Mosquitofish dominate amphibian and invertebrate community development in experimental wetlands

Authors


Correspondence author. E-mail: christopher.shulse@modot.mo.gov

Summary

  1. Restored and constructed habitats can play important conservation roles. Predators help shape communities in these habitats through complex interactions with prey, other predators and biotic and abiotic characteristics of the environment. However, introduced predators can have dramatic effects that may be difficult to predict.
  2. Using regression models, we compared influences of introduced invasive western mosquitofish Gambusia affinis to those of two naturally colonizing predators (crayfish and dragonflies), and vegetation, on three anuran species in experimentally constructed wetlands. Using analyses of covariance, we also examined influences of mosquitofish and vegetation on aquatic invertebrate communities.
  3. We found that mosquitofish reduced abundances of grey treefrogs Hyla versicolor and H. chrysoscelis and boreal chorus frog Pseudacris maculata, but had no significant influence on green frog Lithobates clamitans. Mosquitofish also reduced invertebrate abundance, but their effect on richness was less clear. Vegetation cover did not significantly increase most anuran or invertebrate abundances. However, vegetation increased invertebrate richness. After fish removal, invertebrate abundance increased. Fish removal may have facilitated chorus frog re-colonization into wetlands with low abundance of invertebrate predators.
  4. Our results indicate that mosquitofish are detrimental to wetland communities, and we recommend that managers avoid stocking mosquitofish. We also encourage temporary or drainable wetlands to prevent mosquitofish persistence if colonization occurs. Implementing these recommendations will improve the conservation potential of restored wetlands.

Introduction

A major challenge for restoration ecologists involves predicting pathways of ecological succession in the presence of multiple biotic and abiotic conditions. Predators play key roles in shaping natural communities through interactions with prey and other predators (Van Buskirk 1988; Griffen 2006). These interactions are often complex, thereby making discernment of mechanisms generating natural community patterns and structure difficult (DeWitt & Langerhans 2003). Most prey species are consumed by multiple predators, but prey responses to different predators are not the same. The reaction by prey to one predator may make it more or less vulnerable to another, depending on the nature of interactions between the two predators (Sih, Englund & Wooster 1998). These interactions are important during restoration because as succession proceeds, food webs develop based upon conditions present at a site, some of which can be manipulated by the restoration ecologist. For example, wetland hydroperiod plays a major role in shaping wetland communities (Pechmann et al. 1989) so designing wetlands with temporary or permanent hydroperiods will have a direct impact on community composition (Pechmann et al. 2001).

Introduced predators can dramatically alter community development, particularly if the predator is invasive and prey do not possess adaptive traits to reduce mortality (Nyström et al. 2001). Introduced fish have been implicated in aquatic community disruptions. Eastern Gambusia holbrookii (Girard 1859) and western G. affinis mosquitofish are small poeciliids native to the south-eastern United States, but introduced throughout the world because of their purported effectiveness at controlling mosquitoes (Pyke 2008). Mosquitofish readily consume invertebrates, small fish and amphibian eggs and larvae (Pyke & White 2000; Richard 2002), and they can alter the composition of the aquatic invertebrate community (Hurlbert, Zedler & Fairbanks 1972). Mosquitofish are the most widespread fish in the world (Pyke 2008), and the IUCN lists them among the 100 worst invasive species (Lowe, Browne & Boudjelas 2000). Introductions have been associated with amphibian declines in California, Australia and China (Lawler et al. 1999; Pyke & White 2000; Karraker, Arrigoni & Dudgeon 2010), and negative effects have been recorded in experiments using eggs and larvae of amphibian species within their native range (Grubb 1972; Baber & Babbitt 2004; Stanback 2010).

Dragonfly naiads and crayfish are top invertebrate predators in many wetlands. Dragonflies are carnivorous and consume other aquatic invertebrates and small fish (Merrill & Johnson 1984; Van Buskirk 1988). They are also efficient consumers of larval amphibians (Smith 1983; Semlitsch & Gibbons 1988). Crayfish are highly omnivorous and consume detritus, vegetation, invertebrates, carrion, fish eggs and young, and amphibian eggs and larvae (Momot 1995; Dorn & Wojdak 2004). Although dragonflies are generalist predators (Wallace et al. 1987), their trophic impact is likely to be narrower than crayfish, which can directly impact multiple trophic levels (Dorn & Wojdak 2004). Introduced crayfish can disrupt aquatic communities and have been implicated in amphibian declines (Gamradt & Kats 1996; Axelsson et al. 1997). However, the results from other studies suggest that crayfish are inefficient predators of larval amphibians (Fauth 1990; Lefcort 1996). Nevertheless, crayfish can destroy vegetation (Axelsson et al. 1997), thus lowering habitat complexity and potentially contributing to reduced amphibian abundance.

