The Interactive Effects of Phenotype and Environment on Dispersal

by

Celina Beatrice Baines

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Ecology and Evolutionary Biology University of Toronto

© Copyright by Celina Beatrice Baines 2019

The Interactive Effects of Phenotype and Environment on Dispersal Celina Beatrice Baines

Doctor of Philosophy

Department of Ecology and Evolutionary Biology University of Toronto

2019 Abstract

Dispersal is the movement of organisms between habitat patches that creates the potential for gene flow. This movement of individuals and their genes across space impacts the dynamics of metapopulations and metacommunities. For example, dispersal influences the risk of metapopulation extinction, and affects the coexistence of species with competitors, predators, and parasites. Dispersal therefore influences local and regional species richness and community composition. Dispersal is a complex behaviour characterized by high levels of variation between patches and among individuals. The central question of my dissertation is, what produces this variation? I addressed this using three approaches: quantitative review, theoretical modeling, and empirical investigations. In chapter 2, I conducted a meta-analysis of body size-dependent dispersal. I found that size had a positive effect on dispersal distance, but its average effect on emigration and immigration did not differ from zero. However, there were high levels of heterogeneity in this effect within each stage of dispersal. In chapter 3, I developed an individual based model of how dispersal evolves to respond to both body condition and population density.

I found that dispersal evolved to be a positive threshold function of density, and the value of this threshold depended on condition. In chapters 4-6, I used the backswimmer, Notonecta undulata, as a model empirical system to explore the effects of phenotype and the environment on

ii dispersal. In chapter 4, I conducted a mark-release-recapture study which found that dispersal probability was determined by the interaction between population density, body mass, and sex.

In chapter 5, I conducted an experiment which demonstrated that habitat conditions experienced during development have carryover effects on dispersal in later life stages. Finally, in chapter 6, I investigated the effects of parasites on dispersal. I found that risk of parasitism induced dispersal in healthy backswimmers, but infected backswimmers had low dispersal ability. This impacted dispersal in natural environments: parasites eliminated density dependence in backswimmer dispersal. My thesis provides novel information about how dispersal is influenced by phenotype, the environment, and their interaction. These results alter our understanding of the consequences of dispersal for biological processes, including the spatial distribution of biodiversity.

iii

Acknowledgements

There are many, many people who have taught me, helped me, and mentored me during my PhD, and who have generally made my experience more enjoyable. First, I want to thank Shannon McCauley. I first met Shannon as a first-year undergraduate, when I worked as her summer field assistant. Shannon is a wonderful and enthusiastic teacher, and she really sparked a passion in me for ecological research. When I had the opportunity to do a PhD in her lab, I jumped at it, and I’m so grateful I did. Shannon has been a wonderful supervisor; she has guided me through the challenges of ecological research and academia, and supported my intellectual growth and career development. I’m so lucky to have had her as an advisor, and to count her as a close friend.

I also want to thank my supervisory committee. Locke Rowe has been my advocate since I started working in his lab as an undergraduate. He has always believed in me more than I believe in myself, and has challenged me to reach for higher research and career goals. I’m so grateful for his support. I want to thank Ben Gilbert for providing helpful and insightful comments and advice throughout my PhD, and for answering my stats questions. Helen Rodd has provided invaluable support both through her role on my supervisory committee, and as graduate chair. I want to thank her for her insightful comments, for sharing the notes she takes during committee meetings, for providing career advice, and for helping me with a million and one things that have smoothed my path through grad school.

I have had the great pleasure of doing research at the Koffler Scientific Reserve, and I want to thank everyone there for all they have done to make the reserve run so well. I especially want to thank Stephan Schneider who has gone above and beyond to help me conduct experiments, has built the crazy things I need for research (including flight tunnels and ponds), and has made the reserve an efficient, fun, and safe place to do research. Steph has also rooted for me every step of the way, and I’m so grateful he’s my friend. I also want to thank John Stinchcombe, Jenn English, John Jensen, Tobias Mankis, and Olivera Joksimovic for logistical (and sometimes physically strenuous) support at the reserve.

There are many people to thank for keeping the department running smoothly, and making my life easier in countless ways. At UTM, I want to thank Carolyn Moon, Stephanie do Rego, Amy Yeung, and Bob Liange. And in the EEB department, I want to thank Kitty Lam, Olivera Joksimovic, and again Helen Rodd. iv

Many people provided help in the lab and field during my PhD. Ilia Ferzoco has been an amazing research partner. She has made invaluable contributions with her diligent observations on the natural history of backswimmers, and she has spent countless hours helping me in the field. Much of the work in my thesis could not have been possible without her. I also want to thank Shantel Catania, Ariana Longley, David Soliman, Quan Le, Tharusha Wijewardena, Sarah Hasni, Rokhsar Rezaee, Arjan Banerjee, Julienne Bonoan, Racquelle Mangahas, Samantha Hasbum, Bansari Patel, Rosemary Martin, Dachin Frances, Hugh Traquair, and Betty Dondertman for their help in the field and the lab. I also want to thank my collaborators on the theoretical dispersal modeling project, Justin Travis and Greta Bocedi, for being fantastic teachers.

I was lucky to have amazing labmates to share the grad school experience with. Huge thanks to Ilia Ferzoco, Rosemary Martin, Dachin Frances, Sarah French, Rosalind Murray, and Chris Searcy. I can’t imagine better people to work with, laugh with, and complain to. Special thanks to Ilia for being my number one pep-talker and confidence booster, I’ve appreciated the support so much. Shout outs to Ros, for making me laugh every day and being my Toronto soul-sister, and to Rosie for making everything fun, and making every day more beautiful with her insect- themed art.

I want to thank my family for all of their love, patience, and support. They have always believed I could do anything, and supported me even when I chose a career collecting bugs. My parents Betty Dondertman and Frank Baines, and my sister Nicky Baines, have encouraged and helped me every step of the way. I’m so grateful for my mom, who encouraged my love of nature and animals, and who even drove out to help me in the field at the end of a season when I was exhausted and needed extra moral support.

Finally, I want to thank my partner Hugh Traquair. Hugh has listened to me talk endlessly about bugs, helped me work through problems, and helped me in the field many times, including once during a snow storm. Hugh has provided boundless moral and emotional support through all the highs and lows of research and academia. Life is better for sharing it with you.

v

Table of Contents

Acknowledgements ...... iv

Table of Contents ...... vi

List of Tables ...... x

List of Figures ...... xi

List of Appendices ...... xiii

Chapter 1 General Introduction ...... 1

1.1 Overview ...... 1

1.2 What is dispersal and how do we study it? ...... 1

1.3 Causes of dispersal ...... 4

1.3.1 Ultimate causes of dispersal...... 4

1.3.2 Proximate causes of dispersal ...... 7

1.4 Consequences of dispersal ...... 9

1.4.1 Consequences of dispersal for ecological dynamics ...... 9

1.4.2 Consequences of dispersal for evolutionary dynamics ...... 10

1.4.3 Consequences of context- and phenotype-dependent dispersal ...... 11

1.5 Study system ...... 12

1.6 Summary of chapters ...... 14

1.7 References ...... 15

Chapter 2 Exploring Sources of Heterogeneity in Phenotype-Dependent Dispersal ...... 25

2.1 Abstract ...... 25

2.2 Introduction ...... 26

2.3 Methods...... 30

2.3.1 Data collection ...... 30

2.3.2 Statistical analysis ...... 32

2.4 Results ...... 33 vi

2.4.1 Disperser-resident comparison studies ...... 33

2.4.2 Dispersal distance studies ...... 36

2.5 Discussion ...... 37

2.5.1 Heterogeneity in phenotype-dependent dispersal ...... 37

2.5.2 Sources of heterogeneity in phenotype-dependent dispersal ...... 37

2.5.3 Future directions ...... 43

2.5.4 Conclusions ...... 44

2.6 References ...... 45

Chapter 3 The Joint Evolution of Density- and Body Condition-Dependent Dispersal ...... 54

3.1 Abstract ...... 54

3.2 Introduction ...... 54

3.3 The model ...... 57

3.3.1 The landscape...... 57

3.3.2 Body condition ...... 58

3.3.3 Dispersal ...... 58

3.3.4 Reproduction ...... 59

3.3.5 Order of events ...... 60

3.3.6 Simulation experiments ...... 60

3.3.7 Sensitivity analysis...... 60

3.3.8 Realized dispersal patterns ...... 60

3.4 Results ...... 61

3.4.1 The evolution of a condition- and density-dependent dispersal strategy ...... 61

3.4.2 The evolution of emigration rate and dispersal mortality ...... 63

3.4.3 Realized dispersal patterns ...... 63

3.5 Discussion ...... 65

3.6 References ...... 69 vii

Chapter 4 Phenotype-by-Environment Interactions Influence Dispersal: Results From a Mark-Release-Recapture Study in an Insect ...... 75

4.1 Abstract ...... 75

4.2 Introduction ...... 76

4.3 Methods...... 79

4.3.1 Study system ...... 79

4.3.2 Collection methods and survey design ...... 80

4.3.3 Classifying dispersers ...... 82

4.3.4 Statistical analysis ...... 83

4.4 Results ...... 85

4.5 Discussion ...... 90

4.6 References ...... 95

Chapter 5 Natal Habitat Conditions Have Carryover Effects on Dispersal Capacity and Behaviour ...... 100

5.1 Abstract ...... 100

5.2 Introduction ...... 100

5.3 Methods...... 104

5.3.1 Study system ...... 104

5.3.2 Manipulating natal habitat quality ...... 104

5.3.3 Dispersal experiment ...... 105

5.3.4 Measurement of dispersal capacity ...... 106

5.3.5 Statistical analysis ...... 107

5.4 Results ...... 109

5.4.1 Effects of habitat quality on survival, development rate, body size, and body composition ...... 109

5.4.2 Dispersal ...... 111

5.5 Discussion ...... 113

viii

5.6 References ...... 116

Chapter 6 The Effects of Parasites on Host Dispersal ...... 121

6.1 Part 1: Parasitism Risk Increases Host Dispersal Propensity ...... 121

6.1.1 Abstract ...... 121

6.1.2 Introduction ...... 122

6.1.3 Methods...... 124

6.1.4 Results ...... 126

6.1.5 Discussion ...... 127

6.2 Part 2: Parasite Infection Eliminates Density-Dependent Host Dispersal ...... 130

6.2.1 Abstract ...... 130

6.2.2 Introduction ...... 131

6.2.3 Methods...... 132

6.2.4 Results ...... 135

6.2.5 Discussion ...... 138

6.3 References ...... 142

Chapter 7 General Conclusion ...... 148

7.1 Concluding remarks ...... 148

7.2 References ...... 151

Appendix A: Supplementary material for chapter 2 ...... 152

Appendix B: Supplementary material for chapter 3 ...... 159

Appendix C: Supplementary material for chapter 4 ...... 161

Appendix D: Supplementary material for chapter 5 ...... 173

Appendix E: Supplementary material for chapter 6 ...... 186

Copyright Acknowledgements...... 189

ix

List of Tables

Table 1.1...... 2

Table 4.1...... 87

Table 4.2...... 87

Table 6.1...... 127

Table 6.2...... 137

x

List of Figures

Figure 1.1...... 13

Figure 2.1...... 34

Figure 2.2...... 35

Figure 2.3...... 36

Figure 2.4...... 40

Figure 3.1...... 62

Figure 3.2...... 63

Figure 3.3...... 64

Figure 4.1...... 80

Figure 4.2...... 86

Figure 4.3...... 88

Figure 4.4...... 89

Figure 4.5...... 90

Figure 5.1...... 109

Figure 5.2...... 110

Figure 5.3...... 111

Figure 5.4...... 112

Figure 5.5...... 113

Figure 6.1...... 127

Figure 6.2...... 136 xi

Figure 6.3...... 137

xii

List of Appendices

Appendix A: Supplementary material for Chapter 2

Appendix B: Supplementary material for Chapter 3

Appendix C: Supplementary material for Chapter 4

Appendix D: Supplementary material for Chapter 5

Appendix E: Supplementary material for Chapter 6

xiii

Chapter 1 General Introduction

1.1 Overview

In this thesis, I explore the evolution of dispersal behaviour, and how dispersal is influenced by environmental and phenotypic variables. I investigated dispersal using three approaches: i) quantitative review, ii) theoretical modeling, and iii) empirical studies. All of the empirical studies described in this thesis were conducted using the backswimmer, Notonecta undulata, as the study organism. In this general introduction, I define dispersal and describe how researchers study dispersal. I then discuss the ultimate and proximate causes of dispersal, and the consequences of dispersal for ecological and evolutionary dynamics. Finally, I provide a description of the natural history of N. undulata, focusing on aspects of their biology that are relevant to the empirical studies described in this thesis.

