Habitat complexity does not promote coexistence in a size-structured intraguild predation system

Authors


Summary

  1. Size-dependent interactions and habitat complexity have been identified as important factors affecting the persistence of intraguild predation (IGP) systems. Habitat complexity has been suggested to promote intraguild (IG) prey and intraguild predator coexistence through weakening trophic interactions particularly the predation link.
  2. Here, we experimentally investigate the effects of habitat complexity on coexistence and invasion success of differently sized IG-predators in a size-structured IGP system consisting of the IG-predator Poecilia reticulata and a resident Heterandria formosa IG-prey population. The experiments included medium-long and long-term invasion experiments, predator–prey experiments and competition experiments to elucidate the mechanisms underlying the effect of prey refuges.
  3. Habitat complexity did not promote the coexistence of IG-predator and IG-prey, although the predation link was substantially weakened. However, the presence of habitat structure affected the invasion success of large IG-predators negatively and the invasion success of small IG-predators positively. The effect of refuges on size-dependent invasion success could be related to a major decrease in the IG-predator's capture rate and a shift in the size distribution of IG-predator juveniles.
  4. In summary, habitat complexity had two main effects: (i) the predation link was diminished, resulting in a more competition driven system and (ii) the overall competitive abilities of the two species were equalized, but coexistence was not promoted. Our results suggest that in a size-structured IGP system, individual level mechanisms may gain in importance over species level mechanisms in the presence of habitat complexity.

Introduction

In intraguild predation (IGP) systems, a predator (IG-predator) also competes with its prey (IG-prey) for a common resource (Polis, Myers & Holt 1989). Intraguild predation systems are more complex than linear predator–prey systems because the omnivorous predator profits energetically from feeding on the prey, but simultaneously suffers from competition. In IGP systems, the scope for coexistence is lower than in linear food chains, because at high productivity, the predator can suppress its prey without suffering itself (Polis, Myers & Holt 1989; Holt & Polis 1997; Diehl & Feissel 2001).

In many cases, intraguild interactions result from life-history omnivory where juvenile IG-predators feed on the shared resource and adult IG-predators feed on the IG-prey, separating competition and predation over the life cycle (Polis, Myers & Holt 1989). Often, it is the size relationship between two individuals that determine in what way they can interact (competitors or predator–prey), for example, small stages of the IG-predator may compete with the similar-sized IG-prey on which large IG-predator stages feed (Arim & Marquet 2004; van van de Wolfshaar, De Roos & Persson 2006; Persson, De Roos & Byström 2007; Hin et al. 2011). Consequently, the incorporation of size-dependent interactions (size-structure) in IGP theory adds even more complexity to IGP systems, because depending on its size relative to other individuals, an individual can be either competitor or predator (Werner & Gilliam 1984; Wilbur 1988; Rudolf & Lafferty 2011).

Two new mechanisms arise from introducing life-history omnivory and size-structure into IGP systems: (i) a potential recruitment bottleneck in the juvenile IG-predator imposed on it by its competitively superior IG-prey (Werner & Gilliam 1984; Neill 1988) and (ii) a positive feedback loop between the predation on the IG-predator juveniles' competitor and the recruitment of predating adult IG-predators through the competitive release in the IG-predator's juvenile stage (van de Wolfshaar, De Roos & Persson 2006). Furthermore, the extent to which IG-predator and IG-prey are expected to coexist depends on the IG-predator's dependency on the IG-prey as a resource for reproduction. For example, if large IG-predator stages feed on both the shared resource and the IG-prey (incomplete resource shift) and the IG-predator can gain enough energy for reproduction from the shared resource alone, coexistence is even less likely because the predator can persist in the system once the prey is excluded (van de Wolfshaar, De Roos & Persson 2006; Montserrat et al. 2008, 2012; Schröder et al. 2009; Hin et al. 2011).

The strength of the above-mentioned mechanisms (1 and 2) will depend on IG-predator individuals' sizes relative to IG-prey individuals' sizes. Hence, which body size relationships (smaller, equal size or larger) are realized between IG-predator and IG-prey individuals may be decisive for whether any of the two (IG-predator or IG-prey) can invade a resident population of the other, and whether successful invasion will result in coexistence or in extinction of the resident. Population size-structure at the time of invasion can therefore play an important role in determining invasion success and community composition.