We compared the influences of introduced western mosquitofish to two native predators (crayfish and dragonflies) on three amphibian species in experimental constructed wetlands. Grey treefrogs Hyla versicolor/chrysoscelis complex, boreal chorus frogs Pseudacris maculata and green frogs Lithobates clamitans were selected because each species employs different mechanisms to cope with predation (Smith 1983; Van Buskirk 2003). Grey treefrogs and boreal chorus frogs are palatable to fish, but green frogs are not (Kats, Petranka & Sih 1988). Furthermore, chorus frogs prefer temporary wetlands, whereas grey treefrogs will reproduce in both temporary and permanent water, and green frogs require relatively permanent water (Kats, Petranka & Sih 1988). We predicted mosquitofish would have a greater negative impact on hylids than on green frogs, and mosquitofish effects would be greater than those of crayfish and dragonflies. We also examined the influence of mosquitofish on aquatic invertebrates and whether vegetation attenuates fish impacts. We hypothesized mosquitofish would lower invertebrate abundance and richness and vegetation would attenuate predation because habitat complexity can provide refuge for prey (Sass et al. 2006; Hartel et al. 2007).

This research is part of a larger study that aims to improve the conservation potential of restored and constructed wetlands (see Shulse 2011 and Shulse et al. 2012). The current study examines the roles of predators and vegetation in determining amphibian and invertebrate communities following wetland construction. We focus on mosquitofish because of their widespread use and the perception that they are benign to native wildlife (Pyke 2008). Our goals were to investigate whether mosquitofish influence wetland communities differently than native predators and to present wetland management recommendations based on replicated experimentation in the field.

Materials and methods

During October and November 2006, we constructed replicate wetland arrays at three upland grassland habitats in north-eastern Missouri, USA (Fig. 1), managed by the Missouri Department of Conservation (MDC). Six wetlands (23 m diameter, 0·76 m maximum depth) were constructed at each location (= 18). A complete description of wetland designs, placement and surrounding landscapes is given in Shulse et al. (2012). A goal of another study at these wetlands was to examine the influences of within-wetland slope, mosquitofish and vegetation on amphibian metamorph production and species richness (Shulse et al. 2012). Therefore, we randomly assigned one of the six combinations of slope, mosquitofish and vegetation to each wetland (Table 1). Planted wetlands received 50 cordgrass Spartina pectinata divisions spaced evenly apart and radiating from the centre. Non-surviving plants were replaced during autumn 2007. All other vegetation was allowed to develop naturally.

Table 1. Wetland treatment combinations. One wetland of each treatment combination was constructed at each of the three study sites
Treatment combinationSlopeMosquitofishPrairie cordgrass
14:1StockedNot planted
24:1Not stockedNot planted
315:1StockedNot planted
415:1StockedPlanted
515:1Not stockedNot planted
615:1Not stockedPlanted
Figure 1.

Locations of study sites within north-eastern Missouri, USA.

In March 2007, we captured mosquitofish in a Missouri Department of Transportation compensatory mitigation wetland in Audrain County, Missouri, and released them into the three selected wetlands at each MDC location at a rate of 3089 fish ha−1, which is slightly higher than the rate of 2471 fish ha−1 (1000 fish acre−1) recommended by Duryea et al. (1996). This resulted in a founding population of 125 adult mosquitofish per stocked wetland. Fish were re-stocked where samples indicated low populations in spring 2008. Reconnaissance sampling in early spring 2009 revealed healthy mosquitofish populations in all stocked wetlands so no further re-stocking occurred. MDC personnel removed mosquitofish from stocked wetlands at one location (Redman) on 17 September 2009 using the piscicide rotenone (chemical restoration). Rotenone was applied to stocked wetlands at another site (Sears) on 10 March 2010. Rotenone was applied at label rates. Dead mosquitofish were observed in all wetlands. However, during the second sampling period in 2010, mosquitofish were captured in one treated wetland (Sears 1). Therefore, it was assumed that this wetland contained survivors so we considered it fish-stocked for 2010 analyses. The stocked wetlands at the third location (White) were not treated and reconnaissance sampling in early spring 2010 indicated healthy populations.