1.2 What is dispersal and how do we study it?

Dispersal is the movement of individuals or propagules from a natal patch to a breeding patch, or between successive breeding patches (Howard 1960, Matthysen 2012). The key feature of dispersal that distinguishes it from other types of movement, including migration and routine movement (i.e., movement for the purpose of resource exploitation; Van Dyck and Baguette 2005), is that dispersal produces the potential for gene flow (Ronce 2007, Table 1.1). Dispersal does not result in gene flow in every instance because individuals may die during dispersal, or fail to reproduce after settling in a new patch (Bonte et al. 2012). Dispersal cannot always be easily separated from other types of movement (Van Dyck and Baguette 2005). For example, animals may conduct routine foraging movements during dispersal, and animals may conduct migration and dispersal movement simultaneously if they settle in a patch other than their natal (or previous breeding) patch after migration (e.g., Morton 1992).

Dispersal consists of three stages: emigration, transfer, and immigration (Lidicker and Stenseth 1992). Emigration is movement out of a habitat patch. Transfer (also called transience) is movement between habitat patches (i.e., movement within the interhabitat matrix). Immigration

1

(also called settlement or colonization) is movement into and establishment in a habitat patch. Observing emigration requires that the researcher knows the boundaries of habitat patches, from the perspective of the organism of interest. This is simple in some cases. For example, backswimmers (flight-capable, semi-aquatic insects in the family Notonectidae) conduct all of their routine movements within a single water body (Hungerford 1934), and disperse by flight. Emigrants can then be unambiguously identified as individuals that fly out of a body of water. However, for many organisms, habitat patch boundaries are difficult to define. Researchers have developed protocols for identifying emigrants in these taxa. These protocols generally estimate the distance at which routine movements are conducted (i.e., home ranges) and define emigration as movement outside this range (Massot and Clobert 2000, Loe et al. 2010), or identify a behaviour that is specific to emigrants. An example of this latter method is ballooning in agrobiont spiders. Ballooning is a means of dispersal whereby silk is released and is caught by the wind, carrying the spider long distances. Emigrants are then defined as individuals that exhibit this ballooning behaviour, without having to estimate the point at which spiders move across habitat boundaries (Bonte et al. 2003, Bonte et al. 2006).

Table 1.1. A comparison of the three major types of movement: routine, migration, and dispersal. Movement Distance Causes gene type Direction traveled/day Frequency flow Routine Varies* Short Daily No Migration Round trip Long Annually No Dispersal One-way Varies Zero to several Yes times per lifetime *Routine movement can be round trip, one-way, or restricted to a home range.

The study of transience is generally simplified to measuring the magnitude of displacement between the patch from which the individual emigrated and the patch into which it immigrates. Displacement values are often represented as a dispersal kernel, which is the probability density function of the dispersal distances traveled by each individual in a population or species (Nathan et al. 2012). Dispersal kernels are commonly leptokurtic, meaning they peak at short distances

2

but have more long distance dispersal events than Gaussian or exponential kernels (Nathan et al. 2012). However, other distributions have also been observed. For example, McCauley et al. (2009) found that the semi-aquatic insect Notonecta irrorata exhibits a hump-shaped dispersal distance distribution.

An assumption of many studies of transience is that animals travel in straight lines between patches. This assumption is made because most studies collect information about an animal’s location before and after transience, but rarely collect precise location data during the time animals are in the interpatch matrix. Researchers therefore simplify the study of transience to the investigation of displacement values. Telemetry (i.e., remote tracking of organisms, usually via radio signals) has shown that real animals tend to follow curvy or tortuous paths (del Mar Delgado et al. 2010, Prevedello et al. 2010). Factors such as matrix complexity and body size influence the amount of tortuosity in movement paths (Haynes et al. 2006, Prevedello et al. 2010). In a study of the butterfly Proclossiana eunomia, Schtickzelle et al. (2006) found that populations in fragmented landscapes had evolved to have straighter transfer paths through the matrix, suggesting that tortuosity is an evolvable trait. However, data on movement paths is still rare, and therefore little is known about the mechanisms that generate within- and between- individual variation in movement paths.

The study of immigration, like emigration, requires that researchers are able to define habitat patches from the perspective of the organism of interest. Immigrants are generally identified as individuals that move into a distinct patch (or area defined by the researcher), or individuals that display immigration-specific behaviours such as web building or web takeover (Bonte et al. 2011), or the building of new burrows (Braude 2000). A major topic in the study of immigration is habitat selection, the process of choosing a habitat patch in which to settle (Stamps 2001). Habitat selection involves searching for suitable habitat patches, assessing patch quality, and establishment in a patch (Stamps 2001). Individual variation in search behaviour (Stamps 2006), quality assessments (Stamps and Davis 2006, Stamps et al. 2009), and establishment ability (Bonte et al. 2011, Remy et al. 2011) generates variation in habitat selection, and can result in non-random distributions of phenotypes in space (Jacob et al. 2015). For example, McCauley et al. (2010) found that the offspring of dragonflies (Libellula saturata) that colonized isolated

3

patches had higher foraging rates than the offspring of individuals that colonized more connected patches, producing spatial patterning in dragonfly foraging traits.

1.3 Causes of dispersal 1.3.1 Ultimate causes of dispersal

Dispersal is associated with a variety of costs during each dispersal stage. These costs can be split into four categories: energetic, risk, time, and opportunity costs (reviewed in Bonte et al. 2012). Energy is required to develop dispersal structures such as wings and flight muscles (Dixon et al. 1993, Dixon and Kindlmann 1999). Energetic investment in dispersal structures trades-off with investment in reproduction, resulting in reproductive delays, lower fecundity, and lower mating success (Roff 1984, Denno et al. 1989, Fujisaki 1992, Mole and Zera 1993, Crnokrak and Roff 1995). Energy is also required to power locomotion (Girard 2001, Gade et al. 2004, Gade et al. 2006), and for settlement behaviours including competing for territory (Bonte et al. 2011), and building burrows (Braude 2000).

Risk costs may involve an increased risk of mortality, morbidity, or injury. Mortality is often elevated in the matrix relative to suitable habitat (Bowler and Benton 2009, Trochet et al. 2013), and mortality increases with dispersal distance (Bowler and Benton 2009). This increased mortality is hypothesized to be due to increased predation risk (Yoder 2004) and high energy expenditure during transfer (Rand et al. 2006). Flight has been shown to trade-off with immune response, suggesting that animals that disperse by flight are at an increased risk of disease compared to non-dispersers (Adamo and Parsons 2006, Adamo et al. 2008).

Time costs are the result of dispersers spending time traveling rather than conducting fitness- gaining behaviours such as foraging or mate-searching. In some cases, these costs can be substantial, and greatly reduce individual fitness. For example, in marine bryozoans, long dispersal durations cause dispersers to have lower body condition at settlement, and lower reproductive rates (Burgess and Marshall 2011). Opportunity costs include the cost of moving to a patch which has an environment that is not suited to the disperser’s phenotype, and the cost of moving to unfamiliar areas. For example, Brown et al. (2008) found that cliff swallows (Petrochelidon pyrrhonota) that were familiar with their habitat patches had higher survival than

4

immigrants who were not familiar with the habitat. Opportunity costs may be reduced if individuals can select habitats to colonize which match their experience or phenotype (Jacob et al. 2015). For example, (Karpestam et al. 2012) experimentally manipulated the dorsal colour of grasshoppers (Tetrix subulata), and found that the grasshoppers preferentially moved to habitat patches with thermal conditions that were best suited to their manipulated phenotype. This habitat matching allowed the grasshoppers to avoid the fitness costs of settling in patches with non-ideal thermal conditions (Karpestam et al. 2012).

The costs of dispersal are numerous, and can substantially reduce fitness (Bonte et al. 2012). Why then do organisms disperse? The hypotheses that seek to explain the origin and maintenance of dispersal behaviour in populations across evolutionary time fall into three main categories: kin avoidance, risk spreading, and the tracking of high quality habitat (Matthysen 2012). Below, I discuss the theoretical underpinnings and empirical evidence for each of these hypotheses.

1.3.1.1 Kin avoidance

In their seminal paper, Hamilton and May (1977) developed a theoretical model which predicted that when offspring dispersal is under parental control, parents should force some offspring to disperse even when the costs of dispersal are high. Hamilton and May (1977) suggest that this strategy reduces competition among kin at the natal site. However, empirical evidence that kin competition induces dispersal is sparse, and is mainly from vertebrate systems (reviewed in Lambin 2001).

It has also been hypothesized that avoidance of inbreeding may be an ultimate cause of dispersal (Bengtsson 1978). In a long-term study of the great tit (Parus major), Szulkin and Sheldon (2008) found that birds that dispersed longer distance were less likely to breed with a close relative. However, negative correlations between dispersal and the probability of inbreeding do not demonstrate that avoidance of inbreeding is causing dispersal. Direct evidence for inbreeding as an ultimate cause of dispersal is lacking (Lambin et al. 2001). Perrin and Mazalov (1999) developed a theoretical model that predicted that when inbreeding avoidance is the only potential benefit of dispersal, populations should evolve extreme sex-biased dispersal where one sex

5

disperses and the other remains philopatric. Previous authors have suggested that sex-biased dispersal is evidence of inbreeding avoidance (e.g., Greenwood 1980). However, there are alternative hypotheses to explain sex biases, including asymmetries in intrasexual competition and sex-specific dispersal costs (Moore and Ali 1984, Wild and Taylor 2004, Gros et al. 2008). Therefore sex biases in dispersal are not definitive evidence for the evolution of dispersal as a mechanism to avoid inbreeding (Lambin et al. 2001). The current consensus is that while inbreeding avoidance may contribute to selection for dispersal, it is unlikely to be the sole driver of dispersal in most systems (Perrin and Mazalov 1999, Lambin et al. 2001, Szulkin and Sheldon 2008).

1.3.1.2 Risk spreading

In spatiotemporally heterogeneous environments, individuals who place all of their offspring in a single habitat patch risk having low lifetime reproductive success if that patch deteriorates in quality. Risk spreading is the process of distributing offspring across multiple patches to reduce temporal variance in fitness in heterogeneous environments (den Boer 1968). Some authors have interpreted spatial distribution of offspring as a strategy to spread risk (Root and Kareiva 1984, Cronin and Strong 1993). However, Courtney (1986) has suggested that selection acting to reduce variance in fitness is weak, and therefore risk spreading is unlikely to be an important driver of dispersal. Courtney suggests that an alternative hypothesis to explain observations of individuals distributing offspring across space is that these individuals are preventing kin competition among their offspring (Courtney 1986), but current evidence cannot support one hypothesis over the other. A review of risk spreading in insects failed to find any studies with definitive evidence for risk spreading as a driver of dispersal morphology or behaviour (Hopper 1999).

1.3.1.3 Tracking high quality habitat

The final hypothesis for the ultimate cause of dispersal is that dispersal is an adaptation for tracking habitats that are favourable for survival and reproduction. This hypothesis predicts that dispersal can only be beneficial if there is spatial variance in patch quality, so that there is a moderate to high probability of finding a higher quality patch if the current one is low quality (Southwood 1977, Morris 1987). The extent of spatial autocorrelation in patch quality must be 6

smaller than the maximum dispersal distance of the organism, so that the organism is capable of leaving its low quality patch and locating a higher quality patch (Ims and Hjermann 2001). Finally, there must be temporal variance and positive temporal autocorrelation in patch quality, such that a patch that is currently low quality will continue to be low quality long enough that there is a benefit to leaving that patch (Ims and Hjermann 2001). Using a theoretical model, McPeek and Holt (1992) found that dispersal was selected for in metapopulations in which there was spatial variation in environmental conditions that produced spatial variation in fitness. Individuals are predicted to emigrate from patches where they expect to have low survival or reproductive success, in favour of reproducing in patches where their expected lifetime reproductive success (or that of their offspring) is high (McPeek and Holt 1992).

If dispersal is caused by individuals seeking high quality habitat in which to reproduce, then it follows that individuals should preferentially disperse away from low quality habitat. Numerous studies support this prediction. For example, dispersal rates are elevated in habitats with high predation risk (Cronin et al. 2004, McCauley and Rowe 2010), high parasitism risk (Brown and Brown 1992, Sloggett and Weisser 2002), and low food availability (Kuussaari et al. 1996, Kennedy and Ward 2003). Bonte et al. (2014) directly tested the hypothesis that individuals emigrate away from habitat patches where their expected fitness is low. In an experiment on the spider mite Tetranychus urticae, they found that individuals who possessed phenotypes which conferred a fitness advantage in low density patches dispersed from high density to low density patches; whereas, individuals who had a fitness advantage in high density patches remained in high density patches (Bonte et al. 2014). Overall, the literature provides strong evidence that dispersal has evolved in spatiotemporally variable landscapes in order to increase the likelihood of reproducing in habitats which maximize lifetime reproductive success.

