Effects of an invasive plant transcend ecosystem boundaries through a dragonfly-mediated trophic pathway

Trophic interactions can strongly influence the structure and function of terrestrial and aquatic communities through top-down and bottom-up processes. Species with life stages in both terrestrial and aquatic systems may be particularly likely to link the effects of trophic interactions across ecosystem boundaries. Using experimental wetlands planted with purple loosestrife (Lythrum salicaria), we tested the degree to which the bottom-up effects of floral density of this invasive plant could trigger a chain of interactions, changing the behavior of terrestrial flying insect prey and predators and ultimately cascading through top-down interactions to alter lower trophic levels in the aquatic community. The results of our experiment support the linkage of terrestrial and aquatic food webs through this hypothesized pathway, with high loosestrife floral density treatments attracting high levels of visiting insect pollinators and predatory adult dragonflies. High floral densities were also associated with increased adult dragonfly oviposition and subsequently high larval dragonfly abundance in the aquatic community. Finally, high-flower treatments were coupled with changes in zooplankton species richness and shifts in the composition of zooplankton communities. Through changes in animal behavior and trophic interactions in terrestrial and aquatic systems, this work illustrates the broad and potentially cryptic effects of invasive species, and provides additional compelling motivation for ecologists to conduct investigations that cross traditional ecosystem boundaries.


Introduction
Consumer-resource interactions can strongly inXuence community structure and function via both top-down (predator-controlled) and bottom-up (producer-controlled) interactions (e.g., Fretwell 1987;Schmitz 2010). Predator-and producer-driven trophic patterns have been independently documented in both aquatic and terrestrial ecosystems (reviewed in Power 1992;Shurin et al. 2002). However, increasing evidence suggests that ecological interactions can cross these habitat and ecosystem boundaries, resulting in surprisingly strong direct and indirect eVects among species in distinct environments (Wallace et al. 1997;HelWeld and Naiman 2001;Baxter et al. 2004). For instance, over 80 % of all animal species have complex life cycles and undergo ontogenetic shifts in habitat aYnity (Werner 1988), often alternating between the use of terrestrial and aquatic systems (Wilbur 1980;Werner and Gilliam 1984). Given this prevalence of complex life cycles, linkages Communicated by Scott Peacor.

Electronic supplementary material
The online version of this article (doi:10.1007/s00442-012-2357-1) contains supplementary material, which is available to authorized users. across habitat and ecosystem boundaries are more common than previously recognized (McCoy et al. 2009;Wesner 2010).
Due to the propensity of many species to undergo aquatic-to-terrestrial life-stage progression, terrestrial systems may be particularly sensitive to local aquatic trophic dynamics. For instance, lentic Wsh can dramatically increase the Wtness of nearby terrestrial plants via indirect eVects of larval dragonXy consumption in the aquatic environment (Knight et al. 2005). By preying upon the larval life-stage, Wsh reduce the density of emerging adult dragon-Xies, which are important predators of terrestrial plant pollinators. Though the potential for reciprocal trophic eVects between these ecosystems clearly exists, if and how terrestrial trophic interactions inXuence aquatic food webs is largely unexplored (see Carpenter et al. 2005;McCoy et al. 2009;Sato et al. 2011 for exceptions). Furthermore, a more thorough understanding of trophic linkages among habitats may reveal additional direct and indirect eVects of habitat alteration.
Within terrestrial systems, invasive plants can have sig-niWcant consequences for native plant community composition through competition and species replacement at a local scale (e.g., Vitousek et al. 1997;Levine et al. 2003). Native plant-pollinator interaction networks are also inXuenced by plant invasion (e.g., Aizen et al. 2008;Padron et al. 2009;Vila et al. 2009). However, little is known about how invasive plants might alter terrestrial and aquatic food webs simultaneously, despite the potential for strong consumerresource interactions that could inXuence system dynamics via complex life cycle organisms. For example, the Xowers of highly fertile invasive plants may subsidize insect populations by attracting large numbers of pollinators (Brown et al. 2002). Locally abundant insect pollinator prey may then attract a greater abundance of predatory adult dragon-Xies. This terrestrial, bottom-up eVect resulting from increased prey resources could alter the food webs of nearby wetlands through increased dragonXy oviposition and recruitment of their aquatic larvae. DragonXy larvae are voracious consumers of aquatic macroinvertebrates, and large changes in their abundances and foraging pressure could cascade through the aquatic community to inXuence lower trophic level taxa (Benke 1976(Benke , 1978Batzer and Wissinger 1996).
