Habitat Structure, Trophic Structure and Ecosystem Function: Interactive Evects in a Bromeliad–Insect Community

Oecologia (2006) 149:493–504 DOI 10.1007/s00442-006-0467-3

Habitat structure, trophic structure and ecosystem function: interactive eVects in a bromeliad–insect community

Diane S. Srivastava

Received: 14 July 2005 / Accepted: 17 May 2006 / Published online: 1 August 2006 © Springer-Verlag 2006

Abstract Although previous studies have shown that Keywords Aquatic insects · Decomposition · Habitat ecosystem functions are aVected by either trophic struc- size · Habitat complexity · Predation ture or habitat structure, there has been little consider- ation of their combined eVects. Such interactions may be particularly important in systems where habitat and tro- Introduction phic structure covary. I use the aquatic insects in bromeliads to examine the combined eVects of trophic structure Although ecosystem functions, as measured by energy and habitat structure on a key ecosystem function: detri- and nutrient Xux, are known to be aVected by both tal processing. In Costa Rican bromeliads, trophic struc- habitat structure (Klein 1989; Didham et al. 1996; War- ture naturally covaries with both habitat complexity and dle et al. 1997; Gonzalez and Chaneton 2002; Tewks- habitat size, precluding any observational analysis of bury et al. 2002) and trophic structure (McQueen et al. interactions between factors. I therefore designed meso- 1989; Schindler et al. 1997; Carpenter et al. 2001; War- cosms that allowed each factor to be manipulated sepa- dle et al. 2001), there has been little investigation of rately. Increases in mesocosm complexity reduced how habitat structure and trophic structure might predator (damselXy larva) eYciency, resulting in high interact to aVect ecosystem functions. Habitat structure detritivore abundances, indirectly increasing detrital pro- may aVect ecosystem functions not only through direct cessing rates. However, increased complexity also eVects on the species performing the function, but also directly reduced the per capita foraging eYciency of the through indirect eVects acting via other trophic levels. detritivores. Over short time periods, these trends eVec- This study examines how trophic structure mediates tively cancelled each other out in terms of detrital pro- the eVects of two aspects of habitat structure (habitat cessing. Over longer time periods, more complex patterns complexity and size) on detrital processing in a brome- emerged. Increases in mesocosm size also reduced both liad–insect ecosystem. ‘Habitat complexity’ is a term predator eYciency and detritivore eYciency, leading to used in many diVerent ways in the literature; here I no net eVect on detrital processing. In many systems, eco- deWne it as the spatial subdivision of habitat at a scale system functions may be impacted by strong interactions smaller than the mobility of individuals. between trophic structure and habitat structure, caution- Habitat structure could directly aVect ecosystem func- ing against examining either eVect in isolation. tion by changing the density or diversity of species carry- ing out that function (Schwartz et al. 2000; Loreau et al. 2002). Increases in species diversity resulted in higher Communicated by Thomas Miller rates of ecosystem functioning in >60% of studies (Sri- vastava and Vellend 2005) because higher diversity com- D. S. Srivastava (&) munities had complementary functional niches, included Department of Zoology and Biodiversity Research Centre, more facilitative interactions, or were more likely University of British Columbia, 6270 University Blvd., Vancouver, BC, Canada V6T 1Z4, to contain functional keystone species (Loreau et al. e-mail: [email protected] 2002). Increasing habitat complexity can increase both

123 494 Oecologia (2006) 149:493–504 density and diversity by creating new niches for species nitrogen by bromeliads (J. Ngai and D. Srivastava, (MacArthur 1972; Denno and Roderick 1991; Uetz unpublished results). Detrital processing—the loss of 1991) or reducing rates of competitive encounters (Beck intact detritus through insect feeding—is therefore a 2000; Young 2001). Increases in habitat size may result key process in this miniature ecosystem. in increased density and diversity because of greater Growth of bromeliads, conversely, aVects their opportunities for specialization or increased likelihood aquatic insect communities. In bromeliads in northwest of immigration (MacArthur and Wilson 1967). For Costa Rica, increases in bromeliad size covary with example, forest fragmentation in the Amazon results in changes in both habitat and trophic structure. Larger reduced dung decomposition rates, likely due to reduc- bromeliads have lower detritivore densities, reduced tions in density, species richness and body size of dung habitat complexity (deWned as the number of leaves and carrion beetles (Klein 1989). Habitat complexity subdividing a standard volume of water) and increased and size could also aVect the per capita eYciency of incidence of predacious damselXy larvae (Fig. 1). organisms performing the function. For example, Other phytotelm studies show covariance between increasing the complexity of saltmarsh vegetation results habitat size, habitat complexity and insects. Macrofa- in greater eYciency of wolf spiders in herbivore suppres- unal richness in Ecuadorian bromeliads is positively sion—an ecosystem function (Denno et al. 2002). correlated with both bromeliad complexity and brome- Habitat structure could also have indirect eVects on liad size (Armbruster et al. 2002). Larger treeholes are ecosystem functions, by changing either top–down or reported to have higher occurrence of predatory dam- bottom–up control of the trophic level performing the selXy larvae (Fincke 1992) and predatory mosquito function. Habitat complexity and size have often been larvae (Bradshaw and Holzapfel 1988). shown to aVect predation rates, for example by providing absolute or stochastic refuges to prey (Luckinbill 1974; Crowder and Cooper 1982; Beukers and Jones Guzmania Vriesea Damselflies absent Damselflies present 1997; Nemeth 1998; Floater 2001; Almany 2004; Langell- 100 10 otto and Denno 2004). Such predator-mediated eVects ) -1 of complexity on prey densities can cascade through 10 1 multiple trophic levels, aVecting ecosystem functions ) -1 carried out by lower trophic levels (Power 1992; Denno 1 Detritivore 0.1 Complexity et al. 2002; Finke and Denno 2002). Even the threat of density V 0.1 0.01 predation is enough to potentially a ect ecosystem func- insects ml (

