Feedbacks between community assembly and habitat selection shape variation in local colonization


Correspondence author. E-mail:


1. Non-consumptive effects of predators are increasingly recognized as important drivers of community assembly and structure. Specifically, habitat selection responses to top predators during colonization and oviposition can lead to large differences in aquatic community structure, composition and diversity.

2. These differences among communities due to predators may develop as communities assemble, potentially altering the relative quality of predator vs. predator-free habitats through time. If so, community assembly would be expected to modify the subsequent behavioural responses of colonists to habitats containing top predators. Here, we test this hypothesis by manipulating community assembly and the presence of fish in experimental ponds and measuring their independent and combined effects on patterns of colonization by insects and amphibians.

3. Assembly modified habitat selection of dytscid beetles and hylid frogs by decreasing or even reversing avoidance of pools containing blue-spotted sunfish (Enneacanthus gloriosus). However, not all habitat selection responses to fish depended on assembly history. Hydrophilid beetles and mosquitoes avoided fish while chironomids were attracted to fish pools, regardless of assembly history.

4. Our results show that community assembly causes taxa-dependent feedbacks that can modify avoidance of habitats containing a top predator. Thus, non-consumptive effects of a top predator on community structure change as communities assemble and effects of competitors and other predators combine with the direct effects of top predators to shape colonization.

5. This work reinforces the importance of habitat selection for community assembly in aquatic systems, while illustrating the range of factors that may influence colonization rates and resulting community structure. Directly manipulating communities both during colonization and post-colonization is critical for elucidating how sequential processes interact to shape communities.


Predators can play a large role in structuring communities (Hairston, Smith & Slobodkin 1960; Inouye, Byers & Brown 1980; Wilbur 1997). Although traditionally the effects of predators were thought to be mainly consumptive, more recently the importance of non-consumptive effects has been increasingly appreciated (Preisser, Bolnick & Bernard 2005; Orrock et al. 2008; Peckarsky et al. 2008). In particular, predators have been found to strongly affect colonization behaviour, which can lead to variability in community assembly (Resetarits, Binckley & Chalcraft 2005). While most meta-population and meta-community theory assumes that such variation in colonization is generated by stochastic processes (Gotelli & Kelly 1993; Hanksi & Gilpin 1997; Holyoak, Leibold & Holt 2005), empirical studies of colonization behaviour suggest that this is often not the case; organisms often actively choose the habitats they colonize based on habitat quality. Such adaptive habitat selection (e.g. Fretwell & Lucas 1970) can be an important mechanism in producing variation in colonization among habitats across multiple spatial scales (Caffey 1985; Mayor et al. 2009). These decisions about habitat quality have been implicated as a major pathway by which predators have non-consumptive effects on community structure, but are also made in response to a variety of biotic and abiotic gradients (e.g. predation risk, resource availability, conspecifics and competitors; Gaines & Roughgarden 1985; Binckley & Resetarits 2005; Blaustein et al. 2004; Wilson & Osenberg 2002).

While the study of habitat selection initially focused on consequences for individual fitness, more recently a habitat selection perspective has been extended to suggest an alternative paradigm for community assembly (Resetarits et al. 2005). This view suggests that abundance and composition of taxa within communities are influenced by adaptive habitat selection during colonization, and not just differential post-colonization species sorting mechanisms. Indeed, several studies have demonstrated community-wide shifts in colonization site selection in response to variation in habitat quality (e.g. predation risk, Binckley & Resetarits 2005; pesticides, Vonesh & Kraus 2009). However, understanding how habitat selection affects community assembly and structure requires more than demonstrating shifts in colonization by multiple taxa in response to a static stressor in isolation. As species differentially colonize habitats in response to a stressor, the relative quality of stressor vs. stressor-free habitats for future colonists may diverge. If the effects of these differentially assembling communities on habitat selection take precedence over the effects of the static stressor, habitat selection responses of colonists should change over time as communities assemble. We wanted to know whether community assembly does in fact feedback to modify habitat selection responses to an important indicator of habitat quality (i.e. top predators), and whether these feedbacks change temporally as the community assembles. These issues have not been previously examined, and have large implications for our understanding of how habitat selection shapes community structure.

