University of Nevada, Reno

Aquatic Community Interaction Diversity and Mosquito Larvae

A dissertation submitted in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

in

Ecology Evolution and Conservation Biology

by

Will Patrick Lumpkin

Lee A. Dyer Dissertation Advisor

May, 2020

i

Abstract

Mosquitoes comprise a diverse group of small flies (Diptera) in the family

Culicidae which includes an estimated 3,600 described species. Colloquially we know mosquitoes as biting insects that pose a threat to humans and domestic animals as important vectors of disease. Although a minority of the described species of mosquitoes are not known as competent disease vectors, many competent vector species are highly common surrounding human habitations. Despite being important flying insects, mosquitoes undergo an entirely aquatic life cycle as developing larvae and pupae.

During these developmental stages immature mosquitoes are most vulnerable to predation and competition for resources. Their habitats are highly variable in many factors including size, invertebrate diversity, and spatial heterogeneity.

My dissertation research focuses on the larval stages of mosquitoes. The main questions of my research include: 1. What are the important interactions of co-inhabiting invertebrates including predators and competitors, with mosquito larvae? 2. Does environmental heterogeneity in the form of plant complexity influence the structure of invertebrate diversity in aquatic communities? 3. Does interaction diversity affect the abundance of mosquito larvae? My research includes four complimentary approaches to answering these questions. First, I conducted a meta-analysis on the use of natural enemies to control mosquito populations. Second, I developed simulation models to test the effects of plant, herbivore, and enemy diversity, abundance and diet breadths on sampling interaction diversity in artificial communities. Third, I conducted two identical mesocosm experiments with experimental manipulations of plant diversity and structural complexity in order to test the effects of those on aquatic invertebrate diversity and ii mosquito abundance. Finally, I measured plant diversity and environmental heterogeneity in unmanipulated aquatic field environments to test the effects of plant diversity and environmental heterogeneity on invertebrate diversity and mosquito abundance.

The results from my research show several important relationships between environmental heterogeneity, invertebrate diversity, interaction diversity, and mosquito abundance: 1. Natural enemy groups including predators, competitors, parasites, and pathogens can have important negative effects on mosquitoes. 2. Increased predator and competitor diversities reduce larval mosquito abundance through direct and indirect effects. 3. Plant diversity and environmental heterogeneity have positive effects on community invertebrate diversity. 4. Greater interaction diversity in aquatic systems reduces larval mosquito abundance. These results show the importance of protecting and encouraging biodiversity as components of effective larval mosquito control programs.

Careful management of aquatic macroyphyte diversity and environmental heterogeneity will help reduce larval mosquito abundance. iii

Acknowledgments

I have many people to thank for helping me in my doctoral research and guiding me along the way. I would like to thank my committee members Lee Dyer, Angela

Smilanich, Mike Teglas, Steven Juliano, and Beth Leger. Throughout the years they have each helped me focus my curiosities into the meaningful questions and projects that are presented in this dissertation. Their experience and support helped me focus my ideas into practical scientific processes and research designs. Most importantly, this manuscript would not have been possible without their open mindedness and belief in my research questions. I would also like to thank the many friends and colleagues that I have had over the years in the Ecology Evolution and Conservation Biology program. I have been constantly impressed by the diversity of people and diversity of scientific interests encompassing the program. It is truly an inspirational place to conduct research and collaborate with others. iv

Table of Contents

Chapter

1. Introduction…………………………………………………………………………….1

2. A Meta-analysis on the Effectiveness of Biological Mosquito Control………...... …...7

Abstract……………………………………………..…………………………...... 7

Introduction………………………………………..………………………………8

Methods……………………………………………...…………………………...11

Results……………………………………………...………………………….....14

Discussion…………………………………………...…………………………...16

3. Simulated tri-trophic networks reveal complex relationships between species diversity and interaction diversity………………………………….…………...………………….24

Abstract……………………………………………..……………………………25

Introduction………………………………………...…………………..………...26

Methods……………………………………………...……………………..…….29

Results……………………………………………………...…………………….36

Discussion…………………….……………………...…………………...……...45

4. Macrophyte Diversity and Complexity Reduce Larval Mosquito Abundance……….57

Abstract……………………………………………..……………………………57

Introduction…………………………………………..…………………………..58

Methods……………………………………………...………………….………..63

Results………………………………………………...…………………...……..67

Discussion…………………………………………...………….………………..73 v

5. Aquatic Complexity and Interaction Diversity Reduce Mosquito Abundance…..…..78

Abstract…………………………………………….…….………..………...…...78

Introduction………………………………………….…….……..…………...….79

Methods…………………………………………….…………..……………..….83

Results……………………………………………….…………..……………….85

Discussion…………………………………………….……………..……..…….89

6. Conclusions……………………………………………….….……………..………...97

7. Literature Cited…………………………………..………..…..…………………….101 vi

List of Tables

Chapter 2.

1. Important taxa used for classical and neoclassical biological control of mosquitoes…………………………………………………….10

2. Below are the number of effect sizes collected by antagonist category, experimental location (field versus laboratory), container type (artificial, mesocosm, and natural), and by effect measurement target (adult, larval behavior, larval only)..….…………..…..…………..15

3. The mean posterior probability coefficients, standard deviations and 95 % credibility intervals for the effect of total mosquito measurements (A. adult effects, behavioral effects, and larval effects) and larval only effects (B). ………………………………………..…….………….19

Chapter 4.

1. Results from individual Bayesian Regression models based on hypothetical relationships (Figure 1). Estimates with mosquito larvae as response variables were modeled assuming negative binomial distributions.………………………………………….………..72

Chapter 5.

1. The results for the Bayesian linear regression models are shown below including the mean posterior beta coefficients, 95 % credibility intervals, and posterior standard deviations. Below each estimate are the effects of site name as a covariate and the intercepts. Site shows stronger coefficient values for regressions on larval mosquito abundance……………………………………….……..………88 vii

List of Figures

Chapter 2.

1. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects of trophic groups on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development……………………..…...20

2. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects between experimental container types on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B).…...21

3. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects between field versus laboratory- conducted experiments on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B)……………………………………………….... 22

4. Bayesian posterior probability estimates and 95 % credibility Intervals for natural enemy effects between family or taxonomic grouping, on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B)………………………………………………………………………..22

Chapter 3.

1. A randomly selected tri-tropic network produced from one of the 1000 simulations. Each black bar is a node representing a unique species, while the grey bars are edges connecting the black bars and represent observed interactions between those two species. Green sections within some of the black bars represent individuals within that particular species that were present in the community, but not involved in trophic interactions (e.g., plants without herbivores). The width of each edge and node within the network denotes the abundance of sampled interactions or species. Only species that were sampled are shown in this network. Numbers above each node denote the species identification number from that particular simulation………………………………………….……..38 viii

2. Posterior probabilities of: A) mean Chao1 estimates of richness for species and interactions, and B) the mean slope of rarefaction curves for species and interactions. Interactions are displayed in grey, while species are shown in white. The error bars represent the 95% High Density Intervals (HDI). Mean slopes were acquired by calculating the slope of each rarefaction curve when half of the species or interactions were sampled. Chao1 estimates of richness were acquired using the `estimateR' function in the vegan package in R………………………..………………………...……………………39

3. Summary plots of semi-partial correlations between the residuals of species diversity and interaction diversity (these residuals are on the y-axis) and mean consumer diet breadth, species richness, and total abundance (these three parameters are on the x-axis). We investigated this relationship for all three networks (e.g., PH, HE, PHE). The top three panels represent changes in mean diet breadth for each consumer trophic level; mean herbivore and enemy diet breadth were used for the PH and HE networks respectively, while the mean diet breadth for all consumers (herbivores plus enemies) was used for PHE networks. The middle three panels denote community richness for each respective network, which is the total number of species found in all trophic levels. The lower panel displays semi-partial correlations with total community abundance, which equals the sum of all individuals within each trophic level. The solid black lines are least squares regression lines…………………………………………………………...41

4. A path diagram summarizing the standardized path coefficients across all 1000 local communities (χ2 = 3.6, df = 4, P = 0.5; AIC = 36). Each path was chosen based on a priori hypotheses, and compared to competing models using AIC and χ2. Lines ending with an arrow denote positive coefficients, while lines ending with a circle denote negative coefficients. The width of the arrow indicates the relative size of the coefficient………………………………………..42

5. Scatterplots displaying the relationship between the strength of each path coefficient and the number of sampled interactions included in the path analysis (Figure 5), with the exception of paths associated with connectance. The strength of the path coefficient is shown on the y-axis and number of observed interactions included in the model is shown on the x-axis. The solid line represents outcome of linear or polynomial regressions. Path coefficients used in these analyses were significant (P < 0.05)………………………………………………………………44 ix

Chapter 4.

1. Theoretical relationships between aquatic macrophyte diversity and structural complexity elements. Solid arrows represent direct positive effects, bullets represent negative effects and dashed lines represent indirect effects. Macrophyte diversity should contribute to more foraging surfaces and predator avoidance refuges for mosquito competitors but also spatial refuges for ambush predators. Synergistic relationships are also likely but not described here……………………………………………………………………...62

2. Mesocosms were aligned in a 4 × 10 grid pattern with each container spaced 1 meter apart. The 10-meter length was oriented from East to West. (A and B) the total experimental setup and a high-diversity mesocosm planting, respectively. (C) the potassium permanganate treatment from the 2012 experiment…………………..…64

3. Results from the simple path analysis showing a direct positive effect of competitor diversity on predator diversity and a direct negative effect of predator diversity on mosquito larvae (abundance). Fully standardized estimates are shown in bold with the partially standardized over latent variable estimates are in parentheses. The dashed bulleted line shows a negative indirect effect of competitor diversity on mosquito larvae…………….…..……..68

4. The results of the path analysis model that includes all relationships from Figure 3 and also incorporates plant diversity (inverse Simpson’s) effects on predator and competitor diversity.………….……70

5. The structural equation model showing similar results as Fig. 4 and including the latent variable Structure measured by plant diversity (Shannon Entropy) and total plant biomass……………………………...71

Chapter 5.

1. Most sites sampled were in western Nevada and neighboring counties in California. Several sites were in central Nevada, and extreme northwestern Nevada and southwestern Arizona…………..…...86

2. Plant diversity estimates were conducted within 1-meter square plots (bottom right). The blue squares represent size classes used to count individual plant stem densities, higher-density plants with narrower stems were counted in smaller square areas than larger x

less-densely clustered stems. Area counts were then used to estimate total plant stems within each plot based on density and estimated percent of plot cover…………………………………..………87

3. The mean (+/- 1 SE) total plant and invertebrate diversity (inverse Simpson’s) are shown for each distinct habitat type……………….……90

4. Both predator and competitor diversities (inverse Simpson’s) on larval mosquito abundance. Arrows represent positive effects and bullets represent negative effects……………………….………………..92

5. The direct effects of competitor diversity (inverse Simpson’s) on predator diversity and predator diversity on mosquito abundance. The dotted line shows the indirect effect of competitor diversity on mosquito larvae. The direct effect of predator diversity on competitor diversity and competitor diversity on mosquito larvae with an indirect effect of predator diversity on mosquito abundance.…...94

6. Similar to figure 5, but with direct effects of plant diversity (inverse Simpson’s) on predator diversity and competitor diversity. Plant diversity shows positive effects on both predator and competitor diversities. Competitor diversity shows a negative effect on mosquito abundance and predator diversity shows a relatively weak direct positive effect on mosquito abundance……………………..95

7. The final structural equation model showing positive effects of plant diversity on both predator and competitor diversities. Competitor diversity shows a negative effect on mosquito larvae. Site identity shows a positive correlation with mosquito larvae………………………96 1

Introduction

Mosquitoes belong to the family of flies (Diptera) called the Culicidae, which comprises an estimated total of 3,601 described species (Wilkerson et al. 2015). We know mosquitoes collectively as a public health and economic nuisance in large enough quantities, and for their role in vectoring important human and animal diseases, such as malaria, yellow fever, chikungunya virus, dengue viruses, Zika virus, and West Nile virus. Out of the 3,601 described species only a relative handful of species are considered competent vectors of disease. The remaining species are less competent disease vectors, relatively uncommon, or mostly under-researched in the field of medical entomology. As a group however, most species include an adult female stage which is ectoparasitic and therefore at minimum constitutes and public health nuisance worthy of interest.

Most culicids occupy a herbivorous trophic guild as larvae in the aquatic stages, feeding on a variety of algae, phytoplankton, periphyton, and detritus (Merrit et al. 1992).

Although relatively rare among culicids, predacious groups do exist, namely in the genera

Toxorhynchites, Psorophora, and Anopheles. The Toxorhynchites species especially have inspired significant research as biological control agents of competent vector species of mosquito larvae (Schreiber 2007). My research focuses entirely on larval mosquitoes which occupy an herbivorous trophic group, excluding facultative or obligate predator species, except where they are of interest as natural enemy species in biological control experiments.

Mosquitoes are holometabolous and undergo complete metamorphosis including the egg, four aquatic larval instars, an aquatic pupal stage, and adults. Developing larvae 2 are the stage most dependent on the availability of resources and the most susceptible to natural enemies (Hinman 1934). Research on natural enemies has largely been in the context of classical biological controls using aquatic predators (Hinman 1934, Jenkins

1964, Bay 1974, Chapman 1974, Legner et al. 1974, Collins and Washino 1985, Holck

1988, Mogi 2007, Quiroz-Martinez and Rodrigues-Castro 2007). These reviews focus mainly on specific predator groups and their effectiveness at reducing larval mosquito abundance, but include few examples of direct resource competitors, parasites, or pathogens of mosquito larvae. It is also worth noting that these reviews are not quantitative, rather they rely on subjective interpretations of overall results for synthesis of existing studies.

From a broader ecological perspective, predator and competitor diversity and abundance have been show to increase with habitat stability and influence the abundance of mosquitoes in artificial and natural systems. For example, ponds that experience increased hydroperiod (days post disturbance) following experimental manipulations of competitor and predator species showed increases in diversity and abundance of predators with time following disturbance (Schneider and Frost 1996). Chase and Knight

(2003) surveyed wetlands characterized as temporary, semi-permanent and permanent; and also used experimental mesocosms to simulate these categories. Results were similar for wetlands and mesocosms: temporary wetlands maintain populations of competitor species when re-inundated because species were able to survive desiccation, resulting in fewer mosquitoes produced. Permanent wetlands had higher levels of predation and therefore also lower levels of mosquito production. Finally, semi-permanent wetlands maintained fewer levels of competitor and predator species and resulted in the highest 3 numbers of mosquitoes (Chase and Knight 2003). Similar results were reported by

Carlson et al. (2009) for brick-making pits in the highlands of western Kenya. Older and less recently disturbed pits maintained larger populations and diversities of predators and competitor Culicine species, resulting in both direct and indirect negative effects on

Anopheles gambiae (a competent malaria vector) populations (Carlson et al. 2009).

Antagonistic interactions with mosquito larvae include more than direct negative effects, such as direct predator-induced mortality or direct competition for resources.

Instead, a multitude of community-level direct and indirect antagonistic and beneficial interactions likely work together to negatively and even positively affect mosquito larvae.

Some examples include interference competition, exploitative competition, apparent competition, apparent mutualism, indirect mutualism, intraguild predation, and keystone predation; with each unique interaction type being context-dependent on the community composition of species (Holt 1977, Blaustein and Chase 2007, Juliano 2009). Thus, a complete understanding of the effects of species diversity (antagonistic and beneficial) on mosquito larvae requires knowledge of the interaction diversity among species in the system. Interaction diversity defined as the complete diversity of direct and indirect, single and multi-trophic interactions (links) between species (Janzen 1974, Thompson

1996, Dyer et al. 2010).

Since most mosquito species are herbivorous and feed on algae, phytoplankton, periphyton, and detritus, they are commonly associated with emergent aquatic plants, also referred to throughout this dissertation as aquatic macrophytes. There is a longstanding hypothesis that aquatic vegetation provides less primary productivity input to the aquatic food chain than phytoplankton, algae, and periphyton, resulting in a greater proportion of 4 biomass consolidation up to herbivore and upper trophic levels compared to terrestrial systems (Shurin et al. 2006). Instead aquatic macrophytes provide important structural elements and physical heterogeneity for periphyton growth and invertebrate occupancy

(Hutchinson 1957, Wetzel 1983, Gregg and Rose 1985, Brown et al. 1988).

For my dissertation research, I focused on aquatic community-level species interactions of mosquito larvae and their natural enemies including predators, competitors, parasites, and pathogens. I did not focus on a single mosquito taxon and instead examined a diversity of mosquito habitat types and their resident mosquitoes. I also examined effects of habitat structural complexity and plant diversity on aquatic invertebrate communities and mosquitoes, as well as indirect effects on larval mosquito abundance. My overall research approach utilized a meta-analysis, simulated communities, mesocosm experiments, and unmanipulated field observational data collected in natural and artificial habitats.

The meta-analysis I utilized focused on the natural enemy groups that have been used to control larval mosquitoes. Published studies on predators, competitors, parasites, and pathogens of mosquitoes were collected, data were gleaned from those papers, and study effect sizes were analyzed in order to determine the effectiveness of specific groups at controlling mosquito larvae. Effect sizes were compared between taxonomic groups, field versus laboratory experiments, and between artificial and natural habitats. This meta-analysis provides a greater ecological context for traditional biological control and quantifies the effectiveness of using natural enemy species for controlling mosquito larvae. 5

The simulation of communities was focused on tri-trophic networks, estimating interactive effects of variation in three trophic levels (e.g., producers, primary consumers, and secondary consumers), species specialization (e.g., consumer diet breadth), and relative abundances of species. The simulation allowed for insight into the effects of variation in these parameters on interaction and species diversity among all trophic groups. This approach to studying interactions that are likely to be important for mosquitoes yielded a published manuscript: Pardikes, N. A., W. Lumpkin, P. J. Hurtado,

L. A. Dyer 2018. Simulated tri-trophic networks reveal complex relationships between species diversity and interaction diversity. PLoS ONE 13(3):1-20 (there was equal effort for the first authors on this manuscript). This simulation adds depth to understanding the potential relationships between species diversity, diet breadth, abundance, and how these interact to influence the interaction diversity of biotic communities and has theoretical relevance to both terrestrial and aquatic systems.

The experimental approach to examining community interactions and larval mosquito ecology employed a field-conducted mesocosm experiment where I directly manipulated plant diversity and complexity levels to determine the effects of these independent variables on total invertebrate diversity (predators and competitors) and their effects on larval mosquito abundance. A paper from this work has been published: Will

P Lumpkin, Kincade R Stirek, Lee A Dyer, Macrophyte Diversity and Complexity

Reduce Larval Mosquito Abundance, Journal of Medical Entomology, https://doi.org/10.1093/jme/tjaa012. This study demonstrated clear effects of plant complexity and diversity on aquatic community interactions involving mosquito larvae. It was important because it corroborated other studies demonstrating how complex species 6 interactions, including strong direct and indirect effects of competitors and predators, can affect mosquito larval abundance.

The large-scale observational approach in my dissertation research expanded on the hypothesis tests from the mesocosm experiment and examined how mosquitoes are affected by plant diversity and habitat complexity in natural unmanipulated field habitats.

Invertebrates were sampled using an aquatic sweep net and measurements were taken for water depth, plant density, and species. This work was designed to clarify how plant complexity and diversity can scale up from mesocosm-level effects to much larger, more diverse, and more complex habitats. Overall, results from that work will be important for efforts to understand how mosquito larvae in different habitats face very different ecological pressures. 7

A Meta-analysis on the Effectiveness of Biological Mosquito Control

Abstract

Biological control of mosquitoes has been an important topic of interest since the early 20th century. Natural enemy species have primarily been tested for their efficacy in reducing larval mosquito abundance and ultimately reducing adult mosquito populations.

These trials predate the advent of modern pesticides and especially the use of highly mosquito-specific biological and other larvicide products. Still, a strong interest in biological control persists as a component of modern mosquito abatement programs as shown in the widespread use of Gambusia spp. fish for larval control. Unfortunately, the complexity of natural systems and the potential for unforeseen indirect effects of introducing natural enemy species due to the potential interaction diversity can result in smaller than expected control effects of a given natural enemy of larval mosquitoes.

Here we use meta-analysis to quantify the effect size of using natural enemy species against mosquitoes. We test the effect sizes by taxon, habitat type, habitat location, and by trophic group. The results show that all natural enemy types (e.g., competitors, predators, parasites, fungal pathogens, viral pathogens) produced negative effects on mosquito larval performance, behavior, and final adult development. Effect sizes were stronger in field-conducted experiments than in laboratory experiments. Finally, fully natural and artificial experimental container types showed stronger negative effects than mesocosms. These results suggest that continued interest in using natural enemy species for biological mosquito control should remain an important component of mosquito abatement programs and in the ecological research of larval mosquito development.

Mosquito abatement programs should encourage natural enemy populations where 8 possible and avoid broad-spectrum insecticides, which can negatively impact existing populations of natural enemy species of mosquito larvae.

Introduction

Classical and neoclassical biological control of insect pests has historically encouraged the use of exotic natural enemies to reduce pest population levels. The most common natural enemies of insects utilized for biological control include predators, parasites and pathogens that lower densities or regulate the pest species, making them less ecologically, economically, or medically disruptive. Biological control of mosquito populations has enjoyed a long history of investigations into the role of natural enemies and their potential as forces that can control mosquito populations, especially when mosquitoes contribute to medical and economic hardship. This long-standing interest is evident in many extensive literature reviews on the topic (e.g., Hinman 1934, Jenkins

1964, Bay 1974, Chapman 1974, Legner et al. 1974, Collins and Washino 1985, Holck

1988, Mogi 2007, Quiroz-Martínez and Rodrígues-Castro 2007). Empirical biological control studies and reviews both provide information on individual natural enemy species, often with an emphasis on mass rearing and release for the control of larval mosquito populations. Few studies have specifically addressed predation on mosquito eggs and pupae largely because of the paradigm that larvae are most vulnerable because the larval stage accounts for a major proportion of their aquatic ontogeny (Hinman 1934).