Amphibians, mosquitofish and invertebrates were sampled three times within each season using aquatic funnel traps and dip nets. Aquatic funnel traps were deployed for 48 h in 2007 and 2008 and overnight in 2009 and 2010, using two kinds of commercially available minnow traps: collapsible nylon mesh traps (3-mm mesh; 38 × 26 × 26 cm; 6 cm openings) or galvanized steel wire traps (6-mm mesh; 42 cm long; 2·5 cm openings). Two traps of each were used per wetland and placements were staggered so that traps of the same model were directly across from one another at each cardinal direction. Pair direction assignment was random. One dip net (3-mm nylon mesh) sweep was conducted from the water's edge at each cardinal direction and sweeps were ~1·5 m long with the net pressed to the substrate and pulled towards the sampler. During the second 2007 sampling period, a zooplankton canvas D-net with 500-micron mesh bottom was added to the protocol to capture very small organisms. Sweeps of approximately 1·5 m occurred at each ordinal direction using the D-net. This resulted in four dip net sweeps and 4 D-net sweeps, spaced evenly apart, for each wetland during each sampling period after 2007–1. Data from all methods were combined to calculate amphibian, mosquitofish and invertebrate abundances and invertebrate taxa richness at each wetland during each sampling period. All organisms were released unharmed at point of capture after recording. We were unable to distinguish between eastern grey treefrogs and Cope's grey treefrogs in the field so grey treefrogs are considered as the Hyla versicolor/chrysoscelis complex.

Within-wetland vegetation was measured using four 1-m² quadrats spaced at cardinal directions around the perimeter of each wetland. Quadrats were placed at the edge of each wetland to assess vegetation cover within 1 m of the shore and at 3 m from the shore. The percentages of open water, emergent, floating and submerged vegetation were visually estimated within each quadrat. The three categories of vegetation were combined and averaged for all quadrats over all sampling periods within a season at each wetland to calculate an average measure of vegetation cover for the season. Percentage vegetation cover was transformed to the arcsine square root of the proportion for analyses. Development of natural vegetation occurred faster in some non-planted wetlands than in planted wetlands. Therefore, we used vegetation cover as a continuous covariate within our analyses as opposed to a treatment factor (below).

We analysed each year separately to look for overall patterns in abundance or taxa richness. For regressions and ancovas, a single wetland was used as the unit of replication. All statistical analyses were performed using SPSS version 16·0 (2007 SPSS Chicago, IL, USA). To explain relationships between abundances of amphibians, predators and vegetation cover, we developed regression models with negative binomial distributions and log-link functions using the generalized linear model option in SPSS. We used abundances of mosquitofish, crayfish and dragonfly naiads, along with vegetation cover, as independent variables. We conducted Spearman's rank correlation tests between independent variables to avoid including two variables strongly correlated with one another (r ≥ 0·70) in models. Dragonfly abundance and vegetation cover were highly correlated in 2008 and 2009 (Table 2); therefore, in models for these years, we focused our analyses on predators and excluded vegetation cover. Each regression model contained grey treefrog, boreal chorus frog or green frog abundances as dependent variables, and either all four independent variables or the three predator variables (2008 and 2009). Only crayfish, dragonflies and vegetation cover were included in the model for grey treefrogs in 2010 because no grey treefrogs were captured in wetlands containing fish in that year.

Table 2. Spearman's correlation matrix for independent variables in amphibian generalized linear regression models
= 18MosquitofishCrayfishDragonfly
Vegetation
2007−0·13−0·200·14
2008−0·37−0·180·71
2009−0·34−0·560·78
2010−0·44−0·610·68
Mosquitofish
2007 0·16−0·64
2008 0·16−0·63
2009 0·11−0·50
2010 0·05−0·44
Crayfish
2007  −0·34
2008  −0·39
2009  −0·56
2010  −0·63

To test the hypothesis that mosquitofish reduce invertebrate abundances, we used the cumulative number of invertebrates (log10-transformed) captured during all sampling periods each year at each wetland as dependent variables in generalized linear models with mosquitofish as a factor and vegetation cover as a covariate. We excluded crayfish, snails, bivalves and daphniids from ancovas. Data for snails and bivalves were not collected consistently, and daphniids were challenging to quantify at very high numbers. Crayfish grow large enough to escape fish predation (Stein 1977), and even the smallest crayfish we observed were too large for gape-limited mosquitofish to consume.