tions when prey species respond by modifying their for- Detritivore density 0.01 0.001 aging behavior (Lima 1998; Warfe and Barmuta 2004). Complexity (leaves ml V The direct and indirect e ects of habitat structure 0.001 0.0001 are examined here in a bromeliad–insect ecosystem. A 0.1 1 10 100 1000 10000 broad suite of aquatic insect larvae and microbes colo- Bromeliad capacity (ml) nize water trapped by the tightly interlocking leaves of Fig. 1 Bromeliad architecture and the aquatic insect community phytotelm bromeliads (Picado 1913; Laessle 1961; change over a gradient in bromeliad water-holding capacity. Hab- Frank 1983; Richardson 1999; Armbruster et al. 2002). itat complexity (measured as the number of bromeliad leaves subdividing a standard volume of water) decreases as bromeliad The insects include both detritivores and their preda- capacity increases (ANCOVA F1,87 =1,193, P <0.0001, both vari- tors. Phytotelm bromeliads are usually composed of ables log-transformed). Detritivore density also declines as bro- multiple compartments, with each leaf collecting a sep- meliad capacity increases (ANCOVA, F1,52 =32.2, P <0.0001, arate compartment of water and detritus. Most aquatic both variables log-transformed). The two main genera of tank- forming bromeliads, Guzmania (circles) and Vriesea (triangles), insects (except mosquitoes) can move between com- V di er slightly in complexity (ANCOVA F1,87 =11.6, P =0.001, partments by traversing the juncture between overlap- small symbols) but not in detritivore density (genera: F2,51 =0.003, ping leaves. Thus bromeliads can provide highly P =0.96; genera £ capacity: F2,50 =0.54, P =0.46, large symbols). X complex habitats for insects. An useful index of habitat Damsel y occurrence increases with bromeliad capacity (open small symbols damselXies absent, solid small symbols damselXies complexity in this system is the number of leaves subdi- present). Bromeliad complexity and damselXy occurrence data viding a standard volume of water into compartments. are from 92 bromeliads collected in 1997 (22 bromeliads), 2000 Detritus is the basal resource of the insect and micro- (51 bromeliads including 31 previously reported in Melnychuk bial food webs in bromeliads (Richardson 1999), and and Srivastava 2002) and 2002 (19 bromeliads). Detritivore density data were recorded for 58 of these bromeliads (20 bromeliads the main source of nitrogen for epiphytic bromeliads in both 1997 and 2000, and 18 in 2002). Methods for bromeliad (Reich et al. 2003). Processing of detritus by insects, measurements and insect surveys followed those described in when coupled with predation, facilitates the uptake of Melnychuk and Srivastava (2002) 123 Oecologia (2006) 149:493–504 495

This strong covariance between habitat characteristics (a)Top trophic (b) and the insect community precludes any observational level (Expt. 1 & 2) analysis of how each factor aVects detrital processing Present Absent rates. Instead, I designed mesocosms to disentangle the X X Plastic potential in uences of habitat size, habitat complexity, Bromeliad leaf and predator presence on detrital processing rates. These complexity (Expt. 1 & 2) 1 leaf 3 leaves 6 leaves Water line factorial experiments examine the eVects of changing Plastic habitat structure on a constant community of insects, X cup allowing any treatment eVects to be unambiguously Bromeliad size Small 3-leaf attributed to changes in habitat structure rather than (Expt. 2 only) bromeliad mimic covarying changes in insect abundance. Note that detriti- Small Large vore density is thus manipulated solely by changing mes- Fig. 2a, b The experimental design for both experiments de- ocosm size. I predicted that bromeliad structure would scribed in this study. The artiWcial bromeliad mesocosms are aVect detrital processing through both direct eVects on shown from the top (a) or side (b) detritivore insects (changes in detritivore foraging rates) and by indirect eVects via predatory insects (changes in predation rates on detritivores). SpeciWcally, I hypothe- Mesocosms as experimental units sized that: (1) increased habitat complexity—but not size—will impede the mobility of the detritivores and Two experiments were conducted in artiWcial meso- therefore their eYciency in locating resources, directly cosms that mimicked bromeliad structure (Fig. 2). reducing rates of detrital processing; and (2) both Mesocosms were necessary to allow assembly of repli- increased habitat complexity and size will reduce the rate cate insect communities as it is impossible to remove at which predators encounter detritivores, thus reducing all insects from a real bromeliad without destroying the top–down control of detrital processing. water-holding capability of the plant. Insect growth appears roughly comparable between artiWcial and natural bromeliads. For example, the damselXy larvae in Materials and methods the experiments grew in mass from 0 to 2.2% per day (mean =0.45%, SD =0.67, n =46) in mesocosms. In nat- Study site and system ural bromeliads we have measured growth rates of similar range, 0–2.1% per day, but with a higher mean All experiments and surveys were carried out at the (mean =1.20%, SD =0.98, n =10; D. Srivastava, J. Ware Estación Biológica Pitilla (10°59ЈN, 85°26ЈW) in the and J. HuV, unpublished data). Mesocosms were Área de Conservación Guanacaste, north-western placed in a random array on a 1.3£1.3 m table, and