Additionally, targeting adult populations would be ineffective given the likelihood of biting adults already present at that mosquito life stage in each ecosystem and active dispersal of adults to new colonization sites. 9

Invertebrate groups of interest as potential biological mosquito control agents include: Hemipterans, beetles, several families of Odonata, Diptera, copepods, parasitic nematodes, and predatory planarian flatworms (Table 1). Vertebrate predators for biological control of mosquito larvae have almost exclusively consisted of fish in the genus Gambusia, which are used extensively today. The most successful and important contribution to the biological control of larval mosquitoes was the discovery of the entomopathogenic bacteria Bacillus thuringiensis israelensis (Bti) in 1978 (De Barjac

1978) and Bacillus sphaericus (Kellen et al. 1965). Both bacteria are very well suited for biological control, being easily mass-produced, target-specific to mosquito larvae, and are now widely available to mosquito control agencies from pesticide developers.

From the brief list and literature reviews of the potential natural enemies (Table

1), a clear bias can be seen for a classical, ‘mass-production’ and release type approach to biological mosquito control. Significant drawbacks to this approach include many logistical problems in mass rearing, storing, and timing of delivery of live natural enemies (Mogi 2007). Additionally, although many promising taxa have been studied, several relevant ecological considerations should be made in order to successfully utilize natural enemies for successful biological control and to encourage the establishment and success of naturally occurring antagonistic species. Important considerations include predator preference for specific prey, existing species diversity of the mosquito habitat, stability of the aquatic system, density of mosquito larvae, predator and prey positions in the water column, appropriate number of predators, reproductive potential of the prey species, predator-prey coevolution, synchronization, and spatial refuge from predation

(Quiroz-Martínez and Rodrígues-Castro 2007). Furthermore, mathematical models 10

Table 1. Important taxa used for classical and neoclassical biological control of mosquitoes. Taxon Examples Example references Bacteria Bacillus thuringiensis israelensis, De Barjac, H. (1978) Bacillus sphaericus Kellen et al. (1965)

Toothcarps Poeciliidae (Gambusia) Chandra et al. (2008)

Hemiptera Notonectidae Shaalam and Canyon (2009) Belostomatidae Quiroz-Martínez and Nepidae Rodrígues-Castro (2007 Naucoridae Chesson (1984) Veliidae Hydrometridae Odonata Libellulidae Fincke et al. (1997)

Diptera Chaoboridae Focks (2007) Chironomidae Culicidae (Toxorhynchites) Coleoptera Dytiscidae Hydrophilidae Gyrinidae Copopoda Cyclopidae Marten and Reid (2007)

Nematoda Mermithidae Platzer (2007), Petersen and Chapman (1970) Platyhelminthes Planariidae Medved and Legner (1974)

that either explore potential relationships among these variables and control (e.g., Lord

2007) or that are geared towards prediction (e.g., Smith et al. 2008) are typically not part of planning for mosquito biological control. In general, a basic ecological understanding of the system in which natural enemies are to be utilized and or encouraged is essential for successful control. These considerations are important and often context-dependent and all of them require further research. 11

The purpose of this meta-analysis was to build on earlier reviews of biological mosquito control agents and to add a quantitative summary that includes a broader synthesis of the effects of antagonistic species, including predator species as well as competitors, parasites, and pathogens. Meta-analysis was used to quantify these effects across a range of application habitats. Specifically, I asked how do the effects of biological control agents and other antagonistic species vary across different habitats

(e.g., phytotelmata, artificial containers, ponds, and wetlands) and across habitats of different hydroperiods (e.g., temporary, semi-permanent and permanent)? Our analysis included both laboratory and field applications and studies. My hypotheses were: 1. effect sizes for individual antagonistic species and groups vary across these habitat types;

2. Effect sizes would be smaller in larger, more species-diverse habitats, vs. smaller, less diverse habitats, due to greater variation in community-level interactions. The quantitative synthesis of antagonistic species’ effects on mosquito larvae across the different ecological contexts included in this analysis provides a greater foundation for understanding the community ecology related to mosquito control. In particular, this meta-analysis highlights ecological interactions involving mosquito larvae with an eye towards theory-driven approaches to biological mosquito control.

Methods

Literature Search

A meta-analysis was conducted on the literature of biological control of mosquitoes and antagonistic species effects on mosquito larvae. An attempt was made to include studies on effects of predators, competitors, parasites and pathogens in both field and laboratory trials of biological control agents and studies which reported direct and 12 indirect effects of antagonistic species on mosquito larvae. The online citation database

Web of Science was searched using the following Boolean search criteria: mosquito OR

Culicidae AND biological control, mosquito OR Culicidae AND integrated pest management, mosquito OR Culicidae AND competitor, mosquito OR Culicidae AND parasite, mosquito OR Culicidae AND predator. Additional studies were identified from reviews on biological control and natural enemies of mosquito larvae discussed earlier.

Data Collection

Meta-analysis requires that studies report measures of central tendency, variance, and sample sizes for each treatment group. Where presented, these statistics were extracted from the text, tables, or supplemental material. For studies which did not present the required statistical information in the text but revealed it in figures, ImageJ software was used to extract the data. Studies were coded by location (laboratory or field), habitat type (artificial container, phytotelmata, pond, wetland, pasture, tire, petri dish), antagonist taxa (family, genus, species), antagonist manipulation (presence, absence, gradient, removal), antagonist functional group (predator, competitor, parasite, pathogen), mosquito (target) taxa (genus, species), and target species response (e.g., survival, mortality, abundance, development time, behavior, adult size, adult fecundity).

Effect Size Estimates

A popular effect size estimate used in ecology is the log response ratio which is given by the equation:

RR=ln ( X T / X C )

Where XT and XC are the treatment and control means, respectively; and is equivalent to:

RR=ln XT − ln X C 13

The log response ratio provides a proportional effect size for data synthesis and is a useful index for comparing ecological studies of different treatment effects (Hedges et al.

1999). An important limitation of the log response ratio is that it can be biased when comparing effects given from small sample sizes (Lajeunesse 2015). Here we use a modified version of the log response ratio by Lajeunesse (2015) given by the equation:

2 2 Δ 1 ( SDT ) ( SDC ) RR =RR+ − 2 2 2 [ N T X T N C X C ]

Where RR is the uncorrected log response ratio and SDT, SDC, N T , and N C are the standard deviation of the treatment mean, standard deviation of the control mean, treatment sample size, and control sample size, respectively. Because different response measurement types yield different expectations of response sign (i.e., positive versus negative) between treatment and control groups, the resulting effect sizes were switched to either positive or negative to match their expectation. For example, if the effect of the addition of a predator was measured as the resulting abundance of a target prey, the expected response sign would be negative and the negative value was kept, whereas the expected response sign for mortality would be positive and the resulting sign would be switched to negative. All response signs for the addition of an antagonist (predator or competitor) were therefore switched to negative where required for better interpret- ability.

Data Analysis

Data analysis was conducted using Bayesian regression models with JAGS version 4.3.0 (Plummer 2003) using the R package runjags (Denwood 2016).

Traditionally in meta-analysis the combined variance for a particular effect size is 14 calculated using specific calculations for the effect size being used (see, Hedges et al.

1999, Lajeunesse 2015). Here we instead rely on the Bayesian 95 % credibility intervals for the posterior means to provide that measure of variance. A common issue in conducting meta-analysis is accounting for within study sampling bias (dependence).

Common solutions for correcting within study sampling dependence are to ignore dependence entirely, averaging dependency within studies, averaging dependency among specific sampling units, or to simply select a single effect per study (Cheung 2014). Here again, we use Bayesian regression models with study as a nested categorical variable in order to estimate within study sampling dependence.

Results

Summary of Papers and Treatment Effects

The literature search resulted in an initial list of a total of 255 published articles, with a total of 50 studies applicable to the current meta-analysis. The remaining studies included 22 review articles, 21 studies without control data, 5 studies without replication data, 11 studies without variance data, and 86 studies which require further review to determine their suitability for this analysis. An additional 84 studies were collected during the initial search, but on more extensive review they did not fit the categories of interest in this meta-analysis. A total of 474 total effect sizes were collected from the list of 50 accepted studies. These included 43 competitor, 13 fungal pathogen, 2 parasite,

410 predator, and 6 viral pathogen effects (table 1). Effects were further categorized by container type (natural, artificial, or mesocosm), habitat/study location (field or laboratory), and by target response type (larval effects, larval behavior effects, and effects on emerged adult mosquitoes, table 2). 15

Table 2. Below are the number of effect sizes collected by antagonist trophic category, experimental location (field, laboratory), container type (artificial, mesocosm, natural), and by effect measurement target (larval, larval behavior, adult).

Effect Category Effect Count Antagonist Trophic Group Competitor 43 Fungal Pathogen 13 Parasite 2 Predator 410 Viral Pathogen 6

Experimental Locations Field 151 Laboratory 323

Experimental Container Types Artificial 116 Mesocosm 303 Natural 55

Measured Target Effect Group Larvae 340 Larval Behavior 68 Adults 47

Bayesian Regression Models

The results from the Bayesian regression models showed negative effects for all coded groups (Table 3). For antagonist/enemy type (Table 2, Figure 1) with the full 16 effects data (adult, behavioral, and larval effects) the posterior mean beta (β) and credibility intervals (CI) were for competitors (β = -0.395, CI = -0.843 to 0.079), fungal pathogens (β = -0.46, CI = -1.058 to 0.204 ), parasites (β = -0.543, CI = -1.531 to 0.393), predators (β = -0.556, CI = -0.855 to -0.26), viral pathogens (β = -0.385, CI = -1.07 to

0.46), and study (β = -0.013, CI = -0.025 to -0.002). For experimental habitat type (Table

2, Figure 2) they were artificial (β = -2.083, CI = -2.454 to -1.717), mesocosm (β = -

0.434, CI = -0.69 to -0.173), natural (β = -2.422, CI = -2.905 to -1.932), and study (β =

0.01, CI = 0 to 0.021). For experimental location (table 2, figure 3) they were field (β = -

1.143, CI = -1.54 to -0.766), laboratory (β = -0.413, CI = -0.712 to -0.11), and study (β =

-0.009, CI = -0.021 to 0.003). For the effects on larval development only antagonist type

(Table 2, Figure 1) effects were for competitors (β = -0.574, CI = -1.104 to -0.011), fungal pathogens (β = -0.631, CI = -1.367 to 0.186), parasites (β = -0.674, CI = -1.699 to

0.295), predators (β = -0.684, CI = -1.043 to -0.323), viral pathogens (β = -0.592, CI = -

1.535 to 0.474), and study (β = -0.017, CI = -0.032 to -0.002). For experimental container type (Table 2, Figure 2) they were artificial (β = -2.356, CI = -2.753 to -1.95), mesocosm (β = -0.261, CI = -0.575 to 0.038), natural (β = -1.781, CI = -2.305 to -1.227), and study (β = -0.004, CI = -0.016 to 0.009). For experimental location (Table 2, Figure

3) they were field (β = -1.259, CI = -1.729 to -0.787), laboratory (β = -0.583, CI = -0.935 to -0.233), and study (β = -0.01, CI = -0.025 to 0.004).

Discussion

Given the long history of interest in the biological control of mosquitoes (e.g.,

Hinman 1934, Jenkins 1964, Bay 1974, Chapman 1974, Legner et al. 1974, Collins and

Washino 1985, Holck 1988, Mogi 2007, Quiroz-Martínez and Rodrígues-Castro 2007), it 17 is important to quantify the effects of the natural enemies being used and their effectiveness in diverse habitat types. The results here included many manipulative approaches to controlling mosquitoes using natural enemies and should be relevant to any management strategies that involve the addition or exclusion of natural enemies of mosquito larvae, artificial versus natural containers, and laboratory versus field applications. Control efforts that have historically received less attention than predators were effective, including the application of competitors, pathogens, and parasites. These results also supported our first hypothesis that effects would vary across habitat types, but the very high effect sizes in field, artificial, and natural aquatic communities had similar magnitudes. Our second hypothesis that effects would be greater in more controlled laboratory experiments or experiments using artificial containers (e.g., petri dishes, culture plates, etc.) was not completely supported in our results. Instead, both artificial and natural container experiment produced stronger (i.e., five to six times higher) negative results than mesocosms. For all models, the posterior beta coefficients for individual studies and credibility intervals were low enough to assume small study dependence between effect sizes.

At first glance, it is surprising that both laboratory studies and artificial container studies had substantially smaller effect sizes than studies in the field and in artificial and natural aquatic communities. However, given the complexities of community interactions, and the importance of heterogeneity (see other sections of this dissertation), these results are consistent with theory.

For example, higher levels of interaction diversity may be responsible for preventing mosquitoes from dominating any local habitats (see Dyer 2018), and these 18 levels of diversity at large scales are maintained by substantive turnover (beta diversity of interactions) from one site to another (Dell et al. 2019, Lepesquer et al. 2018). If this is true, and interaction diversity is a component of successful biological control, these results also suggest that it is not enough to utilize a single biological control agent applied across a large management area. Multiple agents can contribute to greater complexity of interactions and effective control across a heterogenous landscape. The best approach to increasing such mosquito-antagonist interaction diversity would be a conservation biological control approach (sensu Barbosa 1998), where greater resource diversity

(primary producer richness and habitat complexity) is enhanced, since this can maintain natural interactions as well as classically introduced agents.

Negative effects were observed in all trophic groups including competitors, predators, fungal pathogens, viral pathogens, and parasites (Figure 1). Likewise, all families represented in the collected studies resulted in negative effects (Figure 4). These observations are not unusual in that one significant challenge in conducting meta-analysis is accounting for publication bias and the lack of data presented for insignificant results.

This study relied exclusively on existing published literature through the scientific search engine Web of Science and therefore did not address the problem of significance reporting in publication bias and ignored the effects of unpublished data.

Regardless of the potential for publication bias, several antagonist groups deserve further discussion, including the Culicidae, Cyclopidae, Daphniidae, mixed-plankton, and mixed-plankton-predator groups. The majority of Culicidae that had strong antagonist effects included mosquitoes in the genus Toxorhynchites, a large container inhabiting mosquito genus of obligately predacious as larvae. Toxorhynchites spp. mosquitoes have 19 been of interest in controlling container inhabiting mosquitoes; however, challenges in their use have included difficulty in deployment methods to aquatic

Table 3. The mean posterior probability coefficients, standard deviations and 95 % credibility intervals for the effect of total mosquito measurements (A. adult effects, behavioral effects, and larval effects) and larval only effects (B).

A. Total Effects (adult, behavior, larval) B. Larval Development Effects

Predictor Beta SD CI Beta SD CI Competitors -0.395 0.236 -0.843 to 0.079 -0.574 0.276 -1.104 to -0.011 Fungi -0.46 0.31 -1.058 to 0.204 -0.631 0.376 -1.367 to 0.186 Parasites -0.543 0.452 -1.531 to 0.393 -0.674 0.473 -1.699 to 0.295 Predators -0.556 0.153 -0.855 to -0.26 -0.684 0.185 -1.043 to -0.323 Viruses -0.385 0.375 -1.07 to 0.46 -0.592 0.478 -1.535 to 0.474 Study -0.013 0.006 -0.025 to -0.002 -0.017 0.007 -0.032 to -0.002

Predictor Beta SD CI Beta SD CI Artificial -2.083 0.188 -2.454 to -1.717 -2.356 0.205 -2.753 to -1.95 Mesocosm -0.434 0.132 -0.69 to -0.173 -0.261 0.156 -0.575 to 0.038 Natural -2.422 0.249 -2.905 to -1.932 -1.781 0.274 -2.305 to -1.227 Study 0.01 0.005 0 to 0.021 -0.004 0.006 -0.016 to 0.009

Predictor Beta SD CI Beta SD CI Field -1.143 0.198 -1.54 to -0.766 -1.259 0.241 -1.729 to -0.787 Laboratory -0.413 0.154 -0.712 to -0.11 -0.583 0.179 -0.935 to -0.233 Study -0.009 0.006 -0.021 to 0.003 -0.01 0.007 -0.025 to 0.004 sources, cannibalism during rearing, poor timing of population overlap with target prey species, requirement of a continuous release program, and prey avoidance behaviors

(Trips 1970, Magi 1992, Nannini and Juliano 1997, Mercer et al. 2005, Focks 2007,

Nyamah et al. 2011, Schiller et al. 2019). The results presented here show a negative but small effect of Culicid antagonists relative to other groups. Further efforts to model the interactions between this competitor and pest mosquitoes, combined with large-scale field studies could potentially increase its effectiveness as a biological control agent. 20

Figure 1. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects of trophic groups on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B).

Cyclopoid copepods in the family Cyclopidae are another group that has shared strong interest in biological control of mosquitoes. Our results show the family

Cyclopidae with the largest negative effect on mosquito development over all other groups. A comprehensive review by Marten and Reid (2007) showed that copepods have many biological traits making them especially useful for classical biological control of mosquitoes. These include a strong persistence in water sources due to their entirely aquatic life cycle, strong predation on early instar mosquito larvae, and relatively easy laboratory rearing requirements (Marten and Reid 2007). Additionally, copepods have been demonstrated to perform well at controlling mosquitoes in a variety of field trials under natural conditions and are actively implemented successfully in biological control programs (Brown et al. 1996, Dieng et al. 2002, Marten and Reid 2007).

Finally, our results show that the treatment effects for Daphniidae, mixed plankton, and mixed plankton with predators have more negative relative effects than 21 other groups. Both mixed plankton treatments included Daphnia. These results suggest that there may be important negative effects of these competitors and potentially synergistic or community-level interactions with mosquito larvae and their enemies.

Negative interactions of mosquito larvae with their natural enemies in complex systems is thought to include both direct and indirect interactions, sometimes with enemies even indirectly benefiting a target species depending on ecological context and interaction with other natural enemies (Holt 1977, Holt and Lawton 1994, Juliano 2007, Juliano 2009,

Juliano et al. 2010, Bonsall and Holt 2010).

Figure 2. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects between experimental container types on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B). 22

Figure 3. Bayesian posterior probability estimates and 95 % credibility intervals for enemy effects between field versus laboratory-conducted experiments on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B).

Figure 4. Bayesian posterior probability estimates and 95 % credibility intervals for natural enemy effects between family or taxonomic grouping, on mosquitoes are shown here for all effects (A, adults, larval behavior, larval development) and for only larval development (B).

23

We conclude that the best approaches to mosquito management via biological control will be based on a diversity of interactions with competitors, predators, and parasites. These conclusions are consistent with pest management in agricultural systems where areas composed of higher levels of invertebrate diversity experience lower overall pest dominance (Bianchi et al. 2006). The complexity of interactions resulting from increased plant diversity across systems remains context dependent and the results for pest control are equally complex (Andow 1991). Still, increased total invertebrate diversity has been shown to reduce mosquito abundance (Chase and Knight 2003,

Carlson et al. 2009) and plant diversity and complexity have been shown to reduce larval mosquito abundance through invertebrate community level interactions (Lumpkin et al.

2020). Incorporation of natural enemy species for the control of mosquito populations should utilize native species and monitor for indirect effects in communities depending on the size and scale. 24

Simulated tri-trophic networks reveal complex relationships between species

diversity and interaction diversity

Nicholas A. Pardikes (1, 2)

Will Lumpkin (1)

Paul J. Hurtado (1, 3)

Lee A. Dyer (1)

1 Department of Biology, Program in Ecology, Evolution, and Conservation Biology,

University of Nevada,

Reno, Nevada, United States of America,

2 Czech Academy of Sciences, Institute of Entomology, Ceske

Budejovice, Czech Republic,

3 Department of Mathematics and Statistics, University of Nevada, Reno,

Nevada, United States of America 25

Abstract

Most of earth's biodiversity is comprised of interactions among species, yet it is unclear what causes variation in interaction diversity across space and time. We define interaction diversity as the richness and relative abundance of interactions linking species together at scales from localized, measurable webs to entire ecosystems. Large-scale patterns suggest that two basic components of interaction diversity differ substantially and predictably between different ecosystems: overall taxonomic diversity and host specificity of consumers. Understanding how these factors influence interaction diversity and quantifying the causes and effects of variation in interaction diversity are important goals for community ecology. While previous studies have examined the effects of sampling bias and consumer specialization on determining patterns of ecological networks, these studies were restricted to two trophic levels and did not incorporate realistic variation in species diversity and consumer diet breadth. Here, we developed a food web model to generate tri-trophic ecological networks and evaluated specific hypotheses about how the diversity of trophic interactions and species diversity are related under different scenarios of species richness, taxonomic abundance, and consumer diet breadth. We investigated the accumulation of species and interactions and found that interactions accumulate more quickly; thus, the accumulation of novel interactions may require less sampling effort than sampling species in order to get reliable estimates of either type of diversity. Mean consumer diet breadth influenced the correlation between species and interaction diversity significantly more than variation in both species richness and taxonomic abundance. However, this effect of diet breadth on interaction diversity is conditional on the number of observed interactions included in the models. The results 26 presented here will help develop realistic predictions of the relationships between consumer diet breadth, interaction diversity, and species diversity within multi-trophic communities, which is critical for the conservation of biodiversity in this period of accelerated global change.

Introduction

The devaluation of natural history and taxonomy has added to the failure of ecologists to document biodiversity and subsequently to understand the magnitude and consequences of the growing extinctions caused by global change (Tewksbury et al.

2014). Knowledge of basic natural history is especially important for quantifying biotic interaction diversity, which encompasses most of earth's diversity (Ohgushi et al. 2007) and should be tightly linked to variables such as community stability and ecosystem services (Dyer et al. 2010, Mougi and Kondoh 2012). The loss of interaction diversity is one of the least understood responses to species extinctions, partly because it has not been consistently treated as a response variable in theoretical or empirical studies of biodiversity and because getting good quantitative data on interaction diversity often requires considerable fieldwork over time. Although network approaches have provided more focus on the structure of species interactions within communities, few analyses are based on detailed natural history data that is linked with experimental evidence of observed interactions actually occurring together (e.g., Novotny et al. 2002, Janzen et al.

2005, Ballantyne et al. 2015). In contrast, using a standardized sampling approach allows for a more rigorous and repeatable resolution of interaction networks at any appropriate scale (Dyer et al. 2010), but it is not clear how much sampling is necessary for accurate 27 measurements nor how relevant small local interaction networks are to larger scale network properties (Poisot et al. 2012, Fründ et al. 2016).