As species colonize a habitat, the number and diversity of co-existing taxa increase (Dean & Connell 1987; Siemann, Haarstad & Tilman 1999). Thus, we expected assembly to increase the number and type of biotic factors defining the quality of that habitat over time. Since organisms often base their habitat selection decisions on these multiple conflicting factors (e.g. predators and competitors, multiple predator types, Rosenzweig 1991; Morris 2003), colonization habitat selection patterns could change as more taxa colonize a given habitat. For example, behavioural avoidance of habitats containing a top predator may diminish as predator-free habitats fill with conspecifics, competitors and other predaceous taxa. Since previous studies vary considerably in how much assembly has occurred prior to the assessment of habitat selection, ranging from a few hours to several months of community assembly, variation in assembly might explain variation in the strength of habitat selection measured in past studies (e.g. Åbjörnsson, Brönmark & Hansson 2002; Blaustein et al. 2004; Binckley & Resetarits 2005).

Predators are known to play both a consumptive and non-consumptive role in structuring freshwater communities (Wellborn, Skelly & Werner 1996; Resetarits et al. 2005). These communities are often characterized by isolated habitat patches (e.g. stream reach or pond) linked to each other and surrounding terrestrial communities by species with complex life cycles. These species, including amphibians (Resetarits & Wilbur 1989), dipterans (Blaustein 1998; Eitam & Blaustein 2004), and beetles (Binckley & Resetarits 2005), have been shown to exhibit strong preferences during colonization and oviposition in response to top predators. However, although predaceous fish are frequently found in habitats colonized by other animal taxa, e.g. permanent water bodies (Wellborn et al. 1996), there has been no explicit examination of how the assembled community could alter colonization habitat selection response to these predators.

Here, we use a factorial experiment to test the hypothesis that aquatic community assembly will feedback to affect the strength of habitat selection responses to fish cues through time. Specifically, we examine colonization and oviposition behaviour by insects and amphibians for 2 months in experimental ponds in which we manipulate both the presence of a top predator (fish) and community assembly (reduced vs. ambient). In this experiment, as in nature, fish are non-lethal during habitat selection but can have both non-consumptive and consumptive effects on the post-colonization community over time. This design allows us to (i) quantify the effects of predator cues and assembly on habitat selection, (ii) test whether assembly modifies the habitat selection response to predator cues and (iii) examine whether these effects change through time due to community assembly, while controlling for other factors such as breeding phenology that also change over time.

Materials and methods

To examine the effect of community assembly on colonization habitat selection response to fish cues, we repeatedly sampled a 2 × 2 factorial experiment arranged in a fully randomized block design that allowed colonists/ovipositors (aquatic insects and frogs) to choose between fish or fishless pools with ‘ambient’ or ‘reduced’ community assembly (i.e. new colonists placed in the pools they selected vs. new colonists removed each sample period). Colonization was monitored for 8 weeks from 20 May to 11 July 2008. The experimental array consisted of 32 plastic wading pools (152 cm diameter) arranged in eight spatial blocks (n = 8) of four pools each in an old field at Virginia Commonwealth University’s Inger and Walter Rice Center ( The blocks were spaced c. 20 m apart (pools within a block were 1 m apart), filled with well water and immediately and continuously covered with tight fitting covers of polyethylene insect screen (800 μm mesh; 62·3 openings cm−2; sensuBinckley & Resetarits 2003). Male blue spotted sunfish (Enneacanthus gloriosus from a nearby reservoir; Harrison Lake, Charles City Co., VA, USA) were added to two randomly assigned pools per block (3 per pool, avg. total length ± SD = 7·2 ± 1·1 cm). Enneacanthus gloriosus are known to eat small crustaceans (e.g. ostracods, copepods) and fly larvae (i.e. chironomids, Snyder & Peterson 1999), as well as elicit a strong avoidance response in both aquatic beetles and gray tree frog during oviposition and colonization of experimental ponds (Binckley & Resetarits 2003, 2005). Pools were seeded with c. 35 L of leaf litter from adjacent mixed oak/pine forest floor.