Table 3. Parameter estimates for independent variables in amphibian abundance generalized linear regression models. Significant parameters and their corresponding statistics are in boldface
SpeciesYearParameterβSEWald χ295% CISig.
LowerUpper
  1. a

    No grey treefrogs were captured in wetlands containing mosquitofish in 2010.

Grey treefrog complex2007 Mosquitofish 0·02 0·01 4·38 0·050·001 0·04
Crayfish 0·03 0·01 7·59 0·050·009 0·006
Dragonfly0·110·081·87−0·050·280·17
Vegetation 19·81 5·79 11·73 8·47 31·16 0·001
2008 Mosquitofish 0·06 0·02 11·68 0·090·03 0·001
Crayfish−0·020·012·18−0·050·0070·14
Dragonfly 0·11 0·05 5·57 0·210·02 0·02
2009 Mosquitofish 0·03 0·01 8·29 0·050·009 0·004
Crayfish 0·60 0·27 4·97 1·130·07 0·03
Dragonfly−0·050·032·82−0·110·0090·09
2010aCrayfish−0·010·010·98−0·040·010·32
Dragonfly0·0030·020·03−0·030·040·87
Vegetation2·771·842·27−0·846·380·13
Boreal chorus frog2008 Mosquitofish 0·02 0·006 10·89 0·030·009 0·001
Crayfish−0·040·031·61−0·110·020·21
Dragonfly0·010·070·02−0·0130·150·88
2010 Mosquitofish 0·04 0·01 6·87 0·070·01 0·009
Crayfish0·0010·020·001−0·040·040·97
Dragonfly 0·12 0·05 6·58 0·210·03 0·01
Vegetation1·462·540·33−3·536·440·57
Green frog2008Mosquitofish−0·0030·010·06−0·020·020·80
Crayfish 0·05 0·02 4·91 0·100·006 0·03
Dragonfly 0·12 0·06 4·35 0·240·008 0·04
2009Mosquitofish0·0030·0030·96−0·0030·0090·33
Crayfish 0·10 0·03 8·34 0·170·03 0·004
Dragonfly−0·02−0·020·54−0·060·030·46
2010Mosquitofish0·0080·0062·10−0·0030·020·15
Crayfish 0·07 0·02 13·05 0·110·03 <0·001
Dragonfly0·0070·010·27−0·020·030·60
Vegetation−0·131·540·007−3·152·880·93

To test whether mosquitofish reduce invertebrate richness, we used the cumulative number of invertebrate taxa captured during all sampling periods per year at each wetland as dependent variables in generalized linear models. Mosquitofish presence was included as a factor and vegetation cover as a covariate. Invertebrate richness values included daphniids but excluded crayfish, snails and bivalves. We attempted to identify each invertebrate to family, but we were unable to identify some to this level in the field (See Table S1, Supporting Information). To achieve normal distribution, invertebrate taxa richness values were log10-transformed for 2007. Because comparisons of taxa richness among different assemblages should account for differences in sampling effort and abundance (Gotelli & Colwell 2001), we plotted rarefied richness curves for each year using EstimateS version 8.2 (Colwell 2005). We included daphniids in rarefaction analyses, but we capped the number of daphnia at 100 per sample due to the aforementioned quantification problems. For rarefaction, we defined sample as the total individuals captured by all methods during a single sampling period at each wetland. Finally, we performed Wilcoxon signed rank tests to evaluate invertebrate abundance and taxa richness of chemically restored wetlands before and after treatment.

Results

Regression analyses revealed negative associations between grey treefrogs and mosquitofish during 2007, 2008 and 2009 (all < 0·05; Table 3). Treefrogs were positively associated with vegetation in 2007. No treefrog larvae were captured in wetlands containing mosquitofish during 2010. Treefrog abundance was also negatively associated with crayfish abundance in 2007 and 2009 and larval dragonflies in 2008. Treefrogs were most abundant during 2007 (Fig. 2), but they were captured in only 39% of the wetlands (Fig. 3). During subsequent years, they were captured in roughly half of the wetlands but their abundance dropped and remained at relatively low levels.

Figure 2.