We define interaction diversity as a measure that combines the relative abundance and richness of interactions linking species together into dynamic biotic communities at multiple scales (Dyer et al. 2010, Janzen 1974, Thompson 1996, Thompson 1997, Dáttilo and Dyer 2014). For this metric of diversity, the calculation of richness, diversity indices, and rarefaction diversity is based on experimentally established links between interacting individuals rather than species alone, or alternatively, lists of observations of species found in the same area to determine network nodes and edges. Trophic interactions, such as enemy- herbivore plant interactions, have large effects on all ecosystem attributes and are well studied (Dyer et al. 2010, Gross et al. 2009, Jiang et al 2009), thus tri-trophic webs are suitable systems for examining networks and interaction diversity. Here we focus on this interaction diversity across multiple trophic levels.

Since most communities can never be completely sampled, and the true community values of diversity and other network parameters are impossible to precisely quantify at community scales larger than a hectare, careful sampling approaches are necessary for characterizing interaction diversity (Novotny et al. 2010). Here we simulate a standardized sampling effort that accumulates individual interactions until each interaction has been accounted for. Utilizing this sampling approach mimics existing systematic sampling protocols in the field, such as standardized plots (e.g., Forister et al.

2015), and allows the comparison of interaction diversity across a broad range of community types. Furthermore, our approach permits us to identify differences between the actual community and a subsample of the community, as certain community 28 characteristics may be more sensitive to disparate sampling efforts than others (Fründ et al. 2016, Dormann et al. 2009, Thébault and Fontaine 2010).

Recently, Fründ et al. (2016) investigated the effects of sampling bias on quantifying specialization in bipartite networks and found significant effects of sampling bias on selected properties, while identifying network parameters that are robust to limited sampling. However, this investigation was restricted to two-trophic levels and the range of taxonomic richness and degree of specialization of their simulated communities was narrow. To add to this existing work, we simulated 1000 tri-trophic communities with representative combinations of species richness, taxonomic abundance, and consumer diet-breadth, allowing for a comprehensive investigation into the determinants of interaction diversity across a wide-range of multitrophic communities (Dyer et al.

2010, Dáttilo and Dyer 2014, Darwin 1859, Wallace 1878, Novotny et al. 2006, Novotny et al. 2007, Dyer et al. 2007, Dyer et al. 2012).

The focus of this study was to test specific hypotheses about the relationships between community species diversity, consumer diet breadth, interaction diversity, and network structure. We addressed the following questions with simulation and statistical models:

1. Does interaction diversity asymptote more quickly than species diversity from a

discrete sample size or area?

2. What are the interactive effects of consumer diet breadth and species diversity on

interaction diversity?

3. Are the combined effects of richness, abundance, and diet breadth on

interaction diversity modified by the number of interactions that are observed? 29

We sampled from simulated networks of interacting trophic levels; mimicking field sampling methods outlined in Dyer et al. (2010) and tested relevant paths from a specific structural equation meta model (SEMM, sensu Jiménez-Alfaro et al. 2016) with hypothesized causal relationships between diet breadth and interaction diversity.

Methods

Food web simulation

The goal of this model was to generate a random plant-herbivore-parasitoid tri- trophic food web, with interactions only between adjacent trophic levels. Each community is generated to represent the scale of a single study site and are based on several pre-specified properties as inputs to investigate possible contributions to interaction diversity. Specifically, these inputs are the number of species at each trophic level (i.e., richness; R1, R2, R3), the overall abundance of each trophic level (i.e., abundance; A1, A2, A3), and a diet breadth parameter (α2, α3) for the consumers that determines the diet breadth distribution for that trophic level according to a truncated discrete Pareto distribution (Forister et al. 2015).

The abundance distribution for trophic level i was constructed by taking a random sample of size Ri from a lognormal distribution with μ = 0 and σ = 1, scaled to sum to the prespecified overall abundance Ai, and then rounded to the nearest integer (Magurran

2013). We denoted the abundance of species j in trophic level i as Aij, where Ai _ Pri j.1

Aij. Individual diet breadth values (number of species each consumer has in their diet) were assigned to each species to get an empirical distribution that represents the desired discrete truncated Pareto distribution of specialization within the consumer trophic levels. 30

These values were obtained by calculating density values for a (continuous) Pareto I distribution (truncated at the number of species at the lower trophic level) with survival function (aka complementary CDF) S(y) = (1/y)α.

The lists of resource species that each species potentially consumes were then sampled (with replacement) uniformly from the list of species in the lower, adjacent trophic level. In sampling real systems in the field, individual consumers are assumed to have been found by sampling their resource (i. e., herbivores are detected by inspecting host plants, and parasitoids are found by inspecting host herbivores). Therefore, we assumed each individual parasitoid/enemy is associated with an individual herbivore, and each individual herbivore with an individual plant. In other words, there is never more than one individual consumer on an individual host, though there are several individuals within a species, so you can have multiple interactions occurring between those two species. Interactions among individuals were therefore constructed as follows. Individual herbivores of species j (recall there are A2j such individuals) were assigned a plant species by cycling through the list of species in their diet. Then each individual plant is assigned an individual herbivore, based on these assignments, and we assume only one herbivore individual per plant individual. This is repeated for each herbivore species until no unoccupied plants remain. Individual herbivores that remained in the community from the original distributions were then removed from the community if all potential host plants are occupied. This process was repeated for enemies, assigning them to herbivores under the same one-to-one assumption, and any unassociated parasitoids are removed from the community. This often resulted in fewer individuals and species, compared to the initial generated values of the communities. 31

Our randomly assembled food webs were generated by sampling Ri randomly from the set of integers {3, 4,. . ., 120} and αi randomly over the interval (Tewksbury et al. 2014, Novotny et al. 2002). Total abundances for each trophic level Ai were randomly sampled from the integers {3, 4,. . ., 500}. These initial values represent the potential values in the realized networks but will not necessarily match following the sampling procedure. The specific distributions for species richness, relative abundances, and alpha parameters were based on food web data from sites across the Americas (Dyer et al.

2007). Using this approach, we generated 1000 random food webs.

Food web sampling

The community was subsampled by randomly selecting individual plants and for each subsample, an individual plant had at most one herbivore and at most one enemy associated with that herbivore. Randomly sampled rows from each local interaction food web were used to calculate the cumulative interaction diversity for each sample. Sampled interaction diversity was calculated using the inverse of the Simpson's entropy (1/D) for each cumulative plant-herbivore, herbivore-enemy, and plant-herbivore-enemy interaction. Sampling was completed once all plant individuals within each local community were sampled. Each community differed in the number of species, the numbers of individuals within each species, and the diet breadth assigned to each consumer species. Sampling within the local community occurred without replacement.

In summary, the assumptions for the simulation were: 1) a lognormal distribution of species abundances for all trophic levels; 2) a truncated discrete Pareto distribution of consumer diet breadths; 3) complete detection of all herbivores and parasitoids associated 32 with an individual plant; 4) only one individual of a consumer species per individual of a resource species.

Total network analysis

We quantified network-level connectance to identify how species richness and specialization influence the structure of entire networks; connectance is a commonly used network parameter (Dormann et al. 2009, Thébault and Fontaine 2010). To accomplish this, we assembled three separate, but not mutually exclusive, networks within each individual local community described above. A plant-herbivore (PH), herbivore-enemy

(HE), and plant-herbivore-enemy (PHE) network were assembled separately to quantify connectance and compare outcomes when examining two- or three-trophic-level networks.

A weighted network was constructed from each local community by generating a bipartite matrix with the abundance of interactions that occurred between individuals of each community. PH and HE matrices were built based on each local community to calculate network-level properties concerning two trophic levels. To investigate PHE networks, we generated a matrix of producers (e.g., plants and herbivores) and consumers

(e.g., herbivores and enemies) and quantified network-level properties similarly to the previously mentioned bipartite networks. For each distinct network (e.g., PH, HE, PHE), the R-package "bipartite" (version 2.05) was utilized to quantify connectance (Dormann et al. 2008). In all subsequent network analyses, empty columns and rows were deleted before calculating network-level metrics. These values were integrated with other diversity measurements from our sampling scheme to investigate the desired relationships. 33

Rarefaction analyses

To compare the accumulation rates of species and interactions in a given local community, we used rarefaction curves and the Chao1 estimator of richness (Chao 1984).

We generated rarefaction curves using the `vegan' package (version 2.2.1) in R (Oksanen et al. 2015) and calculated the slope of each rarefaction curve at the number of samples it took to sample half the total richness for each local community. These values allowed us to compare the accumulation rates between species and interactions across a wide range of local communities. We estimated the richness for interactions and species using the

Chao1 non-parametric estimator of richness (Chao 1984). Chao1 estimates of richness were calculated for PH, HE, and PHE networks. Specifically, for the PHE networks, only complete PHE interactions were used. Slopes and estimated Chao1 richness were compared using Bayesian estimation for two groups in the R package "BEST" (Kruschke

2013, Kruschke and Meredith 2015). This method provides an alternative to classic t-tests and creates posterior estimates for group means and 95% high-density intervals (HDI).

Point estimates and 95% HDI were used to identify differences between sampled interactions and species for all 1000 local communities. The mean and standard deviation of the observed differences between interactions and species networks served as priors.

Given the large sample size, the method provides robust posterior probabilities identifying differences between sample means. Differences were considered significant if the 95% HDI did not overlap. All web simulations and network analyses were performed using program R (version 3.3.2, R Core Team 2014). 34

Statistical analysis

Linear regression and structural equation models were used to identify the relative effects of taxonomic diversity and diet breadth on interaction diversity and other network structure metrics. We assessed specific path models to test a previously hypothesized structure equation meta-model. Path coefficients for direct effects were obtained from the structural equation model, whereas indirect effects were calculated as the product of direct effects in any given pathway. For our a priori specified structural equation model, we identified causal relationships to formulate a simple set of paths with three exogenous variables (plant abundance, herbivore diet breadth, enemy diet breadth) predicting four endogenous variables (interaction diversity, interaction density, species diversity, connectance); no latent variables were used. Specifically, on the basis of literature, our own empirical data, and assumptions of the simulations, all exogenous variables were predicted to increase interaction diversity, species diversity, and connectance. In addition, these exogenous variables were expected to have positive effects on connectance via interaction diversity and density. We tested the fit of this model using SAS (PROC

CALIS) and utilized the reticular action model (RAM-a covariance structure model) to specify the models (SAS Institute Inc. 2011). Starting values for the parameter estimates were determined by using a combination of three methods: observed moments of variables, the McDonald method, and two-stage least squares. The estimation method for the model was maximum likelihood, and the Levenberg-Marquardt algorithm was used to iterate solutions for optimization. The χ2 for the absolute index was used to assess the fit of the model, with P > 0.05 (with 2 df) as an indication of a good fit to the data.

Residuals met assumptions for multiple regressions. This approach was utilized for the 35 full communities generated by our simulations as well as for random samples from each community that started at 5 interactions sampled up to 500 interactions sampled and path coefficients were compared from the identical models across these sample sizes.

Comparing coefficients across a range of sample sizes allowed us to investigate how predicted relationships among variables changes as the number of observed interactions increase, which is analogous to changing the size of the plot or local community.

We also used simple linear regression to examine how consumer diet breadth and taxonomic diversity influence the association between interaction and species diversity.

Species diversity was regressed against interaction diversity and the residuals from that model were used as a dependent variable in subsequent linear models. Using residuals as a dependent variable allowed us to identify whether relationships between species and interaction diversity differed under various community conditions, such as specialized or generalized consumers. Linear regressions were performed to identify whether consumer diet breadth, taxonomic abundance, and species richness significantly altered relationships between interaction and species diversity using these residuals. This analysis was implemented for each distinct network (e.g., PH, HE, PHE). The mean observed diet breadth for consumers was utilized as a measure of specialization. Diet breadth was restricted to mean herbivore diet breadth for PH networks, mean enemy diet breadth for HE networks, and the mean diet breadth among herbivores and enemies for

PHE networks. The sum of species richness and taxonomic abundance across all trophic levels in the local network was used for measures of richness and abundance. These analyses were performed using program R (version 3.3.2, R Core Team 2014). 36

Results

Interaction and species rarefaction curves

1000 different local communities were generated and cumulatively sampled (see

Figure 1 for an example network). Interaction and species rarefaction curves for all PH,

HE, and PHE networks yielded variable rarefaction curves among the local communities, between interactions and species, and among the three networks (e.g., PH, HE, PHE) (S1

Figure). Each step in the accumulation curves is analogous to different levels of sampling due to differences in scales (e.g., plots recommended in Dyer et al. 2010) or due to error or limited sampling.

The Bayesian estimation of two groups identified significant differences in the mean Chao1 estimator of richness between interactions and species in PH (HDIsp = 136 -

146, HDIint = 80 - 91, HDIdiff = 48 - 63), HE (HDIsp = 129 - 140, HDIint = 72 - 82,

HDIdiff = 49 - 65), and PHE networks (HDIsp = 198 - 213, HDIint = 89 - 103, HDIdiff =

99 - 118)(Figure 2A). The effect size was similar in PH (Effect size = 0.95) and HE

(Effect size = 0.94) networks and largest for the PHE only network (Effect Size = 1.45).

Mean Chao1 estimates of species richness were consistently greater than interactions in all three networks.

Bayesian estimates of the mean slope at the number of samples it took to accumulate half of the total richness (a value analogous to the Michaelis constant in

Michaelis-Menton enzyme dynamics) differed significantly among species and interaction rarefaction curves, and among the three network types (Figure 2B) (i.e., PH,

HE, PHE). Rarefaction slopes of PH (HDIsp = 0.38 - 0.42, HDIint = 0.61 - 0.64, HDIdiff

= 0.21 - 0.24), HE (HDIsp = 0.60 - 0.63, HDIint = 0.74 - 0.76, HDIdiff = 0.12 - 0.15), 37 and PHE (HDIsp = 0.40 - 0.43, HDIint = 0.81 - 0.83, HDIdiff = 0.39 - 0.43) networks consistently higher than species. The estimated difference between species and interactions was greatest in PHE networks. Effect size was smallest when investigating

HE (Effect Size = -0.73) networks, but greatest within the PHE networks (Effect Size = -

2.73).

Relationships between species and interaction diversity

The correlation between species and interaction diversity was strongest among PH networks (Pearson's Corr. = 0.96, p < 0.001) and gradually decreased with HE (Pearson's

Corr. = 0.93, p < 0.001) and PHE networks (Pearson's Corr. = 0.35, p < 0.001). This pattern remained consistent when the slope and coefficient of determination in linear models was examined (R2) (Figure 3; S1 Table).

Diet breadth, species richness, and species abundance all significantly influenced the association between interaction and species diversity (partly due to the high power associated with large sample sizes), but the strength of the effects differed among the networks being investigated (Figure 3; S2 Table). Communities with greater mean consumer diet breadth (i.e., increased generalization) resulted in more positive residuals between species and interaction diversity in PH networks (β = 1.5, P<0.001) (Figure 3).

Positive residuals in this case signify higher values of interaction diversity then would be expected given the diversity of species. Similar, but larger effects of diet breadth on relationships between species and interaction diversity were observed in HE (β = 3.53, P

< 0.001) and PHE (β = 10.6, P < 0.001) networks (Figure 3). 38

Figure 1. A randomly selected tri-tropic network produced from one of the 1000 simulations. Each black bar is a node representing a unique species, while the grey bars are edges connecting the black bars and represent observed interactions between those two species. Green sections within some of the black bars represent individuals within that particular species that were present in the community, but not involved in trophic interactions (e.g., plants without herbivores). The width of each edge and node within the network denotes the abundance of sampled interactions or species. Only species that were sampled are shown in this network. Numbers above each node denote the species identification number from that particular simulation. 39

Figure 2. Posterior probabilities of: A) mean Chao1 estimates of richness for species and interactions, and B) the mean slope of rarefaction curves for species and interactions. Interactions are displayed in grey, while species are shown in white. The error bars represent the 95% High Density Intervals (HDI). Mean slopes were acquired by calculating the slope of each rarefaction curve when half of the species or interactions were sampled. Chao1 estimates of richness were acquired using the `estimateR' function in the vegan package in R.

The effect of species richness on the relationships between species and interaction diversity was significant for all three networks (Figure 3; S2 Table). Increased species richness was positively associated with the residual values of PH (β = 0.013, P = 0.007) and HE networks (β = 0.039, P<0.001). Relationships to PHE network (β = 0.23,

P<0.001) residuals displayed the most pronounced, positive linear relationship with increased species richness. These results revealed that local communities with higher values of species richness yielded more interactions than expected based on the number of species present in the community and that effect is strongest when three trophic levels is considered. The variance explained within each model improved in successively higher 40 trophic levels and was greatest when all three trophic levels were incorporated in the models (S2 Table).

Abundance revealed statistically significant linear relationships with residual values from all three networks, but the strength of these associations were relatively weak compared to diet breadth and species richness. Total abundance in PH (β = 0.0041,

P<0.001) and HE (β = 0.007, P<0.001) networks displayed the weakest association with residual values (S2 Table). Total abundance within PHE networks revealed the largest positive estimate, but the slope was still noticeably small (β = 0.015, P<0.001). In two of three cases (e.g., richness and abundance), variance explained was greatest when all three trophic levels were considered. Changes in consumer diet breadth resulted in the largest estimate, but models that included richness explained the most variance.

Path analysis and the effects of sampling

The best-fit path model, when using all samples, performed significantly better than all other models (Figure 4; χ2 = 3.6, df = 4, P = 0.5; AIC = 36; delta AIC range = 60

- 70). Species diversity showed the strongest positive effect on PHE interaction diversity and as predicted, all other variables positively affected interaction diversity (Figure 4).

Only total plant abundance within the local community negatively affected, though indirectly, interaction diversity. Thus, communities with more plant individuals had lower values of interaction diversity, but that effect was driven primarily through its strong negative effect on species diversity. The effects of consumer diet breadth and 41

Figure 3. Summary plots of semi-partial correlations between the residuals of species diversity and interaction diversity (these residuals are on the y-axis) and mean consumer diet breadth, species richness, and total abundance (these three parameters are on the x- axis). We investigated this relationship for all three networks (e.g. PH, HE, PHE). The top three panels represent changes in mean diet breadth for each consumer trophic level; mean herbivore and enemy diet breadth were used for the PH and HE networks respectively, while the mean diet breadth for all consumers (herbivores plus enemies) was used for PHE networks. The middle three panels denote community richness for each respective network, which is the total number of species found in all trophic levels. The lower panel displays semi-partial correlations with total community abundance, which equals the sum of all individuals within each trophic level. The solid black lines are least squares regression lines. 42

Figure 4. A path diagram summarizing the standardized path coefficients across all 1000 local communities (χ2 = 3.6, df = 4, P = 0.5; AIC = 36). Each path was chosen based on a priori hypotheses, and compared to competing models using AIC and χ2. Lines ending with an arrow denote positive coefficients, while lines ending with a circle denote negative coefficients. The width of the arrow indicates the relative size of the coefficient. connectance on interaction diversity were both negligible. While species diversity had a strong positive effect on interaction diversity, more species-diverse communities had lower levels of connectance. Local plant abundance within communities had strong negative effects on species diversity and connectance, but weak direct effects on interaction diversity. 43

To understand the sensitivity of each path coefficient to the number of observations included in the path analysis, path coefficients were derived from SEMs that used random samples from each simulated community that started at 5 interactions and increased up to 500 interactions (Figure 5). Due to issues with generating balanced samples for SEM, connectance was not included in this model and therefore the structure of the path model differed from that shown in Figure 4. We consider the random samples to be analogous either to actual sampling in a biotic community or to smaller scale communities that are derived from a regional pool of species and potential interactions.

As the number of observed interactions included in each SEM increased, the strength of each path coefficient varied considerably (Figure 5). The degree to which each path coefficient changed differed among the coefficients and all responded in a non- linear fashion. The direct positive effect of species diversity on interaction diversity decreased significantly as more observations were included in the path analysis (Figure

2A). Mean herbivore diet breadth maintained a positive effect on interaction diversity across all observed interactions, but its effect was greatest at intermediate values and decreased as the number of observations exceeded 100 (Figure 2B). The direct effect of enemy diet breadth on interaction diversity was strongest at intermediate (~200-300) numbers of observed interactions (Figure 2C), while the effect of local plant abundance on species diversity increased consistently across sampled interactions (Figure 2D). The effect of local plant abundance ultimately yielded a positive influence on species diversity, and this difference from the model with full webs (Figure 4) was most likely due to the absence of connectance from the path analysis. The effect of plant abundance 44 on interaction diversity also increased as the number of observations increased, but never resulted in a positive effect (Figure 2E).

Figure 5. Scatterplots displaying the relationship between the strength of each path coefficient and the number of sampled interactions included in the path analysis (Figure 5), with the exception of paths associated with connectance. The strength of the path coefficient is shown on the y-axis and number of observed interactions included in the model is shown on the x-axis. The solid line represents outcome of linear or polynomial regressions. Path coefficients used in these analyses were significant (P < 0.05). 45

Discussion

The interest in interaction diversity as a metric of biodiversity has developed separately from natural history studies that attempt to rigorously document interactions at local and regional scales (Dyer et al. 2012). Interaction diversity and other network parameters, such as connectance, have been gleaned from loosely constructed networks

(e.g., from literature searches or brief observational studies), and these parameters have been utilized as measures relevant to network structure and resilience. But these networks are not realistic since local networks do not include all possible edges among nodes that are present (Poisot et al. 2012). One reason to examine how relationships among node and edge diversity and network parameters can change with sampling effort or area sampled is to assess the relevance of network analyses based on these putatively empirical regional networks (Schleuning et al. 2012). Such scaling and sampling issues cannot be ignored when this regional network view of interaction diversity is utilized to assess issues associated with relationships between biodiversity, productivity, ecosystem function, and extinction.

Our food web simulation generates hypotheses relevant to the power of sampling actual interactions and calculating the diversity of interacting individuals across a variety of ecological communities. The clearest patterns that emerged and are worth pursuing with empirical data were: 1) randomly assembled networks produce accumulation curves for interaction diversity that reach an apparent asymptote more quickly than species diversity, so interaction diversity may be more practical to estimate than species diversity in real ecosystems: 2) this is especially true at intermediate sample sizes (or local community sizes (Dyer et al. 2010)), where local species diversity is the best predictor of 46 local interaction diversity at multiple sampling scales; 3) consumer diet breadth, defined by a truncated Pareto distribution, may disrupt the strong relationship between interaction and species diversity, as more generalized communities will have higher interaction diversity; 4) species diversity and local plant abundance are also likely to predict other tri-trophic network parameters, such as connectance; and 5) local network parameters are likely to be quite different from the regional networks, and this relationship changes as the networks grow in size.