Accordingly, two recent experimental studies on size-structured IGP systems investigating the invasion success of an IG-predator into a resident IG-prey population showed that species interactions can differ substantially depending on body size relationships (Schröder et al. 2009; Montserrat et al. 2012). In both studies, IG-predator and IG-prey never coexisted. This is in correspondence with what is expected when the IG-predator can mature on the shared resource alone as they could in these two experimental systems (van de Wolfshaar, De Roos & Persson 2006). Montserrat et al. (2012) further showed that, in a community with reciprocal (mutual) IGP, invasion success of the IG-predator depended on the stage structure of the resident IG-prey population where the presence of predacious IG-prey stages prevented IG-predator invasion. Schröder et al. (2009) showed that the invasion success of an IG-predator into a resident size-structured IG-prey system was strictly dependent on invader body size such that small IG-predators could not invade, whereas large IG-predators could invade. The lack of coexistence after successful invasion was explained by that adult predation released the juvenile competitive bottleneck in the IG-predator while imposing a recruitment bottleneck in the IG-prey.

Habitat complexity may promote coexistence in IGP systems (Janssen et al. 2007). Based on a meta-analysis, Janssen et al. (2007) concluded that the presence of habitat complexity weakens the predation link between IG-predator and IG-prey (see also: Persson & Eklöv1995; Finke & Denno 2006), which is one possible mechanism by which habitat complexity may promote coexistence between IG-prey and IG-predator. This conclusion is largely based on the theory on the effects of omnivory on interaction strength and stability of food web modules (McCann, Hastings & Huxel 1998; Maser, Guichard & McCann 2007).

However, the studies included in Janssen et al. (2007) meta-analysis on the effects of habitat complexity on IGP systems did not explicitly account for size-structured IGP systems and were based only on short-term experiments. The projection of inference from short-term to long-term dynamics maybe treacherous (Briggs & Borer 2005; Persson & De Roos 2012), and experimental studies of the effects of habitat complexity on long-term multi-generation dynamics of IGP systems are lacking. Furthermore, in size-structured populations, the presence of habitat complexity will often lead to spatial segregation of life stages because of behavioural responses of juveniles to interspecific or intraspecific predators, which may lead to increased competition among retreating juveniles (Werner & Hall 1988; Persson & Eklöv 1995). Thus, the effect of habitat complexity on overall interaction strength and coexistence is less straightforward to predict for size-structured IGP systems than for less complex IGP systems, as were considered by Janssen et al. (2007). Here, we address these gaps in knowledge by investigating the effects of habitat complexity on invasion success and coexistence in an experimental, long-term, size-structured IGP system.

Our study system consisted of the common guppy Poecilia reticulata as IG-predator and the least killifish Heterandria formosa as IG-prey. Schröder et al. (2009) showed that in the absence of habitat complexity, the invasion success of the IG-predator in this system strictly depended on invader body size. Coexistence between IG-predator and IG-prey was never observed. Here, we predict that the addition of habitat complexity should first affect the invasion of large IG-predators negatively because of reduced predation efficiency when vulnerable IG-prey can retreat into refuges. This is supported by Schröder et al. (2009) observations that a juvenile recruitment bottleneck in the IG-prey population occurred when large IG-predators were introduced in the absence of habitat complexity.

Second, because of the predation risk that large IG-predators impose on small IG-prey and IG-predators, we predict juveniles to increase their use of refuges leading to increased competition intensity among juveniles. Coexistence between species would depend on how habitat complexity affected the possibility for niche segregation between the two species. Overall, we predict that the opposing effects that habitat complexity would have on predatory (decrease) and competitive (increase) intensity, respectively, would lead to no effect of habitat complexity on coexistence.

Materials and methods

Experimental System

We used the viviparous poeciliid fish Poecilia reticulata and Heterandria formosa that occur naturally in freshwater streams and ponds in the coastal regions of Northern Brazil, Venezuela, Guyana, Barbados and Trinidad, and in North America in the coastal plains from North Carolina to Florida, respectively. Poecilia reticulata males reach sizes of 19 mm and females of 41 mm (Reznick & Miles 1989). Heterandria formosa males grow to a maximum length of 20 mm and females up to 35 mm (Frank 1977). Size at birth differs slightly with P. reticulata juveniles (6–9 mm) (Cheong et al. 1984; Reznick & Miles 1989) being larger than those of H. formosa (5–8 mm). The generation times of the two species are 10 weeks for P. reticulata and 7 weeks for H. formosa. For more details on origin, accommodation and breeding biology, see the methods section in Schröder et al. (2009). Poecilia reticulata exhibits life-history omnivory, large females (>18 mm) feeding on juvenile H. formosa (Fig. 4 and Schröder et al. 2009) and the shared resource and was therefore the IG-predator in our system, with H. formosa as its IG-prey.