Mean abundance trends for anurans over all sampling periods. The mean abundance of chorus frogs was 102 in 2008. The scale of the Y-axis has been capped at 30 to more clearly illustrate trends. Fish stocked: = 9 for 2007–2009 and = 4 for 2010. Fish unstocked: = 9 for 2007–2009 and = 14 for 2010.

Figure 3.

Occurrence trends for anurans over all four study years at all wetlands. Green frogs became more common as wetlands aged, grey treefrog occurrence was relatively stable, and chorus frog occurrence was variable, possibly due to predator population fluctuations.

Boreal chorus frog tadpoles were never captured in large numbers (i.e. >15) in wetlands containing mosquitofish. We did not perform a regression analysis for chorus frogs in 2007 because their larvae were captured in only three wetlands. No fish, crayfish or dragonflies were captured in these wetlands during the first two sampling periods when chorus frogs were breeding. Chorus frogs were negatively associated with mosquitofish in 2008 (= 0·001) and 2010 (= 0·009), and they were captured in 67% of wetlands in 2008 and 44% in 2010. Only two larval chorus frogs were captured in 2009 and they occurred in a fish-free wetland. Chorus frogs were also negatively associated with dragonflies in 2010 (= 0·01). The peak abundance for chorus frogs occurred during 2008 (Fig. 2). Although their larvae were nearly absent in 2009, both their abundance and occurrence increased sharply in 2010 (Figs 2 and 3).

Green frogs did not occur in enough numbers to perform regression analyses in 2007, but their abundance was consistently negatively associated with crayfish during 2008, 2009 and 2010 (all < 0·05). Green frogs were also negatively associated with dragonflies in 2008 (= 0·04). There were no statistically significant relationships between green frog abundance and mosquitofish. Green frog abundance and occurrence increased over the course of the study and peaked during the first sampling period of 2010 (Figs 2 and 3).

Analyses using ancovas revealed that invertebrate abundance was significantly reduced in the fish-stocked wetlands during all four sampling years (2007: F1,15 = 13·25, = 0·002; 2008: F1,15 = 21·07, < 0·001; 2009: F1,15 = 55·15, < 0·001; 2010: F1,15 = 15·60, = 0·001). Mean invertebrate abundance was consistently higher in fish-free wetlands throughout the duration of our study (Fig. 4). ancovas also indicated that mosquitofish significantly reduced invertebrate taxa richness during the first 3 years (2007: F1,15 = 6·9, = 0·02; 2008: F1,15 = 19·1, = 0·001; 2009: F1,15 = 14·61, = 0·002), but not in 2010 (= 0·56). The vegetation cover covariate had no significant effects on invertebrate abundance during any year, but it did significantly increase taxa richness in all years except 2007 before natural vegetation cover had developed (2007: = 0·48; 2008: F1,15 = 10·04, = 0·006; 2009: F1,15 = 12·35, = 0·003; 2010: F1,15 = 5·05, = 0·04). Average invertebrate taxa richness generally increased in all wetlands throughout the duration of the study, but most fish-free wetlands were consistently richer during all sampling periods (Fig. 5).

Figure 4.

Mean invertebrate abundance trends over all sampling periods. Wetlands with fish consistently contained fewer invertebrates. Fish stocked: = 9 for 2007–2009 and = 4 for 2010. Fish unstocked: = 9 for 2007–2009 and = 14 for 2010.

Figure 5.

Mean invertebrate taxa richness over all sampling periods. Although this figure illustrates a clear difference in richness between the two treatments, the rarefied richness curves based on individuals suggest that differences are due to a sampling effect – there were simply fewer individuals of most taxa to capture in the wetlands containing fish. Fish stocked: = 9 for 2007–2009 and = 4 for 2010. Fish unstocked: = 9 for 2007–2009 and = 14 for 2010.

The rarefied richness curves illustrate the higher individual abundances in fish-free wetlands, but they also reveal that in 2007 and 2009, taxa richness reached levels in fish-stocked wetlands nearly as high as in those without fish, even though fewer individuals were captured (Fig. 6a). Because the rarefaction curves for fish treatments fail to approach an asymptote, and vastly different numbers of individuals were captured in the two treatments, we also plotted rarefied richness based on samples (Fig. 6b). The sample-based curves suggest that taxa richness was somewhat higher within fish-free wetlands, but during 2007, taxa richness was similar for the two treatments across samples and in 2010, taxa richness was nearly equal at samples below 10.