The interaction diversity model

Our approach to simulating tri-trophic networks provides randomly assembled quantitative communities that can be separated into discrete bipartite networks nested within a randomly assembled community. This provides an opportunity to investigate how the number and position of trophic interactions influences network-level properties from a discrete sampling procedure (Jordán and Osváth 2009). Furthermore, the design of the simulation model provides generous flexibility allowing researchers to modify foundational building blocks of ecological communities, including distributions of consumer diet breadths. Finally, it provides insight into how sample size or spatial scale can affect network properties. The addition of a third trophic level separates our approach from previous simulated network data (e.g., Fründ et al. 2016, Thébault and Fontaine

2010, Blüthgen 2010, Tylianakis et al. 2010), and for both modeling and empirical approaches to ecological networks, expanding to more complex interaction networks should be a focus as it can provide additional information on network dynamics and function. Our model showed that the number and position of trophic levels that are being analyzed, especially when considering plant, herbivore, and natural enemy communities, 47 influence network-level properties. Samples from studies that incorporate higher trophic levels are completely dependent on the successful sampling of associated hosts. This can have significant impacts on the observed structure and diversity of a sampled network. As more trophic levels are included in a network, the dependencies of sampled (or included) interactions increase, which exacerbates problems with large regional networks that actually do not exist locally.

The simulation of tri-trophic networks developed “complete” networks that were assembled with only one assumption - networks consisted of consumers with restricted diets and included realistic numbers of species and interactions (based on empirical interaction diversity data). Our goal was to generate a network that is more consistent with standard neutral assumptions (no assembly rules) combined with niche-based assumptions (specialization), rather than following an abundance-based simulation null model (Dormann et al. 2009). In the future, the flexibility of our simulation model, which allows the manipulation of richness, abundance, and diet breadth for each trophic level included in the community, will incorporate other assumptions, such as assembly rules

(Keddy 1992, Weiher and Keddy 2001), or to omit the assumption of restricted consumer diet. However, our utilization of a truncated Pareto distribution for host range is well supported in plant-arthropod networks (Forister et al. 2015) and provides a realistic measure of host specialization in multitrophic networks that include plants, insect herbivores, and parasitoid natural enemies. The manipulation of richness, abundance, and diet breadth and their distributions, allows for a useful tool to compare observed data to simulated data from the model. This can help with determining the importance of diet breadth distribution or degree of specialization versus other factors in sampled networks 48 when exploring relationships between diversity, network processes, and network patterns

(Fründ et al. 2016, Staniczenko et al. 2013, Peralta et al. 2014). Finally, the subsampling approach can generate smaller networks that are a more realistic representation of interacting species in local food webs (Staniczenko et al. 2017).

Species and interaction rarefaction curves

Few studies have attempted to compare rarefaction curves for species and interactions across a wide range of multitrophic communities (but see Vasquez et al.

2009, Burkle and Knight 2012, López-Carretero et al. 2014). Rarefaction is used to easily compare measures of richness between communities in which the sampling effort is different and can be useful to help identify the completeness of sampling that has occurred in a community (Gotelli and Colwell 2001). It is assumed, though never tested, that given the substantially more potential interactions than species, interactions should accumulate much more slowly than species when sampling from a discrete sample area.

However, many interactions never occur (i.e., they are forbidden or never observed) and it is possible that interactions are characterized by a more kurtotic distribution than species, which should result in interactions obtaining an apparent asymptote more quickly than species (Dyer et al. 2010, González-Varo and Traveset 2016). In other words, similar to species distributions, interactions are typically dominated by a few, abundant connections, with many singleton or rare interactions. Therefore, the shape of rarefaction curves may be highly influenced by the abundance distributions, taxonomic richness, and host range of consumers in multi-trophic communities.

The assumptions of our model and the focus on more specialized consumers clearly impose some limits to the generality of our results. For example, the values of 49 interaction richness yielded by this simulation may be considerably lower than species richness due to our high levels of host specialization. A truncated Pareto distribution involves few generalist and many specialist species, and this increase in more limited trophic interactions reduces the number of unique interactions that occur when there are no assembly rules or differences in densities for consumers of different diet breadths. On the other hand, if specialists are always more abundant than generalists, this distribution can increase the number of unique interactions locally. Other networks (e.g., plant- pollinator) have revealed higher numbers of interactions than species (e.g., plants and pollinators) (Gibson et al. 2016, Chacoff et al. 2012, Fang and Huang 2016), but these mutualistic communities are normally characterized by more generalized interactions, have often been regional networks (i.e., large scale), and the networks are almost always based on all visitors rather than true pollinators (Vázquez and Aizen 2004, Petanidou et al. 2008, King et al. 2013, Valdovinos et al. 2016). Furthermore, these communities ignore more subtle factors that affect network parameters and specialization, such as adaptive foraging (King et al. 2013). The model's generality may be reduced in other ways, but using a truncated Pareto distribution of host specialization may be the best approach to studying antagonistic interactions, especially those involving plants, insects, and parasitoid natural enemies (Dyer et al. 2007). However, the simulation approach is adaptable, and any distribution of host utilization is possible, and modified assumptions would be necessary for communities other than plant, insect herbivore, and parasitoid communities. 50

Associations between species diversity and interaction diversity

As expected, we observed a strong positive correlation between species and interaction diversity, but this relationship was more stable than anticipated across the diverse range of communities, scales, and sample sizes. We hypothesized that consumer diet breadth and other community parameters (e.g., richness and abundance) should have altered the correlation between interaction and species diversity more than what we observed. Specifically, more specialized communities (higher α-parameters) result in lower positive correlation coefficients (fewer links per node) due to the decrease of interactions that involve generalist species. Based on our simulations, notable changes in the correlation coefficient or slope among species and interaction diversity across a wide range of combinations of community parameters were observed (Figure 3, S1 and S2

Tables), but the effects were weaker than expected. More specialized communities displayed more negative residuals, which suggests that there are fewer interactions than expected based on the number of species present in the community. Although this effect was small, it supports the hypothesis that generalized interactions are rare, but have large effects on interaction diversity locally (Dyer et al. 2010). Generally, community parameters (e.g., richness, abundance, diet breadth) had little effect on the relationship between species and interaction diversity, probably due to the lack of assembly rules and low numbers of generalists. The main parameters that altered the associations between species and interaction diversity were the number of trophic levels.

An important contribution of our simulation is that it included more than two trophic levels in an effort to understand how the position and number of trophic levels in a community can drive relationships between species and interaction diversity. Many 51 network studies have been limited to plant-pollinator or plant-herbivore networks, yet communities are far more complex, and patterns of interaction diversity and network topology from two-trophic- level analyses are likely different from more realistic multi- trophic communities. Our results revealed that when incorporating three trophic levels, the community parameters (e.g., diet, richness, and abundance) all have stronger impacts on the relationship between species and interaction diversity. This is likely due to the contingent nature of sampling partners at lower trophic levels to acquire individuals at higher trophic levels. In other words, the likelihood of sampling enemies is founded on the likelihood of sampling an herbivore, which results in a propagation of effects, changing the probability density functions of interactions differently from species density functions.

While this is an unavoidable sampling artifact, it is important to consider when drawing conclusions about the observed structure of a multi-trophic ecological network.

Thus, when investigating more than two trophic levels, the impacts of consumer specialization and species richness are magnified in driving food web patterns and decrease associations between species and interaction diversity (Beaver 1985). Utilizing interaction diversity, as a metric of biodiversity, to help with conservation and management issues will be most useful when more than two trophic levels are investigated. Otherwise, species diversity should be a reasonable proxy for interaction diversity when a community is dominated by only plants and herbivores since disparities between interaction and species diversity are lowest for two trophic levels. 52

Effects of primary productivity, diet breadth, species diversity, and number of observed interactions on network structure

We observed considerable variance in interaction diversity in the assembly of

1000 tri-trophic communities, with the only constraint on consumer diet-breadth distributions. This variance was due to both random effects and partly due to the deterministic effects of the manipulated parameters. By utilizing a path analysis framework we were able to identify direct and indirect effects of multiple community parameters on interaction diversity. Under this framework, species diversity, and to a lesser extent consumer diet breadth revealed the strongest direct effects determining interaction diversity. As expected, species diversity had a strong positive effect on interaction diversity. The effect of herbivore and enemy diet breadth were similarly positive but not very strong. These results are not what we originally predicted given that we expected interaction diversity to be an emergent consequence of distributions of consumer specialization and taxonomic richness (Beckerman et al. 2006). As stated previously, this weak effect of diet breadth was likely due to the highly skewed truncated

Pareto distribution.

The relationship between connectance and interaction diversity was relatively weak and shows dissimilar relationships with other variables in the path analysis. This result suggests that connectance and interaction diversity are measuring different qualities of ecological communities and are determined by different factors within a community.

Further, if the goal is to conserve biodiversity within a community, connectance does not appear to be a good predictor of diversity of interactions and is negatively related to species diversity (Beckerman et al. 2006, Winemiller 1989). However, connectance 53 facilitated both the indirect effects of species diversity and local plant abundance on interaction diversity.

Using a similar path analysis (i.e., without connectance), we found that the number of observations included in the model biases the strength of all path coefficients, and this could also be viewed as a scaling issue - lower numbers of observations in our model are analogous to more localized assemblages within a community. Studies investigating these sampling or scaling effects on ecological network parameters are rare, but they are important because ecological networks are especially vulnerable to sampling effects as well as scale (Fründ et al. 2016, Dormann et al. 2009, Nielsen and Bascompte

2007). One of the more interesting patterns was the sharp decline in the effect of species diversity on interaction diversity as the number of observed interactions increased. This suggests that the effect of species diversity on interaction diversity can be overestimated when the number of observations is insufficient, and that when sufficient observations accumulate the relationship between species and interaction diversity becomes weaker.

Similar to what previous studies have found, diet breadth of both herbivores and predators was sensitive to sampling bias (Fründ et al. 2016). However, identifying the effects of consumer specialization on interaction diversity may be difficult given the nonlinear relationship with the sample size or changes in scale.

Conclusions

While this model will be useful for developing basic hypotheses concerning the drivers of trophic interaction diversity, there are details in our model that merit further work. We utilized this simulation to test hypotheses about accumulation patterns of species and interactions, but this modeling approach is also appropriate for investigating 54 spatial scaling of interactions and species. A great deal of progress has been made towards understanding species diversity, but we lack even a rudimentary understanding of the determinants and spatial or temporal dynamics of interaction diversity. Food web simulations may be particularly useful to investigate more about the relationships between local and regional interaction diversity (Cornell and Lawton 1992, Ricklefs and

Schluter 1993), which will provide insight into the ability of the preponderance of large regional networks that are used to address big issues in ecology and conservation.

In conclusion, we demonstrated that in highly specialized communities, trophic interactions accumulate more quickly than species. We showed that diet breadth and taxonomic richness both interact to influence relationships between species and interaction diversity. Importantly, this model demonstrated that the position and number of trophic levels being investigated strongly impacted correlations among species and interaction diversity, which is critical for biodiversity research and conservation efforts.

Interaction and species diversity are most divergent when incorporating more than two trophic levels, so utilizing interaction diversity as a metric of biodiversity will be useful for multi-trophic investigations for both applied and basic research questions such as spatiotemporal dynamics, a biogeographical theory of species interactions, and the effects of climate change on biological networks. 55

Supporting information

S1 Table. Direct relationship between species and interaction diversity as estimated by correlation coefficients, and the beta coefficients from linear regressions between species and interaction diversity. (PDF)

S2 Table. Beta coefficient and R2 for linear regression of residuals from linear regression between species and interaction diversity and the variable of interest (diet breadth, species richness, abundance). (PDF)

S1 Fig. Rarefaction curves for interactions and species from 1000 simulated communities.

Rarefaction curves were generated using a modified version of the `rarecurve' function in the R-package, vegan. This modification permitted sampling of species and interactions within each community with replacement 500 times. Rarefaction curves were generated for all three networks within each community: Plant-Herbivore (PH), Herbivore-Enemy

(HE), and Plant- Herbivore-Enemy (PHE). PHE networks include each unique PHE interaction, excluding PH interactions that were not involved in a HE interaction. (PDF)

S1 File. R-Code to generate tri-trophic networks used in this analysis. (R)

S2 File. Raw accumulation data from 1000 model simulations used in this manuscript.

(ZIP)

S3 File. Analysis and descriptive statistics for each complete tri-trophic network. No accumulation data is included, only the final, observed network. (ZIP) 56

Acknowledgments

We thank Matt Forister and members of the Dyer and Forister laboratories for discussions the research as well as earlier versions of this manuscript. This work was made possible through funding by the Department of Defense Strategic Environmental

Research and Development Program grant [RC-2243], the National Science Foundation

(DEB 1344250 and DEB 1442103), and the Earthwatch Institute

(http://eu.earthwatch.org/) (LAD).

Author Contributions

Conceptualization: Nicholas A. Pardikes, Will Lumpkin, Lee A. Dyer.

Data curation: Nicholas A. Pardikes, Will Lumpkin, Paul J. Hurtado.

Formal analysis: Nicholas A. Pardikes, Paul J. Hurtado, Lee A. Dyer.

Funding acquisition: Lee A. Dyer.

Investigation: Nicholas A. Pardikes, Lee A. Dyer.

Methodology: Nicholas A. Pardikes, Will Lumpkin, Paul J. Hurtado, Lee A. Dyer.

Project administration: Nicholas A. Pardikes, Will Lumpkin, Lee A. Dyer.

Supervision: Paul J. Hurtado, Lee A. Dyer.

Validation: Nicholas A. Pardikes, Paul J. Hurtado.

Visualization: Nicholas A. Pardikes, Lee A. Dyer.

Writing ± original draft: Nicholas A. Pardikes, Will Lumpkin, Paul J. Hurtado, Lee A.

Dyer.

Writing ± review & editing: Nicholas A. Pardikes, Paul J. Hurtado, Lee A. Dyer. 57

Macrophyte Diversity and Complexity Reduce Larval Mosquito Abundance

Will P. Lumpkin (1)

Kincade R. Stirek (2)

Lee A. Dyer (1)

1. Department of Biology, University of Nevada Reno, 1664 N. Virginia St, Reno, NV

89557-0314, 2 Department of Biological Sciences,

2. Towson University, 8000 York Road, Towson, MD 21252, and 3 Corresponding author, e-mail: [email protected]

Subject Editor: Kristen Healy

Received 4 October 2019; Editorial decision 10 January 2020

Abstract

The role of aquatic arthropod diversity and community interactions of larval mosquitoes are important for understanding mosquito population dynamics. We tested the effects of aquatic macrophyte diversity and habitat structural complexity in shaping the predator and competitor invertebrate communities associated with mosquito larvae.

Experimental mesocosms were planted with live aquatic macrophytes and allowed to be naturally colonized by local invertebrates. Results indicated a positive effect of macrophyte diversity on competitor diversity and a negative effect on predator diversity.

In turn, predator diversity negatively impacted mosquito abundance through a direct effect, while competitor diversity showed an indirect negative effect on mosquito larval abundance through its positive effect on predator diversity. The enhancement of aquatic macrophyte diversity and structural complexity has practical applications for the 58 reduction of mosquito populations in managed systems where complete source elimination is not possible.

Introduction

A thorough understanding of the interactions influencing patterns of diversity and ecosystem services that affect pests and disease vectors is important for improved biological control of harmful insects in natural and managed ecosystems (Altieri and

Letourneau 1982, Altieri and Nicholls 2004, Stireman et al. 2005, Tscharntke et al. 2005,

Tylianakis and Romo 2010, Dyer and Letourneau 2013, Dassou and Tixier 2016).

Research on trophic interactions within the context of integrated pest management (IPM) have largely focused on agricultural systems; however, less research has investigated the importance of understanding and managing natural aquatic food webs as part of effective mosquito (Diptera: Culicidae) control. Historical and contemporary mosquito control efforts have largely relied on habitat source reduction or elimination and reliance on chemical control approaches (Floore 2006). Efforts at biological mosquito control have focused on the use of single enemy species such as larvivorous fish (Chandra et al. 2008) and individual species of invertebrates (Chapman 1974, Collins and Washino 1985, Mogi

2007, Quiroz-Martinez and Rodrigues-Castro 2007, Carlson et al. 2009). While these approaches may be effective for certain systems and for limited time periods, populations of enemy species can collapse, resulting in resurgences of the target nuisance species

(Barbosa 1998). Introduced natural enemies may also consume existing natural enemies and contribute to a net loss of natural predators (Blaustein 1992). Instead, a broader ecological framework may lead to more sustainable control programs for not only mosquitoes but economic and agricultural pests as well (Barbosa 1998, Rusch et al. 59

2017). Understanding aquatic community-level interactions of mosquito larvae is important considering current increases in construction and rehabilitation of wetland systems near human populations, either for environmental conservation and aesthetic value (Willott 2004) or stormwater pollution control and hydrologic capacity (Russell

1999, Thullen et al. 2002).

While aquatic invertebrate diversity has been shown to reduce the incidence of mosquito outbreaks (Chase and Knight 2003, Carlson et al. 2009), little focus has been given to the role of aquatic macrophyte diversity for the stability of aquatic invertebrate communities. Greater plant diversity in managed terrestrial systems can increase natural enemy species (Schmidt et al. 2005, Bianchi et al. 2006), increase overall long-term invertebrate diversity (Haddad et al. 2011), and improve resistance to disturbance

(Tilman 1994) and nutrient utilization (Tilman et al. 1996). More importantly, the interaction diversity of biotic communities enhances community stability (Mougi and

Kondoh 2012), influences the evenness of such communities, and is associated with and overall reduction of dominant pest groups (Dyer and Letourneau 2013). This interaction diversity is defined as the number of unique interactions between species in a community and can include both direct and indirect multi-trophic links between species (Janzen

1974, Thompson 1996, Dyer et al. 2010). Important and naturally occurring species interactions that negatively affect mosquito larvae are not limited strictly to predator-prey interactions as employed in classical biological control programs. Antagonistic interactions also include competition for resources (Barrera 1996, Braks et al. 2004), parasitism (Wise de Valdez 2006), indirect competitive interactions (Alto et al. 2005), and oviposition inhibition (Bentley and Day 1989). Interaction types (i.e., overall 60 interaction diversity) likely affect prey abundances in a context-specific manner and may not necessarily result in direct antagonistic outcomes for prey abundance (Holt 1977,

Holt and Lawton 1994, Carlson et al. 2009, Juliano 2009).

Studies in both terrestrial and aquatic systems have demonstrated clear associations between structural complexity and diversity across trophic levels. In terrestrial systems, habitat complexity enhances primary productivity by creating spatial refugia for invertebrate predators, protecting them from intraguild predation and facilitating trophic cascades (Finke and Denno 2006). A meta-analysis on terrestrial habitat complexity showed that predator density is dramatically increased in more structurally complex habitats (Langellotto and Denno 2004). In lotic river systems, structural complexity can influence diversity by providing structural refuge for both predator and prey species, dependent on feeding behaviors, and by altering hydrodynamic patterns (Cardinale et al. 2002). In artificial mesocosms designed to represent lentic phytotelmata and allow direct manipulation of structural complexity, Srivastava (2006) found strong links between structural complexity and trophic structure, suggesting that neither trait should be studied in isolation. On larger scales, vegetation management intended to enhance waterfowl habitat by reducing dominant plant monocultures has been shown to increase invertebrate biodiversity in wetlands, reducing overall mosquito production in those environments (Batzer and Resh 1992). Predator functional response determinants (i.e., searching, handling time, and satiation rates) are also influenced by structural complexity (Alexander et al. 2012, Barrios-O’Neil et al. 2015, Cuthbert et al.

2018) reducing predation rates in structurally complex habitats and increasing predation risk in structurally simple habitats (Alexander et al. 2015). Alternatively, habitat 61 complexity can have little effect on predator functional response in ambush predator species (Alto et al. 2005).

In this study, we investigate the effects of structural complexity created by aquatic macrophytes and inert structures (sticks) using experimental aquatic mesocosms in an outdoor field system. Our general hypothesis is that greater macrophyte structural complexity should provide greater surface area for phytoplankton and algae growth and spatial refuge for predator avoidance, as well as ambush predator resting sites, causing increased overall invertebrate diversity within the mesocosms. Specifically, our first hypothesis is that greater plant diversity will increase both predator and competitor diversities (Scherber et al. 2010). Secondly, greater arthropod diversity (predators and competitors) should reduce the dominant abundance of any single species, including mosquito larvae (Figure 1). 62

Figure 1. Theoretical relationships between aquatic macrophyte diversity and structural complexity elements. Solid arrows represent direct positive effects, bullets represent negative effects and dashed lines represent indirect effects. Macrophyte diversity should contribute to more foraging surfaces and predator avoidance refuges for mosquito competitors but also spatial refuges for ambush predators. Synergistic relationships are also likely but not described here. 63

Methods

Study System

Experiments were conducted at the University of Nevada Reno Agricultural

Fields in Southern Washoe County Nevada. Three common local species of plants were used including Eleocharis acicularis (Cyperaceae; common spikerush), Typha latifolia

(Typhaceae; broadleaf cattail) and a submerged broad-leaf plant Veronica anagallis- aquatica (Plantaginaceae; water speedwell). These species are common throughout the region where Culex spp. Mosquitoes are present. Dead Coyote Willow (Salix exigua;

Salicaceae) branches were used as an additional experimental structural element. Coyote

Willow is often an emergent vegetation of shallow wetlands with submerged portions being a mix of dead and living branches.

Experimental Design

Experimental mesocosms (Figure 2) were used to test the effects of plant diversity and habitat complexity on the aquatic invertebrate community of naturally colonizing species. Mesocosms were planted with one of four initial ‘plant’ and ‘stick’ treatments with 10 replications of each: low plant diversity, low plant diversity with sticks, high plant diversity, high plant diversity with sticks. The sticks helped create another dimension of complexity (Finke and Denno 2002, Gibb and Parr 2013). Each mesocosm consisted of plastic storage containers measuring approximately 38-liter volume. Ten centimeters of unfertilized topsoil was added to each mesocosm as a plant substrate. All plants were harvested from local wetlands and washed thoroughly in order to minimize introductions of invertebrates from the plants and their rootstock. Each of the 40 mesocosm containers were randomly assigned a level of diversity treatment of either a 64 monoculture of nine individuals of Typha latifolia (low diversity, n = 20) or an even mix of three each of all three plant species (high diversity/highest complexity, n = 20) with a total of nine plant individuals. Ten containers of each plant treatment were then assigned an additional complexity treatment level, which consisted of two dead willow branches crossed diagonally in the mesocosm. The resulting configuration consisted of 10 units of each of the four diversity/complexity treatments listed above, assigned randomly to containers positioned one meter apart in a 4 × 10 meter configuration (10 meters oriented from East to West). A single experimental site was used for all 40 mesocosms in a common garden in order to reduce site-specific factors (e.g., elevation, humidity, proximity to invertebrate source populations, etc.), and to control for local adaptation of invertebrates (e.g., Kawecki and Ebert 2004).

