K. M. Jenkins, Ecosystem Management, University of New England, Armidale, NSW, 2351, Australia (fax +61 2 67732769; e-mail email@example.com).
1Water extraction from arid-zone rivers increases the time between floods across their floodplain wetlands. Less frequent flooding in Australian arid-zone rivers has impaired waterbird and fish breeding, killed riparian vegetation and diminished invertebrate and macrophyte communities. Restoration currently focuses on reinstating floods to rejuvenate floodplain wetlands, yet indicators to measure the success of this are poorly developed.
2We explored the application of criteria for ecologically successful river restoration to potential restoration of floodplain wetlands on the Darling River, arid-zone Australia. Using emergence of micro-invertebrates from resting eggs as an indicator, we compared responses of taxa richness, densities and community composition in floodplain lakes with different inundation histories.
3Increased drying of floodplain lakes reduced the number of micro-invertebrate taxa. Several key taxa were absent and faunal densities (particularly cladocerans) were reduced when the duration of drying increased from 6 to 20 years.
4A conceptual model of the ecological mechanisms by which restoration of flooding regime could achieve the target of preserving micro-invertebrate community resilience predicts that reducing the dry period between floods will minimize losses of viable resting eggs. Protection of this ‘egg bank’ permits a boom in micro-invertebrates after flooding, promoting successful recruitment by native fish and waterbirds.
5Synthesis and applications. In arid-zone rivers, micro-invertebrate densities and community composition are useful indicators of the impact of reduced flooding as a result of water extraction. Critical to successful native fish recruitment as their first feed and as prey for waterbirds, micro-invertebrates are a potential early indicator of responses by higher trophic levels. Taxon richness, density and key taxa present after flooding, all indicators of resilience, can be incorporated into targets for arid-zone river restoration. For example, one restoration target may be microcrustacean densities between 100 and 1000 L−1 within 2–3 weeks after spring flooding. These criteria can be applied to measure the ecological success of restoration projects seeking to recover natural flood regimes. Given the high economic cost of water in arid zones, convincing demonstrations of the ecological success of environmental water allocations are crucial.
Almost half the world's land area lies in the arid zone, characterized by low and variable rainfall (Comín & Williams 1994; Thoms, Beyer & Rogers 2006) yet supporting extensive arid-zone wetlands and rivers that drain or flow through these dry regions (defined in Boulton 2000). In the largely arid Murray–Darling Basin, Australia, 24 700 wetlands larger than 1 ha cover 6306 000 ha (Kingsford, Curtin & Porter 1999). Although many of these wetlands connect to rivers during floods, flood frequency has been reduced by 20–81% as a result of water resource developments, and the length of time that floodplain wetlands remain dry is increasing (Kingsford 2000; Jackson et al. 2001; Thoms, Beyer & Rogers 2006). For example, the Macquarie Marshes, a 200 000-ha Australian Ramsar floodplain wetland that flooded naturally every 1–2 years, has not had a significant flood for 6 years (Jenkins, Boulton & Ryder 2005) and has reduced in size by 50% (Kingsford & Thomas 1995). Irrigation has dried out many other significant floodplain wetlands, including the California and Nevada wetlands in North America, the Aral Sea in central Asia, the Senegal Delta and Hadejia-Nguru wetlands in Africa, and the Doñana wetland in southern Spain (Lemly, Kingsford & Thompson 2000; Tockner & Stanford 2002).
However, drying is unavoidable: habitats contract and fragment as a result of water loss, prompting resistant strategies (retreats to refugia, diapause) and death because of stranding and desiccation (Stanley, Fisher & Grimm 1997). Targets and indicators for arid-zone river restoration must incorporate this variability and ability to recover after flooding that typifies the ‘boom–bust’ ecology of arid-zone rivers and avoid interpreting ‘bust’ populations as ‘unhealthy’ (Sheldon 2005). Although native biota in arid-zone rivers are extremely resilient, responding rapidly to flooding despite years without water (Kingsford, Curtin & Porter 1999; Jenkins & Boulton 2003), the ecosystem processes and resilience of algae, invertebrates, vegetation, fish and waterbirds are adversely affected by loss of floods (Boulton & Lloyd 1992; Ellis, Crawford & Molles 1998; Kingsford & Johnson 1998; Dahm et al. 2003; Valett et al. 2005). Depressed pulses of micro-invertebrates with increased duration of drying deprive fish, filter-feeding waterbirds and macro-invertebrates of a vital food supply and, as such, may be good early warning indicators of impending impacts in arid-zone floodplain wetlands.