Here, we investigate if and how purple loosestrife (Lythrum salicaria), a common, noxious invasive plant of wetland-terrestrial interfaces, aVects terrestrial and aquatic trophic interactions in a manipulative pond study. Purple loosestrife can occur in high densities and dominate large expanses of wetland habitat, is highly fertile, and attracts large numbers of terrestrial, pollinating insects (Brown et al. 2002). Uninvaded wetlands typically have low Xoral densities relative to those invaded by purple loosestrife, through the displacement of native plants with no or low Xowering (e.g., cattails, Mal et al. 1997;or native congener of L. salicaria, Brown et al. 2002). Thus, loosestrife is functionally diVerent from the native plants it replaces, and this may have important community-level consequences. We chose purple loosestrife for this study to (1) investigate, in a basic ecological sense, the degree to which experimental diVerences in Xoral densities can cross the terrestrialaquatic ecotone, and (2) provide a realistic ecological scenario in which such eVects may occur. We hypothesized that the introduction of Xowering purple loosestrife plants to artiWcial wetlands would stimulate a series of terrestrial trophic interactions in which bottom-up eVects of loosestrife plants on secondary consumers would in turn generate top-down eVects in the aquatic community, cascading down to inXuence the abundance and diversity of zooplankton communities (Fig. 1). Although Polis et al. (1997) argue that these types of community alterations could result from direct inputs of resources across an ecotone, in this system we expect loosestrife Xoral resource density to indirectly inXuence the aquatic community via changes in the frequency of dragonXy oviposition events and reproduction. By experimentally manipulating Xoral density, we addressed this gap in knowledge of trophic interactions at the terrestrial-aquatic boundary (e.g., Polis and Strong 1996;Baxter et al. 2005) by focusing on dragonXies, hypothesized key players in this system.

Methods
Eight artiWcial wetlands were created at Washington University in the Tyson Research Center, St. Louis, MO, USA, in June of 2009. Each wetland was located in an old Weld habitat within a matrix of oak-hickory forest, and comprised a central vinyl stock tank (»1,300 L capacity) and four smaller surrounding pools (»100 L capacity) Wlled with well water on June 12. These artiWcial wetlands were each positioned within 80 m of existing water bodies across the landscape (mean distance to existing water: 45 m, range 27-77 m). The experimental wetlands were an average of 306 m from each other (range 163-516 m). Neither distance to nearest water nor distance to nearest experimental wetland inXuenced any of the response metrics (P > 0.15 in all cases). The central tanks were stocked on June 22 with approximately equivalent amounts of six species of aquatic macrophytes and three species of snails. We collected zooplankton and phytoplankton from local ponds using an 80 m plankton net, and used aliquots of this mixture to inoculate each central tank. The remainder of the aquatic community, including amphibians, odonates, dipterans, coleopterans, and hemipterans, was permitted to assemble naturally via dispersal from the local species pool. We assume that these methods provide each mesocosm with the same species pool, but allow for some natural stochasticity in colonization and extinction dynamics. We therefore expect that initial community composition was variable among replicates, but do not expect that initial composition would be biased with respect to experimental Xoral treatment.
We placed 25 loosestrife plants in pots in each of the four small pools around each central pond on June 12. Plants were grown from cuttings derived from Wve parent plants. Plants from the Wve parent lineages were divided equally among the wetland replicates to account for potential genetic eVects. The artiWcial wetlands were randomly assigned one of Wve loosestrife Xowering treatments: 0 % Xowers (n = 2), 25 % Xowers (n = 1), 50 % Xowers (n = 2), 75 % Xowers (n = 1), or 100 % Xowers (n = 2). Once the loosestrife plants began Xowering (July 6), we maintained these treatments until September 1 by removing the appropriate number of Xowers at each wetland by hand three times per week, relative to the number of open Xowers at the 100 % Xowers treatment wetlands. By September 1, the loosestrife plants were Wnished Xowering. We did not clip Xowering stalks outright because we wanted to maintain equivalent plant structure at all wetlands, which is known to aVect odonate oviposition behavior (Remsburg and Turner 2009). Occasionally, we removed some Xowers at one of the 100 % Xowers wetlands in order for the two replicates to have an equal number of open Xowers. The spatial separation of the experimental loosestrife pools from the central sampling pool prevented the input of plant litter and pollen to the central pool. Each wetland was surrounded by fencing to prevent loosestrife decimation by deer herbivory. Thus, the spatial scale of each artiWcial wetland represented initial stages of loosestrife invasion, with small patches easily accessible to deer. The fencing may have allowed species interactions to occur that would typically manifest at moderate stages of loosestrife invasion that are less accessible to deer.