Experiments were performed in an aquarium system of 56 aquaria (80 L) and equipped with air supply, thermostat, biofilters, UV water sterilizer and 15 W neon lights (14 h light/10 h dark regime). Salt was added to prevent infection with ectoparasites (conductivity, 900–1000 μS cm−1). Feeding was regulated by computer-controlled micro feeders. Populations were fed four times a day with two bouts of 9·45 mg ± 0·41 (mean ± 1 SD) pelletized food (SERA Microgran) (75·3 mg per day = ‘medium’ in Schröder et al. 2009). As refuges, we used green plastic thread (Eheim EHFI FIX), loosely packed into four balls per aquarium (two floating and two sunk to the bottom). The aquarium system is the same as that used by Schröder et al. (2009) and is described in more detail there.

Long-term invasion experiment in the presence of refuges

To investigate the long-term effects of habitat structure on size-dependent invasion success and coexistence, we introduced juvenile and adult IG-predators (P. reticulata) into resident IG-prey (H. formosa) populations in the presence of refuges and monitored those communities over a time span corresponding to five IG-predator generations and eight IG-prey generations. In this experiment, we followed the same procedure as Schröder et al. (2009), but including the addition of habitat structure in the form of refuges and the use of only one productivity level [‘medium’ in Schröder et al. (2009)].

For the juvenile invader (XS) treatment, five juvenile IG-predators were used (mean length ± SD; 6·9 mm ± 0·7). For the adult-invader (XL) treatment, two IG-predator males (18 mm ± 1·4) and two IG-predator females (29·6 mm ± 1·6) were used. Only fish with normal swimming and feeding behaviour 48 h after handling were selected for the experiment. We stocked eight with refuges equipped aquaria (80 L) with IG-prey (H. formosa) populations at an approximate equilibrium density (c. 120 individuals) and equilibrium size-structure as estimated in previous studies (Schröder et al. 2009). The day after stocking, juvenile IG-predators were introduced into assigned aquaria kept in rearing cages for acclimatization. Twenty-4 hours later, both adult and juvenile invaders (P. reticulata) were released into assigned aquaria (day 0 of the experiment, replication n = 4). IG-prey and IG-predator populations were sampled 12 times from February 2008 to March 2009 with increasing sampling intervals. For each species' life-history stages, the number of individuals was counted and individual lengths (mouth to beginning of the tail fin) were measured on a computer screen after photographing. Invasion success was defined as successful recruitment by the invader. Aquaria were cleaned once per month. Water level, pump function and feeder function were controlled regularly.

Intermediate-term invasion experiment with intermediate-sized invaders in the presence and absence of refuges

To investigate the size dependency of the two mechanisms (juvenile competitive bottleneck and positive predation-recruitment feedback in the IG-predator) that may underlay the invasion outcome observed in experiment I, we introduced intermediately sized IG-predators (P. reticulata) into size-structured IG-prey (H. formosa) populations in the presence and the absence of refuges.

Overall, we followed the same procedure as in experiment I, with the following departures. We chose three size classes – S (mean length ± SD; 10·67 mm ± 0·88), M (12·97 mm ± 0·81) and L (14·36 mm ± 0·87) – for the invading IG-predator. All treatment combinations (2 refuge × 3 invader sizes) were replicated five times. From experiment I, the invasion outcome for predaceous IG-predators was known. Therefore, invasion outcome was assumed to be predictable when at least two mature individuals had reached predaceous size (>18 mm). After 3 months, all invaders were either extinct or had reached predaceous size. IG-predators were counted and measured once a week for 8 weeks and twice throughout the last 4 weeks. IG-prey populations were sampled every 4th week. The initial IG-prey (H. formosa) population density in this experiment was reduced to half of that used in experiment I (the size-structure was kept the same) because of an initial H. formosa decline to this level observed in all treatments of experiment I. For the smallest invader class (S), controls (n = 5) with only P. reticulata (SC) and controls with only H. formosa (n = 2) for each refuge treatment were conducted.