This study investigated the association of emergence of micro-invertebrates from resistant resting stages with time since flooding. Does their resilience, as indicated by the response of taxon richness and density to flooding after dry periods, diminish with increased time since flooding? If drought-resistant micro-invertebrate resting stages are depleted because of a lack of flooding, we hypothesized that communities from lakes last flooded 20 years ago would have fewer taxa and lower densities than those from lakes last flooded 6 years ago. With these data, we wished to assess the potential application of criteria for ecologically successful river restoration (Jansson et al. 2005; Palmer et al. 2005) to the restoration of arid-zone floodplain wetlands via increased flooding.
study sites and flood history
Teryaweynya Lakes comprise an ephemeral floodplain lake system associated with 150 km of floodplains and channels on the lower reaches of the 2740-km Darling River in south-western New South Wales, Australia (see Fig. S1, Supplementary material). Sixty per cent of the 650 000 km2 catchment is semi-arid or arid (Thoms & Sheldon 2000). Inundation of these lakes relies on floods in the Darling River of long duration (> 30 days) at heights above 8·5 m, or 25 000 ML day−1 at Wilcannia. Records (1890–2000) indicate flood frequencies for individual lakes ranging from 1 in 10 to 1 in 50 years.
Although there are historical records available for flows of the Darling River, there are no gauge records for flows to individual lakes. Fortunately, many landholders have kept diaries of flood events for the creeks, anabranches and lakes of the lower Darling River dating back to the late 1800s. We compiled flood history records from a report by a local historian (Withers 1996) and interviews with local landholders. Lakes were clustered into four groups based on time since the last flood pulse: 6, 20, 50 and 106 years. We selected lakes that last flooded relatively recently (6 and 20 years) because we wanted to relate our findings to the hydrological effects of river regulation that are increasing the time between medium-sized floods from 6 (4–8) towards 20 (10–20) years (Thoms & Sheldon 2000).
Three lakes that last flooded 6 years (Pelican Lake, Dead End Lake and an unnamed lake hereafter referred to as Little Victoria Lake) and 20 years ago [Brick Kiln Lake, Albemarle 1 (unnamed) Lake and Brennans Lake] were selected to maximize interspersion across a broad geographical region. Although care was taken to exclude lakes known to have flooded from rain events in the last 20 years, it was later discovered that Brick Kiln Lake had flooded briefly from rainfall in 1990. The sampled lakes ranged from 60 to 433 ha and up to 3 m deep. The prevailing wind was to the north-east, evidenced by landholder records and geomorphology of lake dunes.
A hierarchical sampling design was used to investigate the effects of flood history (FH) and wind direction (WD) on the response of micro-invertebrates to flooding (see Fig. S1 in the supplementary material). Lakes (LK) were nested within flood history. Repeated measures were carried out on subjects (samples) on days 1, 3, 7, 10, 14, 21 and 28 after flooding (flood day, FD) (see below). The FH factor included two levels, with three lakes last flooded 6 years previously and three lakes that last flooded 20 years previously. Within each lake there were two sites, upwind and downwind, in case adult and resting stages were blown preferentially during drying. WD and FD were fixed factors and lakes were a random factor. Sites were located at similar altitudes towards the edge of lakes. Within each 20 × 20-m site, five samples were collected randomly (see Fig. S1 in the supplementary material).
sediment collection and inundation
Lake sediments were collected in March and April 1996 using methods described in Jenkins & Boulton (2003). Overlying detritus and vegetation were included in the samples. Sampling points that fell directly on perennial vegetation (e.g. spiny lignum Muehlenbeckia horrida) were moved left 1 m. Sediments to 10 mm depth were placed carefully in plastic microcosms (165 × 165 × 115 mm) for transport to a glasshouse.