We observed all small pools equally for visiting insects during peak activity (0900-1500) for eight weeks (July 6 to September 1, 2009). All insects were identiWed in the Weld to species or morphospecies [see Electronic supplementary material (ESM) 1], and their behavior was recorded in the form of time spent at the pool and number of Xowers visited. Insects were also classiWed into one of Wve size categories, ranging from very small (e.g., some sweat bees and syrphid Xies) to very large (e.g., carpenter bees and some butterXies). Each pool was observed for 10 min once per week during the eight weeks of the experiment, for a total of 320 observation minutes at each wetland. Pools were observed individually in order to accurately assess small Xoral insect visitors, and data were pooled for each wetland. Each wetland was also observed for dragonXy activity (10 min per week for eight weeks) in sunny, hot weather in random order. DragonXy individuals were counted and identiWed to species (ESM 2) through binoculars, and their behaviors were recorded as the amount of time spent Xying, perching, or ovipositing. DragonXy abundance and behaviors (e.g., number of oviposition events, quantiWed as when an individual approached the tank and repeatedly dipped her abdomen in the water) were summed over the total observation time for each wetland.
At the end of the experiment in October 2009, each of the eight central tanks was thoroughly sampled for zooplankton and macroinvertebrates using methods similar to those employed by Chase et al. (2009). By sampling over ten weeks after the Wrst oviposition events were observed (July 21), we provided ample time for developing dragonXy larvae to inXuence the aquatic community. We exhaustively sampled invertebrates within two 0.2 m 2 chimney samplers and collected and preserved all individuals in ethanol for later identiWcation in the laboratory. We sampled tank walls Fig. 1 Conceptual diagram of the hypothesized trophic pathway in this study. We predicted that introductions of the proliWcally blooming invasive plant purple loosestrife (L. salicaria) would attract insect pollinators and, subsequently, adult dragonXies, which prey upon the smaller insects. Through increased dragonXy oviposition and greater abundance of predaceous larval dragonXies in the wetland, these bottom-up terrestrial trophic interactions could then translate to top-down eVects on the aquatic zooplankton community for surface-dwelling invertebrates with sweeps of a 25 cm wide rectangular net from the bottom to the top of each tank wall at four locations in each tank. We sampled zooplankton from each central tank with Wve collections using an integrated tube sampler to sample the entire water column. These Wve samples were combined (»15 L total) and were Wltered through an 80 m mesh zooplankton net into a 50 mL sample for later laboratory identiWcation. All taxa were identiWed in the laboratory using standard keys and guides. When identiWcation to species level was not possible, taxa were identiWed to morphospecies (ESM 3 and 4).
To determine the relationships between the loosestrife Xower treatments and insect visitors, dragonXy abundance, dragonXy oviposition, and trophic levels of the aquatic community, we performed individual regressions using JMP (version 4.0.4). We used a partially replicated regression design to maximize our statistical power (Cottingham et al. 2005). There was a large amount of variation in the abundance of zooplankton among treatments (range: 87-1245 sampled zooplankton individuals per tank). To ensure that diVerences among treatments in zooplankton richness were not caused by diVerences in the number of individuals, we conducted an individual-based rarefaction analysis on the zooplankton data by sampling down to the lowest common abundance value. Because the number of loosestrife Xowers naturally Xuctuated over the course of the season, and we manipulated the number of Xowers to maintain our intended Xoral treatments, we used the percent of loosestrife Xowers (i.e., the treatment) in our analyses, because it is a clear and constant independent variable. A single wetland (25 % Xowers treatment) was excluded from all aquatic community analyses due to accidental contamination by Wsh larvae during zooplankton inoculation.