Juvenile competition

To investigate the juvenile competitive bottleneck, we specifically assessed the competitive abilities of IG-prey (H. formosa) and IG-predator (P. reticulata) juveniles (the size class we expected to compete the most), investigating both intra- and interspecific effects.

IG-prey juveniles were 6·72 mm ± 0·5 (mean length ± SD) and IG-predator juveniles were 7·1 mm ± 0·5 (mean length ± SD) at the start. We applied three treatments for each species: intraspecific competition was estimated at two densities, 15 (low) and 30 (high) individuals per aquarium, and interspecific competition was investigated at the high density (mix, 15 H. formosa + 15 P. reticulata). All competition trials were run in the presence and absence of refuges. The growth of juveniles over a period of 3 weeks was used as a measure of competition. Once per week, juvenile lengths were measured and aquaria cleaned. The experiments were executed in the above-described aquarium system. The densities chosen were derived from a pre-experiment with H. formosa to ensure an adequate level of competition (affecting growth but not causing starvation).

Predatory capture rate

To investigate specifically the size-dependent predation link between IG-predator (P. reticulata) and IG-prey (H. formosa), capture rates of adult female P. reticulata of different sizes (20, 25, 27 and 30 ± 0·5 mm) on juvenile H. formosa of different sizes (6, 9, 12 ± 0·5 mm) were estimated in the absence and presence of refuges. The same experiment was performed for adult Hformosa (20 and 25 ± 0·5 mm) feeding on juvenile P. reticulata (6, 9, 12 ± 0·5 mm). Prey–predator size ratios in terms of body length ranged from 0·2 to 0·6.

Plastic aquaria (10 L) were filled with water originating from the aquarium system. The water temperature ranged from 23 to 24 °C, and the light regime was the same as for all other experiments. Aquaria were visually isolated by plastic intercepts. Ten prey individuals and two predators were randomly assigned to aquaria. During a 24-h starvation and acclimatization period, predators were restricted to a transparent plastic cylinder. Thereafter, predators were released, and the number of surviving prey after 24 h was recorded. Background mortality of prey was assessed by controls with 10 prey individuals only. H. formosa and P. reticulata individuals used in the capture rate experiments were taken from breeding aquaria kept in the aquarium system and fed with a standardized amount of food (10 ± 0·4 mg per day) 2 days before the experiment started. One submerged large (23·5 ± 0·36 g) and two floating small (6·7 ± 0·17 g) balls of green plastic thread served as refuges (Eheim EHFI FIX).

Statistical Analysis

All statistical analysis and computations were processed in R2·11·1 (R Development Core Team 2010). The critical α-level was always set to 0·05. All hypothesis testing was two-sided.

To statistically assess the influence of refuges on the invasion success of juvenile and adult invading IG-predators (P. reticulata), the data on invasion success and coexistence from experiment I were pooled with the relevant data (‘medium’ productivity) from Schröder et al. (2009). Invasion success was analysed by fitting a general linear model (GLM) with binomial error distribution. Invader size and refuge availability were used as explanatory variables and invasion outcome was the binomial response variable with success or failure as values.

We analysed decline rates such as the decline in IG-prey (H. formosa) recruits and decline in total IG-prey abundance using nonlinear regressions (nls) for parameter estimation and anova for testing for treatment differences. In the regression process, an exponential function [y = a × e ^ (b × x)] with decline rate (b) and time (x) was fitted. Invader (IG-predator) decline rates over the first 90 days were compared over all invasion experiments (I, II and Schröder et al. 2009) and analysed in the same manner.

Statistics for the juvenile competition experiment were calculated for each species separately. Juvenile growth rates were investigated by fitting a linear model (lm, y = a + x × g) with growth rate (g) and time (x). A linear model was chosen because only the initial growth phase (<3 weeks) was considered. The growth rates extracted from the linear models were analysed in a two-way anova with competitive environment (low/high/mix) and refuge (present/absent) as explanatory variables. Furthermore, we calculated the skew in the size distributions in all replicates after 2-week testing for treatment differences using a two-way anova with log(x + 2)-transformed data. Where two-way anovas indicated significant differences, the Tukey HSD post hoc test for pairwise comparisons was used to identify treatment groups that differed significantly.