Samples were flooded in a glasshouse from June–September 1996 to coincide with winter–spring flooding in the lower reaches of the Darling River. They were flooded in five blocks of 12 samples, with one sample from each treatment and lake (2 WD × 2 FH × 3 LK) included in each block. All treatments and blocks were randomly assigned positions in the glasshouse. Flooding was simulated using deionized water (1936 mL) because it was not logistically possible to inundate samples with floodwater from lakes 1500 km away. Water quality was measured in one randomly selected microcosm from each treatment replicate (2 WD × 2 FH × 3 LK) on flood days 1, 7 and 28. Water samples (100 mL) were collected in acid-washed polyethylene bottles, filtered (GF-F Whatman glass fibre filters, 0·7 µm) and frozen before analysis of soluble reactive phosphorus (SRP) (Murphy & Riley 1962), dissolved oxides of nitrogen (NOx-N) (Wood, Armstrong & Richardson 1967) and conductivity. Water temperature was measured in each treatment twice daily throughout the experimental period. On days 1, 3, 7, 10, 14, 21 and 28, all samples were poured though a 56-µm mesh, and invertebrates were live counted under a microscope and returned to their microcosms. Full details of processing are described in Jenkins & Boulton (2003).
All data were analysed by repeated-measures analyses of variance (anova; see Table S1 in the supplementary material). Residuals were examined to check for heterogeneous variances and non-normality (Quinn & Keough 2002). Log10(x + 1) transformations were applied when necessary to homogenize variances. Dependent variables included taxonomic richness and densities of rotifers and cladocerans. All anovas were carried out using systat for Windows, version 9·0 (systat Incorporated, Evanston, IL). Tukey's tests were employed for multiple comparisons where significant differences among transformed means were detected. Equations for the explained variance of each term in the anova model were determined from estimated mean squares (see Table S1 in the supplementary material) (Quinn & Keough 2002).
Data were classified using twinspan (two-way indicator species analysis; Hill, Bunce & Shaw 1975) to identify assemblages characteristic of flood history. Analysis was done using all 98 taxa recorded in the samples. Of the 420 samples, 16 empty samples were excluded. To complement the twinspan, community compositions were also compared using non-metric multidimensional scaling (NMDS) computed with primer (Clarke & Warwick 2001) on a Bray–Curtis similarity matrix of fourth-root transformed data. Composite samples of the abundances for each taxon from five samples in upwind and downwind sites were analysed, reducing the data set to 42 samples (2 flood histories × 3 lakes × 7 flood days). Coordinates of each composite sample in a lake were plotted in chronological order, generating succession trajectories (cf. Boulton & Lake 1992) to assess change in community composition over time. The taxa that characterized each flood history treatment in each lake were identified using SIMPER (similarity percentages; Clarke & Warwick 2001).
Within 1 day of re-wetting, concentrations of NOx-N and SRP increased in samples from lakes that last flooded 6 compared with 20 years ago (Fig. 1). Conductivity increased in both flood histories from day 1 to 28, and was greater in lakes that had been dry for 6 years (Fig. 1). Dissolved oxygen also increased, generally reaching saturation by day 14. Water temperatures varied from 9 to 20 °C, comparable with those recorded in lakes (12·5–19·2 °C) near the study area (Jenkins & Boulton 2003).
Invertebrate taxon richness varied dramatically between lakes with different flood histories. In lakes dry for 6 years, 38 taxa were detected, 12 more than in lakes dry for 20 years. Eighteen taxa were unique to lakes dry for 6 years. Although eight of these occurred in fewer than five samples, several common taxa (the rotifers Asplanchna, Polyarthra, Filinia and Conochilus and calanoid copepods) were absent in lakes dry for 20 years. The six taxa that only occurred in the lakes dry for 20 years were from one sample and were all rotifers.
Significantly higher numbers of taxa emerged from lakes flooded 6 compared with 20 years previously, accounting for 23·5% of the variation (P < 0·001; Fig. 2) (term 1; Table 1). Taxon richness varied significantly over flood day (10·5% variation, term 7; Table 1), with fewer taxa on day 1 compared with days 7 (Tukey's P= 0·005) and 10–28 (Tukey's P < 0·001). Numbers of taxa on day 3 were lower than those on all other flood days (P < 0·006) except day 7. Taxon richness increased sharply over time in the lakes last flooded 6 years ago (Fig. 2).