Costa Rica. The mid-elevation tropical rainforest kept on a shaded outdoor veranda, that is, at ambient surrounding the station has high densities of bromel- conditions similar to those in the adjacent rainforest iads, particularly in the genera Guzmania and Vrie- but with no natural detrital or insect inputs. sea (Melnychuk and Srivastava 2002). The species pool for bromeliad-dwelling insects in this region is Experiment 1: complexity, predation in excess of 50 species. Aquatic insects occur only as and decomposition rate larvae in the bromeliad; the adults are all winged and terrestrial except in the case of a rare hydrophilid In this experiment, two levels of predation (predator beetle (aquatic larvae and adults). The dominant present or absent) were crossed with three levels of detritivores in the system are larvae of tipulids mesocosm complexity (one, three or six leaves), with (Trentepholia spp., undescribed: Diptera), chirono- six replicates per treatment. mids (especially Polypedilum sp.: Diptera), and scirt- ArtiWcial bromeliads (Fig. 2) were constructed from ids (Coleoptera). The dominant predator in the green plastic cut into leaf shapes and sequentially glued system is Mecistogaster modesta Selys (Pseudostig- together using aquarium-safe silicon, replicating as matidae: Odonata), common in bromeliads > 100 ml much as possible the architecture of natural bromel- in capacity (Fig. 1). This damselXy larvae accounts iads. The outer leaf of each plastic bromeliad was glued for 85–90% of total predator biomass in bromeliads to the inside of a 220 ml plastic cup, this mesocosm size where it occurs; the remaining predator biomass is will be referred to as “small”. Bromeliad mesocosms composed of tabanids, tanypodine chironomids, were constructed out of one, three or six leaves, all ceratopogonids and hydrophilid beetles. with a total mesocosm volume of 95 ml water (i.e. the 123 496 Oecologia (2006) 149:493–504 one-leaf mesocosms had one large compartment scirtid beetle larvae (unidentiWed sp. 4 and 5 mm whereas the six-leaf mesocosms had six small compart- long). Detritivores were randomly distributed among ments). Recently fallen dead leaves, representing a mesocosm compartments. Twenty-four hours after variety of species, were collected from nearby rainfor- all detritivores were added, one damselXy larva (M. est and dried at low temperature for 2 h. The dried modesta, 13–16 mm long, excluding caudal lamellae) leaves were hand-crumbled and separated using soil was added to the central compartment of half the sieves into three size categories: large (2 cm–7 mm), bromeliads. A second round of detritivore insects medium (7 mm–850 m), and Wne (850–250 m). was added mid-experiment, to simulate natural colo- A total of 2.2 g (weighed §0.002 g) of leaves was nization: a tipulid larva (12–20 mm long) was added added to each mesocosm, based on surveys of detrital to all mesocosms on day 14, and on day 25 a scirtid to volume ratios in natural bromeliads (D. Srivastava, larva (5–6 mm long) was added. DamselXies were not unpublished data). The size of detrital fragments added mid-experiment as oviposition by M. modesta added to each individual leaf was: 45.5% large frag- is rare in the wet season; oviposition occurs primarily ments, 45.5% medium and 9% Wne, simulating the in April and May at this site, and the larval period size distribution of detritus in natural bromeliads (D. lasts more than half a year (D. Srivastava and J. Ngai, Srivastava, unpublished data). Detritus was distrib- unpublished results). uted between leaves of the same mesocosm to obtain The experiment was destructively sampled on day a coeYcient of variation of 0.333, similar to that 33. This experimental duration allowed measurable observed in natural bromeliads (D. Srivastava, loss of detrital mass while minimizing loss of detriti- unpublished data). On 28 September 1999, each mes- vores through pupation (all detritivores in this study ocosm received debris, 85 ml settled stream water and have larval stages greater than 1 month). Plastic leaves 10 ml water extracted from nearby natural bromeliads were removed from each mesocosm, separated, and all as a microbial inoculant (all insects were Wrst larvae were retrieved. Body lengths of larvae were removed from this water under £10 magniWcation). measured, and converted to biomass using empirically Water levels were maintained during the course of derived relationships. Pupal cases were recorded. The the experiment by adding appropriate amounts of quantity of detritus remaining at the end of the experi- stream water. ment was measured by sieving the detritus-water mix- Three of the most common species of detritivo- ture through an 850 m soil sieve. Particles retained on rous insect larvae were added to each mesocosm at the sieve were dried for 2 days at 55°C and weighed densities comparable to that observed in nature (§0.0001 g). (Table 1). Insects were added between 30 September Growth rates were estimated for tipulid larvae, the and 2 October (day 1 staggered to allow for harvest- only detritivore species with enough diVerence in ini- ing over a period of days): ten chironomid larvae tial size between individuals to allow individuals to be (Polypedilum sp., 4 mm), one tipulid larva (an unde- tracked through time. I restricted the analysis to bro- scribed Trentepholia sp., 11–12 mm long), and two meliads without predators; otherwise growth and