Predatory capture rate data in experiment IV were non-normally distributed and therefore analysed using the nonparametric Kruskal–Wallis test with juveniles consumed as response variable and refuge (present/absent) and prey–predator body size ratio as explanatory variables. When the Kruskal–Wallis test was significant, it was followed by a post hoc test, based on the Tukey method, to test for significant pairwise treatment differences conducting pairwise U-tests (Wilcox's sum rank test) and comparing the results with a statistic based on the studentized range (Q) (Sokale & Rohlf 1995, Box 13·5). Sample sizes for large H. formosa females preying on P. reticulata juveniles were too small for proper statistical analyses.

Results

Long-Term Invasion Experiment in the Presence of Refuges

For invading juvenile (XS) and adult (XL) IG-predators (P. reticulata), the overall invasion success rate in the presence of refuges was 50%, irrespective of invader size (Figs 1 and 2). No long-term coexistence occurred as IG-prey (H. formosa) was always driven to extinction when IG-predator invasion succeeded. Where the invasion of the IG-predator did not succeed, IG-prey populations first decreased to thereafter rebound to densities around 50 individuals per aquarium (Fig. 1). Pooling the invasion success data with the data from Schröder et al. (2009) showed that the invasion success depended on invader body size (GLM, df = 1, = 0·0408) and not on the presence or absence of refuges (df = 1, = 1), while the interaction of the two was significant (df = 1, = 0·0086). Without refuges, invading adult IG-predators (P. reticulata) succeeded to establish a population, while invading juvenile IG-predators went extinct. With refuges, invading adult as well as juvenile IG-predators succeeded in 50% of the replicates. In the presence of refuges, the IG-predator decline rate was significantly higher (anova, F1,6 = 13·97, = 0·0096) where invasion failed (Fig. 1). IG-prey abundances decreased over 200 days but rebounded and stabilized in all replicates where only the IG-prey survived (Fig. 1). The rate of IG-prey decline depended on both invader body size (two-way anova, F1,12 = 6·27, = 0·0277) and refuges (F1,12 = 16·51, = 0·0016), while the interaction of the two variables was not significant (F1,12 = 0·7575, = 0·4012). The decline rate was lower with refuges present and in treatments with invading juvenile IG-predators. The number of IG-prey recruits (≤9·5 mm) decreased faster in the absence of refuges (two-way anova, F1,12 = 18·00, = 0·001) but was independent of invading IG-predator body size (F1,12 = 2·04, = 0·18). The interaction term between invader body size and refuges was not significant (F1,12 = 1·25, = 0·29). Nevertheless, the difference in IG-prey recruits' decline rates between invading adult and invading juvenile IG-predator treatments was larger without refuges than with refuges.

Figure 1.

Changes in total Heterandria formosa (IG-prey, filled circles) and total Poecilia reticulata (IG-predator, open circles) abundance over time in the presence of refuges for each replicate. Left column: adult invasion. Right column: juvenile invasion. Crossed open circles: original invaders. Dashed – dotted lines: reproduction events during the first 250 days.

Figure 2.

(a) Invasion outcome for all invading IG-predator size classes (XS through XL) with (grey bars) and without (black bars) refuges. Invasion success is given as per cent of replicates where IG-predator invasion was successful (survival of at least two mature predaceous invaders until experiment termination). (b) Mean decline rate for all invading IG-predator size classes (XS through XL) with (grey bars) and without (black bars) refuges over the first 90 days of the experiment. SC are control treatments with IG-predators (Poecilia reticulata) of size S. Standard Error bars show ± 1 SE. No refuges XS and no refuges XL data are from Schröder et al. (2009).