Figure 6.

Rarefied taxa richness curves for 2007–2010 based on individual abundance (a) and samples (b). The sample-based curves suggest richness was higher in fish-free wetlands, but individual-based curves reveal many taxa were present in both treatments, albeit at lower abundances in wetlands with fish. Fish stocked: = 27 for 2007–2009 and = 12 for 2010. Fish unstocked: = 27 for 2007–2009 and = 42 for 2010.

A Wilcoxon signed rank test revealed a statistically significant increase in invertebrate abundance (excluding crayfish, daphniids, snails and bivalves) following rotenone applications to fish-stocked wetlands, = 12, Z = −2·20, = 0·03, with a large effect size (r = 0·64). The median invertebrate abundance in fish-stocked wetlands was 50·5 in 2009 prior to treatment. In 2010, after treatment, the median increased to 248·5. Invertebrate abundances were low in the three untreated fish-stocked wetlands in 2009 and 2010 (2009: mean = 34·0, range = 22–49; 2010: mean = 28·3, range = 17–35).

There was also a statistically significant increase in invertebrate taxa richness following rotenone application to fish-stocked wetlands, = 12, Z = −2·03, = 0·04, with effect size r = 0·59. The median richness in fish-stocked wetlands was 7·5 in 2009 prior to treatment, and in 2010, after treatment, the median increased to 11·5. However, invertebrate taxa richness also increased in the three untreated fish-stocked wetlands from 2009 to 2010 (2009: mean = 8·7, range = 7–10; 2010: mean 11·3, range = 9–13).

Discussion

Amphibian Community Development

Our results illustrate the dramatic role aquatic predators play in wetland community development. While natural predators altered amphibian communities over time, introduced mosquitofish impeded community development from the outset. Chorus frogs and grey treefrogs appeared to be particularly sensitive to mosquitofish and our results may reflect avoidance by breeding adults, predation, trophic effects or a combination thereof. Nevertheless, metamorph production data recorded at the same wetlands during 2007 and 2008 using terrestrial pitfall traps and drift fences reinforce our results (Shulse et al. 2012). Although some models in this study revealed negative influences on hylids from invertebrate predators, none were as consistent as those observed for mosquitofish.

Our wetlands contained water during all four study years. As a result, wetlands without mosquitofish developed high populations of invertebrate predators and by 2010, almost all contained ranid larvae. Heightened competition from ranids (Faragher & Jaeger 1998; Boone, Semlitsch & Mosby 2008) and susceptibility to invertebrate predators (Skelly 1995; Smith & Van Buskirk 1995) may explain why larval chorus frogs were nearly absent by 2009. Some hylids can detect fish and invertebrate predators in wetlands (Resetarits & Wilbur 1989, 1991; Binkley & Resetarits 2008). During diurnal early spring reconnaissance trips in 2007 and 2008, chorus frogs called selectively from fish-free wetlands, but in 2009, chorusing had nearly ceased in all wetlands. Instead, frogs chorused from nearby ephemeral swales and ditches (Shulse, personal observation). Chorus frogs prefer fish-free wetlands with vegetation (Shulse et al. 2010, 2012), but dragonflies also appear to prefer similar habitat (Table 2). Dragonfly populations increased as wetlands aged (Fig. 7c), as did vegetation cover (See Fig. S1, Supporting Information). By 2009, chorus frogs may have avoided the wetlands, even those with high cover, in favour of nearby ephemeral, low-cover aquatic habitat containing few predators and little competition. During 2010, chorusing frogs returned in limited numbers, but mostly to chemically restored wetlands (Shulse, personal observation). Our chorusing observations were validated by capture results. In 2010, the highest abundances of chorus frogs were captured in wetlands that were either 1) chemically restored prior to the breeding season or 2) had relatively low predatory insect populations, illustrating that these anurans detect both fish and invertebrate predators. Other studies have shown that fish removal leads to increased breeding of fish-sensitive anurans (Brönmark & Edenhamn 1994; Vredenburg 2004). Early colonizing amphibians may have a hierarchy of breeding habitat preferences with fish avoidance as the strongest filter, followed by aquatic invertebrate predators. Based on the results of Shulse et al. (2012), this hierarchy can be extended below invertebrate predators to include vegetation cover followed by within-wetland slope. Our observations suggest that the primary habitat trade-off for breeding chorus frogs is between exposure to predators and subjection to breeding site stochasticity.