Colonization and oviposition was measured by submerging the cover screens several centimetres below the water which allowed aquatic colonization on top of the screen while preventing fish predation from below (sensuBinckley & Resetarits 2003). Fish (and any previous colonists) could still access the surface as needed since the screen rose above the water line near the pool margins. Screens were submerged for 48 h of every 4 days (i.e. 2 days submerged, 2 days pulled taut out of the water, except for 3 June which was followed by 4 days out of the water). Colonizing invertebrates and insect egg masses were collected from screens, counted and identified to family or species. Additionally, we opportunistically submerged covers for 48 h if rain was forecasted overnight to better sample oviposition in pools by anurans. Between 20 May and 11 July, we submerged screens to capture colonizing taxa on 13 sample periods with nine additional periods for frog eggs. Frog eggs were removed from screens and photographed for later counting using imagej image processing software ( and standardized by hand counts. After counting and identification, colonists and eggs from the ambient assembly treatments were immediately placed underneath the screen of the pool they had colonized. Colonists of reduced assembly treatments were released to nearby aquatic habitats or collected for vouchers. The most common aquatic beetles were keyed to species (Ciegler 2003). A random subsample of chironomid larvae were mounted on slides and identified to genus (Merritt, Cummins & Berg 2008). To confirm the efficacy of the assembly removal treatment and total effects of fish post-colonization, aquatic community (i.e. beneath screen) data were collected at the end of the experiment via standard sweeps of each pool perimeter with nested aquarium nets (net opening = 14·0 × 13·2 cm; mesh size, large = 0·79 mm2; small = 0·013 mm2) after thoroughly mixing the pool contents. Phytoplankton abundance was estimated using a relative measure of in vivo chlorophyll a using a handheld fluorometer (AquafluorTM, Version 1.3, Turner Designs, 2004).

Data analysis

We took two approaches to analysing the data. First, we examined the independent and combined effects of fish and assembly (i.e. removal treatment) on the total counts for each taxon of incoming colonist over the experiment using Generalized Linear Mixed Model analysis (GLMM) (Bolker et al. 2009). For completeness we also performed a repeated measures GLMM on the raw data, with similar results (Appendix S1, Supporting information). Counts were modelled as having a negative binomial (Littell, Henry & Ammerman 1998) or Poisson error distribution, or in a few cases were square-root transformed and modelled with a normal error distribution. The best fit distribution was selected using AICc for small sample sizes (Burnham & Anderson 1998; Bolker et al. 2009). Secondly, we examined if the effect of assembly rate on the response of colonists to fish changed over the experiment. We calculated the avoidance of fish pools (measured as the log response ratio [L = ln [(eggs or colonists in fish pools +1)/(eggs or colonists in fishless pools +1)], Hedges, Gurevitch & Curtis 1999] for each date that colonization/oviposition occurred in at least one pool in each ‘assembly’ treatment per block, and then regressed L for both assembly treatments with the days from the start of the experiment. Specifically, we tested the hypothesis that as colonists accrued in the assembling communities over the experiment, fish pools would become more preferred by colonists, but in reduced assembly communities preference would remain unchanged (slopeassembly ≠ slopeno assembly in L regression). We also used repeated measures GLMM to look for main effects of assembly and nonlinear effects of time on L. Response ratios were modelled using a normal error distribution. For all analyses, the best-fit model was selected using AICc (Burnham & Anderson 1998). All analyses were performed using SAS GLIMMIX procedure (version 9.1; SAS Institute Inc., Cary, NC, 2002–2003).