Figure 2. Mesocosms were aligned in a 4 × 10 grid pattern with each container spaced 1 meter apart. The 10-meter length was oriented from East to West. (A and B) the total experimental setup and a high-diversity mesocosm planting, respectively. (C) the potassium permanganate treatment from the 2012 experiment.

65

The experiment was conducted once in both 2011 and 2012 from June 1 through

August 31. Mesocosms were left to be colonized by invertebrates naturally and the entire aquatic community was harvested at the end of each experiment. Consistent water volumes were maintained by daily filling to a standardized drain hole level. The 2012 experiment differed only in that the mesocosms were set up 1 week prior to the start of the experiment, to allow plants to become established, and then treated with potassium permanganate for 12 h and then flushed with clean water. This treatment is toxic to aquatic invertebrates and was designed to further inhibit inadvertent introductions resulting from the plant harvesting and setup process. Following the experiment period, the contents of each container were poured through an aquatic net with a 500-micron mesh to collect invertebrates. Plants were harvested by cutting off at the soil level, separated by species and dried at 60°C for 48 h prior to weighing. Individual plant species biomass was used in calculating plant species diversities.

All invertebrates were identified to the family level, given morphospecies identifications and assigned to a trophic group. Representative biomass estimates were calculated for each morphospecies by collectively weighing all individuals of a single species from mesocosm samples with the largest representative individual counts. Inverse

s 2 Simpson’s diversity D=1 /∑ pi was calculated for most diversity indices. The final latent i =1 variable model included plant diversity as the Shannon diversity/entropy index

s

H=− ∑ pi ln pi. Based on observation and the literature, a predator guild was created by i=1 66 grouping morphospecies that are largely predacious together and similarly, the competitor guild included morphospecies that are periphyton and filter algae eaters, as are most mosquito larvae.

Statistical Analysis

Data were analyzed based on hypothesized relationships as summarized by a meta path model (Figure 1). Analyses were conducted using the R programming language (R

Core Team 2018). To examine effects of manipulated variables (treatment) on diversity of different guilds and mosquito abundance and to provide full estimates of type II error, we utilized Bayesian mixed model analysis of variances (BANOVA package in R, Wedel

2017), with experiment year as a random effect. Successful model convergence was tested using default Geweke diagnostic P-values (GDp) and the Heidelberger and

Welch’s diagnostic P-values (HWDp).

Bayesian linear regression models using JAGS version 4.3.0 (Plummer 2003) were implemented with the runjags R package (Denwood 2016). All prior coefficient probability densities were given uninformative neutral distributions with a mean of 0 and a precision of 0.0001 (i.e., β i ~ N(0, 100)). Simulations with normally distributed residuals (excluding mosquito larvae) included three simulation chains, an adaptation, and burn-in of 6,000 simulations and a total of 30,000 samples. Mosquito larvae as a response variable were modeled assuming a negative binomial distribution, with three simulation chains, an adaptation and burn-in of 8,000 simulations, and a total of 30,000 samples.

Subset path models from our meta-model were analyzed using Bayesian structural equation models (R package blavaan, Merkle and Rosseel 2018). Uninformative 67

Gaussian priors were used for all Beta terms (β i ~ N(0, 0.0001). Regression coefficients were sampled from normal distributions. At the time of this analysis, the blavaan package lacked a straightforward method of passing negative binomial response variables directly to JAGS. Therefore, the mosquito larvae response variable was log-transformed (larvae = ln (mosquitolarvae + 1)) for these analyses.

Results

A single mosquito species Culex tarsalis Coquillett (Diptera: Culicidae) was observed in this study. This is a common species that readily colonizes small stagnant water sources and is primarily a filter feeder that consumes detritus and microorganisms suspended in the water column (Merritt et al. 1992). Additional species within the same generalized trophic group as mosquito larvae (hereafter ‘competitors’) included two species of midge larvae (Chironomidae), water boatmen (Corixidae), two species of small nematodes, Ostracods (Ostracoda), mayfly naiads (Baetidae), and Daphnia. Predators included two species of predacious diving beetle adults (Dytiscidae) and one species of dytiscid larvae, dragonfly naiads (Aeshnidae), damselfly naiads (Coenagrionidae), and cyclopoid copepods (Figure 3). Pupal stages of midge and other fly species were excluded from analysis due to lack of identification resources.

Individual mesocosms in both experiments experienced high variability in plant growth and mortality rates among all three plant species. For both experiments, the majority of Veronica anagallis-aquatica plantings died very early during the experiments. Although the initial experimental design was a four-treatment factorial design with two levels of plant diversity and the presence or absence of willow sticks, the final plant densities and diversities within treatments were too variable and not consistent 68

Figure 3. Results from the simple path analysis showing a direct positive effect of competitor diversity on predator diversity and a direct negative effect of predator diversity on mosquito larvae (abundance). Fully standardized estimates are shown in bold with the partially standardized over latent variable estimates are in parentheses. The dashed bulleted line shows a negative indirect effect of competitor diversity on mosquito larvae. with a true factorial design. Bayesian analysis of variance showed no differences in overall invertebrate diversity between levels of either treatment (effect size = 0.000, 95% credibility interval [CI] −0.065 to 0.003, GDp = 0.649, HWDp = 0.322). Invertebrate richness was significantly higher for the 2012 versus the 2011 experiment (effect size =

0.217, 95% CI 0.175–0.228, GDp = 0.239, HWDp = 0.713, Fig. 2). 69

Based on the Bayesian regression models, plant diversity had the following effects on total invertebrate diversity (posterior mean Beta [β] = 0.440, CI = −0.0448 to

0.916), predator diversity (β = −0.0385, CI = −0.301 to 0.0.223), competitor diversity (β

= 0.273, CI = −0.040 to 0.590), and mosquito abundance (β = −0.0530, CI = −0.391 to

0.246). Predator diversity had a negative effect on mosquito abundance (β = −0.810, CI =

−1.170 to 0.460) as did competitor diversity (β = −0.679, CI = −1.14 to 0.267). The separate effects of total plant, predator, and competitor biomass on mosquito abundance were relatively small compared to the diversity effects. For example, total plant biomass

β = 0.00505, CI = −0.00870 to 0.0163), predator biomass (β = −0.058, CI = −0.090 to

0.0238), and competitor biomass (β = −0.00333, CI = −0.00999 to 0.00372).

The structural equation models supported the hypothesis that increases in both predator and competitor diversities have a negative effect on mosquito larvae. These were expressed as a direct negative effect of predator diversity and an indirect negative effect of competitor diversity on mosquito larvae (Figures 3 to 5). The best fit parsimonious model (Figure 3) supported these causal hypotheses with standardized estimated posterior path coefficients for competitor diversity on predator diversity (0.465, CI = 0.357–0.926), predator diversity on mosquito larvae (−0.287, CI = −0.608 to −0.204), and an indirect negative effect of competitor diversity and mosquito larvae (−0.134, CI = −0.401 to

−0.122). Model 2 (Figure 4) is consistent with the more parsimonious model (Fig. 3) but includes the direct effects of plant diversity (inverse Simpson) on predator diversity

(−0.095, CI = −0.324 to 0.125) and competitor diversity (0.131, CI = −0.079 to 0.295). A third model (Figure 5) includes the latent variable ‘Structure’ which is measured by total plant biomass and plant diversity calculated as the Shannon diversity/entropy index. The 70 standardized estimated posterior path coefficients for this model with the latent variable are: plant diversity on structure (0.649, CI = null, parameter fixed), total plant biomass on structure (0.443, CI = −1.53 to 3.96), structure on predator diversity (−0.096, CI = −1.52 to 1.04), structure on competitor diversity (0.154, CI = −0.693 to 1.30), competitor diversity on predator diversity (0.469, CI = 0.319 to 0.94), predator diversity on larval mosquito abundance (−0.287, CI = 0.613 to −0.198), and the indirect path of competitor diversity through predator diversity in larval mosquito abundance (−0.135, CI = −0.400 to 0.135).

Figure 4. The results of the path analysis model that includes all relationships from Figure 3 and also incorporates plant diversity (inverse Simpson) effects on predator and competitor diversity. 71

Figure 5. The structural equation model showing similar results as Figure 4 and including the latent variable Structure measured by plant diversity (Shannon Entropy) and total plant biomass. 72

Table 1. Results from individual Bayesian Regression models based on hypothetical relationships (figure 1). Estimates with mosquito larvae as response variables were modeled assuming negative binomial distributions.

Response Predictor Posterior Mean 95 % CI SD Variable Predator Plant diversity 0.13 -0.14 – 0.40 0.14 Diversity (intercept) (0.73) (0.27 – 1.2) (0.24)

Predator Competitor 0.54 0.26 – 0.83 0.15 Diversity Diversity (0.14) (-0.30 – 0.60) (0.23)

Competitor Plant Diversity -0.04 (-0.24 – 0.16) 0.10 Diversity (1.49) (1.14 – 1.84) (0.18)

Mosquito Plant Diversity -0.0092 -0.41 – 0.36 0.20 Larvae (-0.50) (-0.71 – 0.63) 0.35

Mosquito Predator -0.78 (-1.13 – -0.43) 0.18 Larvae Diversity (0.45) (0.15 – 0.75) (0.16)

Mosquito Competitor -0.32 -0.82 – 0.13 0.24 Larvae Diversity (0.38) (-0.24 – 1.13) (0.34) 73

Discussion

This study was designed to evaluate how plant diversity and structural complexity influenced the abundance and diversity of mosquito predator and competitor communities and how those, in turn, affected overall larval mosquito densities. Our results did not provide unconditionally strong support for the hypothesis that plant diversity should increase overall aquatic invertebrate diversity (predators and competitors) at this scale.

However, based on the simple regression models, plant diversity had a positive effect on competitor diversity and a negative effect on predator diversity and mosquito abundance.

The most complex and diverse mesocosms had 0.109 ± 0.096 (95% CI) more competitor species than the least complex, lowest diversity mesocosms. Such an increase in competitor diversity can have substantive community effects if scaled up to larger communities, affecting the overall functional effects on focal species such as mosquitoes

(e.g., Ebeling et al. 2017). Both predator and competitor diversity had negative effects on mosquito abundance, supporting our second hypothesis. When more specific causal hypotheses were tested via structural equation models, there was support for an indirect negative effect of competitor diversity on mosquito larvae, resulting from a stronger positive effect of competitor diversity on predator diversity. Finally, both plant diversity and structure had small (relative to the other regression paths tested) positive effects on competitor diversity. Although not experimentally tested here, plant diversity and structure may have increased competitor diversity through increased surface area for grazing or spatial refuge from both predation and competition for resources.

All predator groups present in this study are known predators of mosquito larvae and can be grouped as either actively searching predators (Dytiscid beetles and Cyclopoid 74 copepods) or resting ambush predators (Coenagrionidae and Aeshnidae). The latter group was more abundant in our experiment and likely contributed a more negative effect on mosquito larval abundance in both the linear regression and SEM models. Ambush predators related to these groups have been shown to consume mosquito larvae in laboratory and field studies. In particular, Coenagrionid damselflies consume Cx. Tarsalis mosquito larvae in laboratory studies and field trials (Miura and Takahashi 1988) as well as studies in larger anthropogenic wetlands and pastures (Reisen et al. 1989). Dragonflies in the family Libellulidae are well documented natural predators of Anopheles sinensis

Wiedemann (Diptera: Culicidae) larvae in laboratory trials and in rice fields (Mogi 2007).

Competitors were dominated by four main groups, Ostracoda, Daphnia,

Ephemeroptera, and freshwater snails. Also present but rare were nematodes, Chironomid midges, and Corixids. Competitor diversity was associated with substantive decreases in mosquito larvae (Table 1), but structural equation models showed that competitor diversity had a smaller direct effect on mosquito larvae compared to a stronger indirect effect through a positive effect on predator diversity (Figures 3–5). Both Daphnia and

Ostracods are active filter feeders that continuously move through the water column filtering out suspended matter and phytoplankton and occupy the same trophic level and feeding regime as filter-feeding Culex spp. larvae. As active mobile filter feeders both groups would be vulnerable to resting ambush predators present in this study. From this mesocosm experiment, we suggest that dynamics between competitors can be driven by indirect competition via predators. Ostracods and Daphnia likely served as alternative prey species sustaining resting ambush predator levels, making mosquito larvae more vulnerable to attack (i.e., apparent competition, Holt 1977, Holt and Lawton 1994). 75

In our study, mesocosms were created with live plants in a factorial design intended to simulate two levels of plant diversity, crossed with an additional structural element of dead willow branches. Although the addition of willow branches did not influence the diversity of these mesocosms, plant biomass and entropy-based measures of diversity did influence the latent variable that summarized complexity. Alternative studies have used plastic plant analogs to simulate aquatic macrophyte structural complexity in artificial mesocosms. For example, Srivastava (2006) used plastic leaves to manipulate habitat structure in artificial pitcher plant mesocosms. Similarly, Warfe and

Barmuta (2006) manipulated structural complexity by constructing mesocosms with aquatic macrophyte analogs, which included several homemade plant analogs and commercially available aquarium plants.

Similar to results reported here, Carlson et al. (2009) showed a strong positive effect of habitat age on both culicine competitor abundance and predator abundance resulting in an indirect negative effect of habitat age on larvae of the malaria vector,

Anopheles gambiae Giles (Diptera: Culicidae). These changes were due to the successive colonization of brick making pits by invertebrates and a resulting increase in diversity through time as pits became older and aquatic communities more complex after initial construction. The current study did not address habitat age, as all mesocosms were set up and harvested after 90 d. However, both predator and competitor diversities showed strong negative effects on Cx. tarsalis larval abundance, a competent vector of several common arboviruses, including West Nile virus. Based on results reported here, as well as the Carlson et al. (2009) results, we suggest that future studies should incorporate 76 long-term manipulations of aquatic macrophyte communities and also test for interactions between habitat complexity and habitat age.

Our results are relevant to mosquito populations in many different parts of the world, as aquatic habitat created by anthropogenic structures are commonly used by mosquitoes. Mesocosms also provide a convenient alternative to naturally occurring habitats as study systems for mosquito community dynamics because the experimental units can be controlled more consistently. In urban environments, sources of aquatic habitat, such as old tires and other urban debris, are often mesocosm-sized, or smaller.

While applied strategies for mosquito control often emphasize source reduction or elimination of small urban water sources such as buckets or tires, suitable mosquito habitats are provided by a wide variety of permanent structures, ranging from mesocosm- sized storm drains to large wetland systems constructed for stormwater retention, water quality treatment, and managed habitats created for the enhancement of biodiversity within urban-rural landscapes. For many of these systems, source elimination may not be feasible and a better understanding and different management objectives for the biodiversity and structure of aquatic systems may serve as an important component of effective mosquito control strategies. Vegetation management through mechanical or chemical controls may be the most straightforward approach to the manipulation of permanent aquatic ecosystems for the reduction of mosquito populations; however, the effects on individual mosquito species can be unpredictable and are often context-specific o individual systems. For example, mowing emergent vegetation in stormwater wetlands during peak growing seasons can increase West Nile virus infection rates in vector species (Mackay et al. 2016), while increased salt grass cover in seasonal salt marshes 77 increases mosquito abundance but also overall macroinvertebrate diversity (Szalay and

Resh 2000). Overall management practices should consider individual mosquito species traits, wetland design, macroinvertebrate community composition, and regional land-use practices (Hanford et al. 2019).

After decades of efforts and ideas for merging applied and basic research in plant–animal interactions (Gliessman 1990, Dyer and Gentry 1999, Courchamp et al.

2015), it is still difficult to see any measurable utility coming from merging paradigms, consensus, or opinions in basic research (Dyer and Letourneau 2013). The applied issue of mosquito control should be more tractable, especially in urban environments that are dominated by simpler, manipulable ecosystems, or at the urban-rural interface, where expanding developments create simple aquatic habitats, similar to the mesocosms that are a staple of basic ecological research. For example, the idea of ‘diversity cascades’ (e.g.,

Polis 1994, Siemann 1998, Eveleigh et al. 2007) is complex and it is difficult to generate clear predictions of how diversity at one trophic level affects diversity at other trophic levels in more complex systems. But the mesocosms used here have uncovered some clear diversity cascades (sensu Dyer and Letourneau 2013) that are relevant to the small aquatic ecosystems that are important habitat for mosquitoes across terrestrial landscapes. 78

Aquatic Complexity and Interaction Diversity Reduce Larval Mosquito Abundance

Abstract

The successful control of mosquito populations relies on effective surveillance of larval mosquito habitats, identification of larval mosquito populations, and assessment of potential non-target effects of any specific abatement action. Larval sources are often identified by proximity to adult mosquito activity and by that species’ habitat preferences. Common but not universal to mosquito habitats are the presence of plants, algae, phytoplankton, and a community of non-mosquito invertebrates. This community can consist of antagonistic enemy species as well as controphic species that compete with mosquito larvae for resources. Their influences on larval mosquitoes (positive or negative) may result from the diversity of interactions forming the full community of plankton, competitors, and predators. Aquatic macrophytes provide structural elements and environmental heterogeneity which provide niche space for invertebrate diversity and interaction diversity. This study used observational data collected from natural field habitats and explores the role of aquatic macrophytes in structuring invertebrate communities containing mosquito larvae. Plant diversity positively effects predator and competitor diversity and competitor diversity in particular had a negative effect on larval mosquito abundance. Plant, invertebrate, and interaction diversity of aquatic systems contribute in a variety of ways to the abundance of larval mosquitoes. Simple measurements of plant diversity and observations of predator and competitor groups are necessary tools for sound mosquito abatement programs. Increasing plant diversity would be a complementary method to reduce mosquito populations. Protection of biodiversity and discouragement of plant monocultures in managed wetlands will 79 increase the potential interaction diversity of systems and reduce reliance on chemical and physical control of mosquito populations.

Introduction

Early detection and population monitoring of pests is of primary importance for

IPM strategies in managed and natural landscapes (Myers et al. 1998, Myers et al. 2000,

Brockerhoff et al. 2006, Ehler 1998). Monitoring tools for insect pests are nearly as diverse as the range of insect groups of concern and those for monitoring mosquito

(Diptera: Culicidae) populations are no exception, with a myriad of tools and techniques for monitoring all stages of development from egg through adult and specialized tools developed to detect specific genera and individual species (reviewed by Service 1976).

One reason that a wide array of tools is necessary is that culicids are a diverse group of flies that can have important consequences for human health and quality of life. In large enough quantity they can be a social and economic nuisance, and many species are responsible for vectoring important human and animal diseases.

Many challenges for understanding mosquito-borne disease dynamics result from the extensive diversity of larval mosquito habitats, which differ in factors such as size, complexity, natural versus anthropogenic, freshwater versus saltwater, and permanent versus temporary. Examples of large habitats include shallow ponds, freshwater wetlands and marshes, lake edges, saltwater marshes, and flowing or stagnant streams.

Small natural containers include phytotelmata (e.g., treeholes, palms, flower bracts, and bromeliads) and from animal origin (e.g., crab holes and shells). Anthropogenic containers include car tires, water cisterns, buckets, cans, rain gutters, and flower vases. 80

More comprehensive examples along with species associations are reviewed by Laird

(1988) and Rueda (2008).

Despite these challenges, understanding larval mosquito habitats is critical for understanding and predicting adult mosquito population dynamics. Any good approach to understanding population dynamics of mosquitoes will include some modeling component, with (not mutually exclusive) approaches including deterministic and stochastic analytical models, individual based models, spatial models, discrete models, and simulation models, among others. For these and other modeling approaches, including conceptual models, good parameter estimates related to the larval stages are necessary. Mosquitoes are engaged in a diverse mix of ecological interactions as developing larvae. Variation in habitat types, nutrient availability, hydroperiod, and species interactions can all influence important developmental traits that ultimately affect adult populations. For example, competition for resources can influence larval abundance (Murrell and Juliano 2008, Kroeger et al. 2013), mortality (Knight et al.

2004), adult size (Stav et al. 2005, Hardstone et al. 2012), and development time (Stav et al. 2005, Hardstone et al. 2012). Predators can also affect larval abundance (Fry et al.

1994, Schreiber 1996, Dieng et al. 2002, Nyamah et al. 2011, Soumare and Cilek 2011), mortality (Marten et al. 1994, Marten et al. 2000, Dieng et al. 2002, Calliari et al. 2003,

Soumare 2004, Quiroz-Martinez et al. 2005, Pernia et al. 2007), development time

(Hechtel et al. 1997, Nannini and Juliano 1997, Knight et al. 2004, Bond et al. 2005,

Griswold and Lounibos 2005, Griswold and Lounibos 2006, Stav et al. 2005, Fisher et al. 2012, Ohba et al. 2012, Roberts 2012), larval feeding behaviors (Juliano and

Reminger 1992, Hechtel et al. 1997, Gimonneau et al. 2010, Roberts 2014), and adult 81 size (Bond et al. 2005, Stav et al. 2005, Griswold and Lounibos 2006, Fisher et al. 2012,

Ohba et al. 2012). As a guild, mosquito predators have been the subject of a great deal of ecological research, especially regarding their potential in classical biological control strategies (Chandra et al. 2008, Focks 2007. Kumar and Kwang 2006. Lacey and Orr

1994, Marten et al. 1994, Mogi 2007, Quiroz-Martinez and Rodrigues-Castro 2007).

Although there are fewer empirical examples, it has been shown that habitats with greater competitor and predator diversity will tend to be less dominated by large numbers of mosquitoes or primary disease vectors (Chase and Knight 2003, Carlson et al. 2009,

Lumpkin et al. 2020).