1.3.2 Proximate causes of dispersal

Many empirical studies have investigated the effects of ecological conditions (biotic and abiotic) and phenotypic traits on dispersal. Evidence from these studies demonstrates that dispersal is influenced by environmental conditions including temperature, humidity, population density, predation risk, parasite presence/density, and sex ratio (reviewed in Clobert et al. 2001, Bowler and Benton 2005, Clobert et al. 2012). Additionally dispersers are a non-random sample of

7

populations with respect to phenotypes including sex, body size, body condition, personality, metabolic rate, and hormone levels (reviewed in Clobert et al. 2001, Bowler and Benton 2005, Clobert et al. 2009, Cote et al. 2010, Clobert et al. 2012, Mabry et al. 2013). The dispersal literature is characterized by a high level of variability: studies come to variable and sometimes contradictory conclusions about the association between dispersal and ecological conditions, and between dispersal and phenotype (Ims and Hjermann 2001, Clobert et al. 2009). For example, the effect of population density on dispersal is positive in some studies, negative in others, and several have observed no association between density and dispersal (reviewed in Matthysen 2005). Similarly, some studies have observed male-biased dispersal, some have observed female- biased dispersal, and others have found no evidence for a sex bias (reviewed in Mabry et al. 2013). The reasons for this variability have not been resolved, though several hypotheses have been proposed.

Most of the hypotheses that attempt to explain variability in dispersal patterns focus on the relationship between dispersal and a single environmental factor or phenotype. For example, previous authors have suggested that variation in density-dependent dispersal is due to variation across species in the shape of the function relating fitness to density. This variation could be the result of differences across species in the importance of Allee effects, the benefits of group living, or the strength of intraspecific competition (Bowler and Benton 2005, Matthysen 2005). These hypotheses are distinct from hypotheses that have been proposed to explain variability in relationships between dispersal and other factors. For example, variability in the relationship between dispersal and sex has been hypothesized to be due to differences across species in mating strategy (Greenwood 1980), or the magnitude of sex differences in spatiotemporal variation in fitness (Gros et al. 2009). These previous hypotheses (and others put forth to explain variability between dispersal and other variables) may contribute to variability in dispersal patterns; however, a more parsimonious explanation is provided by the “multiple, interacting causes of dispersal” hypothesis. This hypothesis posits that the fitness consequences of dispersal are dependent on many environmental conditions and phenotypes simultaneously. Consequently, dispersal behaviour should depend on interactions between multiple environmental conditions and/or phenotypes (Ims and Hjermann 2001, Matthysen 2012). Since dispersal is difficult to study (Nathan 2001), researchers often investigate dispersal behaviour in a small number of

8

individuals, and therefore lack the statistical power required to measure interactive effects. If multiple studies measure the same effect without considering interacting variables, they will likely come to different, and potentially contradictory conclusions about dispersal patterns. The multiple, interacting causes of dispersal hypothesis therefore provides a general explanation for the observed variability in dispersal. The goal of this dissertation is to test the main prediction of this hypothesis: that interactions among multiple environmental variables and/or phenotypic traits produce variability in dispersal behaviour. This was an expansive goal, and therefore to accomplish it I focused the scope of my investigations to a subset of the variables that influence dispersal. I asked i) how habitat quality influences body size and body condition, and in turn how habitat quality and size/condition interact to influence dispersal, and ii) how the effect of population density on dispersal is determined by ecological context and the phenotypic distributions of populations.

1.4 Consequences of dispersal 1.4.1 Consequences of dispersal for ecological dynamics

Dispersal reduces the risk of metapopulations going extinct by enabling the movement of immigrants into populations that would not persist in a closed system. Populations that experience stochastic extinction can be re-colonized by individuals from other patches in the metapopulation, reducing the probability of whole metapopulation extinction (Levins 1969). Small populations are at particularly high risk of extinction from stochastic forces, and as the result of Allee effects. Immigration into these populations increases population size and provides a source of genetic variation, and thereby decreases the probability of stochastic extinction and reduces the strength of Allee effects (rescue effects; Brown and Kodric-Brown 1977). When habitat conditions within a patch (i.e., abiotic conditions, food availability, etc.) are too poor to maintain populations, higher quality patches can provide a source of immigrants that can increase the probability of persistence for populations in low quality “sink” habitats (source-sink dynamics; Pulliam 1988). Dispersal allows individuals to track high quality habitat in spatiotemporally heterogeneous landscapes, which increases the likelihood that species persist at a regional scale (spatial insurance; Loreau et al. 2003). Dispersal can also maintain biodiversity by promoting the coexistence of species with their competitors, predators, and parasites. Species

9

that are driven locally extinct when they co-occur with natural enemies can persist regionally by dispersing to spatial refuges that lack these enemies (Sabelis and Diekmann 1988, Comins and Hassell 1996, Amarasekare 2010). Empirical evidence supports the prediction that dispersal among patches within a metacommunity maintains biodiversity (Staddon et al. 2010).

Dispersal is also the mechanism by which species shift their ranges, including following the introduction of foreign species (Renault et al. 2018). Many species have shifted their ranges towards the poles in response to climate change (Hickling et al. 2006, Chen et al. 2011), and for some species these shifts may be critical for avoiding extinction as the climate continues to change. The magnitude of those range shifts has been predicted to be related to the dispersal behaviour and dispersal ability of species (Angert et al. 2011). For example, flight-capable insect species are expected to shift faster than related species that lack wings. However, the question of how species traits influence range shifts has not been resolved in the literature.

1.4.2 Consequences of dispersal for evolutionary dynamics

Dispersal is the process by which gene flow occurs (Ronce 2007), and therefore it has important consequences for evolutionary dynamics. Gene flow introduces genetic variation and reduces the amount of genetic divergence between populations caused by drift, and so can influence the genetic structure of metapopulations (Bohonak 1999). Gene flow may impede local adaptation by flooding populations with locally maladaptive alleles, or untested allele combinations (Wade and Goodnight 1998, Rasanen and Hendry 2008). Conversely, the introduction of novel alleles or allele combinations may promote local adaptation when these alleles are beneficial, or they cause populations to move to new adaptive peaks (Slatkin 1987, Holt and Gomulkiewicz 1997). Local adaptation may also be promoted by phenotype-habitat matching, whereby individuals settle in patches with environments that are suitable for their phenotype, causing assortative mating, and genetic divergence between patches with different environments (Edelaar et al. 2008). Gene flow constrains speciation (Slatkin 1987), however, there is still ongoing debate about whether speciation can occur in sympatry, where there are no barriers to gene flow (Edelaar et al. 2008, Foote 2018).

10

1.4.3 Consequences of context- and phenotype-dependent dispersal

Dispersal is influenced by a variety of environmental variables and phenotypes (reviewed in Bowler and Benton 2005). Context and phenotype-dependent dispersal can have different consequences for ecological and evolutionary dynamics than dispersal that is random with respect to environmental conditions and phenotype. For example, Vuilleumier and Possingham (2006) demonstrated that when habitat patches vary in their rates of colonization, the risk of metapopulation extinction is higher than when colonization rates are equal across patches. Unequal dispersal rates are common in natural systems, because organisms are more likely to emigrate from low quality patches and immigrate into high quality patches (Baines et al. 2014, Pintar et al. 2018).

The effect of immigrants on the dynamics of the populations they colonize will depend on phenotypic traits. For example, populations colonized by marine bryozoans in low body condition have lower population growth rates than populations colonized by individuals in high body condition (Burgess and Marshall 2011). The rate of gene flow in metapopulations will also depend on the phenotypes of dispersers. For example, if dispersers have lower fecundity than non-dispersers (e.g., Zera 1984), the magnitude of gene flow will be lower than would be expected assuming that dispersal is not associated with phenotype (Benard and McCauley 2008). This will have cascading effects on local adaptation, speciation, and the genetic structure of metapopulations.

Phenotype-dependent dispersal may also influence evolution via sorting of phenotypes across space (Shine et al. 2011). For example, studies of the invasion of cane toads (Rhinella marina) in Australia have found that these toads exhibit phenotype-dependent dispersal. Toads are more likely to move to the forefront of the invasion if they have longer legs, longer dispersal periods, and straighter dispersal movements (Phillips et al. 2006, Lindstrom et al. 2013). Individuals with these dispersal-promoting traits are over-represented at the invasion front, which has caused assortative mating with respect to dispersal ability, resulting in the evolution of highly dispersive toads. As a consequence, the rate of invasion through Australia has increased over time (Phillips et al. 2006, Lindstrom et al. 2013).

11

1.5 Study system

Chapters 4, 5 and 6 in this thesis describe empirical studies of the same species, the backswimmer Notonecta undulata (Figure 1.1). Notonecta spp. (Heteroptera: Notonectidae) are semiaquatic insects that live in ponds, streams, and lakes (Hungerford 1934). They complete their life cycle entirely in the aquatic environment. They swim under the water, but breathe air at the surface (Essenberg 1915). Notonecta undergo incomplete metamorphosis; they have five juvenile instars before adult maturation (Hungerford 1934). Both adult and juvenile Notonecta are predacious on a variety of aquatic and semi-aquatic prey (including zooplankton, Diptera and Odonata larvae, amphibians, fish, and other Heteropterans), but also scavenge for terrestrial prey trapped on the water surface (Essenberg 1915).

Notonecta undulata are broadly distributed across North America (Hungerford 1934), and are typically found in fishless ponds (Streams 1987). The empirical studies described in this thesis were conducted at the University of Toronto’s Koffler Scientific Reserve (KSR; King, Ontario, Canada; 44°01ʹN, 79°32ʹW). In this region, N. undulata lays eggs primarily in the spring months (approximately March – June; C. Baines, personal observation). Female N. undulata can oviposit multiple times across the breeding season. Adults die at the conclusion of the breeding period (C. Baines, personal observation). Eggs hatch approximately two weeks after being laid (Hungerford 1934). Juveniles develop through the summer and reach adult maturity in the period from late June to mid-August, depending on water temperature and habitat quality (C. Baines, personal observation). Notonecta undulata overwinters in the adult stage (Hungerford 1934). We have observed mating interactions occurring in March immediately after ice thaw, but since notonectids have been observed to be actively swimming under the ice during winter (S. McCauley, personal observation), mating may begin before ice thaw. Notonecta undulata is generally univoltine in the region of my study site, but some populations have two generations per year under ideal conditions (C. Baines, personal observation).

Notonecta undulata is flight capable in the adult stage (Hungerford 1934). Juveniles are wingless, and restricted to the body of water in which they were oviposited. Adult notonectids use flight only as a means of dispersal. Routine behaviours such as foraging are conducted entirely in the aquatic environment (Hungerford 1934). Therefore, dispersers can be 12

unambiguously defined as individuals that fly out of a water body. All individuals in the populations studied here appear to be have fully developed wings and wing muscles for their entire adult lives (C. Baines, unpublished data), but dispersal behaviour is plastic. Dispersal in this species is one-way movement; once notonectids disperse by flight away from their natal ponds, they do not return (C. Baines, personal observation).

Figure 1.1. A) Dorsal view of a juvenile N. undulata. Juvenile notonectids lack wings. Photo by: Bansari Patel. B) Dorsal view of an adult N. undulata. The fragile wings of notonectids are covered and protected by hemelytra. Photo by: Bansari Patel. C) Anterior-ventral view of a live adult N. undulata in water. The notonectid is shown in a stereotypical posture: ventral side facing upward with the posterior end of the body angled up, and the hind legs extended. Photo by: Shannon McCauley. Note that photos are not shown to scale. 13

The dispersal kernel has not been described for this species. However, a previous study demonstrated that the dispersal kernel of a related species, N. irrorata, is hump-shaped, and ranges from approximately 0 to 1200 m, with a peak ~600 m from the nearest source population (McCauley et al. 2009). The maximum dispersal distance observed for N. undulata is ~1000 m (C. Baines, unpublished data).

Notonecta undulata are commonly infected with Hydrachnidia mites. In the region of my study site (KSR), incidence of mite infection ranges from 0 – 80% of notonectids infected within a given pond (Baines and McCauley, unpublished data). Hydrachnidia larvae parasitize aquatic insects, while the adults are predaceous, mainly on insect eggs and larvae (Di Sabatino et al. 2000). Hydrachnidia are generally not host-specific, but parasitize several species or families in multiple insect orders including Heteroptera, Odonata, Coleoptera, and Diptera (Smith 1988, Di Sabatino et al. 2000). Parasitic mites reduce survival and fecundity in their insect hosts (reviewed in Smith 1988). Parasitic larval mites tend to attach to insect larvae without feeding, and begin to engorge soon after the host undergoes its final moult to the adult form, when the host integument is vulnerable (Smith 1988, Di Sabatino et al. 2000). In N. undulata, Hydrachnidia typically attach to the dorsum beneath the hemelytra, or on the underside of the hemelytra (C. Baines, personal observation). The parasitic larvae can remain attached to their hosts for several weeks (Smith 1988).

