) sd Females 1 Juveniles Young Old Mean female Body weight (g) sd Totals P days (Table 3). These different sampling regimes were associated with a highly significant difference (P = 0.000) in the age structure of the rats collected (Table 7). Many more juvenile males were collected by trapping to extinction in the fragments than by the repeated, brief, low-level sampling at Pureora. The disproportionate representation of juvenile males in the prolonged Fragments samples was associated with a significant decline by g, se = 3.52 g (P \ 0.001) in the average male body weight of rats collected from the Fragments compared with the Pureora males (Table 7). The same difference did not reappear in the female body weights. Not all immigrant rats were marked, which raises the question of whether the marked rats were typical of the immigrant group as a whole. Our data suggest that they were. During the first eradication in the Fragments, the shift towards juvenile males in distribution of age/gender classes was already evident among the 206 rats caught after day 7 compared with the 160 rats caught on days 1 6 (P = 0.000, Table 8). If these marked rats do indeed represent the unmarked immigrants during that period, there should be no difference in age distribution between Table 8 Distribution of age/gender groups in samples of rats collected on the first 6 days of trapping compared with subsequent days, Fragments programme Group Eradication 1 Eradication 2 Days 1 6 7? 1 6 7? Males 1 Juveniles Young Old Females 1 Juveniles Young Old Totals P the marked rats and the 127 unmarked rats caught on days 7?; that was so (P = 0.168, Table 9). By the second eradication, the proportion of juvenile males caught was not different in the later versus early days of trapping (P = 0.191, Table 8). The marked rats no longer represented the immigrants, because total sample of rats collected during the second eradication, conducted late in the season, included fewer juveniles and more young adults than in the marked group (P = 0.005). Genetic diversity and population differentiation A total of 493 ear samples from all fragments and 29 hair samples from three fragments (Table 1) were successfully genotyped at nine microsatellite loci identifying nine pairs of hair-hair or hair-ear matches. Eight loci had between zero and \10% null alleles. One locus, D10Rat20, had [10% null alleles at all sites, so we removed it from subsequent population analyses, as did Abdelkrim et al. (2010) for the same reason. Allele frequencies and Hardy Weinberg Equilibrium (HWE) summary statistics for each fragment are given in Table 10. The mean number of alleles per locus varied by fragment from 5.9 to Fragments with private alleles had, on average, higher numbers of alleles per locus, and these private alleles were all low frequency (\0.05). All fragments had similar observed heterozygosities ranging from Only Fragment 5, the area from which fewest rats

13 Eradication and reinvasion by Rattus rattus in forest fragments 2403 Table 9 Distribution of rats classed by both RhB status and age/gender group collected during the first eradication compared with the unmarked rats collected after day FST Group Marked Unmarked Days All days 7? Males 1 Juveniles Young Old Females 1 Juveniles Young Old 9 27 Totals P were collected (Table 3) showed a significant departure from HWE. For our entire collection (n = 493 genotyped rats), F ST values ranged between and 0.096, indicating low to moderate genetic differentiation between fragments. Plotting pair-wise F ST values against the corresponding linear separations of the eight fragments showed no significant correlation between genetic and geographic distance (Fig. 2). Principal Components Analysis based on pairwise genetic differences showed no correlation between rats caught during the first versus the second eradication attempts (Fig. 3). The cumulative percentage of variation explained by the first three PCA axes was very low (24.13, and 59.09%) and subsequent components added progressively less, suggesting a ball-like data cloud. Analysis using STRUCTURE 3.2 also showed little genetic structure between fragments. Models that subdivided the total dataset into K [ 1 F ST value Distance between pairs, km Fig. 2 Pair-wise comparisons of all F ST values plotted against the cross-country distances between each pair of fragments were not supported when compared with the null hypothesis of no substructure. Hair/hair and hair/ear matches We collected 78 hair samples from 230 hair tubenights, of which 29 (Table 1) from three fragments returned genotypes of good enough quality for subsequent analysis. Hair samples from the other five fragments were not analysed. The overall P sib score was 0.004, which is below the 0.01 reported as the upper limit required for individual genetic assignment and population estimation (Mills et al. 