Most taxa altered their colonization and oviposition behaviour in response to fish, but the direction of response and its dependence on assembly conditions varied. Hydrophilid beetles (total = 5240, ≥45%Enochrus ochraceus [Melsheimer 1844], ≥28%Berosus infuscatus [LeConte 1855], ≥7%Tropisternus collaris striolatus [LeConte 1855]) and Culex mosquitoes (698 egg rafts) avoided fish regardless of assembly history: fish pools received 32% fewer hydrophilid adults and 52% fewer mosquito egg rafts compared to fish free pools (Table 1, Fig. 1). Of the two hydrophilid species numerous enough to analyse separately, one mirrored this pattern of fish avoidance: fish pools received 79% fewer E. ochraceus than fish free pools (F1,28 = 35·15, P < 0·001), but Berosus infuscatus did not avoid fish (F1,21 = 0·51, P = 0·49). In contrast, chironomid midges (≥90%Chironomus spp., 10 974 egg masses) preferred fish regardless of assembly history: chironomids laid 70% more egg masses in pools containing fish than in pools without fish (Table 1, Fig. 1). Finally, the response of both dytiscid beetles (3034 adults, ≥83%Copelatus punctulatus [Aubé, 1938] and Hyla frogs (Hyla chrysoscelis, [Cope 1880] 59 509 eggs) to fish depended on assembly history (Table 1, Fig. 1), but the effect of this dependence differed among the two taxa. Dytiscid adults more strongly avoided fish in reduced assembly pools: 55% fewer in fish pools with reduced assembly vs. 17% fewer in fish pools with ambient assembly. Copelatus showed an identical response to dytiscids as a whole (Assembly × Fish, F1,21 = 4·58, P = 0·04, 52% reduction in reduced assembly pools and 14% reduction in ambient assembly). Hyla also appeared to avoid fish in reduced assembly pools (39% fewer eggs in pools that contained fish), but prefer fish in ambient assembly pools (92% more eggs in fish pools).

Table 1.   Generalized linear mixed model statistical output testing (a) the effect of fish and assembly on total colonization/oviposition summed over the experiment (Sum) and (b) the effect of assembly and time on strength of oviposition/colonization response to fish, repeatedly measured by log response ratios (L), over the experiment. Estimates of F statistics are analogous to type 3 SS in anova. Counts are of chironomid and Culex egg masses (Diptera), dytiscid and hydrophilid adults (Coleoptera) and individual Hyla eggs (Anura). Significant effects for each taxa are in bold
EffectsCommon taxa
  1. aBlock included in the Sum model. bBlock included in the L model.

A × F1,280·050·8171,280·250·6211,285·860·0221,210·640·4321,214·610·044
A × T12,1681·030·4233,120·810·51011,1520·750·69010,1400·860·57213,392·160·032
Figure 1.

 Oviposition and colonization response to fish cues and assembling aquatic community summed over the experiment (a–d) and change in preference for fish pools in reduced and ambient assembly pools over time (e–i). Total mean count per pool ± SE for (a) chironomid egg masses, (b) Culex egg rafts (Culicidae), (c) dytiscid beetle adults, (d) hydrophilid beetle adults, (e) Hyla eggs (Hylidae). Mean log response ratio (L) per pool per date ± SE for (e) chironomid egg masses, (f) Culex egg rafts, (g) adult dytiscid beetles, (h) adult hydrophilid beetles and (i) Hyla eggs. Solid circles and lines = ambient assembly pools, open circles = reduced assembly pools. (N = 8 pools per treatment). Regression lines are plotted through block means when slope ≠ 0 (< 0·05). *< 0·05, **< 0·01, ***< 0·001.

Taxa also showed various responses to our manipulation of assembly. Both hydrophilid and dytiscid beetles avoided reduced assembly pools (29% and 33% fewer adult colonists respectively; Table 1, Fig. 1). In contrast, chironomids, Culex and anurans showed no preference based on assembly treatment (Table 1, Fig. 1), although repeated measures analysis suggested that chironomids slightly preferred ovipositing in reduced assembly pools (18% more egg masses per date, Appendix S1, Table S1, Supporting information).