1.6 Summary of chapters

The central question of my thesis is, what produces variation in dispersal among habitat patches and among individuals? I address this in five chapters using quantitative review, individual based modeling, and empirical studies. In chapter 2, I use a meta-analytic approach to test how dispersal is related to body size and body condition. I then review the evidence for the “multiple, interacting causes of dispersal” hypothesis, which proposes that dispersal patterns are not fixed, but vary across contexts because of interactive effects acting on dispersal capacity and behaviour. In chapter 3, I develop an individual based model (IBM) of the evolution of dispersal in response to the joint effects of body condition and population density. I investigate how the cost of dispersal influences the dispersal strategy that evolves, as well as emigration rates, and dispersal mortality. I also explore how dispersal strategies translate to realized dispersal patterns. 14

In chapters 4-6 I describe the results of empirical studies conducted on the backswimmer Notonecta undulata. In chapter 4, I present the results of a mark-release-recapture study conducted on backswimmers in a set of 36 semi-natural ponds. I investigated the interactive effects of environmental variables (population density and sex ratio) and phenotypic traits (body mass and sex) on emigration probability and dispersal distance. In chapter 5, I present the results of an experiment I conducted to investigate the carryover effects of natal patch quality on adult dispersal behaviour. I raised juvenile backswimmers under different habitat quality treatments, and when the backswimmers reached adult maturity, I measured their emigration rates under three different adult habitat quality treatments. I tested whether natal habitat conditions influenced their dispersal response to habitat quality as adults. Finally, in chapter 6, I present the results of a series of studies investigating the effects of ectoparasitic mites on dispersal in backswimmers. I tested whether cues indicating the risk of parasitism increased emigration probability in uninfected backswimmers. I also tested whether infection by mites reduced the dispersal ability of backswimmers. Finally, I used a mark-release-recapture study to test how the combined effects of parasitism risk and parasite infection influenced dispersal behaviour in a natural system.

1.7 References

Adamo S.A., and N.M. Parsons (2006). The emergency life-history stage and immunity in the cricket, Gryllus texensis. Animal Behaviour. 72: 235-244. Adamo S.A., J.L. Roberts, R.H. Easy, and N.W. Ross (2008). Competition between immune function and lipid transport for the protein apolipophorin III leads to stress-induced immunosuppression in crickets. Journal of Experimental Biology. 211: 531-538. Amarasekare P. (2010). Effect of non-random dispersal strategies on spatial coexistence mechanisms. Journal of Animal Ecology. 79: 282-293. Angert A.L., L.G. Crozier, L.J. Rissler, S.E. Gilman, J.J. Tewksbury, and A.J. Chunco (2011). Do species' traits predict recent shifts at expanding range edges? Ecology Letters. 14: 677-689. Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287.

15

Benard M.F., and S.J. McCauley (2008). Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist. 171: 553-567. Bengtsson B.O. (1978). Avoiding inbreeding: at what cost? Journal of Theoretical Biology. 73: 439-444. Bohonak A.J. (1999). Dispersal, gene flow, and population structure. The Quarterly Review of Biology. 74: 21-45. Bonte D., J.V. Borre, L. Lens, and M. Jean-Pierre (2006). Geographical variation in wolf spider dispersal behaviour is related to landscape structure. Animal Behaviour. 72: 655-662. Bonte D., N. De Meester, and E. Matthysen (2011). Selective integration advantages when transience is costly: immigration behaviour in an agrobiont spider. Animal Behaviour. 81: 837-841. Bonte D., A. De Roissart, N. Wybouw, and T. Van Leeuwen (2014). Fitness maximization by dispersal: evidence from an invasion experiment. Ecology. 95: 3104-3111. Bonte D., H. Van Dyck, J.M. Bullock, A. Coulon, M. Delgado, M. Gibbs, V. Lehouck, E. Matthysen, K. Mustin, M. Saastamoinen, N. Schtickzelle, V.M. Stevens, S. Vandewoestijne, M. Baguette, K. Barton, T.G. Benton, A. Chaput-Bardy, J. Clobert, C. Dytham, T. Hovestadt, C.M. Meier, S.C. Palmer, C. Turlure, and J.M. Travis (2012). Costs of dispersal. Biological Reviews. 87: 290-312. Bonte D., N. Vandenbroecke, L. Lens, and J.P. Maelfait (2003). Low propensity for aerial dispersal in specialist spiders from fragmented landscapes. Proceedings of the Royal Society B: Biological Sciences. 270: 1601-1607. Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Bowler D.E., and T.G. Benton (2009). Variation in dispersal mortality and dispersal propensity among individuals: the effects of age, sex and resource availability. Journal of Animal Ecology. 78: 1234-1241. Braude S. (2000). Dispersal and new colony formation in wild naked mole-rats: evidence against inbreeding as the system of mating. Behavioral Ecology. 11: 7-12. Brown C.R., and M.B. Brown (1992). Ectoparasitism as a cause of natal dispersal in cliff swallows. Ecology. 73: 1718-1723.

16

Brown C.R., M.B. Brown, and K.R. Brazeal (2008). Familiarity with breeding habitat improves daily survival in colonial cliff swallows. Animal Behaviour. 76: 1201-1210. Brown J.H., and A. Kodric-Brown (1977). Turnover rates in insular biogeography: effect of immigration on extinction. Ecology. 58: 445-449. Burgess S.C., and D.J. Marshall (2011). Are numbers enough? Colonizer phenotype and abundance interact to affect population dynamics. Journal of Animal Ecology. 80: 681- 687. Chen I.-C., J.K. Hill, R. Ohlemüller, D.B. Roy, and C.D. Thomas (2011). Rapid range shifts of species associated with high levels of climate warming. Science. 333: 1024-1026. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock (eds.) (2012). Dispersal ecology and evolution. Oxford University Press, Oxford. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols (eds.) (2001). Dispersal. Oxford University Press, Oxford. Clobert J., J.F. Le Galliard, J. Cote, S. Meylan, and M. Massot (2009). Informed dispersal, heterogeneity in animal dispersal syndromes and the dynamics of spatially structured populations. Ecology Letters. 12: 197-209. Comins H.N., and M.P. Hassell (1996). Persistence of multispecies host-parasitoid interactions in spatially distributed models with local dispersal. Journal of Theoretical Biology. 183: 19-28. Cote J., J. Clobert, T. Brodin, S. Fogarty, and A. Sih (2010). Personality-dependent dispersal: characterization, ontogeny and consequences for spatially structured populations. Philosophical Transactions of the Royal Society of London B: Biological Sciences. 365: 4065-4076. Courtney S.P. (1986). Why insects move between host patches: some comments on 'risk- spreading'. Oikos. 47: 112-114. Crnokrak P., and D.A. Roff (1995). Fitness differences associated with calling behaviour in the two wing morphs of male sand crickets, Gryllus firmus. Animal Behaviour. 50: 1475- 1481. Cronin J.T., K.J. Haynes, and F. Dillemuth (2004). Spider effects on planthopper mortality, dispersal, and spatial population dynamics. Ecology. 85: 2134-2143.

17

Cronin J.T., and D.R. Strong (1993). Substantially submaximal oviposition rates by a Mymarid egg parasitoid in the laboratory and field. Ecology. 74: 1813-1825. del Mar Delgado M., V. Penteriani, E. Revilla, and V.O. Nams (2010). The effect of phenotypic traits and external cues on natal dispersal movements. Journal of Animal Ecology. 79: 620-632. den Boer P.J. (1968). Spreading of risk and stabilization of animal numbers. Acta Biotheoretica. 18: 165-194. Denno R.F., K.L. Olmstead, and E.S. McCloud (1989). Reproductive cost of flight capability: a comparison of life history traits in wing dimorphic planthoppers. Ecological Entomology. 14: 31-44. Dixon A.F.G., S. Horth, and P. Kindlmann (1993). Migration in insects: cost and strategies. Journal of Animal Ecology. 62: 182-190. Dixon A.F.G., and P. Kindlmann (1999). Cost of flight apparatus and optimum body size of aphid migrants. Ecology. 80: 1678-1690. Edelaar P., A.M. Siepielski, and J. Clobert (2008). Matching habitat choice causes directed gene flow: a neglected dimension in evolution and ecology. Evolution. 62: 2462-2472. Essenberg C. (1915). The habits and natural history of the backswimmers Notonectidae. Journal ofAnimal Behavior. 5: 381-390. Foote A.D. (2018). Sympatric speciation in the genomic era. Trends in Ecology and Evolution. 33: 85-95. Fujisaki K. (1992). A male fitness advantage to wing reduction in the oriental chinch bug, Cavelerius saccharivorus Okajima (Heteroptera: Lygaeidae). Research in Population Ecology. 34: 173-183. Gade G., L. Auerswald, and H.G. Marco (2006). Flight fuel and neuropeptidergic control of fuel mobilisation in the twig wilter, Holopterna alata (Hemiptera, Coreidae). Journal of Insect Physiology. 52: 1171-1181. Gade G., L. Auerswalk, R. Predel, and H.G. Marco (2004). Substrate usage and its regulation during flight and swimming in the backswimmer, Notonecta glauca. Phsyiological Entomology. 29: 84-93.

18

Girard I. (2001). Field cost of activity in the kit fox, Vulpes macrotis. Physiological and Biochemical Zoology: Ecological and Evolutionary Approaches. 74: 191-202. Greenwood P.J. (1980). Mating systems, philopatry and dispersal in birds and mammals. Animal Behaviour. 28: 1140-1162. Gros A., T. Hovestadt, and H.J. Poethke (2008). Evolution of sex-biased dispersal: The role of sex-specific dispersal costs, demographic stochasticity, and inbreeding. Ecological Modelling. 219: 226-233. Gros A., H.J. Poethke, and T. Hovestadt (2009). Sex-specific spatio-temporal variability in reproductive success promotes the evolution of sex-biased dispersal. Theoretical Population Biology. 76: 13-18. Hamilton W.D., and R.M. May (1977). Dispersal in stable habitats. Nature. 269: 578-581. Haynes K.J., J.T. Cronin, and A. de Roos (2006). Interpatch movement and edge effects: the role of behavioral responses to the landscape matrix. Oikos. 113: 43-54. Hickling R., D.B. Roy, J.K. Hill, R. Fox, and C.D. Thomas (2006). The distributions of a wide range of taxonomic groups are expanding polewards. Global Change Biology. 12: 450- 455. Holt R.D., and R. Gomulkiewicz (1997). How does immigration influence local adaptation? A reexamination of a familiar paradigm. The American Naturalist. 149: 563-572. Hopper K.R. (1999). Risk-spreading and bet-hedging in insect population biology. Annual Review of Entomology. 44: 535-560. Howard W.E. (1960). Innate and environmental dispersal of individual vertebrates. American Midland Naturalist. 63: 152-161. Hungerford H.B. (1934). The genus Notonecta of the world (Notonectidae-Hemiptera). University of Kansas Science Bulletin. 21: 5-195. Ims R.A., and D.O. Hjermann (2001). Condition-dependent dispersal. In: Dispersal (eds. Clobert J, E Danchin, AA Dhondt, and JD Nichols). Oxford University Press Oxford, pp. 203- 216. Jacob S., E. Bestion, D. Legrand, J. Clobert, and J. Cote (2015). Habitat matching and spatial heterogeneity of phenotypes: implications for metapopulation and metacommunity functioning. Evolutionary Ecology. 29: 851-871.