This existing research on mosquito larval population and community ecology provides some support for the idea that protecting and enhancing biodiversity of aquatic systems can be an important component of ecosystem stability and reducing the adverse effects of large mosquito populations. Increased diversity of primary producers

(macrophyte diversity) can enhance biodiversity of successively higher trophic groups in terrestrial systems (Dyer and Letourneau 2013) and can increase nutrient utilization

(Tilman et al. 1996) and stability of systems to endure disturbance events (Tilman and

Downing 1994) in terrestrial ecosystems. Although macrophytes are common in aquatic systems, primary productivity is dominated by microscopic algae, periphyton, and phytoplankton, resulting in trophic biomass increasing toward intermediate trophic levels

(i.e., primary consumers and their predators) (Shurin et al. 2006). These nutrient sources are consumed by detritivores, grazers, and filter-feeding invertebrates, including many species of mosquito larvae (Merritt 1992). Aquatic macrophytes however provide structural elements and surface area for periphyton growth (Hutchinson 1957, Wetzel 82

1983, Gregg and Rose 1985, Brown et al. 1988) and can stabilize nutrient uptake in eutrophic freshwater systems (Brix 1997).

An important theoretical determinant of species diversity is environmental heterogeneity, which has a long and complex history in ecological research as a factor influencing species diversity (MacArthur and MacArthur 1961, Pianka 1967). A more important accompaniment to species diversity is the diversity of interactions and potential interaction diversity (Janzen 1974, Thompson 1996, Dyer et al. 2010). Interaction diversity is defined here as the diversity of direct and indirect interaction links between species and their relative abundances, and the interaction types can include beneficial, harmful, mutual, and multi-trophic interactions (Janzen 1974, Thompson 1996, Dyer et al. 2010). Environmental heterogeneity is not defined by a single ecosystem parameter, but includes a range of resource types (e.g., nutrient availability, soil characteristics, elevational gradients, inert physical structure, and plant complexity (Stein and Kreft

2015). As in terrestrial systems, environmental heterogeneity could provide ecosystem- stabilizing mechanisms resulting in more stable invertebrate communities. These relationships are also relevant to spatial variance in networks, where there can be high turnover in interaction diversity from one site to another, given sufficient habitat heterogeneity (Dell et al. 2019).

In this study we explore environmental heterogeneity in the form of aquatic macrophyte complexity resulting from plant diversity and the role that plant diversity and complexity play in structuring freshwater aquatic invertebrate communities and larval mosquito abundance. Our main hypothesis is that greater plant diversity will support a greater number of aquatic invertebrates. Secondly, we expected that greater overall 83 invertebrate diversity will result from more predator and competitor species. Finally, mosquito abundance should be lower in less structurally complex, more diverse habitats.

Methods

Study System

Survey sites (32 total) were identified from a variety of aquatic systems including wetlands, ponds, lake margins, river marsh areas, and irrigated pastures. Sites were selected by their general appearance for the likelihood of harboring aquatic invertebrates and mosquito larvae. Most aquatic sites were located in the Truckee Meadows region of southern Washoe County, Nevada and the surrounding areas in Northern California with a small number of surveys conducted in central Nevada and Southern Arizona (Figure 1

Map). Sample site surveys were conducted along the shorelines of aquatic sites within the typical range of emergent or submerged vegetation, open bodies of water or areas with visible water movement were not sampled (e.g., rivers, creeks, open lake areas).

Site surveys were conducted from May through October from 2011 to 2015.

Survey Plots

Individual survey plots (58 total) consisted of 1-meter square sites delineated using a square measuring frame. Depending on the size of each aquatic site, one to five individual subplots were sampled for each site. Subplots were separated by using a random number generator to select a separation by meters. For sites with very shallow transitions toward the center, subplots were oriented from the shoreline towards the center of the site. For sites that were either narrowly shaped or had steep transitions in depth, subplots were oriented along the shoreline. For each subplot measurements were taken including water depth at the center of the plot, individual plant species height, 84 percent plot cover in 5% increments, stem width, and estimated density by counting the

2 2 2 2 number of stems within a single reference square (1 cm❑, 25 cm❑, 100 cm❑, 625 cm❑ ,

2 2 (1/16 plot), 2,500 cm❑(1/4 plot), and 10,000 cm❑(entire plot) (figure2). Unique plant density counts were used to estimate total individual stems, total stem area, and total stem volume per plot. Invertebrates were collected by taking two sweeps through each plot using a round 25 cm 500-micron aquatic sweep net. Invertebrates were stored in 70 % ethanol and later identified to the family level and given a morphospecies identification.

Statistical Analysis

Total invertebrate diversities for each plot were calculated as morphospecies

s 2 diversity with Inverse Simpson's diversity (D=1 /∑ pi ) using the total estimated plant i =1 volumes for each morphospecies. Separate morphospecies diversities were also calculated for families that are potential predators of mosquito larvae (predators) and for families which generally occupy the same trophic group as mosquito larvae, feeding on small plankton, algae, and periphyton (competitors).

All data analysis was conducted using JAGS version 4.3.0 (Plummer 2003).

Models were tested using Bayesian linear regression (R package runjags, Denwood 2016) and Bayesian structural equation models (R package blavaan, Merkle and Rosseel 2018).

Uninformative Gaussian priors were utilized for all Beta terms with a mean of 0 and precision of 0.0001 (βᵢ ~ N(0, 0.0001). Models were run with 3 simulation chains, a burn in of 8,000 simulations, and a total of 30,000 simulations. Diversity indices were modeled in their existing form while individual mosquito counts were modeled assuming 85 a negative binomial distribution for Bayesian linear regressions and log transformed larvae=ln ( mosquitolarvae+1) for Bayesian structural equation models.

Results

A total of 32 aquatic sites were surveyed resulting in a total of 58 individual site survey plots, with an average number of subplots per site of 1.844. Site habitat types included small to large stormwater retention systems, ornamental ponds, river marshes, irrigated pastures, wetlands, snowmelt ponds, and still water lakeshore areas. Irrigated pastures showed the highest levels of plant diversity while storm water infrastructure the lowest. Invertebrate diversity was highest in lakeshore, pond, and stormwater systems

(figure 3). The most commonly collected mosquitoes were in the genus Culex and a small proportion of Culiseta spp. and Aedes spp. were collected.

The full results from the Bayesian linear regression models including regressions for each site and intercept are shown in table 1. Plant diversity had positive effects on total invertebrate diversity (posterior mean Beta (β) = 0.780, credibility interval (CI) = -

0.674 – 2.177), predator diversity (β = 1.349, CI = -0.0623 – 2.585), competitor diversity

(β = 0.380, CI = -0.549 – 1.272), and a negative effect on mosquito abundance (β = -

4.0536, CI = -9.464 – 0.860). Total invertebrate diversity showed a positive effect on mosquito abundance (β = 0.727, CI = -1.170 – 2.500), as did competitor diversity (β =

0.398, CI = -1.475 – 2.290), while predator diversity had a smaller negative effect on mosquito abundance (β = -0.0670, CI = -1.929 – 1.860). Nested site effects were below

0.10 (β < 0.100) for regressions with plant diversity on total invertebrate diversity, predator diversity, and competitor diversity, and with predator diversity on mosquito abundance; but were larger for regressions with mosquito abundance and total 86 invertebrate diversity (site, β = 0.270, CI = -0.134 – 0.648) and competitor diversity (site,

β = 0.243, CI = -0.0218 – 0.511), (table 1). This suggests that for these two examples site identity of repeated within sites subplots was important for determining mosquito abundance. For example, the addition of each subplot resulted in 0.243-0.270 additional mosquito larvae.

Figure 1. Most sites sampled were in western Nevada and neighboring counties in California. Several sites were in central Nevada, and extreme northwestern Nevada and southwestern Arizona. 87

Figure 2. Plant diversity estimates were conducted within 1-meter square plots (bottom right). The blue squares represent size classes used to count individual plant stem densities, higher-density plants with narrower stems were counted in smaller square areas than larger less-densely clustered stems. Area counts were then used to estimate total plant stems within each plot based on density and estimated percent of plot cover.

The Bayesian structural equation models supported hypotheses of more complex causal relationships between variables. Most models supported positive parameter estimates for predator diversity effects on competitor diversity and mosquito abundance, plant diversity effects on competitor and predator diversities, and competitor diversity effects on mosquito abundance (Figures 4 - 6). The most complete model (figure 7) yielded coefficients supporting hypotheses that plant diversity has a positive effect on predator diversity (β = 0.169, CI = -0.193 – 0.954), a positive effect on competitor diversity (β = 0.141, CI = -0.181 – 0.635), predator diversity has a positive effect on mosquito abundance (β = 0.103, CI = -1.005 – 2.449), competitor diversity has a negative effect on mosquito abundance (β = -0.226, CI = -4.688 – 0.221), and a relatively large site effect on mosquito abundance (β = 0.328, CI = 0.089 – 0.554). 88

Table 1. The results for the Bayesian linear regression models are shown below including the mean posterior beta coefficients, 95 % credibility intervals and posterior standard deviations. Below each estimate are the effects of site name as a covariate and the intercepts. Site shows stronger coefficient values for regressions on larval mosquito abundance.

Response Variable Predictor Posterior 95% CI Posterior Mean SD (beta)

Invertebrate Plant Diversity 0.780 -0.674 – 2.177 0.734 Diversity Site 0.0588 -0.0411 – 0.159 0.0517 Intercept 1.670 -0.448 – 3.754 1.0821

Predator Diversity Plant Diversity 1.349 -0.0623 – 2.585 0.649 Site 0.0831 -0.00532 – 0.172 0.0548 Intercept -0.0404 -1.914 – 1.804 0.957

Competitor Plant Diversity 0.380 -0.549 – 1.272 0.469 Diversity Site 0.0277 -0.0361 – 0.0918 0.0330 Intercept 1.109 -0.244 – 2.441 0.691

Mosquito Larvae Plant Diversity -4.0536 -9.464 – 0.860 2.632 Site 0.0800 -0.203 – 0.381 0.150 Intercept 3.360 -2.631 – 9.564 3.108

Mosquito Larvae Invertebrate 0.727 -1.170 – 2.500 0.975 Diversity Site 0.270 -0.134 – 0.648 0.208 Intercept -3.381 -9.190 – 3.076 3.250

Mosquito Larvae Predator -0.0670 -1.929 – 1.860 0.949 Diversity Site 0.0564 -0.216 – 0.305 0.130 Intercept -0.833 -4.841 – 3.339 2.076

Mosquito Larvae Competitor 0.398 -1.475 – 2.290 0.946 Diversity Site 0.243 0.0218 – 0.511 0.124 Intercept -2.036 -6.169 – 1.842 2.0292 89

Discussion

This study was designed to evaluate the role of aquatic plant diversity in structuring habitat complexity and invertebrate community diversity using simple observational measurements, including water depth, plant density, and plant diversity

(using plant morphospecies). Plant diversity was measured using the estimated spatial volume of plants in the water column, and was an effective predictor of predator and competitor diversities and mosquito abundance. The data presented here supported our first hypothesis that increased plant diversity would positively influence invertebrate diversity as well as both predator and competitor diversities and those effects were consistent across multiple plant species. Plant diversity did not have a direct positive effect on mosquito abundance and the structural equation models that provided estimates of indirect effects of plant diversity on mosquito abundance did not yield strong standardized path coefficients compared to other models (figure 6). Finally, the results supported hypotheses that competitor diversity had a direct negative effect on mosquito abundance while predator diversity had a much smaller direct positive effect. The pattern of high variance components for the nested site effect mosquito abundance suggests that there were unexplained factors influencing mosquito abundance. One explanation may be that some sites may or may not have been treated as part of existing mosquito abatement programs and this study did not differentiate between treated or untreated sites. 90

Figure 3. The mean (+/- 1 SE) total plant and invertebrate diversity (inverse Simpson’s) are shown for each distinct habitat type.

Strong competitor and predator populations resulting from habitat age (Carlson et al. 2009) or habitat permanence (Chase and Knight 2003) can reduce larval and adult mosquito densities in both natural and anthropogenic habitats. Protection from vector species and vector-borne diseases has received considerable attention in the literature as a 91 welcome effect of enrichments or protection of biodiversity, including hypotheses of zooprophalaxis and the dilution effect hypothesis. These hypotheses focus on the ideas that a greater diversity of hosts, some of which are less competent reservoirs, may ultimately reduce the dominance of specific infected vectors or reduce the likelihood of them finding human or domestic animal hosts (Johnson et al. 2009, Randolph and

Dobson 2012). For example, the presence of livestock has been implicated in reducing the risk of human malaria in some parts of the world by providing alternate hosts for

Anopheline mosquitoes (Dobson et al. 2006); however, their presence has also been shown to produce the opposite effect (Bouma and Rowland 1995). These concepts have gained the greatest support in the literature in the ecology of Lyme borreliosis (Randolph and Dobson 2012) where it is suggested that the diversity of less-competent host- reservoir populations of rodents leads to a lower proportion of Lyme disease infected ticks (LoGiudice et al. 2003, Keesing et al. 2006). It has, however, been shown that even with a lower proportion of infected ticks, their overall abundance may offset the reduced proportion of infected ticks, resulting in greater overall numbers of infected ticks and a greater risk for Lyme exposures (Randolph and Dobson 2012). It is clear that a great deal of additional research is needed on community diversity and how it affects both insect vectors and the diseases they host. 92

Figure 4. Both predator and competitor diversities (inverse Simpson’s) on larval mosquito abundance. Arrows represent positive effects and bullets represent negative effects.

The plant diversity metric used in this study is also a measure of environmental heterogeneity (plant complexity) in that it quantifies the physical space occupied by plant species. As predicted by theories of environmental heterogeneity, plant diversity showed positive effects on both competitor and predator diversities. A similar mosquito study used experiential mesocosms to simulate levels of plant diversity and habitat complexity effects on invertebrates, showing that plant diversity and complexity increased predator diversity, reducing larval mosquito abundance (Lumpkin et al. 2020). In some studies, environmental heterogeneity across small islands creates greater beta diversity, and in this case island size is less important than heterogeneity across islands (Jinliang et al.

2018). In other systems however, habitat complexity has been shown to correlate with 93 invertebrate abundance but not with diversity (McAbendroth et al. 2005). A meta- analysis on spatial patterns of species diversity showed that environmental heterogeneity was less important in influencing diversity than climate and primary productivity measures (Field et al. 2009). A significant challenge in comparing the effects of heterogeneity between studies is the overwhelming number of measures and methods used to describe environmental heterogeneity (Stein and Kreft 2015). The few empirical studies examining these relationships with mosquitoes (e.g., the present study, Carlson et al. 2009, Lumpkin et al. 2020), suggest that environmental heterogeneity are potentially very important for maintaining diverse aquatic communities and lower mosquito densities.

What is the role of interaction diversity in reducing mosquito abundances? Is it the same as overall arthropod community diversity? We quantified interaction diversity here between plants, predators, competitors, and mosquito larvae. Because we did not find a meaningful relationship between competitors and predators with these data, it is unlikely that any measure of interaction diversity will be more predictive of mosquito abundance than examining plant diversity and each arthropod guild individually. The actual relationships between species may however be obscured through a combination of direct and indirect effects, where simple measurements of group diversities (predators and competitors) may not reveal more complex interactions. Measurements of plant, predator, and competitor diversity do provide valuable insight for the possibility of potential interaction diversity within systems. For example, results from the structural equation models suggest that each additional plant species an additional 0.147 competitors would be present resulting in 0.200 fewer mosquito larvae (figure 7). These 94 diversity of interactions may ultimately reduce mosquito abundance but should be examined more formally via models (see earlier sections of this dissertation) or empirical approaches.

Figure 5. The direct effects of competitor diversity (inverse Simpson’s) on predator diversity and predator diversity on mosquito abundance. The dotted line shows the indirect effect of competitor diversity on mosquito larvae. The direct effect of predator diversity on competitor diversity and competitor diversity on mosquito larvae with an indirect effect of predator diversity on mosquito abundance. 95

Figure 6. Similar to figure 5, but with direct effects of plant diversity (inverse Simpson’s) on predator diversity and competitor diversity. Plant diversity shows positive effects on both predator and competitor diversities. Competitor diversity shows a negative effect on mosquito abundance and predator diversity shows a relatively weak direct positive effect on mosquito abundance.

Aquatic systems that provide habitat for mosquito larvae are perhaps as diverse as the species of mosquitoes and other invertebrates that colonize them. In constructed and natural wetland systems, land managers have ample tools for influencing diversity and reducing mosquito abundance that do not rely entirely on chemical control methods. For example, reducing plant monocultures and overall vegetation density has been shown to 96 reduce adult mosquito emergence in managed wetlands (Jiannino and Walton 2004,

Willot 2004). Increasing plant diversity and encouraging greater interaction diversity within aquatic systems may provide an important additional tool for the control of mosquito populations. Larval mosquito surveillance programs should incorporate simple plant and invertebrate diversity counts aimed at estimating the potential interaction diversity of aquatic systems and how those may reduce adult mosquito abundance.

Figure 7. The final structural equation model showing positive effects of plant diversity on both predator and competitor diversities. Competitor diversity shows a negative effect on mosquito larvae. Site identity shows a positive covariance with mosquito larvae. 97

Conclusions

The results of this dissertation contribute to our knowledge of mosquito ecology in several important respects. In sum, I have shown that: 1) a variety of species and guilds of natural enemies can effectively reduce larval mosquito abundance, 2) increases in both plant diversity and complexity can increase predator and competitor diversities, 3) increases in both predator and competitor diversities can ultimately reduce larval mosquito abundance, 4) interaction diversity is an important attribute of mosquito larval ecology, and 5) simulated tri-trophic networks indicate that unique interactions may accumulate faster than unique species. Each of these contributions is discussed below.

The meta-analysis provides a quantified summary of the strengths of different taxa that have been tested experimentally in the laboratory and field to negatively impact mosquito abundance. The results suggest that many taxa are capable of negatively affecting larval and adult mosquitoes. An unexpected result was that field-conducted experiments resulted in stronger negative effects than more controlled laboratory- conducted experiments. Experiments conducted in either natural or artificial containers produced stronger negative effects versus mesocosms. These results suggest that continued field-conducted use of biological control agents should be explored for the reduction of larval and adult mosquito abundance. Modern biological control efforts should aim to utilize locally native species whenever possible.

While the simulated tri-trophic networks did not specifically address interaction diversity in the context of larval mosquito communities, this project is nonetheless applicable to mosquito ecology, especially in light of other results reported here suggesting that diverse interactions are likely to have substantive effects on mosquito 98 densities. Even basic methodologies can be modified based on this insight. For example, perhaps the most essential tool for applied mosquito ecology research and surveillance is the basic and universally adopted larval mosquito dipper. The user searches out mosquito larvae by taking repeated dips of water from suspected mosquito sources. In doing so they will often collect a wide variety of other invertebrate species. With relatively simple training individuals can learn to recognize predator and competitor groups within these basic samples providing opportunities to estimate the potential interaction diversity with the system. This can at least provide a qualitative knowledge of the relative vulnerability of larval mosquitoes due to potential effects of interaction diversity among predators, competitors, and larvae.

The mesocosm experiments add value to understanding aquatic communities associated with mosquitoes by showing that greater plant diversity and structure increase predator diversity and that increases in both competitor and predator diversities reduce larval mosquito abundance. Predator diversity exerted a direct negative effect on mosquitoes, while competitor diversity caused indirect negative effects through positive effects on predator diversity. This interaction suggests an apparent competition where invertebrates in the same trophic group as mosquito larvae provided an alternative prey source to mosquito predators, resulting in an enhanced predator diversity and increased risk to mosquito larvae. These relationships are also applicable to land managers and mosquito control agencies, where the management of plant and invertebrate diversity can be used as an additional tool for controlling dominant pest species. In addition to this result, the meta-analysis suggests that mesocosm studies produce smaller effect sizes, so it’s possible that these interactions are even stronger at larger scales. 99

In fact, I examined interactions at larger scales: the observational study was explicitly designed to expand on the mesocosm concepts and determine the effects of plant diversity and structural complexity on a larger spatial scale and over a wider range of habitat types and sizes. The results showed a slightly ambiguous relationship between predators and competitors. Some models showed increases in predator diversity having a positive effect on competitor diversity and a smaller but still positive effect on mosquito abundance. Regardless, there were clear direct positive effects of plant diversity on both predator and competitor diversities. Predator diversity had small effects (with high variation) on mosquito abundance, but competitor diversity showed a strong negative effect on mosquito abundance. These results are also valuable for agencies in that the enhancement of plant diversity and structural complexity can increase total invertebrate diversity of aquatic systems.

This full dissertation revealed many interesting aspects of how mosquito larvae are affected by plant diversity, habitat complexity, predator and competitor diversities, and interaction diversity. An additional and important component structuring these relationships and aquatic communities is the types and diversity of phytoplankton, periphyton, and detritus involved in these systems. For my dissertation research I did not have the ability to effectively study this trophic level. I will continue to study the fascinating world of mosquito ecology and hope to examine interaction diversity of entire aquatic systems and including aquatic autotrophs; both approaches should provide a much richer and more complete picture of how a broader range of interactions effect mosquito larvae. This dissertation does, however, provide practical concepts and tools to help aid land managers, conservationists, government agencies, and mosquito control 100 agencies in reducing the impact of nuisance and vector mosquito species through the enhancement of plant diversity, habitat structure, predator diversity, and competitor diversity. 101

Literature Cited

Alexander, M. E., J. T. A. Dick, N. E. O’Connor, N. R. Haddaway, and K. D. Farnsworth. 2012. Functional responses of the intertidal amphipod Echinogammarus marinus: effects of prey supply, model selection and habitat complexity. Mar. Ecol. Prog. Ser. 468:191–202.

Alexander, M. E., H. Kaiser, O. L. F. Weyl, and J. T. A. Dick. 2015. Habitat simplification increases the impact of a freshwater invasive fish. Environ. Biol. Fishes 98: 477–486.

Altieri, M. A., and D. K. Letourneau. 1982. Vegetation management and biological control in agroecosystems. Crop Prot. 1:405–430.

Altieri, M., and C. Nicholls. 2004. Biodiversity and pest management in agroecosystems, 2nd ed. Food Products Press, New York, NY.

Alto, B. W., L. P. Lounibos, S. Higgs, and S. A. Juliano. 2005. Larval competition differentially affects arbovirus infection in Aedes mosquitoes. Ecology 86:3279–3288.

Andow, D. A. 1991. Vegetational diversity and arthropod population response. Annual Review of Entomology 36:561-586.