Colonization rates (# of individuals colonizing during a given sample period) varied over the experiment (i.e. colonization was pulsed) (Table S1, Fig. S1, Supporting information). However, contrary to our prediction, the effect of ambient assembly on response to fish did not increase over the experiment for any taxa (Fig. 1). However, chironomids in reduced assembly treatments showed an increased preference for fish pools over the experiment (by 150%, Regression, F1,11 = 8·34, P = 0·02, R2 = 0·43, m = −0·02 ± 0·006, Fig. 1). The strength of response to fish did not change linearly over the season for beetles and Hyla (Table 1, Fig. 1).

Sweep samples of the pools at the end of the experiment (15 July) revealed that our reduced assembly treatment successfully excluded most adult beetles (97%) and hydrophilid larvae (100%) from post-colonization communities, but was unsuccessful in excluding chironomids or odonates. Unfortunately, sweeps were ineffective at sampling tadpoles. Numerous pools were observed to contain tadpoles, but none were found in our samples which were processed after pools were emptied. Fish pools contained 704% more chironomid larvae but 90% fewer dytiscid larvae. Chlorophyll a concentration in the water column per pool (under the net) did not differ by treatment (final GLMM, fish: F1,26 = 0·22, P = 0·65; assembly: F1,26 = 0·04, P = 0·85, fish × assembly, F1,26 = 0·25, P = 0·62). Water depth did not differ among treatments (final GLMM, assembly, F1,21 = 2·23, P = 0·15; fish, F1,21 = 0·09, P = 0·77; mean depth ± SD, 16·9 ± 1·8 cm), nor did afternoon water temperature (Kruskal–Wallis, χ2 = 0·32, d.f. = 3, P = 0·96; mean temperature ± SD, 25·5 ± 0·50 °C). All fish pools had fish at the end of the study (mean ± SD, 2·6 ± 0·7 per pool).


Our results support the hypothesis that community assembly can modify the non-consumptive effects of a top predator on colonization by prey. However, both the non-consumptive effects of fish and the dependence of this effect on assembly were taxa specific. The preferences of both Hyla and dytiscid beetles for fish pools depended on assembly. Dytiscids avoided fish pools when assembly was reduced, but did not avoid fish pools that received ambient levels of colonization. Frogs trended towards avoidance of fish pools when assembly was reduced, and seemed to prefer fish pools at ambient levels of assembly. On the other hand, hydrophilid beetles and mosquitoes strongly avoided, and chironomid midges strongly preferred, colonizing pools containing fish at both reduced and ambient levels of assembly. Post-colonization, the abundance of dytiscid beetle larvae was strongly reduced and midge larvae enhanced in fish pools. These results suggest that by altering abundance and composition of assembling and post-colonization communities, fish alter quality of fish vs. fishless habitats for future colonists. Furthermore, the relative quality of these habitats appear to change with community assembly, and feedback to modify some habitat selection responses of prey with complex life histories. These findings have important implications for understanding when and how the effects of predators structure communities: for example, top predators may shift from having net negative to net positive effects on prey colonization rates as communities assemble and the effects of competitors and other predators take precedence over the direct effects of fish.

Given the cumulative nature of community assembly, we expected the impact of assembly on habitat selection responses to fish to increase over time. Instead, we observed that the effect of assembly on habitat selection responses to top predators (when there was one) appeared either to occur almost immediately (i.e. dytiscids) or to have a complex interaction with time (i.e. Hyla) (Fig. 1). The nearly immediate modification of dytiscid response to fish suggests a low threshold response to assembly rather than a gradual change over time. Behavioural responses to important stimuli are often more sensitive to the presence rather than the degree of the stimuli. For example, Rieger, Binckley & Resetarits (2004) suggest that ovipositing Hyla femoralis strongly avoided even one predatory fish, despite the fact that predation risk for larvae increased only when fish density was >1. Thus in our experiment, it seems even a short period of assembly was sufficient to affect the response of colonizing dytiscids to fish. On the other hand, the change in the effect of assembly on response to fish over time in Hyla suggests a more complex relationship between habitat selection, assembly and time. In this case, the erratic effect of assembly may have resulted from nonlinearity in the assembly process itself: for example, Hyla oviposition was very strongly pulsed (Appendix S1, Fig. S1, Supporting information). Thus, abrupt temporal fluctuations in colonization may lead to especially complex effects of assembly on habitat selection response to fish.