19

Karpestam E., L. Wennersten, and A. Forsman (2012). Matching habitat choice by experimentally mismatched phenotypes. Evolutionary Ecology. 26: 893-907. Kennedy P.L., and J.M. Ward (2003). Effects of experimental food supplementation on movements of juvenile northern goshawks (Accipiter gentilis atricapillus). Oecologia. 134: 284-291. Kuussaari M., M. Nieminen, and I. Hanski (1996). An experimental study of migration in the Glanville fritillary butterfly Melitaea cinxia. Journal of Animal Ecology. 65: 791-801. Lambin X., J. Aars, and S.B. Piertney (2001). Dispersal, intraspecific competition, kin competition and kin facilitation: a review of the empirical evidence. In: Dispersal (eds. Clobert J, E Danchin, AA Dhondt, and JD Nichols). Oxford University Press Oxford, pp. 110-122. Levins R. (1969). Some demographic and genetic consequences of environmental heterogeneity for biological control. Bulletin of the Entomological Society of America. 15: 237-240. Lidicker W.Z., and N.C. Stenseth (1992). To disperse or not to disperse: who does it and why? In: Animal Dispersal: Small mammals as a model. (eds. Stenseth NC, and WZ Lidicker). Chapman and Hall London, pp. 21-36. Lindstrom T., G.P. Brown, S.A. Sisson, B.L. Phillips, and R. Shine (2013). Rapid shifts in dispersal behavior on an expanding range edge. Proceedings of the National Academy of Sciences. 110: 13452-13456. Loe L.E., A. Mysterud, V. Veiberg, and R. Langvatn (2010). No evidence of juvenile body mass affecting dispersal in male red deer. Journal of Zoology. 280: 84-91. Loreau M., N. Mouquet, and A. Gonzalez (2003). Biodiversity as spatial insurance in heterogeneous landscapes. Proceedings of the National Academy Sciences. 100: 12765- 12770. Mabry K.E., E.L. Shelley, K.E. Davis, D.T. Blumstein, and D.H. Van Vuren (2013). Social mating system and sex-biased dispersal in mammals and birds: a phylogenetic analysis. PLoS One. 8: e57980. Massot M., and J. Clobert (2000). Processes at the origin of similarities in dipsersal behaviour among siblings. Journal of Evolutionary Biology. 13: 707-719.

20

Matthysen E. (2005). Density-dependent dispersal in birds and mammals. Ecography. 28: 403- 416. Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J, M Baguette, TG Benton, and JM Bullock). Oxford University Press Oxford, pp. 3-18. McCauley S.J., T. Brodin, and J. Hammond (2010). Foraging rates of larval dragonfly colonists are positively related to habitat isolation: results from a landscape-level experiment. The American Naturalist. 175: E66-E73. McCauley S.J., C.J. Davis, J. Nystrom, and E.E. Werner (2009). A hump-shaped relationship between isolation and abundance of Notonecta irrorata colonists in aquatic mesocosms. Ecology. 90: 2635-2641. McCauley S.J., and L. Rowe (2010). Notonecta exhibit threat-sensitive, predator-induced dispersal. Biology Letters. 6: 449-452. McPeek M.A., and R.D. Holt (1992). The evolution of dispersal in spatially and temporally varying environments. The American Naturalist. 140: 1010-1027. Mole S., and A.J. Zera (1993). Differential allocation of resources underlies the dispersal- reproduction trade-off in the wing-dimorphic cricket, Gryllus rubens. Oecologia. 93: 121-127. Moore J., and R. Ali (1984). Are dispersal and inbreeding avoidance related? Animal Behaviour. 32: 94-112. Morris D.W. (1987). Spatial scale and the cost of density-dependent habitat selection. Evolutionary Ecology. 1: 379-388. Morton M.L. (1992). Effects of sex and birth date on premigration biology, migration schedules, return rates and natal dispersal in the mountain white-crowned sparrow. The Condor. 94: 117-133. Nathan R. (2001). The challenges of studying dispersal. Trends in Ecology and Evolution. 16: 481-483. Nathan R., E. Klein, J.J. Robledo-Arnuncio, and E. Revilla (2012). Dispersal kernels: review. In: Dispersal ecology and evolution (eds. Clobert J, M Baguette, TG Benton, and JM Bullock). Oxford University Press Oxford, pp. 187-210.

21

Perrin N., and V. Mazalov (1999). Dispersal and inbreeding avoidance. The American Naturalist. 154: 282-292. Phillips B.L., G.P. Brown, J.K. Webb, and R. Shine (2006). Invasion and the evolution of speed in toads. Nature. 439: 803. Pintar M.R., J.R. Bohenek, L.L. Eveland, W.J. Resetarits, and C. Seymour (2018). Colonization across gradients of risk and reward: Nutrients and predators generate species-specific responses among aquatic insects. Functional Ecology. 32: 1589-1598. Prevedello J.A., G. Forero-Medina, and M.V. Vieira (2010). Movement behaviour within and beyond perceptual ranges in three small mammals: effects of matrix type and body mass. Journal of Animal Ecology. 79: 1315-1323. Pulliam H.R. (1988). Sources, sinks, and population regulation. The American Naturalist. 132: 652-661. Rand P.S., S.G. Hinch, J. Morrison, M.G.G. Foreman, M.J. MacNutt, J.S. Macdonald, M.C. Healey, A.P. Farrell, and D.A. Higgs (2006). Effects of river discharge, temperature, and future climates on energetics and mortality of adult migrating Fraser River sockeye salmon. Transactions of the American Fisheries Society. 135: 655-667. Rasanen K., and A.P. Hendry (2008). Disentangling interactions between adaptive divergence and gene flow when ecology drives diversification. Ecology Letters. 11: 624-636. Remy A., J.F. Le Galliard, G. Gundersen, H. Steen, and H.P. Andreassen (2011). Effects of individual condition and habitat quality on natal dispersal behaviour in a small rodent. Journal of Animal Ecology. 80: 929-937. Renault D., M. Laparie, S.J. McCauley, and D. Bonte (2018). Environmental adaptations, ecological filtering, and dispersal central to insect invasions. Annual Review of Entomology. 63: 345-368. Roff D.A. (1984). The cost of being able to fly: a study of wing polymorphism in two species of crickets. Oecologia. 63: 30-37. Ronce O. (2007). How does it feel to be like a rolling stone? Ten questions about dispersal evolution. Annual Review of Ecology, Evolution, and Systematics. 38: 231-253.

22

Root R.B., and P.M. Kareiva (1984). The search for resources by cabbage butterflies (Pieris rapae): ecological consequences and adaptive significance of Markovian movements in a patchy environment. Ecology. 65: 147-165. Sabelis M.W., and O. Diekmann (1988). Overall population stability despite local extinction: the stabilizing influence of prey dispersal from predator-invaded patches. Theoretical Population Biology. 34: 169-176. Schtickzelle N., G. Mennechez, and M. Baguette (2006). Dispersal depression with habitat fragmentation in the bog fritillary butterfly. Ecology. 87: 1057-1065. Shine R., G.P. Brown, and B.L. Phillips (2011). An evolutionary process that assembles phenotypes through space rather than through time. Proceedings of the National Academy of Sciences. 108: 5708-5711. Slatkin M. (1987). Gene flow and the geographic structure of natural populations. Science. 236: 787-792. Sloggett J.J., and W.W. Weisser (2002). Parasitoids induce production of the dispersal morph of the pea aphid, Acyrthosiphon pisum. Oikos. 98: 323-333. Southwood T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of Animal Ecology. 46: 336-365. Staddon P., Z. Lindo, P.D. Crittenden, F. Gilbert, and A. Gonzalez (2010). Connectivity, non- random extinction and ecosystem function in experimental metacommunities. Ecology Letters. 13: 543-552. Stamps J. (2001). Habitat selection by dispersers: Integrating proximate and ultimate approaches. In: Dispersal (eds. Clobert J, E Danchin, AA Dhondt, and JD Nichols). Oxford University Press Oxford, pp. 230-242. Stamps J.A. (2006). The silver spoon effect and habitat selection by natal dispersers. Ecology Letters. 9: 1179-1185. Stamps J.A., and J.M. Davis (2006). Adaptive effects of natal experience on habitat selection by dispersers. Animal Behaviour. 72: 1279-1289. Stamps J.A., V.V. Krishnan, and N.H. Willits (2009). How different types of natal experience affect habitat preference. The American Naturalist. 174: 623-630.

23

Streams F. A. (1987). Foraging behavior in a notonectid assemblage. American Midland Naturalist. 117: 353-361. Szulkin M., and B.C. Sheldon (2008). Dispersal as a means of inbreeding avoidance in a wild bird population. Proceedings of the Royal Society B: Biological Sciences. 275: 703-711. Trochet A., D. Legrand, N. Larranaga, S. Ducatez, O. Calvez, J. Cote, J. Clobert, and M. Baguette (2013). Population sex ratio and dispersal in experimental, two-patch metapopulations of butterflies. Journal of Animal Ecology. 82: 946-955. Van Dyck H., and M. Baguette (2005). Dispersal behaviour in fragmented landscapes: Routine or special movements? Basic and Applied Ecology. 6: 535-545. Vuilleumier S., and H.P. Possingham (2006). Does colonization asymmetry matter in metapopulations? Proceedings of the Royal Society B: Biological Sciences. 273: 1637- 1642. Wade M.J., and C.J. Goodnight (1998). Perspective: the theories of Fisher and Wright in the context of metapopulations: when nature does many small experiments. Evolution. 52: 1537-1553. Wild G., and P.D. Taylor (2004). Kin selection models for the co-evolution of the sex ratio and sex-specific dispersal. Evolutionary Ecology Research. 6: 481-502. Yoder J.M. (2004). The cost of dispersal: predation as a function of movement and site familiarity in ruffed grouse. Behavioral Ecology. 15: 469-476. Zera A. (1984). Differences in survivorship, development rate and fertility between the long- winged and wingless morphs of the waterstrider, Limnoporus canaliculatus. Evolution. 38: 1023-1032.

24

Chapter 2 Exploring Sources of Heterogeneity in Phenotype-Dependent Dispersal

2.1 Abstract

Dispersal influences a variety of biological processes including gene flow and metapopulation persistence. The effects of dispersal on these processes is determined by both the number of dispersers, and the phenotypic composition of the disperser pool. Therefore, understanding the effects of dispersal on ecological and evolutionary processes will require investigating the effects of phenotype on dispersal. However, there is apparently contradictory evidence for the direction and magnitude of the effect of phenotype on dispersal. In this chapter, I conduct a meta-analysis of studies measuring the effects of body size and body condition (two traits that influence individual performance) on dispersal in animals. I found that there is substantial heterogeneity in the effect of these traits on dispersal. I tested four hypotheses previously proposed to explain this heterogeneity. I found that the mean effect of size/condition on dispersal distance was positive, but that the mean effect of size/condition on emigration and immigration did not differ from zero, demonstrating that the effects of phenotype vary across dispersal stages. Heterogeneity in phenotype-dependent dispersal was not associated with taxonomic group, and could not be explained by the spatial scale at which the study was conducted. I then reviewed evidence for the hypothesis that the heterogeneity observed in phenotype-dependent dispersal is due to interactive effects among multiple factors that influence dispersal. I found no evidence that the effect of phenotype on dispersal differs between the sexes. However, there is preliminary evidence to support the hypothesis that interactions between phenotype and habitat quality produce variability in phenotype-dependent dispersal. The results of this meta-analysis demonstrate that there is substantial heterogeneity in phenotype-dependent dispersal, which may in turn affect ecological and evolutionary dynamics. Future studies should make use of recent technological advances for observing dispersal to conduct high-powered studies measuring the effects of multiple phenotypic and environmental variables on all three stages of the dispersal process.

25

2.2 Introduction

Recent studies have demonstrated that dispersal that is non-random with respect to phenotype has important effects on ecological and evolutionary dynamics (Clobert et al. 2009, Edelaar and Bolnick 2012, Bolnick and Otto 2013, Lowe and McPeek 2014). For example, the offspring of dragonflies that disperse long distances have higher foraging rates than those that disperse short distances, producing spatial patterning in the effects of dragonflies on prey communities (McCauley et al. 2010). In addition, there are numerous examples of highly dispersive phenotypes (e.g., large or aggressive individuals) building up at range fronts, which has had a variety of effects including accelerating the pace of invasions (Phillips et al. 2006, Lindstrom et al. 2013), and altering interspecific dynamics (Duckworth and Badyaev 2007). Intraspecific associations between phenotypic traits and dispersal are therefore key components of spatial dynamics.

Studies of dispersal behaviour have demonstrated that dispersal propensity and ability are influenced by a number of phenotypes including sex (Bowler and Benton 2009, Baines et al. 2017), personality (Dingemanse et al. 2003), body size (Chelgren et al. 2006), and body condition (Eraud et al. 2011, Baines et al. 2015). However, this literature is characterized by a high level of variability; studies have come to differing conclusions about the magnitude and direction of the effect of the above phenotypes on dispersal (Bowler and Benton 2005, Clobert et al. 2009), even within a single species or population (e.g., Massot and Clobert 2000). As a result, conclusions about the impact of phenotype-dependent dispersal on the spatial dynamics of one species cannot be generalized to related species. I argue that this variability has also hindered the incorporation of phenotype-dependent dispersal into models of spatial dynamics. Theoretical models have generally made the simplifying assumption that there is no intraspecific variance in dispersal traits (e.g., Kendall et al. 2000, Best et al. 2007, Thompson and Gonzalez 2017), despite the fact that evidence for such variation has existed for decades (Howard 1960). An estimate of the mean and variance in these dispersal-phenotype relationships, and the sources of heterogeneity in these relationships will therefore improve our understanding of spatial dynamics in metapopulations and metacommunities, and aid efforts to incorporate realistic dispersal behaviour into spatial models.