Ballantyne, G., K. C. Baldock, P. G. Willmer. 2015. Constructing more informative plant-pollinator networks: visitation and pollen deposition networks in a heathland plant community. Philosophical Transactions of the Royal Society B: Biological Sciences 282: 2015-1130.

Barbosa, P. A. (eds.) 1998. Conservation biological control. Academic Press, San Diego, CA.

Barrera, R. 1996. Competition and resistance to starvation in larvae of container‐ inhabiting Aedes mosquitoes. Ecol. Entomology. 21:117–127.

Barrios-O’Neil, D., J. T. A. Dick, M. C. Emmerson, A. Ricciardi, and H. J. Maclsaac. 2015.Predator-free space, functional responses and biological invasions. Funct. Ecol. 29:377–384.

Batzer, D. P., and V. H. Resh. 1992. Wetland management strategies that enhance waterfowl habitats can also control mosquitoes. J. Am. Mosq. Contr. Assoc. 8:117–125.

Bay, E. C. 1974. Predator-prey relationships among aquatic insects. Annual Review of Entomology 19:441-453. 102

Beaver, R. A. 1985. Geographical variation in food web structure in Nepenthes pitcher plants. Ecological Entomology 10:241-248.

Beckerman, A. P., O. L. Petchey, P. H. Warren. 2006. Foraging biology predicts food web complexity. Proceedings of the National Academy of Sciences 103:13745-13749.

Becnel, J. J. and M. A. Johnson. 2000. Impact of Edhazardia aedis (microsporidia) on a seminatural population of Aedes aegypti (Diptera: Culicidae). Biological Control 18:39- 48.

Bentley, M. D., and J. F. Day. 1989. Chemical ecology and behavioral aspects of mosquito oviposition. Annu. Rev. Entomol. 34:401–421.

Bianchi, F. J. J. A., C. J. H. Booij, and T. Tscharntke. 2006. Sustainable pest regulation in agricultural landscapes: a review on landscape composition, biodiversity and natural pest control. Proceedings of the Royal Society 273:1715–1727.

Blaustein, L. 1992. Larvivorous fishes fail to control mosquitoes in experimental rice plots. Hydrobiologia 232:219-232.

Blaustein, L. and J. M. Chase. 2007. Interactions between mosquito larvae and species that share the same trophic level. Annual Review of Entomology 52:489-507.

Blüthgen, N. 2010. Why network analysis is often disconnected from community ecology: a critique and an ecologist's guide. Basic and Applied Ecology 11:185-195.

Bond, J. G., J. I. Arredondo-Jimenez, M. H. Rodrigues, H. Quiroz-Martinez, T. Williams. 2005. Oviposition habitat selection for a predator refuge and food source in a mosquito. Ecological Entomology 30:255-263.

Bonsall, M. B. and R. D. Holt. 2010. Apparent competition and vector-host interactions. Israel Journal of Ecology and Evolution 56:393-416.

Bouma, M. J. and M. Rowland. 1995. Failure of passive zooprophylaxis: cattle ownership in Pakistan is associated with higher prevalence of malaria. Transactions of the Royal Society of Tropical Medicine and Hygiene 89:351-353.

Bowatte, G., P. Perera, G. Senevirathne, S. Meegaskumbura, M. Meegaskumbura. 2013. Tadpoles ad dengue mosquito (Aedes aegypti) egg predators. Biological Control 67:469- 474.

Braks, M. A. H., N. A. Honório, L. P. Lounibos, R. Lourenço-de-Oliveira, and S. A. Juliano. 2004. Interspecific competition between two invasive species of container mosquitoes, Aedes aegypti and Aedes albopictus (Diptera: Culicidae), in Brazil. Ann. Entomol. Soc. Am. 97:130–139. 103

Brix, H. 1997. Do macrophytes play a role in constructed treatment wetlands? Water Science Technology 35(5):11-17.

Brockerhoff, E. G., D. C. Jones, M. O. Kimberley, D. M. Suckling, T. Donaldson. 2006. Nationwide survey for invasive wood-boring and bark beetles (Coleoptera) using traps baited with pheromones and kairomones. Forest Ecology and Management 228:234-240.

Brown, C. L., T. P. Poe, J. R. P. French III, D. W. Schloesser. 1988. Relationships of phytomacrofauna to surface area in naturally occurring macrophyte stands. Freshwater Science 7(2):129-139.

Brown, M. D. 1996. Evaluation of Mesocyclops aspericornis (Cyclopoida: Cyclopidae) and Toxorhynchites speciosus as integrated predators of mosquitoes in tire habitats. Journal of the American Mosquito Control Association 12(3):414-420.

Burkle, L. A. and T. M. Knight. 2012. Shifts in pollinator composition and behavior cause slow interaction accumulation with area in plant-pollinator networks. Ecology 93:2329-2335.

Cardoso, H. C. B., B. Queiroz da Silva, T. Brandao de Assis, L. C. S. Lopez. 2013. Effects of predation by the copepod Mesocyclops ogunnus of the sex ratios of mosquito Aedes albopictus. Hydrobiologia 705:55-61.

Carlson, J. C., L. A. Dyer, F. X. Omlin, J. C. Beier. 2009. Diversity cascades and malaria vectors. Journal of Medical Entomology 46(6):460-464.

Calliari, D., K. Sanz, M. Martinez, G. Cervetto, M. Gomez, C. Basso. 2003. Comparison of the predation rate of freshwater cyclopoid copepod species on larvae of the mosquito Culex pipiens. Medical and Veterinary Entomology 17:339-342.

Cardinale, B. J., M. A. Palmer, and S. L. Collins. 2002. Species diversity enhances ecosystem functioning through interspecific facilitation. Nature. 415: 426–429.

Chacoff, N. P., D. P. Vazquez, S. B. Lomascolo, E. L. Stevani, J. Dorado, B. Padron. 2012. Evaluating sampling completeness in a desert plant-pollinator network. Journal of Animal Ecology 81:190-200.

Chandra, G., I. Bhattacharjee, S. N. Chatterjee, A. Ghosh. 2008. Mosquito control by larvivorous fish. Indian Journal of Medical Research 127:13-27.

Chao, A. 1984. Nonparametric estimation of the number of classes in a population. Scandinavian Journal of Statistics 265-270. 104

Chapman, H. C. 1974. Biological control of mosquito larvae. Annual Review of Entomology 19:33-59.

Chase, J. M. and T. M. Knight. 2003. Drought-induced mosquito outbreaks in wetlands. Ecology Letters 6:1017-1024.

Chesson, J. 1984. Effects of notonectids (Hemiptera: Notonectidae) on mosquitoes (Diptera: Culicidae). Environmental Entomology 13:531-538.

Cheung, M. W-L. 2014. Modeling dependent effect sizes with three-level meta-analysis: A structural equation modeling approach. Psychological Methods 19(2):211-229.

Collins, F. H. and R. K. Washino. 1985. Biological control of mosquitoes. Journal of The American Mosquito Control Association. 6(6):25-41.

Cornell, H. V., J. H. Lawton. 1992. Species interactions, local and regional processes, and limits to the richness of ecological communities: a theoretical perspective. Journal of Animal Ecology 1-12.

Courchamp, F., J. A. Dunne, Y. Le Maho, R. M. May, C. Thébaud, and M. E. Hochberg. 2015. Fundamental ecology is fundamental. Trends Ecol. Evol. 30: 9–16.

Cuthbert, R. N., J. T. A. Dick, and A. Callaghan. 2018. Interspecific variation, habitat complexity and ovipositional responses modulate the efficacy of cyclopoid copepods in disease vector control. Biol. Control 121: 80–87.

Dáttilo, W., L. Dyer. 2014. Canopy openness enhances diversity of ant-plant interactions in the Brazilian Amazon rain forest. Biotropica 46:712-719.

Darwin, C. 1859. On the origin of species by means of natural selection, or, the preservation of favoured races in the struggle for life. 15th ed. London: J. Murray.

Dassou, A. G., and P. Tixier. 2016. Response of pest control by generalist predators to local-scale plant diversity: a meta-analysis. Ecol. Evol. 6: 1143–1153.

De Barjac, H. 1978. Un nouveau candidat a la lute biologique contre les moustiques : Bacillus thuringiensis var. israelensis. Entomophaga 23(4):309-319.

Dell, J., D. Salcido, W. Lumpkin, L. A. Richards. S. Pokswinski, E. L. Loudermilk, et al. 2019. Interaction diversity maintains resiliencey in a frequently disturbed ecosystem. Frontiers in Ecology and Evolution 7:145.

Denwood, M. J. 2016. Runjags: an R package providing interface utilities, model templates, parallel computing methods and additional distributions for MCMC models in JAGS. J. Stat. Soft. 71: 1–25. 105

Dieng, H., M. Boots, N. Tuno, Y. Tsuda, M. Takagi. 2002. A laboratory and field evaluation of Macrocyclops distinctus, Megacyclops viridis and Mesocyclops pehpeiensis as control agent so ft eh dengue vector Aedes albopictus in a peridomestic area in Nagasaki, Japan. Medical and Veterinary Entomology 16:285-291.

Dobson, A., I. Cattadori, R. D. Holt, R. S. Ostfeld, F. Keesing, K. Krichbaum, J. R. Rohr, S. E. Perkins, P. J. Hudson. 2006. Sacred cows and sympathetic squirrels: the importance of biological diversity to human health. PLOS Medicine 3(6)714-718.

Dormann, C. F., B. Gruber, J. Fründ. 2008. Introducing the bipartite package: analyzing ecological networks. Interaction 1:0±2413793.

Dormann, C., J. Fründ, N. Blüthgen, B. Gruber. 2009. Indices, graphs and null models: analyzing bipartite ecological networks. The Open Ecology Journal 2:7-24.

Duquesne, S., I. Kroeger, M. Kutyniok, M. Liess. 2011. The potential of cladocerans as controphic competitors of the mosquito Culex pipiens. The Journal of Medical Entomology 48(3):554-560.

Dyer, L. A., and G. Gentry. 1999. Predicting natural‐enemy responses to herbivores in natural and managed systems. Ecol. Appl. 9: 402–408.

Dyer, L. A., M. S. Singer, J. T. Lill, J. O. Stireman, G. L. Gentry, R. J. Marquis, et al. 2007. Host specificity of Lepidoptera in tropical and temperate forests. Nature 448:696.

Dyer, L. A., T. R. Walla, H. F. Greeney, J. O. Stireman III, R. F. Hazen. 2010. Diversity of interactions: A metric for studies of biodiversity. Biotropica 42: 281-289.

Dyer, L., D. Wagner, H. Greeney, A. M. Smilanich, T. Massad, M. Robinson, et al. 2012. Novel Insights into Tritrophic Interaction Diversity and Chemical Ecology Using 16 Years of Volunteer-Supported Research. American Entomologist 58:15-19.

Dyer, L. A., and D. Letourneau. 2013. Can climate change trigger massive diversity cascades in terrestrial ecosystems? Diversity 5: 479–504.

Dyer, L. A. 2018. Multidimensional diversity associated with plants: a view from a plant-insect interaction ecologist. American Journal of Botany 105(9):1439-1442.

Ebeling, A., M. Rzanny, M. Lange, N. Eisenhauer, L. R. Hertzog, S. T. Meyer, and W. W. Weisser. 2017. Plant diversity induces shifts in the functional structure and diversity across trophic levels. Oikos 127: 208–219. 106

Ehler, L. E. 1998. Invasion biology and biological control. Biological Control 13:127- 133.

Eveleigh, E. S., K. S. McCann, P. C. McCarthy, S. J. Pollock, C. J. Lucarotti, B. Morin, G. A. McDougall, D. B. Strongman, J. T. Huber, J. Umbanhowar, et al. 2007. Fluctuations in density of an outbreak species drive diversity cascades in food webs. Proc. Natl. Acad. Sci. U. S. A. 104: 16976–16981.

Fang, Q., S-Q. Huang. 2016. Plant-pollinator interactions in a biodiverse meadow are rather stable and tight for three consecutive years. Integrative zoology 11:199-206.

Farajollahi, A., B. Kesavaraju, M. P. Nelder, S. C. Crans, R. Gaugler. 2009. An unusual larval collection and survival of Orthopodomyia signifera in the presence of the predator Toxorhynchites rutilus septentrionalis. Journal of the American Mosquito Control Association 25(3):370-373.

Field, R., B. A. Hawkins, H. V. Cornell, D. J. Currie, A. F. Diniz-Filho, J-F. Guégan, D. M. Kaufman, J. T. Kerr, G. G. Mittelbach, T. Oberdoff, E. M. O’Brien, J. R. G. Turner. 2009. Spatial species richness gradients across scales: a meta-analysis. Journal of Biogeography 36:132-147.

Fincke, O. M., S. P. Yanoviak, R. D. Hanschu. 1997. Predation by odonates depresses mosquito abundance in water-filled tree holes in Panama. Oecologia 112(2):244-253.

Finke, D. L., and R. F. Denno. 2002. Intraguild predation diminished in complex‐ structured vegetation: implications for prey suppression. Ecology 83:643–652.

Finke, D. L., and R. F. Denno. 2006. Spatial refuge from intraguild predation: implications for prey suppression and trophic cascades. Oecologia. 149:265–275.

Fisher, S., D. Pereyra, L. Fernandez. 2012. Predation ability and non-consumptive effects of Notonecta sellata (Heteroptera: Notonectidae) on immature stages of Culex pipiens (Diptera: Culicidae). Journal of Vector Ecology 37(1):245-251

Floore, T. G. 2006. Mosquito larval control practices: past and present. J. Am. Mosq. Control Assoc. 22:527–533.

Focks, D. A. 2007. Toxorhynchites as biocontrol agents. The Journal of the American Mosquito Control Association 23(2):118-127.

Forister, M. L., V. Novotny, A. K. Panorska, L. Baje, Y. Basset, P. T. Butterill, et al. 2015. The global distribution of diet breadth in insect herbivores. Proceedings of the National Academy of Sciences 112:442-447. 107

Fründ, J., K. S. McCann, N. M. Williams. 2016. Sampling bias is a challenge for quantifying specialization and network structure: lessons from a quantitative niche model. Oikos 125: 502-513.

Fry, L. L., M. S. Mulla, C. W. Adams. 1994. Field introductions and establishment of the tadpole shrimp, Triops longicaudatus (Notostraca: Triopsidae), a biological control agent of mosquitoes. Biological Control 4:113-124.

Gibb, H., and C. L. Parr. 2013. Does structural complexity determine the morphology of assemblages? An experimental test on three continents. PloS One 8: 64005.

Gibson, R. H., B. Knott, T. Eberlein, J. Memmott. 2011. Sampling method influences the structure of plant-pollinator networks. Oikos 120:822-831.

Gimonneau. G., J. Bouyer, S. Morand, N. J. Besansky, A. Diabate, F. Simard. 2010. A behavioral mechanism underlying ecological divergence in the malaria mosquito Anopheles gambiae. Behavioral Ecology 21:1087-1092.

Gliessman, S. R. 1990. Agroecology: researching the ecological basis for sustainable agriculture, pp. 3–10. In Agroecology. Springer, New York, NY.

González-Varo, J. P., A. Traveset. 2016. The labile limits of forbidden interactions. Trends in Ecology & Evolution 31:700-710.

Gotelli, N. J., R. K. Colwell. 2001. Quantifying biodiversity: Procedures and pitfalls in the measurement and comparison of species richness. Ecology Letters 4:379-391.

Gregg, W. W. and F. L. Rose. 1985. Influences of aquatic macrophytes on invertebrate community structure, guild structure, and microdistribution in streams. Hydrobiologia 128:45-56.

Griswold, M. W. and L. P. Lounibos. 2005. Competitive outcomes of aquatic container Diptera depend on predation and resource levels. Annals of the Entomological Society of America 98(5):673-681.

Griswold, M. W. and L. P. Lounibos. 2006. Predator identity and additive effects in a treehole community. Ecology 87(4):987-995.

Gross, T., L. Rudolf, S. A. Levin, U. Dieckmann. 2009. Generalized models reveal stabilizing factors in food webs. Science 325:747-750.

Haddad, N. M., G. M. Crutsinger, K. Gross, J. Haarstad, and D. Tilman. 2011. Plant diversity and the stability of foodwebs. Ecol. Lett. 14:42–46. 108

Hanford, J. K., C. E. Webb, and D. F. Hochuli. 2019. Habitat traits associated with mosquito risk and aquatic diversity in urban wetlands. Wetlands 39:743–758.

Hardstone, M. C. and T. G. Andreadis. 2012. Weak larval competition between the invasive mosquito Aedes japonicus japonicus (Diptera: Culicidae) and three resident container-inhabiting mosquitoes in the laboratory. Journal of Medical Entomology 48(2)277-285.

Hedges, L. V., J. Gurevitch, P. S. Curtis. 1999. The meta-analysis of response ratios in experimental ecology. Ecology 80(4):1150-1156.

Hetchel, L. J. and S. A. Juliano. 1997. Effects of a predator on prey metamorphosis: plastic responses by prey or selective mortality? Ecology 78(3):838-851.

Hinman, E. H. 1934. Predators of Culicidae (mosquitoes). The Journal of Tropical Medicine and Hygiene 37(9):129-134.

Holck, A. R. 1988. Current status of the use of predators and parasites for the control of mosquitoes. The Florida Entomologist. 71(4):537-546.

Holt, R. D. 1977. Predation, apparent competition, and the structure of prey communities. Theoretical Population Biology 12:197-229.

Holt, R. D. and J. H. Lawton. 1994. The ecological consequences of shared natural enemies. Annual Review of Ecology and Systematics 25:495-520.

Hutchinson, G. E. 1957. A treatise on limnology Vol 1: Geography, Physics, and Chemistry. Wiley, New York.

Irwin, P. and S. Paskewitz. 2009. Investigation of fathead minnows (Pimephales promelas) as a biological control agent of Culex mosquitoes under laboratory and field conditions. Journal of the American Mosquito Control Association 25(3):301-309.

Janzen, D. H. 1974. The deflowering of central America. Nat. Hist. 83:48–53.

Janzen, D. H. , M. Hajibabaei, J. M. Burns, W. Hallwachs, E. Remigio, P. D. N. Hebert. 2005. Wedding biodiversity inventory of a large and complex Lepidoptera fauna with DNA barcoding. Philosophical Transactions of the Royal Society B: Biological Sciences 360: 1835-1845.

Jenkins, D. W. 1964. Pathogens, parasites and predators of medically important arthropods. World Health Organization, Geneve pp. 1-150.

Jiang, L., H. Joshi, S. N. Patel. 2009. Predation alters relationships between biodiversity and temporal stability. The American Naturalist 173:389-399. 109

Jiannino, J. A. and W. E. Walton. 2004. Evaluation of vegetation management strategies for controlling mosquitoes in a southern California constructed wetland. Journal of the American Mosquito Control Association 20(1)18-26.

Jiménez-Alfaro, B., M. Chytrỳ, L. Mucina, J. B. Grace, M. Rejmánek. 2016. Disentangling vegetation diversity from climate-energy and habitat heterogeneity for explaining animal geographic patterns. Ecology and evolution.

Jinliang, L., M. Vellend, Z. Wang, M. Yu. 2018. High beta diversity among small islands is due to environmental heterogeneity rather than ecological drift. Journal of Biogeography 45:2252-2261.

Johnson, P. T. J., P. J. Lund, R. B. Hartson, T. P. Yoshino. 2009. Community diversity reduces Schistosoma mansoni transmission, host pathology and human infection risk. Proceedings of the Royal Society of London, B 276:1657-1663.

Jordán, F., G. Osváth. 2009. The sensitivity of food web topology to temporal data aggregation. Ecological Modelling 220:3141-3146.

Juliano, S. A. and L. Reminger. 1992. The relationship between vulnerability to predation and behavior of larval treehole mosquitoes: geographic and ontogenetic differences. OIKOS 63:465-476.

Juliano, S. A. 2007. Population Dynamics. The Journal of the American Mosquito Control Association 23(2):265-275.

Juliano, S. A. 2009. Species interactions among larval mosquitoes: context dependence across habitat gradients. Annual Review of Entomology 54:37-56.

Juliano, S. A., L. P. Lounibos, N. Nishimura, K. Greene. 2010. Your worst enemy could be your best friend: predator contributions to invasion resistance and persistence of natives. Oecologia 162:709-718.

Kawecki, T. J., and D. Ebert. 2004. Conceptual issues in local adaptation. Ecology Lett. 7: 1225–1241.

Keddy, P. A. 1992. Assembly and response rules: two goals for predictive community ecology. Journal of Vegetation Science 3:157-164.

Keesing, F., R. D. Hold, R. S. Ostfeld. 2006. Effects of species diversity on disease risk. Ecology Letters 9:485-498.

Kellen, W. R., T. B. Clark, J. E. Lindegren, B. C. Ho. 1965. Bacillus sphaericus Neide as a pathogen of mosquitoes. Journal of invertebrate pathology 7:442-448. 110

King, C., G. Ballantyne, P. G. Willmer. 2013. Why flower visitation is a poor proxy for pollination: measuring single-visit pollen deposition, with implications for pollination networks and conservation. Methods in Ecology and Evolution 4:811-818.

Knight, T. M., J. M. Chase, C. W. Goss, J. J. Knight. 2004. Effects of interspecific competition, predation, and their interactions on survival and development time of immature Anopheles quadrimaculatus. Journal of Vector Ecology 29(2):277-284.

Kroeger, I., M. Liess, F. Dziock, S. Duquesne. 2013. Sustainable control of mosquito larvae in the field by the combined actions of the biological insecticide Bti and natural competitors. Journal of Vector Ecology 38(1):82-89.

Kruschke, J. K. 2013. Bayesian estimation supersedes the t test. Journal of Experimental Psychology 142: 573.

Kruschke, J., M. Meredith. 2015. BEST: Bayesian estimation supersedes the t-test. R Package version 0.5.0 [Internet]. http://cran.cnr.berkeley.edu/web/packages/BEST/vignettes/BEST.pdf

Kumar, R. and J-S. Hwang. 2006. Larvicidal efficiency of aquatic predators: a perspective for mosquito biocontrol. Zoological Society 45(4):447-466.

Kweka, E. J., G. Zhou, T. M. Gilbreath III, Y. Afrane, M. Nyindo, A. K. Githeko, G. Yan. 2011. Predation efficiency of Anopheles gambiae larvae by aquatic predators in western Kenya highlands. Parasites and Vectors 128(4):1-7.