Figure 7.

Mean abundances of mosquitofish (a), crayfish (b) and dragonflies (c) over all sampling periods at all wetlands. Total mosquitofish plummeted after rotenone treatments in late 2009 and early 2010, but numbers increased during summer 2010 at the stocked wetlands. Crayfish numbers were similar in both treatments, suggesting that mosquitofish have little impact on crayfish populations, and dragonfly abundances were consistently lower in fish-stocked wetlands. Dragonflies exhibited a pattern similar to that of most other invertebrate taxa sampled.

Grey treefrogs were also less abundant in fish-stocked wetlands, but they were able to persist in fish-free wetlands throughout the duration of the study. Their high numbers during 2007 indicate that they, like chorus frogs, are early colonizers that prefer wetlands with low predator levels. Adult female grey treefrogs will minimize predation risk to their eggs and larvae by avoiding wetlands containing fish (Binkley & Resetarits 2008). However, larvae will often develop bright red pigment on their tails and altered body shape in the presence of high populations of aquatic invertebrate predators (McCollum & Leimberger 1997). This ‘dragonfly morph’ appears less susceptible to invertebrates than the typical morph (McCollum & Van Buskirk 1996) and may indirectly contribute to their ability to continue to breed in permanent wetlands. We often observed ‘dragonfly morph’ grey treefrog larvae in our wetlands with varying shades and amounts of red pigment. Because these anurans are mid-spring to early summer breeders whose larvae emerge during mid- to late summer, highly ephemeral wetlands that become dry by mid-summer may reduce or eliminate recruitment. However, breeding later in more permanent wetlands may expose larvae to the highest seasonal levels of dragonflies.

The increasing abundance of green frogs over the duration of the study, like invertebrates, probably reflects hydroperiod. Green frogs overwinter as larvae and therefore require permanent or semi-permanent wetlands. Green frogs were negatively associated with crayfish, although it is not clear whether this reflects mortality or avoidance. Anderson & Brown (2009) observed reduced hatching of green frogs in the presence of crayfish, even when the crayfish had no direct access to the eggs. Interestingly, many sparsely vegetated wetlands contained high populations of crayfish. The negative correlations between crayfish and vegetation (Table 2) illustrate the effects that these shredders have upon aquatic vegetation. These effects, in addition to predation, may explain the negative associations between crayfish and anurans. Green frog abundances were never significantly negatively associated with mosquitofish, perhaps reflective of their ability to persist with fish. However, they did not appear to be facilitated by fish as has been demonstrated for bullfrog tadpoles (Werner & McPeek 1994; Adams, Pearl & Bury 2003).

While our study did not reveal strong associations between most anurans and vegetation, grey treefrogs were strongly positively associated in 2007, perhaps reflecting the sparse vegetation present at the time. Treefrogs may have used the planted cordgrass for chorusing or cover. We also found strong positive associations between total amphibian metamorph production and vegetation cover during 2008 at the same wetlands (Shulse et al. 2012). Chorus frog metamorph production during 2008 was also positively associated with vegetation cover, but a model that combined mosquitofish abundance with vegetation cover best explained chorus frog abundance that year (Shulse et al. 2012). Habitat complexity may increase in importance, surpassing other features such as within-wetland slope, as wetlands age (Shulse et al. 2012). Studies have shown that high vegetation cover is important for tadpole survival (Babbitt & Tanner 1998; Baber & Babbitt 2004), but predators that easily penetrate dense vegetation may have lowered the importance of cover in our models. For some anurans, predator population crashes (and manipulations that directly alter predator levels) may be more important modifiers of reproductive success than habitat alterations.

Invertebrate Community Development

Our results for invertebrates are concordant with those of previous studies that have demonstrated that mosquitofish are injurious to aquatic invertebrates (i.e. Hurlbert, Zedler & Fairbanks 1972; Jassby et al. 1977a,b; Lawler et al. 1999). Mosquitofish may alter aquatic communities by selectively feeding on large zooplankton reducing pressure on smaller zooplankton species, phytoplankton and bacteria (Jassby et al. 1977a,b). Studies suggest mosquitofish are primarily zooplanktivores (Garcia-Berthou 1999; Reynolds 2009) and may prefer zooplankton to larval amphibians (Reynolds 2009). We captured a total of 5 daphniids in Sears 1 during sampling period 2010–1 after chemical restoration which nearly eliminated mosquitofish. None were captured during subsequent sampling periods when the fish population recovered, or at any other fish-stocked wetland during the study. After autumn rotenone treatments in 2009 at other wetlands, daphniid captures increased during the following spring to the highest levels recorded (C. D. Shulse & R. D. Semlitsch, unpublished data). Because daphnia are preyed upon by other invertebrates, high abundances shortly after fish removal may reflect ideal conditions before predator populations recover.