The focus of this study is the experimentally rigorous identification of patterns of habitat selection. However, to explore the possible mechanisms driving effects of assembly on colonization and habitat selection responses to fish and the efficacy of our reduced assembly treatments, we sampled post-colonization communities at the end of the experiment. We found our removal treatments were particularly effective at reducing the abundance of adult beetles and larval hydrophilids post-colonization. Based on this result and the published literature, we consider two mechanisms that could lead to reduced avoidance of pools containing top predators as communities assemble. First, colonization of top predator-free pools by competitors and other predators could make habitats without top predators relatively less attractive as assembly occurs. Secondly, colonization of top predator pools by organisms that could serve as alternative prey for the top predator (conspecific and heterospecific) could make those pools more attractive as the community assembles. Pools already containing colonists would be less risky than pools without colonists via the predator dilution effect (Roberts 1996) in which adding other prey reduces per capita predation risk (e.g. assuming a type 2 saturating functional response in which per capita risk decreases with increasing prey density).

Based on our post-colonization sample and other research on similar aquatic communities (Åbjörnsson et al. 1997), dytiscid beetles appeared to avoid fish in reduced assembly pools in part to reduce their individual risk of being consumed (i.e. predator dilution effect), but Hyla may have been attracted to fish in ambient assembly pools to avoid competition. Åbjörnsson et al. (1997) found that dytscid beetles can distinguish between full and hungry predators: Acilius sulcatus decreased its activity in response to starved predatory fish (Perca fluviatilis) but showed no such response to satiated fish. Thus, if fish in ambient assembly pools were more satiated than those in reduced assembly pools, fish in ambient assembly pools may not have elicited as much of an avoidance response. Furthermore, in this study beetles in general showed an overall preference for ambient assembly pools (Fig. 1), which suggests there may be a benefit to colonizing pools with conspecifics even though larvae of these taxa frequently exhibit size-specific cannibalism. Although a similar mechanism could be driving frog behaviour, Hyla have been found, in contrast, to strongly avoid conspecifics (Resetarits & Wilbur 1989). Thus, as fishless pools fill with eggs, frogs may re-evaluate habitat quality to avoid highly competitive environments for their larvae, leading to habitat selection of pools with fish (i.e. ideal free distribution, Fretwell & Lucas 1970). Several previous studies provide supportive correlative evidence that corroborates our findings in the field: they found that the avoidance of fish by ovipositing Hyla was reduced when large numbers of eggs were laid in one night (Binckley & Resetarits 2003, 2008).

Although we found evidence that assembly modifies patterns of the habitat selection response of colonizing taxa, we also found assembly independent effects of predators in many of the focal taxa studied. The patterns of fish avoidance we observed in hydrophilid beetles and mosquitoes are consistent with past studies of predator induced shifts in habitat selection (Fig. 1, Blaustein 1998; Binckley & Resetarits 2005). On the other hand, our result that midges are attracted to fish pools suggests (surprisingly) that they must receive a direct or conditional benefit (or reduced cost) from occupying pools with E. gloriosus (Fig. 1). Since fish often eliminate large predatory invertebrates through consumptive and non-consumptive means (Wellborn et al. 1996; Åbjörnsson et al. 2002; Binckley & Resetarits 2003), chironomids may benefit from reduced predation risk in fish pools. Post-colonization samples suggested that predaceous dytiscid larvae in pools were almost completely eliminated by fish in our experiment (Table 2, Fig. 2) and have been found in some small lentic systems to consume benthic chironomid larvae (Pajunen 1983). Thus, the preference of midges for fish pools may represent a multiple predator effect (Sih, Englund & Wooster 1999), where risk to the midge larvae of being consumed by predatory dytiscid larvae is reduced by the presence of a top predator (i.e. fish), although there was no evidence that the midges responded directly to dytiscid adults.