Intermediate-Term Invasion Experiment with Intermediate-Sized Invaders in the Presence and Absence of Refuges

The results from the invasion experiment with intermediate sized invading IG-predators were consistent with the above. In treatments without refuges, the percentage of replicates where IG-predator invasion succeeded increased with increasing invader body size (Fig. 2a). In contrast, invasion success did not show any IG-predator body size association when refuges were present, instead invasion success was more random (Fig. 2a). Invader (IG-predator) decline rates were compared over all invader treatments including those from experiment I and Schröder et al. (2009) (only data over the first 90 days of all experiments) (Fig. 2b). Mean decline rates differed depending on IG-predator body size (F5,44 = 5·71, < 0·001), the presence of refuges (F1,44 = 5·54, = 0·023) and treatment combinations (interaction: F5,44 = 2·91, = 0·024). In the presence of refuges, decline rates did not differ significantly between invader treatments. In contrast, when refuges were absent, invading IG-predators of small body sizes declined faster than larger invading IG-predators (Fig. 2b and Table S1 in Supporting Information).

Juvenile Competition

The mean growth rate of juveniles in three different competitive situations differed only significantly for the IG-prey (H. formosa) but not for the IG-predator (P. reticulata). Two-way anova analysis showed an effect of refuges (F1,22 = 5·40, = 0·006) and competitive environment (F2,22 = 6·43, = 0·030), but no significant interaction between the two factors (F2,22 = 0·18, = 0·832) for H. formosa (Fig. 1a). Tukey HSD post hoc testing (Table S2 in Supporting Information) revealed that the growth rate of H. formosa was higher with refuges and at low densities. Growth rates at high intraspecific density did not differ significantly from those at interspecific competition.

The analysis of size distributions showed no differences between treatments in H. formosa, while P. reticulata showed a significant interaction term in the full two-way anova model (interaction term: F2,20 = 5·92, = 0·009). The skew in the P. reticulata size data for interspecific competition differed between presence and absence of refuges (Table S3 in Supporting Information), with a right-skew in the presence of refuges and almost no skew in the absence of refuges (Fig. 3d).

Figure 3.

Mean growth rate (a,b) and mean skewness in size distributions (c,d) in IG-prey (Heterandria formosa, a,c) and IG-predator (Poecilia reticulata, b,d) juveniles after 2 weeks in different competitive environments with (grey bars) and without (black bars) refuges. High: intraspecific competition at 30 individuals per aquarium. Low: intraspecific competition at 15 individuals per aquarium. Mix: interspecific competition with 15 individuals of each species. Means ± 1 SE are given.

Predatory Capture rate

The predation effect of the IG-predator (P. reticulata) on the IG-prey (H. formosa) decreased when refuges were provided (Kruskal–Wallis test, = 0·0025) and with increased prey-predator size ratio (Kruskal–Wallis test, < 0·001) (Fig. 4). Poecilia reticulata only imposed a significant predation impact on H. formosa juveniles when both conditions, prey–predator size ratio ≤ 0·33 and refuges absent, were satisfied (Table S4 in Supporting Information). For prey–predator size ratios smaller than 0·33, P. reticulata females' average capture rate was 8–9 (mean 8·69 ± 0·66 SE) H. formosa juveniles per 24 h in the absence of refuges, but only 0–1 (0·57 ± 0·30 SE) H. formosa juveniles in the presence of refuges. No background mortality in controls was detected in H. formosa juveniles.

Figure 4.

Mean number of IG-prey (Heterandria formosa) juveniles consumed by large IG-predator (Poecilia reticulata) females over a gradient of prey–predator – body size ratios and with refuges present (grey bars) or absent (black bars). No background mortality in controls was detected in Heterandria formosa juveniles. Mean ± 1 SE is given by the error bars. na = treatment missing.

Discussion

Overall Invasion Outcomes

Without habitat structure, invasion success was strictly size-dependent (see also: Schröder et al. 2009) and invasion probability increased continuously along an increasing invader body size gradient. This increase in invasion success resulted from size-dependent interactions between individuals. When prey–predator size ratios were lower or equal to 0·33, invading IG-predators imposed a significant and detrimental predation pressure on the resident IG-prey recruits. In contrast, above this threshold of 0·33, interactions were restricted to competition and juvenile IG-predators on their own could not invade. The fact that invasion success increased continuously over all intermediate invader body sizes, even though the initial interaction was competition (invader sizes S, M & L), reflects the weakened (shortened) competitive bottleneck IG-predators experienced, as less growth was required them to reach predacious sizes.

The introduction of habitat complexity totally erased this size dependence in the IG-predator's invasion success. In fact, it caused a decrease in the invasion success of large IG-predators and an increase in the invasion success of small IG-predators but did not affect the likelihood for coexistence. In the following, we discuss the possible mechanisms that may have led to this change in the invasion outcome.