2000). We therefore considered it unlikely that matching hair and tissue genotypes were from siblings, as opposed to two samples from the same individual. We obtained pairs of matching hair samples (identical at all nine loci) from three individual rats that visited two hair tubes within a source area, 45 and 87 m apart in Fragment 8, and 32 m apart in Fragment 7 (mean 54.5 m). The hair tubes were Table 10 Summary statistics of genetic diversity for rats trapped within eight Waikato remnant forest fragments Fragment n Na Priv He Ho * n number of samples, Na mean number of alleles per locus, Priv. mean number of private alleles per locus, Ho mean observed heterozygosity, He mean expected heterozygosity * denotes a significant departure from HWE

14 2404 C. M. King et al. Component 2 E1 E 2 None of the marked rats trapped had left matching samples in hair tubes in source areas. Hence none of the nine hair-ear matches (n = 9) or the 94 carcases marked with RhB identified individual animals moving into a fragment from a source area. Component 1 Fig. 3 Principal components analysis (PCA) performed with GENALEX software using the nine microsatellite loci from rats caught during the first eradication (E1) versus rats caught during the second eradication (E2) always set 1 day and cleared the next day, so all these distances refer to net movements over 24 h. The identities, ages and genders of these rats were unknown because none of the genotypes identified from hair tubes in source areas matched any of the rat ears genotyped from the adjacent (or any) fragment. Four hair samples returned identical matches with an ear sample from four rats trapped; three were caught 27, 28, and 55 m from the tube where the hair was collected within Fragment 2 and one (matching at 8 loci) at 40 m from the hair tube within Fragment 8. In every case, the hair sample was collected on the day before the traps were set, and the rat was collected 2 days later, on the first day of the first eradication. All four pairs of hair-ear matches came from old females that we sampled inside a fragment, both before and after death, presumably within their home ranges. A fifth hair sample matched an ear sample from another old female caught 99 m from the hair tube on the first day of the second eradication in Fragment 2. We cannot distinguish whether this rat was (1) a survivor of the 27 days of the first eradication attempt, or (2) a post-eradication colonist arriving from or via the broad shelterbelt that connected this fragment to the adjacent Te Miro forest reserve. This rat was not detected by the hair tubes and RhB bait stations along the Te Miro forest edge, and was not genetically distinct from the rats caught during the first eradication. The minimum overnight local movements of these five old females averaged 50 m. Only a single old male recorded a hair-ear match, again within Fragment 2. This rat was caught on the second day of the first eradication, in a trap 203 m from the hair tube it had visited 3 days previously. Discussion Ship rats are key pests throughout New Zealand. Conservation managers working to protect endemic species on the mainland have long known how rapidly reinvasion can reverse the beneficial effects of local ship rat eradication programmes (Innes et al. 1999), but the demographic characters of replacement populations have remained obscure. Here we present new data decisively demonstrating three key observations on the re-invasions we observed: (1) the replacement rats were derived from a meta-population so large that our samples, spread over 20,000 km 2, could not define its boundaries; (2) gene flow between forest fragments within the metapopulation was not inhibited by large areas of intervening non-preferred habitat, so a few hundred metres of pasture clearly could not protect our study areas from reinvasion after eradication; and (3) movements of individual rats towards the cleared fragments from outside were not stimulated by human interference on the contrary, we detected substantial numbers of marked rats on the very first day of trapping, suggesting an ongoing dynamic interaction between the rats living in the fragments and in neighbouring areas that was already in progress before our first eradications began to take effect. Demography of the replacement population Marked rats could have picked up the RhB marker dye if they were either (1) resident in a fragment with home ranges extending into a source area, which had wandered outside the fragment, picked up the bait marker and then returned, or (2) immigrants which had newly arrived from a different area outside the fragment. We acknowledge that our methods could not objectively distinguish between these two categories, so the rats collected from the original population comprised, at least at first, an unknown mixture of original residents and immigrants.