Table 1 Comparison of insect densities in natural bromeliads and experimental mesocosms Insect densities (individuals per gram dry detritus)a

Polypedilum Scirtid beetle Trentepohlia Mecistogaster modesta chironomid larvae larvae tipulid larvae damselXy larvae

Natural bromeliads with damselXy larvae 1997 (mean § SE, n =9) 5.96 § 1.43 7.08 § 1.11 1.60 § 0.41 0.56 § 0.17 2000 (mean § SE, n =5) 14.25 § 4.38 7.15 § 3.21 1.72 § 0.59 0.17 § 0.05 2002 (mean § SE, n =11) 6.84 § 2.36 3.27 § 0.97 0.97 § 0.18 0.75 § 0.20 Experimental mesocosms Experiment 1 4.54 0.9–1.4 0.45–0.91 0.45b Experiment 2 10–13 3–5 2 1b a Only data for bromeliads that contained at least one damselXy larva are reported here. Bromeliads were collected in both primary and secondary forest in 1997, entirely in primary forest in 2000, and entirely in secondary forest in 2002. Insect densities in experimental mesocosms are given as a range, where appropriate, to represent net densities before and after mid-experiment addition of further larvae b Predator treatments

123 Oecologia (2006) 149:493–504 497 development estimates could be biased towards size scirtid larvae (one 4–5 mm, two 5–6 mm), and ten chir- classes not detected by the predator. Larval growth onomids (4 mm). A single larvae of M. modesta (12– rates were calculated only for tipulid larvae added at 14 mm) was added to the central compartment of half the start of the experiment; many of the tipulids larvae of the mesocosms on day 4. At the mid-point of the added mid-experiment were later instars and so experiment, day 21, a second round of detritivores (to pupated near the end of the experiment. simulate natural colonization), consisting of three chironomid larvae (4 mm) and two scirtid larvae (5– Experiment 2: size, complexity, predation and decom- 6 mm), was added to all mesocosms. The experiment position rate was harvested after 38–41 days, as described earlier. Mesocosms were carefully searched for cadavers of lar- In this experiment, three levels of the insect commu- vae in this experiment, to quantify non-predation mor- nity (detritivores and predators, detritivores only, no tality. Growth of tipulids in the small mesocosms was insects) were crossed with three levels of mesocosm calculated as in Experiment 1. There were not enough complexity (one, three or six leaves) and two levels of tipulids remaining in the large, one-leaf mesocosms at mesocosm size (small or 50% larger); there were Wve the end of the experiment to permit growth analysis for replicates per treatment. Experiment 2 is thus an the large mesocosms. DamselXy length was measured expanded version of Experiment 1, with the additional at the start and end of the experiment, and converted factors of mesocosm size and no-insect controls. The to wet mass using an empirical relationship (r2 =0.96, no-insect controls allowed the amount of detritus pro- n =46) cessed by insects to be calculated as the diVerence in detritus between treatments with and without insects. Analysis of experiments Bromeliad mesocosms were constructed as described earlier. “Small” mesocosms were con- Loss of detrital mass was measured at the end of both structed from 220 ml plastic cups (diameter 4.7–7.1 cm, experiments. Detrital loss includes eVects of insect pro- height 9.0 cm) with a total water volume of 133 ml. cessing as well as passive leaching and microbial “Large” mesocosms were constructed in 440 ml cups decomposition. In Experiment 2, detrital processing by (diameter 5.5–9.0 cm, height 11.0 cm) with a total insects was calculated as detrital loss in the insect-con- water volume of 195 ml (50% more habitat volume taining replicate minus mean detrital loss in the corre- than small mesocosms). Note that the “small” meso- sponding no-insect control. In Experiment 1, there cosms of Experiment 2 were exactly the same size as were no treatments without insects, so detrital process- the “small” mesocosms of Experiment 1, but were ing could not be directly calculated; instead, detrital Wlled with more water so that the small and large bro- loss was used as the response variable. The results were meliads of Experiment 2 had the same water height similar regardless of whether detrital loss or detrital (6 cm). Any diVerence in water height could have processing was considered the response variable in aVected the distance insects needed to travel out of Experiment 2. water from one compartment to the other, potentially Complexity could have been analyzed as either a confounding eVects of mesocosm volume. Dead leaves categorical or continuous variable in these experi- were obtained from windfall branches of a Quiina schi- ments. By analyzing complexity as a continuous vari- ppii (Quiinaceae) tree; a single detrital source was used able, I am able to examine how—not just if— to reduce between-mesocosm variation in rates of complexity aVects insects and energy Xux. Note that leaching. On 3 October 2000, 1.0 g oven-dried, crum- my hypothesized mechanisms involve directional bled dead leaves (weighed to the nearest 0.002 g) was change in insect behavior over a complexity gradient. added to each mesocosm. Detrital distribution I also avoid any loss in power associated with post hoc amongst compartments followed the same rules as in comparisons of means. Regression analysis is gener- Experiment 1. Insect densities (per gram detritus) were ally recommended over ANOVA because it aVords higher than in Experiment 1 but still within the range greater power and provides more quantitative infor- naturally observed (Table 1). DiVerences between the mation about causal patterns (Cottingham et al. experiments in the abundance of individual species 2005). All data were analyzed with regression models. simply reXect variation between years in the availabil- Categorical terms of predator presence and brome- ity of speciWc instars harvested from natural bromel- liad size were coded as either 1 or 0, and nonlinear iads. The following detritivore insects were added eVects of complexity were tested using a quadratic between 4 and 6 October 2000 (day 1): two tipulid lar- model. Most data was modeled with normal errors. vae (one 15–20 mm long, one 9 –13 mm long), three The pupation and cadaver data had substantial 123 498 Oecologia (2006) 149:493–504 heteroscedasticity, so Poisson errors with a log-link 0.32 function were used in a generalized linear model. The No predatorpredat (a) PredatorPredato pupation but not the cadaver data was initially under- 0.28 dispersed (residual deviance: df =0.65), and this was corrected with an empirical scale parameter. Analy- 0.24 ses used GENSTAT 5, release 3.2 (Lawes Agricul- tural Trust) and R version 1.8.0 (http://www.r- Detritus lost (g) project.org). Detritus lost (g) 0.2

0.16 Results 10 Two experiments were conducted. Experiment 1 exam- (b) 8 ined the eVects of habitat complexity on detrital loss and predation. Experiment 2 examined the eVects of 6 both habitat complexity and habitat size on detrital processing and predation. Similar results between the 4 two experiments indicate robustness of the conclu- sions, whereas diVerent results suggest either low 2 power or strong context-dependence. To facilitate such 0 comparisons, I have grouped results from both experiments under several themes. 1 ) Detritivore abundance (no.) -1 (c) V Net e ects of habitat and predation on detrital 0.8 processing

0.6 In Experiment 1, the eVects of habitat complexity on detrital loss depended on whether predators were present or absent (Fig. 3a). In the presence of predators, 0.4 rates of detrital loss were constantly low, irrespective Tipulid growth (mg d of complexity. In the absence of predators, rates of 0.2 detrital loss were much higher in the one-leaf meso- 01 3 6 cosms than either the three- or six-leaf mesocosms. Bromeliad complexity (no. leaves) The eVects of complexity were thus dependent on tro- Fig. 3 EVects of predation by damselXy larvae (Mecistogaster phic structure, and were non-linear in nature (Table 2: modesta Selys) and bromeliad complexity in small mesocosms 2 predator £ complexity and predator £ complexity (Experiment 1) on a detrital loss over 4 weeks, b Wnal detritivore interactions). abundance, and c growth of tipulid larvae. Tipulid growth is presented only for mesocosms without predators, as growth estimates could otherwise be biased towards size classes not detected by the predator. Lines Best-Wt regression models (see Table 2), Table 2 Reduced regression models (backwards elimination) for symbols mean §SE for each treatment the response variables measured in Experiment 1. Only signiW- cant (P< 0.05) explanatory variables are shown, no variables were In Experiment 2, insects processed the most detritus W marginally signi cant (0.10 > P >0.05). For the response variables in mesocosms of intermediate complexity (Fig. 4a). For of detrital loss and detritivore abundance, the full model included predator, complexity, complexity2, predator £ complexity, and all four treatment combinations in this experiment predator £ complexity2. The full model for tipulid growth, exam- (large and small mesocosms, with and without preda- ined in no-predator mesocosms, included complexity and com- tors), there were therefore strong non-linear relation- 2 plexity ships between complexity and detritus processed Response variable Explanatory variables F ratio P value (Table 3: complexity2 eVect). The form of these non- linear relationships depended on habitat size, Detrital loss Predator x complexity F1,31 =5.73 0.022 Predator x complexity2 F =7.43 0.010 predation and their interaction (Table 3; Fig. 4a). Spe- 1,31 W Detritivore Predator F1,32 =70.19 < 0.0001 ci cally, in the absence of predators, small mesocosms abundance Predator x complexity F1,31 =19.1 < 0.0001 had higher detrital processing rates than large meso- Tipulid growth Complexity F =64.3 0.021 V 1,31 cosms (e ect of combining these treatments: F1,54 =67.2, 123 Oecologia (2006) 149:493–504 499