Table 1. anova results for untransformed taxa richness and log10(x + 1) transformed densities of rotifers and cladocerans. Significant P-values (< 0·05) are shown in bold and variance components are indicated (%VC)
Source of variation
1 Flood history, FH
2 Lakes within FH, L(FH)
3 Wind direction, WD
4 FH × WD
5 L(FH) × WD
6 Samples (residual) Within samples
7 Flood day, FD
8 FD × FH
9 FD × L(FH)
10 FD × WD
11 FD × FH × WD
12 FD × L(FH) × WD
13 Samples × FD (residual)
The greatest variance in taxon richness, for all flood days combined, occurred in the interaction between lakes within flood history and wind direction (52·4%, term 5; Table 1). This reflected the low numbers of taxa in the downwind sites in Dead End Lake and Albemarle 1 Lake, reversing the trends seen in the other two lakes flooded 6 and 20 years ago (Fig. 2). Both the downwind sites with few taxa had low dissolved oxygen because of decaying vegetation in the samples.
Densities of rotifers increased more steeply over time in samples from lakes flooded 6 compared with 20 years ago (term 8; Table 1 and Fig. 3). Flood history accounted for 20·4% of variation in the model, but the similar densities on day 28 coupled with high variation among samples (39·0%) and lakes (10·1%) contributed to the non-significant test of flood history (terms 1, 2 and 6; Table 1). Rotifer densities differed among lakes within flood history (term 2; Table 1 and Fig. 3) because of higher densities in Brick Kiln Lake compared with the other two lakes (Fig. 3). The significant interaction between lakes within flood history and wind direction (term 5; Table 1) reflected reduced densities in the downwind site in Dead End Lake (Fig. 3) and the higher densities in the downwind site in Brick Kiln Lake (Fig. 3). Rotifer densities showed significant temporal variation (term 7; Table 1 and Fig. 3), with lower densities on day 1 than days 14, 21 and 28 (Tukey's P= 0·010, 0·001 and < 0·001, respectively).
Cladocerans were either absent or rare in lakes flooded 20 years previously, whereas they were detected in all sites from the recently flooded lakes, reaching densities of 1000 m−2 in most sites by day 21 (terms 1 and 5; Table 1 and Fig. 4). Again, densities were higher in downwind sites, except in Dead End Lake where cladocerans were not detected until day 21, and then only in low numbers (term 5; Table 1 and Fig. 4). Temporal variation in cladoceran densities was significant (term 7; Table 1). Cladocerans did not occur in samples until day 7, when their densities were 10 m−2, significantly lower than on days 10 (Tukey's P= 0·017) and days 14–28 (P < 0·001) (Fig. 4). Trends in cladoceran densities over time were not consistent between flood history (term 8; Table 1) as densities did not increase in lakes dry for 20 years (Fig. 4).
twinspan classification of the sample data indicated striking changes in community structure over time and between flood history treatments (Fig. 5). Flood days 1–3 from both flood histories, along with flood days 7–28 from lakes last flooded 20 years ago (A–D), were separated in the first division from flood days 7–28 in the more recently flooded lakes (E–G; Fig. 5). Only seven of the 101 samples in the latter group were from lakes last flooded 20 years ago. Groups E–G were characterized by a diversity of rotifers (Cephalodella spp., Polyarthra, Brachionus spp., Filinia, Asplanchna, Conochilus) and crustaceans (Macrothricidae, Chydoridae, Daphniidae, calanoid copepods). Groups A–D separated on the second division into early flood days 1–3 (A and B) and later flood days 7–28 from lakes last flooded 20 years ago (C and D) (Fig. 5). The early flood days were characterized by one protozoan taxon, nematodes and a bdelloid rotifer, while the later flood days were characterized by diverse protozoans, Volvox sp. and, in some samples, low numbers of cladocerans.
Community structure in lakes flooded 6 and 20 years previously was similar 1 day after re-wetting because of seven shared protozoan taxa (Fig. 6). However, 1 month after flooding, the community composition between flood histories diverged (Fig. 6) because of a suite of rotifers (Cephalodella catellina, Brachionus bidentatus, Asplanchna sieboldi, Polyarthra dolichoptera and several bdelloids) and crustaceans (Daphniidae, Macrothricidae, assorted nauplii, Chydoridae and ostracods) characterizing lakes flooded 6 years ago compared with the predominance of protozoans and nematodes in lakes flooded 20 years ago.