26

In this chapter, I focus on a subset of phenotypes: body size and body condition. The effects of size and condition on dispersal are of particular interest because these traits are tightly linked to aspects of individual performance such as competitive ability (Crespi 1986, Johns et al. 2014) and reproductive success (in males: Dyrcz et al. 2005, Hofmann and Henle 2006, and females: Pachkowski et al. 2013). Consequently, dispersal that is biased with respect to size or condition has different impacts on ecological and evolutionary dynamics than non-biased (random) dispersal. For example, McDevitt et al. (2013) found that body size in weasels (Mustela nivalis) was strongly associated with habitat, and body size influenced dispersal probability. This resulted in unidirectional gene flow from good to poor habitats, and fine-scale population genetic structuring (McDevitt et al. 2013). Understanding the effect of phenotype on dispersal is therefore important, however, the empirical evidence is mixed: positive, negative, and null effects of size and condition on dispersal have all been reported (Bowler and Benton 2005), and the source of this variation is still unknown (Ims and Hjermann 2001).

Theoretical predictions for the effect of body size and condition on dispersal are complex. High condition individuals have greater dispersal ability because energy is required to pay the high physiological costs of dispersal (Bonte et al. 2012). For example, high condition animals are better able to meet the energetic demands of locomotion (Cockbain 1961, Beck and Congdon 2000), which may enable them to colonize higher quality patches (silver spoon effects; see Stamps (2006) for a review). Energy reserves are required to build and maintain morphological structures needed for transience (including wings and flight muscles; Harada and Spence 2000, Pfenning et al. 2007). Energy reserves are also required for settlement. For example, sheetweb spiders (Erigone atra) with high body condition are better able to locate and take over the webs of competitors (Bonte et al. 2011). Condition may also influence settlement in non-competitive situations. For example, naked mole rats (Heterocephalus glaber) store large quantities of fat prior to dispersal (O'Riain et al. 1996) which may aid in the performance of energetically costly behaviours required for settlement, such as the creation of new burrows (O'Riain et al. 1996, Braude 2000). Large body size is associated with large energy reserves (Dmitriew et al. 2009), and high locomotive performance (Tejedo et al. 2000, Phillips et al. 2006). Therefore, large size and high condition confer high dispersal ability.

27

The positive effect of body size and condition on dispersal ability may not necessarily result in positive associations between these traits and dispersal probability. Negative size/condition- dependent dispersal is expected when size and condition are related to habitat quality (Pettorelli et al. 2002, Brown and Sherry 2006), such that small/low condition individuals have higher motivation to emigrate. Small/low condition individuals may also have greater dispersal distances, if the extent of spatial autocorrelation in habitat quality is large enough that dispersers must travel long distances to move from low to high quality habitat (Ims and Hjermann 2001). Individuals with inferior competitive ability due to small size/low condition may also be induced to emigrate if they are subject to aggressive behaviour from superior competitors (McCauley 2010, Hewett Ragheb and Walters 2011). Theory predicts that behaviour during each dispersal stage is dependent on whether individuals have sufficient motivation and ability to initiate and/or complete that stage (Stamps 2006, Benard and McCauley 2008). However, it is still unclear how the effects of dispersal motivation balance against the effects of dispersal ability to determine emigration probability, dispersal distance, and immigration success (Benard and McCauley 2008).

Several hypotheses have been put forth to explain the observed variability in phenotype-dispersal relationships (Ims and Hjermann 2001, Matthysen 2012). I narrow these down to four testable hypotheses:

1. Variability in dispersal patterns is the result of phenotypic traits having different effects at different dispersal stages: dispersal consists of three stages – emigration, transience, and immigration. These stages are behaviourally and physiologically distinct. Therefore, a phenotype that has a positive effect on one stage may have a negative effect on another (e.g., small individuals may be more likely to disperse, but large individuals may disperse longer distances). 2. The relationship between phenotype and dispersal varies across spatial scales: dispersal is inherently a spatial process, yet few studies have estimated how dispersal patterns vary across spatial scales. Dispersal patterns are expected to vary across spatial scales if dispersal distance is a nonlinear function of phenotype, or if the effects of phenotype on dispersal vary across an environmental gradient.

28

3. Different metrics of body size and condition have different effects on dispersal: dispersal studies have measured the effect of body mass, body size, body condition, and the size of morphological structures used for dispersal. These phenotypes are often combined in discussions of body size- and body condition-dependent dispersal. However, they likely have different relationships with dispersal propensity and dispersal ability. For example, body size is often an important determinant of dispersal ability (Stevens et al. 2014), but will not indicate the quality of the current habitat (and therefore dispersal motivation) in species in which structural body size is fixed at adulthood. Whereas, the size of energy reserves may influence both dispersal ability and motivation (by indicating habitat quality). 4. Variability in dispersal patterns is the result of interactive effects of phenotypic traits and environmental variables acting on dispersal, i.e., the multiple, interacting causes of dispersal hypothesis: multiple phenotypes and environmental factors influence dispersal at all stages (e.g., body size, sex, population density, predation risk, etc.; Bowler and Benton 2005). If these factors interact, then variation in dispersal patterns will be produced when studies measure the effect of a single factor on dispersal without controlling for the effects of interactors.

The goals of this paper were to i) estimate the mean and variance of the effect of size and condition on dispersal, and ii) test the hypotheses for the sources of heterogeneity in the size/condition-dispersal relationship. To do this, I conducted a meta-analysis of phenotype- dependent dispersal studies. I used a meta-regression approach to evaluate whether heterogeneity in the effect of phenotype on dispersal could be explained by the dispersal stage observed, the spatial scale at which the study was conducted, and the phenotypic trait measured. Several recent studies have investigated the interactive effects of multiple factors on dispersal (e.g., Tarwater and Beissinger 2012, Baines et al. in press). A number of these have reported the effects of phenotype on dispersal separately for each sex; I used a meta-regression approach to test whether males and females have different phenotype-dispersal relationships. However, there are still an insufficient number of studies that have measured the interactive effects of phenotype and environmental factors on dispersal. I therefore conducted a narrative review of the evidence for

29

these interactive effects, rather than using a meta-analytic approach, to test the hypothesis that multiple interacting factors produce heterogeneity in phenotype-dependent dispersal.

2.3 Methods 2.3.1 Data collection

I collected data from studies that measured the effect of body size or body condition on any stage of dispersal (emigration, transience, or immigration) in animals. I searched for studies in the ISI Web of Science database using the search terms (dispersal or *migration) AND (‘body condition’ or ‘body size’ or ‘size’). I included studies of both natal and breeding dispersal. I refined these results to include both observational and manipulative studies that measured i) an aspect of body size or condition, and ii) dispersal status (disperser vs. non-disperser) or dispersal distance, with the condition that size/condition and dispersal were measured on the same individuals. I also searched the reference lists of the included studies. I excluded studies that measured only date of dispersal rather than dispersal status or dispersal distance. I also excluded studies on non-dispersal movement such as annual migrations, or movement associated with life history transitions (e.g., movement of amphibians from aquatic into terrestrial habitats following metamorphosis). If a study measured an effect of size/condition at multiple levels of a factor (e.g., sex, population), I included it if there was no significant interaction between predictors and/or if the paper reported the statistical results for the effect of size/condition on dispersal separately for each factor level. If there was a significant interaction but statistics were not available for each factor separately, the study was excluded. I was unable to collect the data required to calculate effect sizes from several studies, so they were necessarily excluded.

Studies were categorized based on how the relationship between dispersal and phenotype was measured. The first group of studies were those that measured the relationship between size/condition and binary dispersal status. Dispersers could be either known emigrants or known immigrants. Emigrants are individuals that left a pre-defined natal patch or range, and immigrants are individuals that settled in a non-natal patch or range. This group of studies was further divided based on the type of statistical test conducted. Some studies compared the mean size/condition of dispersers and residents, with dispersal status as a predictor variable (disperser- resident comparison studies). Others conducted a logistic regression using dispersal status as the 30

response variable and size/condition as the predictor variable (dispersal status LR studies). For disperser-resident comparison studies, I used Hedge’s g as the effect size metric. Hedge’s g was calculated using the ‘esc’ package in R (Lüdecke 2018). When data were only reported in figures, I extracted mean and variance values using the program DataThief (Tummers 2006). Effect size was positive when dispersers were larger than residents, and negative when dispersers were smaller than residents. Results from four studies were excluded because they reported statistics demonstrating no significant difference in mean size/condition between dispersers and residents, but did not report the direction of the effect. The disperser-resident comparison dataset consisted of 50 effect sizes from 14 papers, representing 17 species. For dispersal status LR studies, I used logit (or log-odds) as the effect size metric. Logit values and their standard errors were collected directly from papers. Dispersal status LR effect sizes were positive when the probability of being a disperser increased with increasing size/condition, and negative when the probability of being a disperser decreased with size/condition. The dispersal status LR dataset consisted of 7 effect sizes from 5 papers, representing 5 species.

The second group of studies measured correlation between size/condition and dispersal distance (dispersal distance studies). I collected r values from the text, or when not reported, I used DataThief to extract data and calculate Pearson’s r. I converted r to Fisher’s Z (which accounts for sample size in its variance measure) using the ‘esc’ package in R (Lüdecke 2018). The dispersal distance dataset consisted of 24 effect sizes from 7 papers, representing 10 species.

I chose to analyze the disperser-resident comparison, dispersal status LR, and dispersal distance groups separately because of fundamental differences in how dispersal is measured in these three groups. In the disperser-resident comparison group, phenotype is used as the dependent variable, and dispersal status is the independent variable, whereas these variables are inverted in the other two study groups. Therefore, their effect sizes cannot be directly compared. The dispersal status LR and dispersal distance groups are also not directly comparable, because the first estimates the probability of engaging in a discrete behaviour (disperse vs. remain philopatric), while the second measures the correlation between phenotype and the duration/distance in which individuals engaged in that behaviour.

31

To test my a priori hypotheses, I collected additional data from each article. I recorded the dispersal stage of each result (emigration or immigration for disperser-resident comparison and dispersal status LR studies; dispersal distance studies measured only transience, by definition). I also divided studies based on whether they measured dispersal in a single patch, or multiple patches. I divided the phenotypic traits into four categories: body mass, body size, body condition index (BCI) and dispersal structure (DS) size. Body mass was a measurement of the mass of the whole or a part of the animal. Body size was any of multiple traits correlated with structural body size including total body length and snout-vent length. BCI was any measure of body mass corrected for body size (the exact formulae for BCI varied across studies; Appendix A: Table A1). DS size was the length of the relevant dispersal structure (wings for flying animals, and legs for walking animals). I also recorded whether the result was associated with males, females, or both sexes together. Finally, I recorded the broad taxonomic group investigated (amphibians, birds, reptiles, mammals, fish, or invertebrates) because taxonomy may be associated with factors such as sociality, territoriality, dispersal ability, etc., that may influence the evolution of dispersal. Methods for measuring phenotype and dispersal also vary across taxonomic groups.

2.3.2 Statistical analysis

Articles often reported multiple effect sizes for different species, populations, sexes, or dispersal stages, or because they measured multiple phenotypes. I dealt with this issue by conducting a three-level meta-analysis using the ‘meta3’ function in the ‘metaSEM’ package in R (Cheung 2015). Three-level meta-analysis accounts for non-independence of effect sizes from the same studies. ‘meta3’ estimates the Q statistic, which tests the null hypothesis of homogeneity in effect sizes. This method also estimates I2, which represents the proportion of the total variation in the effect size that can be explained by each level (in this case, the between-study and within-study levels). The number of articles in the dispersal status LR group was insufficient for running three-level meta-analysis. I therefore report results only for the disperser-resident comparison and dispersal distance groups in the main text. Effect sizes for the dispersal status LR group can be found in the appendix (Appendix A: Figure A4).

32

I first built intercept models to test whether the overall effect size differed from zero for both groups of studies (disperser-resident comparison and dispersal distance studies). The intercept model included article ID as a clustering variable, but had no fixed predictors. I then built mixed- effects models to test each of the hypothesized sources of variation in dispersal: dispersal stage, spatial scale, phenotype metric, and sex. I also built models to test for an effect of taxonomic group. It was only possible to test the effect of dispersal stage (emigration vs immigration) using disperser-resident comparison studies because dispersal distance studies, by definition, measured only transience. I did not build models with all possible predictor variables because of insufficient sample sizes. I included article ID as a clustering variable in all models, to account for non-independence of results from the same study. The significance of each model was tested by evaluating the difference in the log-likelihood of the model to the intercept model, using χ2 tests.