No predator; Small No predator; Large (a) Predator effect Predator; Small Predator; Large -2 Expt. 1 - smallSmall 0.2 (a) Expt. 2 - smallSmall -1 Expt. 2 - largeLarge 0.15 0

0.1 -1 1 0.05 2 by insects (g)

Detritus processed 00 3 -0.05 4 0.03 (b) (b) Mesocosm size effect No predators -1 Predators 0.02 -0.5 growth 0.01 0

0.5 0 (g per 6 weeks) Effect ondetrital processing rate (mg d ) Effect

M. modesta 1

-0.01 1.5 0.8 )

1 2 - (c) - 136 0.6 Bromeliad complexity (no. leaves)

0.4 Fig. 5 a EVects of predation on detrital processing over a gradient in bromeliad complexity, for both experiments. Negative values indicate that predators reduce detrital processing. b EVects of 0.2 habitat size on detrital processing over a gradient in complexity for Experiment 2. Negative values indicate that increasing habitat

Tipulid growth (mg d V 0 size reduces detrital processing. Bars Mean (§ SE) di erences between appropriate treatments 01 3 6 Bromeliad complexity (no. leaves)

Fig. 4 EVects of predation, bromeliad complexity and bromeliad duration. However, there are important similarities size in Experiment 2 on a detrital processing over 6 weeks, and between the experiments. In Experiment 1 mesocosms growth rate of b M. modesta larvae and c tipulid larvae. Tipulid and the similar-sized small mesocosms of Experiment growth is presented only for mesocosms without predators, as 2, predators reduced daily rates of detrital processing growth estimates could otherwise be biased towards size classes not detected by the predator. There were not enough tipulids by comparable amounts (Fig. 5a), and this predator remaining in the large, one-leaf mesocosms at the end of the eVect depended on habitat complexity Tables 2, 3: experiment to permit growth calculations. Lines Best-Wt regres- predation £ complexity interactions). SpeciWcally, pre- sion models (see Table 3) and, in the case of the small and large dation reduced detrital loss in the one-leaf mesocosms mesocosms with predators in (a), are so similar as to be indistin- guishable; symbols mean §SE for each treatment marginally in Experiment 1 (two-tailed t test: t10 =2.00, P =0.073) and substantially in Experiment 2 (t8 =5.06, P =0.0010), but as complexity increased, the eVects of P << 0.0001, Fig. 5b), but in the presence of predators, predation on detrital loss diminished to insigniWcant in V V this di erence disappeared (e ect of combining these six-leaf mesocosms (Experiment 1: t10 =0.89, P =0.39; treatments: F1,53 =0.008, P =0.92; Fig. 5b). Experiment 2: t8 =1.25, P =0.25). At Wrst glance, the eVect of habitat complexity on In summary, detrital processing was reduced by detrital processing appears quite diVerent between the large habitat size and high habitat complexity and— experiments: maximal detrital processing occurred in particularly in small, low-complexity habitats—by pre- one-leaf mesocosms in Experiment 1 but in three-leaf dation. To understand the mechanisms underlying mesocosms in Experiment 2. It will be argued later that these patterns, I now examine the eVects of habitat such diVerences in patterns of detrital processing are structure on the foraging eYciency of both predators an artifact of diVerences between experiments, such as and detritivores. 123 500 Oecologia (2006) 149:493–504