We described and quantiWed diVerences in zooplankton communities among treatments using descriptive and inferential multivariate methods. We Wrst ordinated the zooplankton communities using nonmetric multidimensional scaling (NMDS) based on Bray-Curtis similarity values. Because our experimental design had very low (or no) replication within treatment groups, we could not formally test for diVerences in zooplankton community structure among treatment groups. Instead, we tested for a correlation between diVerence in treatment group (percent of loosestrife Xowers) and Bray-Curtis community similarity via a Mantel's test. Finally, to identify key diVerences in zooplankton communities between treatment groups, we pooled all zooplankton into four major functional/taxonomic groups (rotifers, cladocerans, copepods, or ostracods) and then conducted a SIMPER analysis to identify the groups that contributed the greatest change in Bray-Curtis similarity values among treatments. The SIMPER analysis was restricted to the 100, 50, and 0 % Xower treatment cat-egories, each of which were replicated twice. Multivariate community analyses were conducted in PAST (Hammer et al. 2001) and R (R Development Core Team, version 2.13.1).
Zooplankton communities in similar Xower treatment categories were more similar to one another than those in widely diVering treatments (ESM Fig. 1, observed correlation value = ¡0.28, mean simulated value § SD = 0.0045 § 0.2060, P = 0.0731). DiVerences in zooplankton communities among treatments were primarily driven by diVerences in the abundance of rotifers, with greater rotifer abundance in 0 % Xower treatments compared to 50 and 100 % Xower treatments (Table 1). Copepods and cladocerans were also more abundant in the 0 % Xower treatment than in the 100 % Xower treatment.

Discussion
Organisms, nutrients, and energy are increasingly observed to cross traditional ecosystem boundaries (Power and Rainey 2000;Baxter et al. 2005). Here, we showed the capacity of the invasive wetland plant purple loosestrife (L. salicaria) to spark a series of terrestrial-to-aquatic trophic interactions, enhancing pollinator and predatory adult dragonXy local abundance, increasing dragonXy oviposition events in our experimental wetlands, increasing predatory larval dragonXy abundance, and altering zooplankton species richness as well as community structure in the aquatic community.
We documented a signiWcant positive relationship between the density of purple loosestrife Xowers and the visitation of Xying insects, including many potential pollinators. Previous research has demonstrated that purple loose-strife is highly attractive to pollinating insects, and usurps pollinators from native congeners (Brown et al. 2002). Our results further suggest that, by attracting relatively high levels of pollinating insects where there might otherwise be little insect activity (Fig. 2a), purple loosestrife Xowers potentially created a new resource base for novel trophic interactions. For instance, we documented a concomitant increase in adult dragonXy abundance with increasing loosestrife Xower levels. Adult dragonXies are predators of many small Xying insects (Corbet 1999;Knight et al. 2005). These results suggest that adult dragonXies respond with increased abundance and activity to the presence of Xying insect prey, to the visual cue of the Xowers themselves, or to both. The nature of our experimental design does not allow us to separate and assess the relative importance of these potential eVects. We are conWdent, however, that by removing only Xowers and not plant stems we were  Our results also indicate that the adult dragonXy abundances associated with high-Xower treatments resulted in a higher frequency of dragonXy oviposition events and a subsequently higher abundance of dragonXy larvae in the experimental ponds as compared to low-Xower treatments. This is a key link between the terrestrial and aquatic food webs, with a bottom-up eVect of increased adult dragonXies resulting in higher densities of aquatic predators (dragonXy larvae) and in the potential for a top-down aquatic trophic cascade. Further experiments are required to mechanistically conWrm each step of this cascade beyond what we observed from Xoral manipulation alone.
Finally, we document a strong positive relationship between Xower treatment level and zooplankton species richness. Although we hypothesized that Xower treatment level would inXuence lower trophic levels in the aquatic environment, we did not predict a priori that this eVect would manifest itself as elevated zooplankton richness in high-Xower wetlands, with no overall eVect on zooplankton abundance. Although a bottom-up eVect of increased inputs of pollen or detritus from purple loosestrife is one possible explanation for this result, we consider it to be unlikely for three reasons. First, purple loosestrife is primarily insect, not wind, pollinated. Second, the loosestrife plants were housed in separate aquatic mesocosms and could therefore not deposit detritus in the central mesocosms that were the focus of our aquatic sampling. Third, rareWed zooplankton richness also increased with loosestrife Xower treatment, which suggests that diVerences in richness among the treatments were not driven by diVerences in abundance; this contrasts with a bottom-up (more individuals) eVect on zooplankton.