The spectacular pulses in densities of invertebrates, fish and waterbirds after floods in arid-zone rivers (Kingsford, Curtin & Porter 1999; Jenkins & Boulton 2003) rely on maintenance of resistant populations, either in aquatic refugia or as dormant life stages (Stanley, Fisher & Grimm 1997). Given the importance of micro-invertebrates in the inundated floodplain food webs (see the Introduction), it is clear that restoration efforts must maintain these resistant stages. Depletion of viable micro-invertebrate dormant stages through water extraction and increased duration of drying in arid-zone rivers may break the links between micro-invertebrates and their dependent fauna.
Our results show that, as periods of connectivity between floodplain and regulated rivers become less frequent, resistant micro-invertebrates decline and their resilience after floods is reduced. In this study, cladoceran production fell by more than an order of magnitude as the duration of drying increased from 6 to 20 years. As cladocerans are the preferred prey for many Australian native fish at their first feed and for filter-feeding waterbirds, reduced emergence has important ramifications for the management of native fisheries and waterbirds.
As the duration of the dry period increases, losses of resting eggs inevitably occur as their finite energy reserves are exhausted (Carvalho & Wolf 1989), and they are decomposed by micro-organisms (Moritz 1987) or eaten by predators (De Stasio 1989). Floodplain sediments, where nutrients, seeds and eggs are stored during dry periods, experience high summer temperatures, aeolian transport, water loss, altered terrestrial inputs and breakdown and consumption of organic material. Changes in floodplain sediment storage during increasing dry periods alters ecosystem processes (Ellis, Crawford & Molles 1998; Molles et al. 1998; Dahm et al. 2003; Valett et al. 2005) that, in turn, may also influence the response of animal and plant populations when flooding eventually occurs (Capon & Brock 2006). For example, the reduced pulse of nutrients in lakes dry for 20 years may trigger lower emergence rates, as has been shown for high salinity levels (Brown & Carpelan 1971). Increased accumulation of organic matter associated with reduced flooding in some arid-zone rivers elevated respiration rates during short (4-week) experimental floods (Valett et al. 2005), severely depleting dissolved oxygen levels and potentially inhibiting the emergence of micro-invertebrates (Dana et al. 1988). It is likely that, during dry periods, resting eggs accumulate downwind (lakes dry for 6 years) but that, as the dry period increases, aeolian sediment progressively buries resting eggs, depriving them of suitable hatching cues (Hairston et al. 1995; Gleason et al. 2003).
There is also evidence that cladoceran and copepod eggs can take longer to hatch when diapause is extended (Moritz 1987; Elgmork 1996). If there is a ‘delayed boom’ in cladoceran numbers after a flood pulse in lakes that have not flooded for extended periods (15–50 years), then native fish and waterbirds that breed early may miss the peak in prey availability. In arid-zone rivers, the timing of responses is probably critical to successful recruitment by birds and fish because the inundated ‘window of opportunity’ is so short.
Although our findings are from microcosm experiments, the densities observed in our microcosms (143–2270 L−1 or 1000–100 000 m−2) were within the range recorded in temperate ephemeral ponds (675–107 000 m−2; Wyngaard, Taylor & Mahoney 1991) and in flooded lakes in the study area (100–1000 L−1; Jenkins & Boulton 2003) and nearby wetlands (50–600 L−1; Crome & Carpenter 1988). Further, taxon richness and community succession trajectories converged after flooding in microcosms and lakes (Jenkins & Boulton 2003), supporting the realism of this approach, at least in the short term.
Reductions in micro-invertebrate production and biodiversity have global implications for the restoration of floodplain wetlands and conservation of fish and waterbirds. For example, by 1994 more than 60% of the annual flow in the lower reaches of the Barwon–Darling River was being diverted for irrigation (Thoms, Beyer & Rogers 2006), dramatically reducing flooding to floodplain lakes on Teryaweynya Creek, the Darling River and anabranches. In North America, 90% of the total discharge from rivers is impacted by dams, reservoirs, interbasin transfers and irrigation (Jackson et al. 2001), and some 54% of wetlands has been lost (Spiers 1999; Lemly, Kingsford & Thompson 2000). In Europe, wetland losses exceed 50–65% in many countries and up to 95% for floodplain wetlands (Spiers 1999; Tockner & Stanford 2002). Similarly, in Asia wetland loss is severe, with 85% of the 734 sites in the Asian wetland directory under threat and the complete loss of significant wetlands, such as the Red River delta floodplains in Vietnam that covered 1·75 million ha (Spiers 1999).