Table 3 Reduced regression models (backwards elimination) for EVects of habitat on detritivore foraging eYciency the response variables measured in Experiment 2. Only explanatory variables signiWcant (P< 0.05) or marginally signiWcant (in parentheses, 0.10 > P >0.05) are shown. The full model for all re- High habitat complexity also reduced the foraging sponse variables except damselXy and tipulid growth included eYciency of detritivores. In Experiment 1, in the predator, size, complexity, complexity2, and all two- and three- absence of predators, total detrital processing way interactions between predator, size and complexity (linear decreased as complexity increased yet detritivores and quadratic forms). The full model for both tipulid and damselXy growth included size, complexity, complexity2, size maintained a constant abundance across the complex- £ complexity, size £ complexity2. Tipulid growth was examined ity gradient; in other words, each individual detritivore only in small mesocosms without predators, damselXy growth was consumed less detritus as complexity increased. In the examined only in mesocosms with predators presence of predators, detritivore abundance increased Response variable Explanatory F ratio P value with complexity, yet total detritus processed remained variables constant over the complexity gradient. Thus, again, per capita foraging eYciency must have decreased as com- Detrital processing Complexity F1,53 = 3.14 < 0.0001 Complexity2 F = 35.6 < 0.0001 plexity increased. 1,53 Y Size F1,53 = 6.64 0.013 These reductions in foraging e ciency as a result of Predator F1,53 = 20.8 < 0.0001 high habitat complexity had important consequences £ Predator F1,53 = 3.34 (0.072) for detritivores: slower growth, lower pupation rates, complexity and higher mortality rates. In both experiments, tipu- Predator £ size F1,53 = 3.20 (0.080) X Damsel y growth Complexity F1,26 = 8.59 0.007 lids had slower growth rates in the more complex mes- 2 Complexity F1,26 = 7.09 0.013 ocosms (Figs. 3, 4; Tables 2, 3). In Experiment 2, Tipulid growth Complexity F1,12 = 4.67 0.05 pupation rates of all species declined with increasing (in absence of predators) mesocosm complexity, particularly in the small meso- Pupation rate (all species) Complexity £ F1,58 = 6.32 0.01 size cosms, and non-predation deaths increased with meso- Non-predation mortality Complexity F1,58 = 17.2 < 0.0001 cosm complexity (Table 3; Fig. 6; comparable data is (all species) not available for Experiment 1). A likely cause of non- predation deaths is starvation due to low foraging eYciency, though further experiments are required for conWrmation. In general, each extra compartment in EVects of habitat on predation rates the mesocosm resulted in an additional 1.47 larval cadavers, and a reduction of 1.43 pupal cases. High habitat complexity reduced predation rates. This Predation also aVected pupation rates in Experi- can easily be seen in Experiment 1, in which predators ment 2. In the absence of predators, one-leaf meso- reduced larval abundance by 85% in one-leaf bromel- cosms had higher pupation rates than more complex iads, 50% in three-leaf bromeliads, and not at all in mesocosms (57% of all pupations) but, in the presence six-leaf bromeliads (Fig. 3b, signiWcant predator of predators, had no pupations since as many detriti- £ complexity eVect: Table 2). Larvae were also lost in vores were consumed as larvae. Thus, the low detrital the absence of predators (due to other causes of mor- processing rates in one-leaf mesocosms in Experiment tality and pupation), but Wnal abundances were similar 2 (Fig. 4a) was due to loss of detritivores to pupation across complexity levels. The frequency of pupation when predators were absent, but loss of detritivores to was greater in the longer-running Experiment 2. Pupa- predation when predators were present. The low detri- tion rates were not independent of predator presence, tal processing rates in six-leaf mesocosms (Fig. 4a) was as described later, preventing the estimation of preda- due to high detritivore mortality in both cases. The tion rates as simply the diVerence in abundance combination of both patterns resulted in maximum between treatments with and without predators (as in detrital processing in the three-leaf mesocosms of Experiment 1). However, predation rates could be Experiment 2 (Fig. 4a). measured indirectly by examining the growth rate of the damselXy, M. modesta. DamselXy growth rates were signiWcantly higher in the one-leaf mesocosms Discussion than in more complex mesocosms in Experiment 2 (Fig. 4b, Table 3: complexity eVects). Although dam- In manipulations with bromeliad mesocosms, habitat selXies tended to grow more slowly in large than small structure had both direct (detritivore-mediated) and mesocosms, this size eVect was not signiWcant (Table 3, indirect (predator-mediated) eVects on detrital Fig. 4b, F1,25 =2.72, P =0.11). processing. Increasing habitat complexity reduced the 123 Oecologia (2006) 149:493–504 501 foraging eYciency of detritivores, and so directly reduced probability of detritivores and damselXies co- reduced rates of detrital processing. However, increas- occurring in the same leaf compartment. Co-occur- ing habitat complexity simultaneously increased rates rence probably occurs when detritivores, not damsel- of detrital processing by decreasing the eYciency of Xies, move compartments in search of resources. M. damselXies preying on detritivores. Thus, increasing modesta damselXies are sit-and-wait predators, and habitat complexity resulted in opposing eVects on have been recorded residing in the same leaf compart- detrital processing rates: more surviving detritivores ment for at least 2 weeks at a time, whereas detriti- (due to reduced predation) but less processing per det- vores normally move compartments within 2 weeks (D. ritivore (Fig. 3). These opposing eVects of complexity Srivastava, unpublished observation). EVects of habitat were of similar magnitude in Experiment 1—there was complexity on predation rates have been shown for no net eVect of complexity on detrital loss in the pres- many other aquatic communities (Crowder and Coo- ence of predators (Fig. 3a)—but not in Experiment 2 per 1982; Heck and Crowder 1991; Sebens 1991; (Fig. 4a). Habitat size may have also had similar direct Nemeth 1998; Warfe and Barmuta 2004), as well as ter- and indirect eVects on detrital processing; increasing restrial insect communities (Casas and Djemai 2002). habitat size led to reductions in detrital processing in In a few cases, such as the current study, complexity the absence, but not in the presence, of predators eVects have been documented to cascade through (Fig. 5b). However, the corollaries of reduced predator multiple trophic levels. For example, increases in the and detritivore growth with increased habitat size are complexity of Spartina tussocks improve predator not as well supported as with increased habitat com- eYciency in reducing planthopper abundance, which in plexity. Note that habitat structure aVected not only turn increases Spartina biomass (Denno et al. 2002). detrital processing, but more generally the Xux of Similarly, rock size in stream beds aVects the strength energy from detritus to detritivores to damselXy larvae: of a Wsh–damselXy–chironomid-algae trophic cascade, complex mesocosms had lower growth rates of both as the complex habitat provided by gravel creates ref- detritivores and damselXies. uges for invertebrates from Wsh predation (Power High habitat complexity likely reduced predation 1992). rates by providing detritivores with a stochastic refuge: High habitat complexity also reduced detritivore foraging rates, even in the absence of predators. Com- plexity may reduce the ability of detritivores to Wnd No predator; Small Predator; Small high-quality patches of resources (detritus in the meso- No predator; Large Predator; Large cosms was heterogeneously distributed amongst compartments, as in natural bromeliads). Detritivores in 3 complex bromeliads may also delay moving to a new 2.5 compartment to search for better resources, trading-oV 2 the beneWts of Wnding new resources against the ener- 1.5 getic costs of moving, or the risk of encountering a V 1 predator. Such trade-o s between feeding and predator avoidance have been shown for many organisms