A second, and more likely, explanation is that higher zooplankton richness in high-Xower treatment ponds was mediated by higher densities of dragonXy larvae, possibly via a keystone eVect of predation by dragonXy larvae. We have evidence of large shifts in abundance patterns of zooplankton taxonomic categories (Table 1), which in combination with the positive relationship between Xower treatment and zooplankton richness, is consistent with a keystone eVect. We also found evidence of a shift in zooplankton community composition associated with Xower treatment level, such that similar Xower treatments (and similar abundance of larval dragonXies) resulted in more similar zooplankton communities (Mantel's test and ESM Fig. 1). This general pattern suggests that in this experiment, predation by dragonXy larvae was selective and altered the structure of the zooplankton communities, potentially toward more similar, species-rich communities. This result is further supported by evidence that this change in aquatic community structure was primarily driven by a reduction in the abundance of rotifers. Although signiWcant eVects of dragonXy predation on zooplankton assemblages have been documented relatively rarely and are variable among studies (e.g., Hampton and Gilbert 2001;Burks et al. 2001;Magnusson and Williams 2009), dragonXy larvae are known to prey on rotifers to varying degrees (Hampton and Gilbert 2001;Walsh et al. 2006). This eVect may have been particularly strong in our study, in which a large proportion of dragonXy larvae were of small size.
Other mechanisms may also contribute to the positive relationship between Xoral treatment and zooplankton richness, including increased zooplankton colonization opportunities associated with increased dragonXy oviposition events (Havel and Shurin 2004). Intraguild predation, seen as the suppression of other macroinvertebrate predators by high densities of dragonXy larvae, is also a possible mechanism. However, we did not detect any associations between Xower treatments and the abundance or richness of nonodonate macroinvertebrate predators in this study. Finally, we cannot rule out other indirect eVects of larval odonate abundance on zooplankton richness. The majority of the dragonXy larvae present in our experimental ponds were small in size, supporting eVects on zooplankton and not other taxa.
Overall, our results demonstrate that trophic eVects generated by an emergent wetland invasive Xowering plant are propagated into the aquatic system by a common group of insects with a complex lifecycle: dragonXies. Our results were surprising in that the eVects of purple loosestrife were consistently strong and operated across four distinct levels of trophic interactions. Our study was short-term, highly manipulative, and focused on the assembly phase of the aquatic food web, when dragonXies (for example) may be recruitment limited. Whether eVects such as those observed here also occur in natural habitats in mature wetlands is unknown. However, the importance of assembly history in determining the long-term trajectory of ecological communities suggests that eVects similar to those we document here could potentially have long-lasting consequences for the structure of aquatic communities. Documenting the strength of longer-term reciprocal eVects (e.g., Baxter et al. 2005;Massol et al. 2011) of loosestrife across the aquaticterrestrial ecotone will be an important goal for future work in this and similar systems. Invasive plants have well-documented eVects on terrestrial communities. Purple loosestrife can directly inXuence native plant communities through competition for resources by reducing the colonization success of native species and outcompeting rare species (Hovick et al. 2011). Purple loosestrife can also indirectly reduce native plant Wtness by usurping pollinators (Brown et al. 2002) or altering abiotic conditions that inXuence pollinator visitation, such as the light environment (McKinney and Goodell 2010). Invasive wetland plants like purple loosestrife also have the potential to link and disrupt native terrestrial and aquatic ecosystems (Naiman and Decamps 1997) via allochthonous resource inputs and alterations of aquatic communities (e.g., Schulze and Walker 1997;Bailey et al. 2001;Going and Dudley 2007), leading to an overall alteration of community structure and ecosystem functioning (Hladyz et al. 2011). However, less studied are the multi-trophic interactions that may be propagated across ecosystem boundaries through behavioral responses to resource availability and ontogenetic habitat shifts. Our results suggest that purple loosestrife, as a plant that produces much more Xoral resources than the native plants that it replaces, has the capacity to inXuence the attraction of predatory adult dragonXies that link terrestrial and aquatic trophic interactions. Further experimentation incorporating a native Xoral community as an additional control will help to quantify the relative magnitude of eVects on aquatic communities. Given that purple loosestrife can dominate large areas of wetlands, its invasion might have important implications for species interactions and trophic structure over large scales. Larger-scale studies will help elucidate the landscape implications of loosestrife invasion, particularly in the context of population-level pollinator and dragonXy foraging behavior. The eVects of invasive species may propagate further and through more cryptic pathways than previously appreciated. By broadening our views of ecosystems and collaborating across traditional disciplines, terrestrial and aquatic ecologists together may better understand the intricacies of trophic interactions.