Table 2.   Generalized linear mixed model statistical output testing the effect of fish and assembly on taxa abundance in pools at the end of experiment. Estimates of F statistics are analogous to type 3 SS in anova. Counts are of chironomid larvae (Diptera), dytiscid adults and larvae (Coleoptera), hydrophilid adults and larvae (Coleoptera) and libellulid dragonfly larvae (Odonata). Significant and marginally significant effects for each taxa are in bold
EffectsCommon taxa
Larvae + pupaeAdultsaLarvaeAdultsLarvaeaLarvae
  1. aBlock included in the model.

Assembly1,28 0·090·7681,214·040·0581,28 0·000·9591,2810·690·0301,218·020·0101,284·170·051
Fish1,2820·28<0·0011,212·580·1231,2821·91<0·0011,28 1·620·2141,210·250·6231,280·500·487
A × F1,28 3·060·0911,211·450·2411,28 0·000·9591,28 0·030·8751,210·250·6231,280·300·591
Figure 2.

 Aquatic community composition in pools at end of experiment. Mean count per sample ± SE for (a) dytiscid beetle adults, (b) dytiscid beetle larvae, (c) hydrophilid adults, (d) hydrophilid beetle larvae, (e) chironomid larvae and (f) dragonfly larvae. N = 8. Symbol legend same as Fig. 1. *< 0·05, **< 0·01, ***< 0·001.

Re-examining previous studies of habitat selection in aquatic systems reveals that variation in response to top predators due to community assembly may be widespread; although these patterns may not have been previously recognized because both the nature of colonist removal and timing of habitat selection response measurements are not often explicitly examined. For example, Stav, Blaustein & Margalit (2000) allow assembly to occur and find that on average ovipositing Culiseta longiareolata show no difference in the number of egg rafts laid in pools with and without caged predators (Anax imperator). However, examination of the time series data presented in the paper reveals complete avoidance of caged predators for the first 20 days of the study, followed by a loss and then reversal of predator avoidance by the end of the study. Without a ‘no assembly’ control, it is impossible to say if this pattern is due to assembly per se, but the pattern is suggestive of an interaction between predator avoidance and a temporal process such as assembly. Thus, modification of habitat selection responses by community assembly may be relatively common. In our study, two of the five (i.e. 40%) of the broad taxa examined showed this pattern. As both colonist removal and timing of habitat selection response measurements may drive variation in the effect size of habitat selection to top predators, explicitly including time and assembly status will allow researchers to more accurately assess how habitat selection combines with post-colonization processes (e.g. Vonesh et al. 2009) to shape communities. Disentangling time, assembly and colonization to elucidate the mechanisms driving these patterns will require further creative manipulation of the factors identified to vary with time, which are numerous.

Our results illustrate that assembly plays a complex role in shaping habitat selection responses to top predators. We find that habitat selection behaviour in response to fish of terrestrially dispersing adults shapes community assembly by modifying colonization and oviposition of aquatic habitats, while aquatic assembly causes taxa-dependent feedback that can modify subsequent habitat selection. Thus, the role of habitat selection in shaping communities becomes more complex as communities assemble and colonists no longer respond to a single factor alone (e.g. the presence of fish), but also to the cascade of changes that follow. This work reinforces the importance of habitat selection for community assembly in aquatic systems (i.e. colonization is non-random), while illustrating the range of factors that may influence colonization rates and resulting community structure. Our approach of simultaneously and directly manipulating both natural colonization and post-colonization processes is critical for elucidating how these factors interact to shape communities (see also Vonesh et al. 2009). Future studies should continue these types of experiments to explore more explicitly the mechanisms driving patterns of community assembly and structure over time.


Thanks to R. Niccoli and B. Van Allen for assisting with field work and to C. Asquith, F. Butler, A. Drawdy, J. Hite, T. Jackson, K. Sutton for helping in the lab. D. Chalcraft and H. Vance-Chalcraft provided valuable comments on the manuscript. J. Epler, S. Ganguly, D. Garey and L. Smock provided assistance with chironomid identification. This manuscript was greatly improved by comments from two anonymous referees. This is VCU Rice Center Contribution number 009. This research was approved by VCU IACUC #AM1011 and VDGIF permit #031450 to JRV.