Non-predation deaths (no.) 0.5 (Lima 1998). 0 Increasing habitat size also reduced detrital process- 1 ing rates in the absence of predators. The mechanism is 0.8 not clear here, for detritivores might be expected to easily Wnd detritus within 6 weeks in even the largest 0.6 compartment (though arguably at a slower rate, due to 0.4 reduced density of both insects and detritus). This 0.2 study was designed so that increases in habitat size Pupations (no.) resulted in reduced detritivore density, to mimic pat- 0 163 terns in real bromeliads (Fig. 1). Reduced encounter rates between detritivores in large mesocosms might Bromeliad complexity (no. leaves) have been expected to reduce interference competi- Fig. 6 EVects of bromeliad complexity, size and predator pres- tion, promoting detrital processing, but the opposite ence on the number of detritivore larvae that a died due to non- trend was observed. This experiment does not allow a predation causes or b reached pupation. Values are mean (+ SE) number of occurrences per mesocosm measured over the 6 weeks direct assessment of resource competition, but note that of Experiment 2 detritivores consumed only a fraction of the detritus 123 502 Oecologia (2006) 149:493–504 provided. In general, detritivore species in phytotelmata Understanding the direct and indirect eVects of hab- have more often been shown to have facilitative than itat structure on ecosystem function is not just impor- competitive interactions (Bradshaw and Holzapfel tant in pristine systems like the bromeliad–insect food 1992; Heard 1994a; Paradise and Dunson 1997). web. Whether habitat structure is changed through The patterns in detrital processing observed in this natural or anthropogenic eVects, systems experiencing study may, in addition to the mechanisms already reductions in habitat structure may be expected to addressed, involve disruption of two types of facilita- show concomitant decreases in trophic structure. For tive interactions. First, damselXy predation on detriti- example, extinction of predator species has been vores reduced the fragmentation of detritus, and may linked to reductions in habitat size (Luckinbill 1974; have thus reduced the colonization of detritus by bac- Holt et al. 1999; Brashares et al. 2001) and increases in teria and fungi. Microbial decomposition, which is habitat fragmentation (Gilbert et al. 1998). Similarly, essential to the breakdown of detritus, may be facili- decreases in habitat complexity lead to reductions in tated by insects (Cummins and Klug 1979). Second, predator abundance (meta-analysis: Langellotto and increases in either predation or complexity may have Denno 2004). The bromeliad example shows that, disrupted detrital processing chains. A processing when trophic structure and habitat structure covary, it chain occurs when some detritivores convert intact will be particularly important to understand how the detritus into Wner organic particles (by shredding detri- two factors interact to aVect ecosystem functions. tus or scraping mesophyll layers), facilitating other Despite interest in biotic regulation of ecosystem func- detritivores that collect or Wlter Wne particulate matter tion (Loreau et al. 2002) and the ecological conse- (Heard 1994b). Processing chains have been docu- quences of changed habitat structure (Langellotto and mented in phytotelmata, but only under certain condi- Denno 2004), there is still remarkably little research on tions (Heard 1994a; Paradise and Dunson 1997; the combined eVects of species loss and habitat change Paradise 1999; Daugherty and Juliano 2002). We have on function (Srivastava 2002; Srivastava and Vellend recently found evidence for processing chains in bro- 2005). meliads (B. Starzomski, D. Suen and D. Srivastava, unpublished results). Tipulids are particularly vulnera- Acknowledgments Volunteer assistance by Kate Edwards, ble to damselXy predation and loss of this dominant Michael Melnychuk, Jennifer Passmore and Jessica Ware was vi- tal to this project. Personnel of the Área de Conservación Gua- shredder may have disrupted any processing chain. nacaste provided logistical support, particular thanks to Róger Spatial partitioning of both insects and detritus in com- Blanco, María Marta Chavarría, Calixto Moraga, Petrona Rios, plex mesocosms may have also disrupted potential pro- and Carlos Zuñiga. John Epler, Jon Gelhaus, Dennis Paulson and W cessing chains; this could have resulted in the observed Daryl Suen aided with identi cations. The manuscript was im- Y proved by the careful reading of Jacqueline Ngai, Jordan Rosen- reduction in detritivore e ciency. Such mechanisms feld, Thomas Miller and several anonymous reviewers. Funding can only be tested with additional experiments. was provided by the Biodiversity Research Centre, UBC, and by The two experiments exhibited very diVerent pat- NSERC grants to W.E. Neill and D.S. terns between complexity and detrital loss (processing), despite similar underlying mechanisms. In the absence of predators, maximal detrital loss occurred at References the lowest complexity in Experiment 1. By contrast, Tewksbury JJ et al (2002) Corridors aVect plants, animals, and maximal detrital processing occurred at intermediate their interactions in fragmented landscapes. Proc Natl Acad complexity in Experiment 2. The proximate reasons Sci USA 99:12923–12926 for these diVerences were more pupations (especially Almany GR (2004) DiVerential eVects of habitat complexity, pre- in one-leaf bromeliads) and more non-predation dators and competitors on abundance of juvenile and adult coral reef Wshes. Oecologia 141:105–113 deaths (especially in six-leaf bromeliads) in Experi- Armbruster P, Hutchinson RA, Cotgreave P (2002) Factors inXu- ment 2. Both of these eVects probably occurred encing community structure in a South American tank bro- because of the longer duration of Experiment 2 (38– meliad fauna. Oikos 96:225–234 Beck MW (2000) Separating the elements of habitat structure: 41 days for Experiment 2 vs 33 days for Experiment 1), V V independent e ects of habitat complexity and structural although this cannot be isolated from other di erences components on rocky intertidal gastropods. J Exp Mar Biol between the experiments (year, detritus type and Ecol 249:29–49 amount, detritivore abundances). The contrast Beukers JS, Jones GP (1997) Habitat complexity modiWes the im- W between the two experiments suggests that, in natural pact of piscivores on a coral reef sh population. Oecologia V 114:50–59 bromeliads, net e ects of complexity on detrital pro- Bradshaw WE, Holzapfel CM (1988) Drought and the organiza- cessing may be quite sensitive to seasonal variation in tion of tree-hole mosquito communities. Oecologia 74:507– oviposition rates and detrital inputs. 514 123 Oecologia (2006) 149:493–504 503