Kwekam E, J., G. Zhou, T. M. Gilbreath III, Y. Afrane, M. Nyindo, A. K. Githeko, G. Yan. 2001. Predation efficiency of Anopheles gambiae larvae by aquatic predators in western Kenya highlands. Parasites and Vectors 4(128):1-7

Lacey, L. A. and B. K. Orr. The role of biological control of mosquitoes in integrated vector control. American Journal of Tropical Medicine and Hygiene 50(6):97-115.

Laird, M., 1988. The Natural History of Larval Mosquito Habitats. Academic Press, London, 555 pp.

Lajeunesse, M. J. 2015. Bias correction for the log response ratio in ecological meta- analysis. Ecology 96(8):2056-2063.

Langellotto, G. A., and R. F. Denno. 2004. Responses of invertebrate natural enemies to complex-structured habitats: a meta-analytical synthesis. Oecologia. 139: 1–10. 111

Lederman, J. P., E. L. Suchman IV, W. C. Black, J. O. Carson. 2004. Infection and pathogenicity of the mosquito densoviruses AeDNV, HeDNV, and ApeDNV in Aedes aegypti mosquitoes (Diptera: Culicidae). Journal of Economic Entomology 97(6):1828- 1835.

Legner, E. F., R. D. Sjogren, and I. M. Hall. 1974. The biological control of medically important arthropods. Critical Reviews in Environmental Control 4:85-113.

Lepesquer, C., S. Scherrer, M. C. Viera, M. Aleida-Neto, D. M. Salcido, et al. 2018. Changing interactions among persistent species as the major driver of seasonal turnover in plant-caterpillar interactions. PLoS one 13(9):1-8.

Lord C. C. 2007. Modeling and biological control of mosquitoes. Journal of the American Mosquito Control Association, 23(2 Suppl), 252–264. https://doi.org/10.2987/8756-971X(2007)23[252:MABCOM]2.0.CO;2

LoGiudice, K. R. S. Ostfeld, K. A. Schmidt, F. Keesing. 2003. The ecology of infectious disease: effects of host diversity and community composition on Lyme disease risk. PNAS 100(2):567-571.

López-Carretero, A., C. Díaz-Castelazo, K. Boege, V. Rico-Gray. 2014. Evaluating the spatio-temporal factors that structure network parameters of plant-herbivore interactions. PloS ONE 9:e110430. https://doi.org/10.1371/journal.pone.0110430 PMID: 25340790.

Lumpkin, W. P., K. R. Stirek, L. A. Dyer. 2020. Macrophyte diversity and complexity reduce larval mosquito abundance. Journal of Medical Entomology https://doi.org/10.1093/jme/tjaa012 .

MacArthur, R. H. and J. W. MacArthur. 1961. On bird species diversity. Ecology 42:594-598.

Mackay, A. J., E. J. Muturi, M. P. Ward, and B. F. Allan. 2016. Cascade of ecological consequences for West Nile virus transmission when aquatic macrophytes invade stormwater habitats. Ecol. Appl. 26:219–232.

Magurran, A. E. 2013. Measuring biological diversity. John Wiley & Sons.

Marina, C. F., J. E. Ibarra, J. I. Arredondo-Jimenez, I. Fernandez-Salas, J. Valle, T. Williams. 2003. Sublethal iridovirus disease of the mosquito Aedes aegypti is due to viral replication not cytotoxicity. Medical and Veterinary Entomology 17:187-194.

Marten, G. G., G. Borjas, M. Cush, E. Fernandez, J. W. Reid. 1994. Control larval Aedes aegypti (Diptera: Culicidae) by Cyclopoid copepods in peridomestic breeding containers. Journal of Medical Entomology 31(1):36-44. 112

Marten, G. G., M. Nguyen, B. J. Mason, G. Ngo. 2000. Natural control of Culex quinquefasciatus larvae in residential ditches by the copepod Macrocyclops albidus. Journal of Vector Ecology 25(1):7-15.

Marten, G. G. and J. W. Reid. 2007. Cyclopoid Copepods. The Journal of the American Mosquito Control Association 23(2)65-92.

McAbendroth, L., P. M. Ramsay, A. Foggo, S. D. Rundle, D. T. Bilton. 2005. Does macrophyte fractal complexity drive invertebrate diversity, biomass and body size distributions? OIKOS 111(2):279-290.

Merkle, E. C., and Y. Rosseel. 2018. Blavaan: Bayesian structural equation models via parameter expansion. J. Stat. Soft. 85: 1–30.

Merritt, R. W., R. H. Dadd, and E. D. Walker. 1992. Feeding behavior, natural food, and nutritional relationships of larval mosquitoes. Annu. Rev. Entomol. 37: 349–376.

Medved, R. A. and E. F. Legner. 1974. Feeding and reproduction of the planarian, Dugesia dorotocephala (Woodworth), in the presence of Culex peus Speiser. Environmental Entomology 3(4):637-641.

Mercer, D. R., G. R. Wettach, J. L. Smith. 2005. Effects of larval density and predation by Toxorhynchites amboinensis on Aedes polynesiensis (Diptera: Culicidae) developing in coconuts. The Journal of the American Mosquito Control Association 21(4):425-431.

Micieli, M. V., G. Marti, J. J. Garica. 2002. Laboratory evaluation of Mesocyclops annulatus (Wierzejski, 1892) (Copepoda: Cyclopoidea) as a predator of container- breeding mosquitoes in Argentina. Mem. Inst. Oswaldo Cruz, Rio de Janeiro 97(6):835- 838.

Minakawa, N., K. Futami, G. Sonye, P. Akweywa, S. Kaneko. 2007. Predatory capacity of shorefly, Ochthera chalybescens, on malaria vectors. Malaria Journal 6(104):1-5.

Miura, T., and R. M. Takahashi. 1988. A laboratory study of predation by damselfly nymphs, Enallagma civile, upon mosquito larvae, Culex tarsalis. J. Am. Mosq. Control Assoc. 4: 129–131.

Miyagi, I. 1992. Biological control of container-breeding mosquitoes, Aedes albopictus and Culex quinquefasciatus, in a Japanese island by release of Toxorhynchites sp. Medical and Veterinary Entomology 6(3):290-30.

Mogi, M. 2007. Insects and other invertebrate predators. The Journal of The American Mosquito Control Association. 23(2)93-109. 113

Mougi, A. and M. Kondoh. 2012. Diversity of interaction types and ecological community stability. Science. 337: 349–351.

Murrell, E. G. and S. A. Juliano. 2008. Detritus type alters the outcome of interspecific competition between Aedes aegypti and Aedes albopictus (Diptera: Culicidae). Journal of Medical Entomology 45(3):375-383.

Murrell, E. G. and S. A. Juliano. 2013. Predation resistance does not trade off with competitive ability in early-colonizing mosquitoes. Oecologia 173:1033-1042.

Myers, J. H., A. Savoie, E. van Randen. 1998. Eradication and pest management. Annual Review of Entomology 43:471-491.

Myers, J. H., D. Simberloff, A. M. Kuris, J. R. Carey. 2000. Eradication revisited: dealing with exotic species. TREE 15(8):316-320.

Nannini, M. A. and S. A. Juliano. 1997. Effects of Developmental asynchrony between Aedes triseriatus (Diptera: Culicidae) and its predator Toxorhynchites rutilus (Diptera: Culicidae). The Journal of Medical Entomology 34(4):457-460.

Nielsen, A., J. Bascompte. 2007. Ecological networks, nestedness and sampling effort. Journal of Ecology 95:1134-1141.

Novotny, V., Y. Basset, S. E. Miller, G. D. Weiblen, B. Bremer, L. Cizek, et al. 2002. Low host specificity of herbivorous insects in a tropical forest. Nature 416: 841-844.

Novotny, V., P. Droid, S. E. Miller, M. Kulfan, M. Janda, Y. Basset, et al. 2006. Why are there so many species of herbivorous insects in tropical rainforests? Science 313:1115-1118.

Novotny, V., S. E. Miller, J. Hulcr, R. A. I. Drew, Y. Basset, M. Janda, et al. 2007. Low beta diversity of herbivorous insects in tropical forests. Nature 448:692.

Novotny, V., S. E. Miller, L. Baje, S. Balagawi, Y. Basset, L. Cizek, et al. 2010. Guild- specific patterns of species richness and host specialization in plant-herbivore food webs from a tropical forest. Journal of Animal Ecology 79:1193-1203.

Nyamah, M. A., S. Sulaiman, B. Omar. 2011. Field observation on the efficacy of Toxorhynchites splendens (Wiedemann) as biocontrol agent against Aedes albopictus (Skuse) larvae in a cemetery. Tropical Biomedicine 28(2):312-319.

Ohba, S., M. Ohtsuka, T. Sunahara, Y. Sonoda, E. Kawashima, M. Takagi. 2012. Differential responses to predator cues between two mosquito species breeding in different habitats. Ecological Entomology 37:410-418. 114

Ohgushi, T, T. P. Craig, P. W. Price. 2007. Ecological communities: plant mediation in indirect interaction webs. Cambridge University Press.

Oksanen, J., F. G. Blanchet, R. Kindt, P. Legendre, P. R. Minchin, R. B. O'Hara, et al. 2015. Vegan: community ecology package, R Package version 2.2-1. http://CRAN.R-project.org/package=vegan.

O’Neal, P. A. and S. A. Juliano. 2013. Seasonal variation in competition and coexistence of Aedes mosquitoes: stabilizing effects of egg mortality or equalizing effects of resources? Journal of Animal Ecology 82:256-265.

Peralta, G., C. M. Frost, T. A. Rand, R. K. Didham, J. M. Tylianakis. 2014. Complementarity and redundancy of interactions enhance attack rates and spatial stability in host-parasitoid food webs. Ecology. 2014 95:1888-1896.

Perez-Pacheco, R., C. Rodrigues-Hernandez, J. Lara-Reyna, R. Montes-Belmont, J. Ruiz- Vega. 2005. Control of the mosquito Anopheles pseudopunctipennis (Diptera: Culicidae) in Romanomermis iyengari (Nematoda: Mermithidae) in Oaxaca, Mexico. Biological Control 31:137-142.

Pernia, J., E. Zoppi, De Roa, M. Palacios-Caceres. 2007. Prey-predator relationship between the cyclopoids Mesocyclops longisetus and Mesocyclops meridianus with Anopheles aquasalis larvae. Journal of the American Mosquito Control Association 23(2):166-171.

Petanidou, T., A. S. Kallimanis, J. Tzanopoulos, S. P. Sgardelis, J. D. Pantis. 2008. Long-term observation of a pollination network: fluctuation in species and interactions, relative invariance of network structure and implications for estimates of specialization. Ecology Letters 11:564-575.

Petersen, J. J. and H. C. Chapman. 1970. Parasitism of Anopheles mosquitoes by a Gastromermis sp. (Nematoda: Mermithidae) in southwestern Louisiana. Mosquito News 30(3):420-424.

Pianka, E. R. 1967. On lizard species diversity: North American flatland deserts. Ecology 48(3):333-351.

Platzer, E. G. 2007. Mermithid nematodes. The Journal of the American Mosquito Control Association 23(2):58-64.

Plummer, M. 2003. JAGS: a program for analysis of Bayesian graphical models using Gibbs sampling. In Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DCS 2003), March 20–22, Vienna, Austria. 115

Poisot, T., E. Canard, D. Mouillot, N. Mouquet, D. Gravel, F. Jordan. 2012. The dissimilarity of species interaction networks. Ecology Letters 15: 1353-1361.

Polis, G. A. 1994. Food webs, trophic cascades and community structure. Aust. J. Ecol. 19: 121–136.

Quiroz-Martínez, H., V. A. Rodrígues-Castro, C. Solís-Rojas, M. G. Maldonado-Blanco. 2005. The Journal of the American Mosquito Control Association 21(3):328-330.

Quiroz-Martínez, H. and A. Rodrígues-Castro. 2007. Aquatic insects as predators of mosquito larvae. Journal of the American Mosquito Control Association 23(2):110-117.

R Core Team. R: A Language and Environment for Statistical Computing, version 3.3.2 [Internet]. Vienna, Austria: R Foundation for Statistical Computing; 2014. http://www.R-project.org .

Randolph, S. E. and A. D. Dobson. 2012. Pangloss revisited: a critique of the dilution effect and the biodiversity-buffers-disease paradigm. Parasitology 139(7):847-863.

Reisen, W. K., R. P. Meyer, J. Shields, and C. Arbolante. 1989. Population ecology of preimaginal Culex tarsalis (Diptera: Culicidae) in Kern County, California. J. Med. Entomol. 26: 10–22.

Rey, J. R., S. O’Connell, S. Suarez, Z. Menendez, L. P. Lounibos, G. Byer. 2004. Laboratory and field studies of Macrocyclops albidus (Crustacea: Copepoda) for biological control of mosquitoes in artificial containers in a subtropical environment. Journal of Vector Ecology 29(10:124-134.

Ricklefs, R. and D. Schluter. 1993. Species diversity in ecological communities: historical and geographical perspectives.

Roberts, D. 2012. Responses of three species of mosquito larvae to the presence of predatory dragonfly and damselfly larvae. Entomologia 145:23-29.

Roberts, D. 2014. Mosquito larvae change their feeding behavior in response to kairomones from some predators. Journal of Medical Entomology 51(2):368-374.

Rodrigues de Paula, A., E. S. Brito, C. R. Pereira, M. P. Carrera, R. I. Samuels. 2008. Susceptibility of adult Aedes aegypti (Diptera: Culicidae) to infection by Metarhizium anisopliae and Beauveria bassiana: prospects for dengue vector control. Biocontrol Science and Technology 18(9-10):1017-1025.

Rueda, L. M. 2008. Global diversity of mosquitoes (Insecta: Diptera: Culicidae) in freshwater. Hydrobiologia 595:477-487 116

Rusch, A., R. Bommarco, and B. Ekbom. 2017. Conservation biological control in agricultural landscapes, Adv. Bot. Res. 81: 333–360.

Russell, R. C. 1999. Constructed wetlands and mosquitoes: health hazards and management options-an Australian perspective. Ecol. Eng. 12: 107–124.

SAS Institute Inc. Base SAS® 9.3 Procedures Guide. Cary, NC: SAS Institute Inc.; 2011.

Scherber, C., N. Eisenhauer, W. W. Weisser, B. Schmid, W. Voigt, M. Fischer, E. D. Schulze, C. Roscher, A. Weigelt, E. Allan, et al. 2010. Bottom-up effects of plant diversity on multitrophic interactions in a biodiversity experiment. Nature. 468: 553–556.

Schiller, A., M. Alllen, J. Coffey, A. Fike, F. Carballo. 2019. Updated methods for the production of Toxorhynchites rutilus septentrionalis (Diptera, Culicidae) for use as biocontrol agent against container breeding pest mosquitoes in Harris County, Texas. Journal of Insect Science 19(2)1:6.

Schleuning, M., J. Fründ, A-M. Klein, S. Abrahamczyk, R. Alarcón, M. Albrecht, et al. 2012. Specialization of mutualistic interaction networks decreases toward tropical latitudes. Current biology 22:1925-1931.

Schmidt, M. H., I. Roschewitz, C. Thies, and T. Tsharntke. 2005. Differential effects of landscape and management on diversity and density of ground-dwelling farmland spiders. J. Appl. Ecol. 42: 281–287.

Schneider, D. W. and T. M. Frost. 1996. Habitat duration and community structure in temporary ponds. Journal of the North American Benthological Society 15(1):64-86.

Scholte, E. J., B. G. J. Knols, W. Takken. 2006. Infection of the malaria mosquito Anopheles gambiae with the the entomopathogenic fungus Metarhizium anisopliae reduces blood feeding and fecundity. Journal of Invertebrate Pathology 91:43-49.

Schreiber, E. T., C. F. Hallmon, K. M. Eskridge, G. G. Marten. 1996. Effects of Mesocyclops longisetus (Copepoda: Cyclopidae) on mosquitoes that inhabit tires; influence of litter type, quality, and quantity. The Journal of the American Mosquito Control Association 12(4):688-694.

Schreiber, E. T. 2007. Toxorhynchites. Journal of the American Mosquito Control Association 23(2):129-132.

Service, M. W. 1976. Mosquito ecology field sampling methods. Applied Science Publishers Ltd. London.

Shaalam, E. A. S. and D. V. Canyon. 2009. Aquatic insect predators and mosquito control. Tropical Biomedicine 26(3):223-261. 117

Shurin, J. B., D. S. Gruner, H. Hillebrand. 2006. All wet or dried up? Real differences between aquatic and terrestrial food webs. Proceedings of the Royal Society Biology 273:1-9.

Siemann, E. 1998. Experimental tests of effects of plant productivity and diversity on grassland arthropod diversity. Ecology 79: 2057–2070.

Smith, T., Maire, N., Ross, A., Penny, M., Chitnis, N., Schapira, A., Studer, A., Genton, B., Lengeler, C., Tediosi, F. and De Savigny, D., 2008. Towards a comprehensive simulation model of malaria epidemiology and control. Parasitology, 135(13), pp.1507- 1516.

Soumare, M. K. F., J. E. Cilek, E. T. Schreiber. 2004. Prey and size preference of Mesocyclops longisetus (Copepoda) for Aedes albopictus and Culex quinquefasciatus larvae. The Journal of the American Mosquito Control Association 20(3):305-310.

Soumare, M. K. F. and J. E. Cilek. 2011. The effectiveness of Mesocyclops longisetus (Copepoda) for the control of container-inhabiting mosquitoes in residential environments. The Journal of the American Mosquito Control Association 27(4):376- 383.

Srivastava, D. S. 2006. Habitat structure, trophic structure and ecosystem function: interactive effects in a bromeliad-insect community. Oecologia. 149: 493–504.

Staniczenko, P. P., J. C. Kopp, S. Allesina. 2013. The ghost of nestedness in ecological networks. Nature Communications 4:1391.

Staniczenko, P.P., O. T. Lewis, J. M. Tylianakis, M. Albrecht M, V. Coudrain, A-M. Klein, et al. 2017. Predicting the effect of habitat modification on networks of interacting species. Nature Communications 8:792.

Stav, G., L. Blaustein, Y. Margalit. 2005. Individual and interactive effects of a predator and controphic species on mosquito populations. Ecological Applications 15(2):578-298.

Stein, A. and H. Kreft. 2015. Terminology and quantification of environmental heterogeneity in species-richness research. Biological Reviews 90:815-836.

Stireman, J. O., L. A. Dyer, and R. B. Matlock. 2005. Top-down forces in managed versus unmanaged habitats, pp. 303–323. Ecology of predator-prey interactions.

Szalay, F. A., and V. H. Resh. 2000. Factors influencing macroinvertebrate colonization of seasonal wetlands: responses to emergent plant cover. Freshw. Biol. 45: 295–308. 118

Tewksbury, J. J., J. G. Anderson, J. D. Bakker, J. J. Billo, P. W. Dunwiddie, M. J. Groom, et al. 2014. Natural history's place in science and society. BioScience 64: 300- 310.

Thébault, E., C. Fontaine. 2010. Stability of ecological communities and the architecture of mutualistic and trophic networks. Science 329:853-856.

Thompson, J. N. 1996. Evolutionary ecology and the conservation of biodiversity. Trends in Ecology and Evolution 11:300-303.

Thompson, J. N. 1997. Conserving interaction biodiversity. The Ecological Basis of Conservation. Springer; pp. 285-293.

Thullen, J. S., J. J. Sartoris, and W. E. Walton. 2002. Effects of vegetation management in constructed wetland treatment cells on water quality and mosquito production. Ecol. Eng. 18: 441–457.

Tilman, D. 1994. Competition and biodiversity in spatially structured habitats. Ecology 75:2–16.

Tilman, D. and J. A. Downing. 1994. Biodiversity and stability in grasslands. Nature 367:363-365.

Tilman, D., D. Wedin, and J. Knops. 1996. Productivity and sustainability influenced by biodiversity in grassland ecosystems. Nature 379: 718–720.

Trpis, M. 1970. Adult population estimate of Toxorhynchites brevipalpis breeding in man-made containers in Dar-Es-Salaam, Tanzania. World Health Organization 231:1-7.

Tscharntke, T., A. M. Klein, A. Kruess, I. Steffan‐Dewenter, and C. Thies. 2005. Landscape perspectives on agricultural intensification and biodiversity–ecosystem service management. Ecology Lett. 8:857–874.

Tylianakis, J. M., and C. M. Romo. 2010. Natural enemy diversity and biological control: making sense of the context-dependency. Basic Appl. Ecol. 11: 657–668.

Tylianakis, J. M., E. Laliberte, A. Nielsen, J. Bascompte. 2010. Conservation of species interaction networks. Biological Conservation 143:2270-2279.

Valdovinos, F. S., B. J. Brosi, H. M. Briggs, P. Moisset de Espanés, R. Ramos-Jiliberto, N. D. Martinez. 2016. Niche partitioning due to adaptive foraging reverses effects of nestedness and connectance on pollination network stability. Ecology Letters 19:1277- 1286. 119

Vázquez, D. P., M. A. Aizen. 2004. Asymmetric specialization: a pervasive feature of plant-pollinator interactions. Ecology 85:1251-1257.

Vázquez, D. P., N. P. Chacoff, L. Cagnolo. 2009. Evaluating multiple determinants of the structure of plant-animal mutualistic networks. Ecology 90:2039-2046.

Wallace, A. R. 1878. Tropical nature, and other essays. Macmillan and Company.

Warfe, D. M., and L. A. Barmuta. 2006. Habitat structural complexity mediates food web dynamics in a freshwater macrophyte community. Oecologia. 150: 141–154.

Wedel, D. C. 2017. BANOVA: an R package for hierarchical Bayesian ANOVA. J. Stat. Soft. 81: 1–46.

Weiher, E. and P. Keddy. 2001. Ecological assembly rules: perspectives, advances, retreats. Cambridge University Press.

Wetzel, R. G. 1983. Limnology 2nd ed. Saunders College Publishing, Philadelphia.

Willott, E. 2004. Restoring nature, without mosquitoes? Restor. Ecol. 12:147–153.

Winemiller, K. O. 1989. Must connectance decrease with species richness? The American Naturalist 134:960-968.

Wise de Valdez, M. R. 2006. Parasitoid-induced behavioral alterations of Aedes aegypti mosquito larvae infected with mermithid nematodes (Nematoda: Mermithidae). J. Vector Ecol. 31: 344–354.