Clearly, the extent of anthropogenically ‘parched’ wetlands (Jenkins, Boulton & Ryder 2005) is increasing world-wide. However, reinstatement of increased flood frequency of wetlands disconnected from arid-zone rivers could reverse these trends. As long as regional extinction has not occurred, restoring the natural flood regime should recover the balance between build up vs. loss phases of the egg bank. With prolonged inundation or serial flooding (Puckridge et al. 1998), recovery from low numbers of long-lived resting stages may result because of the short generation time and high reproductive capacity of micro-invertebrates (Wyngaard, Taylor & Mahoney 1991). To guide restoration, we have extended a model of processes affecting egg banks (De Stasio 1989; Hairston et al. 1995) to include effects of the duration of drying in arid-zone wetlands (Fig. 7). This guiding image (criterion 1; Palmer et al. 2005), of a healthy, dynamic arid-zone floodplain wetland, must include recruitment pulses in native fish and waterbirds underpinned by booms in micro-invertebrates, algae, macro-invertebrates and functioning ecosystem processes, as well as bust periods when densities are naturally reduced. By assessing densities of the highly responsive lower trophic levels, we obtain a sensitive indicator of the ecological success of restoration (Palmer et al. 2005) along with its mechanism (Fig. 7; criterion 6 in Jansson et al. 2005).
A density of 100–1000 cladocerans L−1 within 3 weeks of floodplain inundation (Jenkins & Boulton 2003) in spring/summer would meet prey requirements for larval fish (King 2004) and is a measurable improvement in ecological condition (criterion 2; Palmer et al. 2005). This criterion is readily assessed by examination of the viability of the egg bank and production of cladocerans in flooded wetlands (Hairston et al. 1995; Jenkins & Boulton 2003; Gleason et al. 2004; Angeler & Garcia 2005). The third criterion, of achieving a more ‘self-sustaining’ system (Palmer et al. 2005), can be assessed by coupling the indicators in criterion 2 with a measure of the extent of floodplain wetland inundation so that restoration of micro-invertebrate responses occurs over an area that matches the restoration targets for fish and waterbird recruitment. Sustainability might also be presumed if egg bank inputs and losses were equivalent across inundation and drying events.
In a natural arid-zone wetland, floods alternately disrupt aquatic and terrestrial communities, but this is not deemed to cause lasting harm (criterion 4; Palmer et al. 2005) because it is part of the system's boom and bust ecology (Sheldon 2005). In regulated systems, shifts in species composition and accumulation of organic material on increasingly dry floodplains can magnify harmful effects after floods, such as black water events with low oxygen levels or toxic leachates in the short term (Molles et al. 1998; Valett et al. 2005), which may lower emergence rates and kill or repel fish (Gehrke 1991). Using floods as a restoration technique will require research on longer term responses of newly reconnected floodplains to predict responses of ecosystem processes and populations and avoid irreparable harm.
Micro-invertebrates respond to changed water regimes in floodplain wetlands and are useful indicators for pre- and post-assessment (criterion 5; Palmer et al. 2005). Using microcosms, emergence from egg banks can be assessed before and after restoration floods, taking care to replicate at the scale of the process and to address variability across relevant scales. More importantly, we must exploit opportunities to assess restoration success during flood events.
A conceptual model of the ecological mechanisms by which restoration of flooding frequency will produce a ‘self-sustaining’ viable egg bank that generates booms of micro-invertebrates after floods (Fig. 7) helps to illustrate the importance of restoring natural flooding regimes. For most Australian arid-zone floodplain rivers, we recommend that the duration of drying should not extend beyond 10–20 years, as this appears to be an important threshold for micro-invertebrates. Floods every 2–3 years will produce rich communities, matching requirements for aquatic vertebrates that may live 5–10 years and need to reproduce within these time limits. By restoring natural flood frequency, the loss of resting stages during dry periods is predicted to balance egg production and hatching during floods, resulting in timely booms of micro-invertebrates that underpin native fish and waterbird recruitment after floods.
We thank the landholders who generously gave their time to this project and provided field accommodation. We thank the State Wetland Action Group and the Lower Murray Darling Catchment Management Board for funding, the New South Wales National Parks and Wildlife Service for logistical support, Gina Dimcev for laboratory assistance and Ben Wolfenden for contributions to figures. Three anonymous referees provided useful comments on our manuscript. K. Jenkins was supported by an ARC Linkage fellowship while writing this paper.