15 Eradication and reinvasion by Rattus rattus in forest fragments 2405 However, our criteria for achieving an eradication were severe, and allowed for trap evasion (we specified that no rats be detected in either traps or footprint tunnels for three consecutive nights), so we think it likely that few or no rats remained by the end of the first eradication. Dispersal is male-biased in most rodents, including Rattus sp. (Calhoun 1963; Krebs et al. 2007) and is most frequently undertaken by younger, non-breeding individuals (Peakall and Lindenmayer 2006). For example, Stokes et al. (2009) found higher proportions of subadults, and no breeding females at all, in replacement populations of R. rattus collected from eradication sites. If our marked rats fairly represented the immigrants, they would be expected to differ significantly in age, gender ratio and reproductive (but not genetic) characters from unmarked rats. This was true (P = 0.001, Table 6); the marked rats included more immature juvenile males than any other age/gender group, and fewer old adult females (Tables 5, 6), and the difference was established relatively early (after day 7 of the first eradication: Tables 7, 8, 9). The implication is that, although rat populations re-establish rapidly after an eradication operation, the new group can be expected to start with a different demography and a lower reproductive potential compared with that of the original population. These observations have some important implications for management of invasive rats on the mainland, because age distribution and sexual maturity are two of the parameters most affecting population rate of increase. This information gives some interesting clues about the population dynamics of invaders, although it does not help conservation in the short term since rats of all ages and both genders are equally unwelcome conservation pests. In our study, the movements of individual rats revealed by both genetic and RhB techniques were consistent with previous knowledge of male and female ship rat range lengths in New Zealand podocarp-broadleaved forest (summarised by Innes 2005). Five of the six matched hair and ear samples came from old females caught on the first day of a trapping session, and the sixth from an old male caught on the second day. These data suggest that old rats freely ran through both hair tubes and trap tunnels, and so were likely to be caught as soon as the traps were set. Larger, older and socially dominant rats are often the first to explore new devices containing food, because fear of conspecific aggression prevents smaller, younger rats exploring or feeding in a confined space inside a tunnel or bait station (Quy 2003). If the old females had often moved outside the fragments, they might have been more often marked with RhB, but, on the contrary, Table 6 shows that they were the least likely group to be marked. Since RhB dye was available only outside the fragments, this confirms that the old females were the most sedentary group (with smallest home ranges and shortest local movements) resident within the fragments (Innes 2005). Older adult males are much more mobile and hold larger home ranges than old females (Hooker and Innes 1995), perhaps because older males are at greater risk of mating with their own offspring. We did not document any larger movements that could clearly be seen as natal dispersal. In other species of Rattus, dispersal studies have revealed typically small local movements, with occasionally some much longer, e.g., \100 m and up to 1,750 m in R. fuscipes (Bentley 2008) or c.62 m and 11 km in R. norvegicus (Gardner-Santana et al. 2009). Such comparisons depend on assuming no difference in trappability between marked and unmarked rats, but we have no way to test for this. Our data illustrate incidentally the potential bias inherent in using the results of continuous snaptrapping for population analyses. Density estimates calculated from samples of rats taken as by-catch in permanently-set trap lines for carnivores (Dilks et al. 2003), assume that the probability of capture remains constant over the long term, but it does not (Watkins et al. 2010). Likewise, the proportions of age/gender classes (and hence, the reproductive potential) will be estimated differently from small rat populations trapped to extinction than from samples of rats taken from a large population sampled periodically. Genetic differentiation among fragments Table 10 may be compared directly with Abdelkrim et al. (2010: table 1), who document the same statistics for 274 rats from 5 km 2 Puketi forest (Northland) and 39 from two outlying fragments each separated by 3.5 km from Puketi in opposite directions. The values for observed and expected mean heterozygosity were comparable across all samples. The lack of any significant correlation between genetic and geographic distance among our 8