I examined contour-enhanced funnel plots for evidence of publication bias against non- significant results (Peters et al. 2008). Publication bias manifests at the level of a study. Therefore, for studies that reported more than one result, I averaged the effect sizes and their associated standard errors, and plotted the average in the funnel plot. All analyses were conducted in R version 3.4.3 (R Core Team 2017).

2.4 Results

2.4.1 Disperser-resident comparison studies

The average effect size for disperser-resident comparison studies was not significantly different from zero (z = 0.19, p = 0.85; Figure 2.1). The effect sizes were heterogeneous (Q = 366.15, df = 49, p < 0.0001). Within-study and between-study effects explained 61% and 30% of the total variation, respectively. None of the predictors tested (dispersal stage, spatial scale, phenotype metric, or sex) explained a significant amount of heterogeneity in effect sizes for disperser- resident comparison studies (Figure 2.2, Appendix A: Table A2). Effect sizes did not vary significantly across taxonomic groups (Appendix A: Figure A1, Table A2). I did not detect evidence for publication bias against non-significant results (Appendix A: Figure A5).

33

Figure 2.1. Effect size estimates (Hedge’s g) ± 95% confidence intervals for each result in the disperser-resident comparison studies dataset. The overall effect is not significantly different from zero (p = 0.85). Point size is inversely proportional to the standard error of each result.

34

Figure 2.2. Effect sizes (Hedge’s g) as a function of A) dispersal stage, B) spatial scale, C) phenotype metric, and D) sex for studies in the disperser-resident comparison dataset. Data are effect size estimates ± 95% confidence intervals derived from three-level meta- analytic regression models. The dashed vertical line indicates an effect size of zero. Abbreviations: BCI = body condition index, DS = dispersal structure size, ‘Studies’ indicates the number of studies represented in each category, and ‘Results’ indicates the number of results in each category. The number of studies in each category does not sum to the total in (C) and (D) because some studies reported effect sizes for multiple categories. Note that the x-axis scale varies across panels. Point size is inversely proportional to the standard error of each category.

35

2.4.2 Dispersal distance studies

The average effect size for dispersal distance studies was significantly positive (z = 2.07, p = 0.038; Figure 2.3). The effect sizes were heterogeneous (Q = 89.87, df = 23, p < 0.0001). Within- study and between-study effects explained 50% and 28% of the total variation, respectively. None of the predictors (spatial scale, phenotype metric, or sex) explained a significant amount of heterogeneity in effect sizes for dispersal distance studies (Appendix A: Table A3, Figure A2). Effect sizes did not vary significantly across taxonomic groups (Appendix A: Figure A3, Table A3). I did not detect evidence for publication bias against non-significant results (Appendix A: Figure A6).

Figure 2.3. Effect size estimates (Fisher’s Z) ± 95% confidence intervals for each result in the dispersal distance studies dataset. The overall effect is significantly positive (p = 0.038). Point size is inversely proportional to the standard error of each result.

36

2.5 Discussion 2.5.1 Heterogeneity in phenotype-dependent dispersal

The results of this meta-analysis demonstrate that there is a substantial degree of heterogeneity in phenotype-dispersal relationships. This was true both of studies that compared the phenotypes of dispersers and residents, and those that measured the correlation between phenotype and dispersal distance. This is the first quantitative confirmation of the observation made by several authors (Ims and Hjermann 2001, Bowler and Benton 2005, Benard and McCauley 2008, Clobert et al. 2009), that the effects of size and condition on dispersal vary to a large extent across studies. Because of the high variance in effect sizes, and the relatively small number of studies, the power of my analyses to detect significant differences was low to moderate. This limits my ability to make definitive conclusions about the mean effect of phenotype on dispersal (overall and within categories such as taxonomic groups). However, heterogeneity in effect sizes is interesting because it indicates substantial variability in dispersal behaviour within and across species, and suggests that mean effect sizes may not relay meaningful information about dispersal. Below, I discuss the results of the analyses exploring the sources of this heterogeneity.

2.5.2 Sources of heterogeneity in phenotype-dependent dispersal

2.5.2.1 Dispersal stage

The mean effect sizes for emigration and immigration did not differ from zero, but the mean effect size for transience was positive. This is not evidence that there is no effect of size/condition on emigration or immigration, but rather that positive and negative relationships are both common in these stages of dispersal. The difference between dispersal stages can be explained by hypotheses regarding the different rules that govern behaviour during each stage. The decision to emigrate is expected to depend on the motivation to disperse, as well as each individual’s assessment of its own dispersal ability (i.e., the likelihood of successfully completing dispersal and settling in a new site; Southwood 1977). As discussed above, dispersal ability is a positive function of size and condition, but the relationship between dispersal motivation and phenotype depends on the ecological context (Benard and McCauley 2008). This

37

context dependency is likely responsible for the variability we see in phenotype-emigration relationships (see section 2.5.2.4).

The positive relationship between size/condition and dispersal distance supports a hypothesis put forward by Stamps (2006). This hypothesis proposes that individuals should remain in the transience stage until they find a high quality settlement site, or are constrained by the high costs of dispersal (i.e., time, energetic, and opportunity costs; Bonte et al. 2012). Individuals who are large and/or are in good condition have high dispersal ability, and therefore are predicted to spend longer amounts of time in the transience phase and move longer distances, on average, in order to locate a high quality site in which to breed. Individuals with low dispersal ability may be forced to accept lower quality settlement sites than those with high ability, because they are unable to cope with the high costs of long distance dispersal (silver spoon effects; Stamps 2006).

The association between phenotype and immigration probability/success is a product of the effect of phenotype on both emigration probability and transience success (i.e., surviving and completing emigration and transience). It is therefore unsurprising that the phenotype- immigration relationship is variable. It is interesting however, that there is a trend toward a negative average effect size for phenotype-emigration relationship, but a positive average effect size for phenotype-transience relationships, and a trend toward a positive average effect size for phenotype-immigration relationships. This suggests the hypothesis that negatively biased emigration is counterbalanced by positively biased transience to determine the phenotypes of immigrants. There is a dearth of theory and empirical investigation into whether the phenotypes of immigrants are a biased subset of the phenotypes of the entire metapopulation. This represents a significant gap, because only immigrants can impact demographic processes and produce gene flow, not failed emigrants or transients. Future theoretical and empirical studies should investigate the effects of phenotype on the entire dispersal process, including immigration, to make direct connections to ecological and evolutionary processes.

2.5.2.2 Spatial scale

My results do not support the hypothesis that phenotype-dispersal relationships vary across spatial scales. However, my analysis had low power to test this question since relatively few studies investigated dispersal at the scale of a single habitat patch, and only one study 38

investigated dispersal at multiple spatial scales (Tarwater and Beissinger 2012). Since dispersal is inherently a spatial process, the failure to explicitly account for spatial scale in dispersal studies is concerning (Ims and Hjermann 2001). As technology for tracking large numbers of dispersers becomes cheaper and more accessible for research, researchers should invest in large- scale, spatially explicit studies that measure phenotype and movement at different spatial scales. This will provide new insights into the effects of spatial scale on our understanding of dispersal processes.

2.5.2.3 Phenotype metric

Overall, phenotype metric did not explain a significant amount of variability in effect sizes, but there was a trend toward BCI having a negative average effect size, and other metrics (body mass, body size, and dispersal structure size) having positive average effect sizes. This is likely an artifact of the fact that BCIs are artificial constructs that are often unrelated to individual performance and to reliable indicators of condition such as body fat content (Schamber et al. 2009, Wilder et al. 2016). On the basis of this evidence, I caution against using BCIs in studies of dispersal behaviour, and suggest researchers use alternative methods. If mass and structural size are highly correlated, researchers can choose a single phenotypic trait (e.g., the trait that is measured with least disruption to the organism). If the trait of interest is the size of energy reserves, researchers should use traits such as body fat content, which is a more reliable indicator of energy reserves, and is more directly relevant to dispersal because fat is often used as fuel for dispersal (O'Riain et al. 1996, Gade et al. 2004).

2.5.2.4 Multicausality of dispersal

Assuming organisms make dispersal decisions to promote fitness, individuals should disperse when the quality of their current habitat is lower than the quality of potential settlement patches (Southwood 1977, Morris 1987, Johnson and Gaines 1990). Habitat quality is determined by a myriad number of abiotic and biotic variables including temperature, moisture level, population density, and the density of parasites and predators. These factors have all been found to influence dispersal behaviour (Brown and Brown 1992, O'Connor et al. 2007, Baines et al. 2014). The fitness consequences of dispersal will also depend on phenotypic traits that influence an individual’s ability to emigrate, navigate the matrix, and settle in a new patch (Clobert et al.

39

2009). Therefore, individuals must integrate information from both internal and external sources in order to make dispersal decisions. If these variables interact in their effects on fitness or dispersal ability, then dispersal behaviour is also expected to depend on their interaction. As a consequence, variability in dispersal patterns will be observed in phenotype-dependent dispersal if researchers do not account for interactive effects. Variability may even be observed within a single species across multiple populations, or through time, if aspects of habitat quality such as temperature and predator density vary across space or time.

Figure 2.4. Habitat quality influences dispersal directly, and indirectly through effects on phenotype. Solid lines indicate positive relationships between two variables, and dashed lines indicate negative relationships between two variables.

Although some variation in size and condition is heritable (e.g., as a result of heritable variation in growth rate, or metabolic or foraging efficiency; Nilsson et al. 2009, Nachappa et al. 2010), a large proportion of variation in size and condition is determined by habitat quality (Pettorelli et al. 2002, Brown and Sherry 2006, Dmitriew et al. 2009). Habitat quality also influences dispersal

40

behaviour directly (Johnson 1965, Kuussaari et al. 1996, Sloggett and Weisser 2002, Baines et al. 2014). Therefore, the relationship between dispersal and phenotype will depend on the association between phenotype and habitat quality, and in turn, their effects on dispersal capacity and motivation (Figure 2.4). However, most studies of dispersal have observed natural populations, and are unable to disentangle the direct effects of habitat quality on dispersal from indirect effects acting via phenotype.

The interactive effects of size/condition and habitat quality on dispersal have been measured in only five studies published to date. Two of these studies manipulated food availability to create groups of animals that differed in body condition, and then conducted experiments in which body condition group was crossed with habitat quality treatment (Baines et al. 2015, chapter 5 of this document and published as Baines et al. 2018). Baines et al. (2015) found that dispersal in the semi-aquatic insect Notonecta undulata was positively associated with body condition (body fat content), regardless of the level of predation risk. In chapter 5, we found that body mass was positively associated with dispersal in N. undulata, regardless of resource availability. The other three studies were observational studies of natural metapopulations. One of these studies reported that in high quality, moist habitats, mole-rat dispersers had the same mass as residents, but in low quality, arid habitats, there was a trend toward dispersers being heavier than residents (Spinks et al. 2000). Two observational studies investigated the interaction between phenotype and population density. Hanski et al. (1991) found that in years when population density was low, dispersing shrews were smaller than residents. When population density was high, there was a trend toward dispersers being larger than residents (Hanski et al. 1991). Baines et al. (chapter 4 and published as Baines et al. in press) found that when population density was high, dispersing N. undulata were much smaller than residents, and when population density was low, this relationship was weakened. Two additional observational studies found that the association between body condition and dispersal varied across populations (Tarwater and Beissinger 2012), and across years within a single population (Massot and Clobert 2000), suggesting that unmeasured variables that changed across space and time produced variability in the phenotype- dispersal relationship. Taken together, the studies described above provide support for the hypothesis that size/condition-dependent dispersal is influenced by habitat quality. This evidence is preliminary, since I could find only five studies that tested this interaction, and three of these

41

were studies of the same species. However, given the fact that i) the effects of phenotype on fitness (and therefore dispersal motivation) depend on ecological context, and ii) habitat quality influences phenotype, this is a hypothesis that warrants further study.