Our rarefied richness curves (Fig. 6) and invertebrate taxa captured (Table S1, Supporting information) illustrate that most taxa present in fish-free wetlands were also present in those with fish, although many at comparatively very low numbers. This suggests that our ancova results illustrating lower richness in fish-stocked wetlands, along with differences in mean richness between the two treatments (Fig. 5), may be partly explained by mosquitofish reducing richness simply through reducing abundance (i.e. a sampling effect). Culicids, chironomids, gerrids, amphipods and hydrachnids were also very rare in fish-stocked wetlands. Unlike daphnia, these were also relatively uncommon in fish-free wetlands. Mosquitofish likely prefer daphnia, but they appear to prey indiscriminately on most aquatic invertebrates once daphnia populations are depleted.

Stewart & Downing (2008) found macroinvertebrate richness and abundance increased along with vegetation in constructed wetlands. While our results also indicate that invertebrate richness is bolstered by vegetation, we found no evidence that vegetation increased overall invertebrate abundance. Reynolds (2009) found that high levels of both aquatic invertebrates and vegetation cover reduced mosquitofish predation on anuran larvae. However, our results illustrate that aquatic invertebrate abundance is severely reduced by mosquitofish, suggesting vegetation cover provided insufficient refuge for invertebrates.

Management Implications

Complex interactions between predators, wetland hydroperiod and successional processes shape wetland communities, but mosquitofish break down natural processes, alter populations of other predators and grazers and impact multiple trophic levels. Our results suggest that outside their native range and ecosystems, mosquitofish reduce the ecological value and conservation potential of wetlands, particularly those restored or created as compensatory mitigation for the destruction of natural wetlands. Some invertebrates and amphibians responded positively to both fish removal and low invertebrate predator populations. These conditions are likely to be similar to those that occur when a wetland re-fills after drying. However, rotenone can have negative consequences for amphibians (Fontenot, Noblet & Platt 1994) so caution is warranted.

Building wetlands of varying sizes and depths creates hydroperiod diversity across the landscape similar to natural conditions (Semlitsch 2002; Petranka et al. 2007; Shoo et al. 2011), and constructing drainable ponds will allow managers to control hydroperiods. These non-toxic approaches will ensure that mosquitofish and other aquatic predators are occasionally eliminated in some pools, thereby increasing the ecological value of restored and constructed wetland complexes. Furthermore, native predators should be encouraged to colonize through management regimes that mimic pre-settlement conditions and natural successional processes. However, non-native predators can have devastating impacts and they should not be stocked or allowed to persist.

Our experimental approach at replicated environments goes beyond traditional laboratory or observational investigations to provide a unique comparison between the impacts of an invasive introduced predator to those of naturally colonizing predators. We encourage researchers, ecologists and managers to work collaboratively to incorporate experimentation into restoration projects. Doing so will yield valuable information that will improve the ecological value of restoration projects, reduce the threat of invasive species and increase the conservation potential of restored habitats.

Acknowledgements

We thank the Missouri Department of Transportation, Missouri Department of Conservation, University of Missouri Division of Biological Sciences and University of Missouri Department of Civil and Environmental Engineering. Thanks to S. Becker, K. Kettenbach, L. Rehard, D. Kuschel, A. Robertson, D. Lund, A. Leary, B. McMurray and G. Schmitz for assistance in the field. Thanks to K. Smith and an anonymous reviewer for insightful comments on an earlier version of this manuscript. This project was funded by a United States Environmental Protection Agency Region VII Grant CD-98769101-0, a Missouri Department of Conservation Wildlife Diversity Fund Grant and a Missouri Department of Transportation Research and Development Grant (RI 07-005). Organisms were captured under Missouri Department of Conservation Wildlife Collector's Permits 13438, 13769, 14120 and 14533 and University of Missouri Animal Care and Use Protocol 4189.

Ancillary