Bradshaw WE, Holzapfel CM (1992) Resource limitation, habitat Holt RD, Lawton JH, Polis GA, Martinez ND (1999) Trophic segregation, and species interactions of British tree-hole rank and the species-area relationship. Ecology 80:1495–1504 mosquitos in nature. Oecologia 90:227–237 Klein BC (1989) EVects of forest fragmentation on dung and car- Brashares JS, Arcese P, Sam MK (2001) Human demography and rion beetle communities in Central Amazonia. Ecology reserve size predict wildlife extinction in West Africa. Proc R 70:1715–1725 Soc Lond B Biol Sci 268:2473–2478 Laessle AM (1961) A micro-limnological study of Jamaican bro- Carpenter SR et al (2001) Trophic cascades, nutrients, and lake meliads. Ecology 42:499–517 productivity: whole-lake experiments. Ecol Monogr 71:163– Langellotto GA, Denno RF (2004) Responses of invertebrate 186 natural enemies to complex-structured habitats: a meta-ana- Casas J, Djemai I (2002) Canopy architecture and multitrophic lytical synthesis. Oecologia 139:1–10 interactions. In: Tscharntke T, Hawkins BA (eds) Multi- Lima SL (1998) Non-lethal eVects in the ecology of predator-prey trophic level interactions. Cambridge University Press, Cam- interactions. BioScience 48:25–34 bridge, pp 174–196 Loreau M, Naeem S, Inchausti P (eds) (2002) Biodiversity and Cottingham KL, Lennon JT, Brown BL (2005) Knowing when to ecosystem functioning: synthesis and perspectives. Oxford draw the line: designing more informative ecological experi- University Press, Oxford ments. Front Ecol Environ 3:145–152 Luckinbill LS (1974) The eVects of space and enrichment on a Crowder LB, Cooper WE (1982) Habitat structural complexity predator–prey system. Ecology 55:1142–1147 and the interaction between bluegills and their prey. Ecology MacArthur RH (1972) Geographical ecology. Harper and Row, 63:1802–1813 New York Cummins KW, Klug MJ (1979) Feeding ecology of stream inver- MacArthur RH, Wilson EO (1967) The theory of island biogeog- tebrates. Annu Rev Ecol Syst 10:147–172 raphy. Princeton Univeristy Press, Princeton Daugherty MP, Juliano SA (2002) Testing for context-depen- McQueen DJ, Johannes MRS, Post JR, Stewart TJ, Lean DRS dance in a processing chain interaction among detrius-feed- (1989) Bottom-up and top-down impacts on freshwater pe- ing aquatic insects. Ecol Entomol 27:541–553 lagic community structure. Ecol Monogr 59:289–309 Denno RF, Roderick GK (1991) InXuence of patch size, vegeta- Melnychuk MC, Srivastava DS (2002) Abundance and vertical tion texture, and host plant architecture on the diversity, distribution of a bromeliad-dwelling zygopteran larva, Me- abundance, and life history styles of sap-feeding herbivores. cistogaster modesta, in a Costa Rican rainforest (Odonata: In: Bell SS, McCoy ED, Mushinsky HR (eds) Habitat struc- Pseudostigmatidae). Int J Odonatol 5:81–97 ture: the physical arrangement of objects in space. Chapman Nemeth RS (1998) The eVect of natural variation in substrate and Hall, New York, pp 169–196 architecture on the survival of juvenile bicolor damselWsh. Denno RF, Gratton C, Peterson MA, Langellotto GA, Finke DL, Environ Biol Fishes 53:129–141 Huberty AF (2002) Bottom-up forces mediate natural-ene- Paradise CJ (1999) Interactive eVects of resources and a process- my impact in a phytophagous insect community. Ecology ing chain interaction in treehole habitats. Oikos 85:529–535 83:1443–1458 Paradise CJ, Dunson WA (1997) Insect species interactions and Didham RK, Ghazoul J, Stork NE, Davis AJ (1996) Insects in resource eVects in treeholes: are helodid beetles bottom-up fragmented forests: a functional approach. Trends Ecol Evol facilitators of midge populations? Oecologia 109:303–312 11:255–260 Picado C (1913) Les broméliacées épiphytes considerées comme Fincke OM (1992) Consequences of larval ecology for territorial- milieu biologique. Bull Sci Fr Belg 47:215–360 ity and reproductive success of a neotropical damselXy. Ecol- Power ME (1992) Habitat heterogeneity and the functional sig- ogy 73:449–462 niWcance of Wsh in river food webs. Ecology 73:1675–1688 Finke DL, Denno RF (2002) Intraguild predation diminished in Reich A, Ewel JJ, Nadkarni NM, Dawson T, Evans RD (2003) complex-structured vegetation: implications for prey sup- Nitrogen isotope ratios shift with plant size in tropical bro- pression. Ecology 83:643–652 meliads. Oecologia 137:587–590 Floater GJ (2001) Habitat complexity, spatial interference, and Richardson BA (1999) The bromeliad microcosm and the assess- “minimum risk distribution”: a framework for population ment of faunal diversity in a Neotropical forest. Biotropica stability. Ecol Monogr 71:447–468 31:321–336 Frank JH (1983) Bromeliad phytotelmata and their biota, espe- Schindler DE, Carpenter SR, Cole JJ, Kitchell JF, Pace ML cially mosquitoes. In: Frank JH, Lounibos LP (eds) Phytotel- (1997) InXuence of food web structure on carbon exchange mata: terrestrial plants as hosts of aquatic insect between lakes and atmosphere. Science 277:248–251 communities. Plexus, Medford, pp 101–128 Schwartz MW, Brigham CA, Hoeksema JD, Lyons KG, van Gilbert F, Gonzalez A, Evans-Freke I (1998) Corridors maintain Mantgem PJ (2000) Linking biodiversity to ecosystem func- species richness in the fragmented landscapes of a microeco- tion: implications for conservation ecology. Oecologia system. Proc R Soc Lond B 265:577–582 122:297–305 Gonzalez A, Chaneton EJ (2002) Heterotroph species extinc- Sebens KP (1991) Habitat structure and community dynamics in tion, abundance and biomass dynamics in an experimen- marine benthic systems. In: Bell SS, McCoy ED, Mushinsky tally fragmented microecosystem. J Anim Ecol 71:594– HR (eds) Habitat structure: the physical arrangement of ob- 602 jects in space. Chapman and Hall, New York, pp 211–234 Heard SB (1994a) Pitcher-plant midges and mosquitoes: a pro- Srivastava DS (2002) The role of conservation in expanding bio- cessing chain commensalism. Ecology 75:1647–1660 diversity research. Oikos 98:351–360 Heard SB (1994b) Processing chain ecology: resource condition Srivastava DS, Vellend M (2005) Biodiversity-ecosystem func- and interspeciWc interactions. J Anim Ecol 63:451–464 tion research: is it relevant to conservation? Annu Rev Ecol Heck KL Jr, Crowder LB (1991) Habitat structure and predator– Evol Syst 36:267–294 prey interactions in vegetated aquatic systems. In: Bell SS, Uetz GW (1991) Habitat structure and spider foraging. In: Bell McCoy ED, Mushinsky HR (eds) Habitat structure: the SS, McCoy ED, Mushinsky HR (eds) Habitat structure: the physical arrangement of objects in space. Chapman and Hall, physical arrangement of objects in space. Chapman and Hall, New York pp 281–299 New York, pp 325–348

123 504 Oecologia (2006) 149:493–504

Wardle DA, Zackrisson O, Hörnberg G, Gallet C (1997) The Warfe DM, Barmuta LA (2004) Habitat structural complexity inXuence of island area on ecosystem properties. Science mediates the foraging success of multiple predator species. 277:1296–1299 Oecologia 141:171–178 Wardle DA, Barker GM, Yeates GW, Bonner KI, Ghani A Young K (2001) Habitat diversity and species diversity: testing (2001) Introduced browsing mammals in New Zealand natu- the competition hypothesis with juvenile salmonids. Oikos ral forests: aboveground and belowground consequences. 95:87–93 Ecol Monogr 71:587–614

123