Species Interactions and Habitat Complexity

The decrease in large IG-predator invasion success in the presence of habitat structure can be related to that the large IG-predators' attack rate was, as predicted, substantially reduced (by almost an order of magnitude). This decrease in predation efficiency annulled the negative effect that large IG-predators had on IG-prey recruits in the absence of habitat structure. The outcome is in correspondence with other studies that have shown that habitat complexity (e.g. vegetated littoral zone of lakes) reduces the strong size dependency in attack rate as compared to less complex habitats (e.g. pelagic) (Persson & Crowder 1997). Habitat complexity is known to decrease the negative effect of IG-predators on IG-prey (Persson & Eklöv 1995; Finke & Denno 2006; Janssen et al. 2007), which, as we show here for a size-structured IGP system, negatively affected the IG-predator's ability to indirectly facilitate its own recruitment through predation. Consequently, habitat structure decreased the invasion success of large IG-predators and caused extinction when the IG-prey could not be suppressed sufficiently by predation.

In contrast, the increase in invasion success of small IG-predators in the presence of habitat structure can hardly be explained by reduced predation of large IG-prey on small IG-predator (mutual IGP) as no mortality effect of large IG-prey on small IG-predators was present (Fig. 4 and see also Schröder et al. 2009). Still, behavioural responses of juvenile IG-prey and IG-predator in terms of increased refuge use may still be present. Habitat structure affects not only prey-predator encounter rates but also the spatial distribution of small and large individuals (Werner & Hall 1988; Persson & Eklöv 1995). Small vulnerable individuals retreat into refuges to avoid predation, while large individuals are restricted to the open water and the refuges' periphery. Correspondingly, we expected that the addition of habitat structure should lead to a spatial redistribution of IG-prey and IG-predators with juveniles using the refuges (Briggs & Borer 2005; Schröder et al. 2009; Nilsson et al. 2011).

This size-dependent redistribution of life stages may, in turn, affect competitive interactions. Experimental studies have shown that foraging relationships between competing species may be equalized or even reversed in structurally complex environments due to that structural complexity may interfere with foraging performance (Winfield 1986; Persson 1991). Our competition experiments revealed a negative interspecific competitive effect of small IG-predators on small IG-prey mean growth in systems with habitat structure both present and absent, whereas no negative interspecific competitive effect of small IG-prey on mean growth of small IG-predators was present in any treatment. At the same time, the combined presence of small IG-prey and habitat structure resulted in a change in the size distribution of small IG-predators, and a lower decline rate of small IG-predator invaders in the presence of habitat structure in the invasion experiments. The change in skewness of the size distribution was attributed to that a major fraction of small IG-predators grew substantially faster in the habitat structure-IG-prey treatment than in the other treatments, whereas a minor fraction grew more slowly.

The fact that small IG-predators did not succeed to invade the resident IG-prey population in the absence of refuges can be related to that the competitive effect of the IG-prey could play out to a maximum extend. Introducing habitat structure may lead to spatial segregation between life stages, specifically a spatial separation between larger individuals of the resident IG-prey population and small IG-predators. In the presence of large predacious individuals, juveniles of both IG-predator and IG-prey have been shown to increase their refuge use (Nilsson et al. 2011, B. Reichstein personal observation).

Moreover, Nilsson et al. (2011) showed that small guppies still spent as much as 50% of their time outside the refuge in the presence of large cannibalistic guppies, whereas small Heterandria spent almost all their time in the refuge (B. Reichstein, personal observation). We suggest that the partial use of the nonrefuge habitat by small IG-predators may reflect individual variation in refuge use that may be reinforced by the presence of small IG-prey and have caused the negatively skewed size distribution in small IG-predators. Individual level mechanisms may thus be important in the complex habitat because IG-predator individuals experience the environment in various ways, with some being positively (exploiting the refuge periphery) and others negatively (staying in the refuge) affected. Such spatial effects depend on specific behavioural responses and may therefore have positive, neutral or negative effects making it difficult to predict the outcome of species interactions (Finke & Denno 2006; Janssen et al. 2007). Griffen & Byers (2006) also showed that increased habitat complexity may significantly alter the interaction landscape when species interactions are habitat specific. When populations are size-structured, the interaction landscape will further be shaped not only by species and habitat specific responses, but also by size-specific responses, and the resulting interaction landscape will be dynamic over time as size relationships change continuously.

Habitat Structure and Coexistence

Habitat complexity has been suggested to promote coexistence in IGP systems. Janssen et al. (2007) argued that if the reduced interaction strength (weakened predation) they observed in their meta-analysis of short-term IGP experiments carried over to long-term dynamics, this could explain the persistence of IGP in nature. Our study is the first to investigate the long-term effect of habitat structure on the dynamics of a size-structured IGP system.

Habitat structure significantly decreased the predation link as suggested by Janssen et al. (2007), but habitat structure did, as we predicted, not promote the long-term coexistence of IG-prey and IG-predator. These results are in correspondence with theoretical studies that suggest that coexistence in size-structured IGP systems is fragile when IG-predator and IG-prey are relatively similar in size and the IG-predator is not depending on the IG-prey for maturation (van de Wolfshaar, De Roos & Persson 2006; Hin et al. 2011). Still, these theoretical studies only considered the effects of habitat structure indirectly through a stage-dependent resource shift in the IG-predator, and we still lack a theoretical analysis of the effects of habitat structure per se on coexistence in IGP systems.

In the presence of habitat structure, IG-predator invasion success was not size-dependent because small IG-prey could avoid predation, and the initial dynamics were hence restricted to competition where IG-predator invaders, regardless of their body size, were forced to pass through a competitive bottleneck that determined juvenile survival and adult reproductive output. The inability of IG-predators to successfully invade in half of the trials can only be explained by negative competitive effects by the resident IG-prey population as predation by the IG-prey on the IG-predator does not occur. The competition experiments suggest that the performance of a fraction of small IG-predators was improved as a result of the combined presence of refuges and small IG-prey. This increased performance of a fraction of small IG-predators may explain the successful invasion of IG-predators in half of the trials, and also that this success was independent of invader size. The fact that invasion at the same time failed in half of the trials points to that the performance of small IG-predators compared to IG-prey individuals was not very different. Furthermore, successful IG-predator establishment was dependent on substantial reproductive outputs; hence, it can be hypothesized that whether the IG-predator or the IG-prey took over was a function of the strength of recruitment in the IG-predator (Fig. 1).

The extinction of IG-prey in half of the trials, on the other hand, can be explained in the following way: if IG-predators escaped the competitive bottleneck, successful reproduction led to an increase in the number of refuging individuals that increased competition among juveniles. Increased juvenile densities were expected to have a negative effect on IG-prey juvenile growth that reintroduced a juvenile recruitment bottleneck in the IG-prey favouring the IG-predator. Additionally, very high juvenile densities could possibly have resulted in the displacement of the IG-prey from the refuge reintroducing intraguild predation. In fact, successful invasion was, as mentioned above, enhanced by high reproductive output of the first and second generation invaders accompanied by a steep decline in IG-prey abundance. Hence, a numerical domination by IG-predator juveniles could directly and indirectly reintroduce the juvenile recruitment bottleneck in the IG-prey thereby demoting coexistence.

In conclusion, habitat complexity may interfere with intrinsic size-dependent processes creating a new interaction landscape that changes with time and numerical abundances. Initially, when IG-predator abundances were low, we observed a refuge-mediated neutralization of trophic interactions, an effect that opened a window where either IG-prey or IG-predator could win. IG-predator persistence depended foremost on invader survival and maturation but also on a sufficient reproductive output that could reintroduce a juvenile recruitment bottleneck in the IG-prey. To predict the outcome of species interactions in such a setting where spatial and temporal heterogeneity are integrated is challenging, and more knowledge about the effects of size-dependent interactions on changes in interaction landscape and interaction strength in space and time is needed.

Acknowledgements

The authors thank Sara Rundqvist and Anja Wenzel for help with aquarium maintenance. Lars Lundmark and William Larsson provided technical support for the feeders. Karin Nilsson supplied us with Poecilia reticulata from her stock. The research was supported by grants from the Swedish Research Council to L. Persson. The experimental design and the use of vertebrates were in accordance with institutional standards and were approved by the Swedish Animal Welfare Organization (Dnr A 95–04).

Ancillary