The effects of size/condition on dispersal may also depend on other phenotypic traits. The only other trait regularly reported in size/condition-dependent dispersal studies was sex. Previous studies have observed both male and female biased dispersal (reviewed in Greenwood 1980, Mabry et al. 2013). Theory suggests that sex-biased dispersal will occur if there are sex differences in the costs and/or benefits of dispersal (Perrin and Mazalov 2000, Wild and Taylor 2004, Gros et al. 2008, Gros et al. 2009). This may also cause the effect of size/condition on dispersal to differ between the sexes. For example, if females have lower tolerance for the costs of dispersal (e.g., because of dispersal-fecundity trade-offs; Roff 1984, Zera 1984), then dispersal may be uncommon in females, and restricted to females with large energy reserves. Whereas males may be more likely to disperse (Baines et al. 2017), and if intrasexual competitive ability is size-dependent, small males may be more likely to disperse (Lawrence 1987). However, the results of my meta-analysis do not provide evidence for sex differences in phenotype-dependent dispersal. Future studies should test the hypothesis that sex differences in the costs and/or benefits of dispersal influence both which sex disperses more, and the relative strengths of phenotype-dependent dispersal in each sex.

Gyllenberg et al. (2008) have proposed that the observed inconsistency in the dispersal-condition relationship may be due to variation in the relationship between dispersal ability and condition among taxa. They argue that negative condition-dependent dispersal occurs in species in which low condition individuals have higher dispersal ability, and vice versa. However, to my knowledge, no current evidence exists for low condition individuals being better dispersers in any taxonomic group, which suggests that this hypothesis cannot explain the high degree of variability observed. Bonte and de la Pena (2009) proposed that variation in dispersal mortality and environmental stochasticity may produce variation in dispersal-condition associations. This hypothesis is unable to explain cases in which the dispersal-condition relationship fluctuates within species across short spatial or temporal scales (Massot and Clobert 2000, Tarwater and Beissinger 2012), but may contribute to the variation we observe across species. Both Gyllenberg et al. (2008) and Bonte and de la Pena (2009) rely on kin selection to drive dispersal 42

evolution (at least in some parts of parameter space). However, empirical evidence for kin selection influencing dispersal evolution is lacking for most taxa (Lambin et al. 2001). I argue that the hypothesis that interactive effects are producing variability in phenotype-dependent dispersal relationships can apply to a wider variety of taxonomic groups, and can more easily explain cases in which the effect of phenotype on dispersal varies across short spatial or temporal scales.

2.5.3 Future directions

Most previous studies of dispersal have focused on investigating the effects of a single variable on a single dispersal stage at one spatial scale. These studies also tend to be very low powered (see Appendix A). This is not surprising given how difficult is to track dispersers in natural systems. For example, Thompson (1991) marked and released 2653 adult damselflies at a single pond, and subsequently searched for dispersers at the release pond as well as a nearby pond. Although he resighted 480 marked individuals, only 1 female and 7 males were dispersers (Thompson 1991). This is typical of dispersal studies, where large amounts of effort yield low numbers of known dispersers. However, as technology for tracking individuals improves and becomes less expensive, researchers are increasingly investing in large studies spanning longer distances and time periods (e.g., Debeffe et al. 2012, Selonen et al. 2012). Below, I list recommendations for future dispersal studies that have been highlighted by my review of the literature:

1. Follow the entire dispersal process. As the result of logistical constraints on tracking dispersal, many studies are able to measure only a single dispersal stage (e.g., studies that observe immigration into a habitat patch, but are unable to determine where immigrants were born or how far they traveled). This makes it difficult to determine the cumulative effects of phenotype throughout the dispersal process. As the required technology becomes more accessible, researchers should design studies that track many individuals from their birth place to the site of reproduction. 2. Test for interactions. Previous studies have demonstrated that interactions among phenotypic traits and environmental variables influence dispersal. In addition to the studies measuring interactions between size/condition and habitat quality or sex reviewed

43

here, previous studies have also provided evidence for interactions among environmental variables (e.g., Fellous et al. 2011, Hammill et al. 2015), and between phenotype and social variables (e.g., Léna et al. 1998, Tarwater and Beissinger 2012). However, these studies are still relatively rare. We are still unable to determine whether the same interactive effects are important across species, or if species (or larger taxonomic groupings) have idiosyncratic dispersal patterns. Moreover, several phenotypes and environmental variables have been demonstrated to influence dispersal, but few studies have had sufficient power to test the effects of more than two of these variables simultaneously. Obtaining data to build “interaction maps” of dispersal behaviour would improve our understanding of this key biological process, and allow researchers to build models to predict the consequences of dispersal after environmental change (e.g., climate change or habitat fragmentation). 3. Consider spatial scale. Dispersal is inherently a spatial process, yet few studies have investigated whether the spatial scale at which dispersal is measured influences our conclusions. This is especially problematic for studies that define dispersers as individuals that move more than a set distance from their birthplace (e.g., Massot and Clobert 2000, Kennedy and Ward 2003, Chaput-Bardy et al. 2010). By not considering the spatial scale of movement, we risk miscategorising routine movement (i.e., movement for activities such as foraging; Van Dyck and Baguette 2005) as dispersal, and obscuring real dispersal patterns.

2.5.4 Conclusions

Phenotype-dependent dispersal has the potential to strongly skew our estimates of metapopulation connectivity and gene flow, and impact ecological and evolutionary dynamics. It is therefore critical that we understand how phenotype influences dispersal, and how this may depend on factors such as ecological context and the spatial scale at which this relationship is measured. In this paper, I conducted a meta-analysis which demonstrates that there is substantial heterogeneity in phenotype-dependent dispersal, and tested four hypotheses that have been proposed to explain this heterogeneity. I found support for the hypothesis that the effects of phenotype vary across dispersal stages; the average effect size for transience was positive, while the average effect sizes for emigration and immigration did not differ from zero. My meta-

44

analysis did not provide support for the hypotheses that phenotype-dependent dispersal varies across spatial scales, or depends on how size/condition was measured; however, I suggest that more effort needs to be invested into measuring dispersal across multiple spatial scales, and that researchers should avoid using body condition indices in dispersal studies. Finally, I found preliminary support for the hypothesis that interactions between size/condition and environmental variables (including population density) contribute to the observed variation in dispersal patterns. However, I argue that high-powered studies that investigate the effects of multiple factors on dispersal are required to fully elucidate dispersal patterns.

2.6 References

Baines C.B., I.M. Ferzoco, and S.J. McCauley (2017). Sex-biased dispersal is independent of sex ratio in a semiaquatic insect. Behavioral Ecology and Sociobiology. 71: 119. Baines C.B., and S.J. McCauley (2018). Natal habitat conditions have carryover effects on dispersal capacity and behavior. Ecosphere. 9: e02465. Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287. Baines C.B., S.J. McCauley, and L. Rowe (2015). Dispersal depends on body condition and predation risk in the semi-aquatic insect, Notonecta undulata. Ecology and Evolution. 5: 2307-2316. Beck C.W., and J.D. Congdon (2000). Effects of age and size at metamorphosis on performance and metabolic rates of Southern Toad, Bufo terrestris, metamorphs. Functional Ecology. 14: 32-38. Benard M.F., and S.J. McCauley (2008). Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist. 171: 553-567. Best A.S., K. Johst, T. Munkemuller, and J.M.J. Travis (2007). Which species will successfully track climate change? The influence of intraspecific competition and density dependent dispersal on range shifting dynamics. Oikos. 116: 1531-1539. Bolnick D.I., and S.P. Otto (2013). The magnitude of local adaptation under genotype-dependent dispersal. Ecology and Evolution. 3: 4722-4735. Bondrup-Nielsen S. (1992). Emigration of meadow voles, Microtus pennsylvanicus: the effect of sex ratio. Oikos. 65: 358-360.

45

Bonte D., and E. de la Pena (2009). Evolution of body condition-dependent dispersal in metapopulations. Journal of Evolutionary Biology. 22: 1242-1251. Bonte D., N. De Meester, and E. Matthysen (2011). Selective integration advantages when transience is costly: immigration behaviour in an agrobiont spider. Animal Behaviour. 81: 837-841. Bonte D., H. Van Dyck, J.M. Bullock, A. Coulon, M. Delgado, M. Gibbs, V. Lehouck, E. Matthysen, K. Mustin, M. Saastamoinen, N. Schtickzelle, V.M. Stevens, S. Vandewoestijne, M. Baguette, K. Barton, T.G. Benton, A. Chaput-Bardy, J. Clobert, C. Dytham, T. Hovestadt, C.M. Meier, S.C. Palmer, C. Turlure, and J.M. Travis (2012). Costs of dispersal. Biological Reviews. 87: 290-312. Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Bowler D.E., and T.G. Benton (2009). Variation in dispersal mortality and dispersal propensity among individuals: the effects of age, sex and resource availability. Journal of Animal Ecology. 78: 1234-1241. Braude S. (2000). Dispersal and new colony formation in wild naked mole-rats: evidence against inbreeding as the system of mating. Behavioral Ecology. 11: 7-12. Brown C.R., and M.B. Brown (1992). Ectoparasitism as a cause of natal dispersal in cliff swallows. Ecology. 73: 1718-1723. Brown D.R., and T.W. Sherry (2006). Food supply controls the body condition of a migrant bird wintering in the tropics. Oecologia. 149: 22-32. Chaput-Bardy A., A. Gregoire, M. Baguette, A. Pagano, and J. Secondi (2010). Condition and phenotype-dependent dispersal in a damselfly, Calopteryx splendens. PLoS One. 5: e10694. Chelgren N.D., D.K. Rosenberg, S.S. Heppell, and A.I. Gitelman (2006). Carryover aquatic effects on survival of metamorphic frogs during pond emigration. Ecological Applications. 16: 250-261. Cheung M.W.L. (2015). metaSEM: An R package for meta-analysis using structural equation modeling. Frontiers in Psychology. 5: 1521.

46

Clobert J., J.F. Le Galliard, J. Cote, S. Meylan, and M. Massot (2009). Informed dispersal, heterogeneity in animal dispersal syndromes and the dynamics of spatially structured populations. Ecology Letters. 12: 197-209. Cockbain A.J. (1961). Fuel utilization and duration of tethered flight in Aphis fabae Scop. Journal of Experimental Biology. 38: 163-174. Coll M., and B. Yuval (2004). Larval food plants affect flight and reproduction in an oligophagous insect herbivore. Environmental Entomology. 33: 1471-1476. Conrad K.F., K.H. Willson, K. Whitfield, I.F. Harvey, C.J. Thomas, and T.N. Sberratt (2002). Characteristics of dispersing Ischnura elegans and Coenagrion puella (Odonata): age, sex, size, morph and ectoparasitism. Ecography. 25: 439-445. Crespi B.J. (1986). Size assessment and alternative fighting tactics in Elaphrothrips tuberculatus (Insecta: Thysanoptera). Animal Behaviour. 34: 1324-1335. Debeffe L., N. Morellet, B. Cargnelutti, B. Lourtet, R. Bon, J.M. Gaillard, and A.J. Mark Hewison (2012). Condition-dependent natal dispersal in a large herbivore: heavier animals show a greater propensity to disperse and travel further. Journal of Animal Ecology. 81: 1327-1337. Dingemanse N.J., C. Both, A.J. van Noordwijk, A.L. Rutten, and P.J. Drent (2003). Natal dispersal and personalities in great tits (Parus major). Proceedings of the Royal Society B: Biological Sciences. 270: 741-747. Dmitriew C., J. Carroll, and L. Rowe (2009). Effects of early growth conditions on body composition, allometry, and survival in the ladybird beetle Harmonia axyridis. Canadian Journal of Zoology. 87: 175-182. Dubois G.F., P.J. Le Gouar, Y.R. Delettre, H. Brustel, and P. Vernon (2010). Sex-biased and body condition dependent dispersal capacity in the endangered saproxylic beetle Osmoderma eremita (Coleoptera: Cetoniidae). Journal of Insect Conservation. 14: 679- 687. Duckworth R.A., and A.V. Badyaev (2007). Coupling of dispersal and aggression facilitates the rapid range expansion of a passerine bird. Proceedings of the National Academy of Sciences. 104: 15017-15022.

47

Dyrcz A., M. Wink, A. Kruszewicz, and B. Leisler (2005). Male reproductive success is correlated with blood parasite levels and body condition in the promiscuous aquatic warbler (Acrocephalus paludicola). The Auk. 122: 558-565. Edelaar P., and D.I. Bolnick (2012). Non-random gene flow: an underappreciated force in evolution and ecology. Trends in Ecology and Evolution. 27: 659-665. Eraud C., A. Jacquet, and P. Legagneux (2011). Post-fledging movements, home range, and survival of juvenile Eurasian collared-doves in western France. The Condor. 113: 150- 158. Fellous S., E. Quillery, A.B. Duncan, and O. Kaltz (2011). Parasitic infection reduces dispersal of ciliate host. Biology Letters. 7: 327-329. Gade G., L. Auerswalk, R. Predel, and H.G. Marco (2004). Substrate usage and its regulation during flight and swimming in the backswimmer, Notonecta glauca. Phsyiological Entomology. 29: 84-93. Greenwood P.J. (1980). Mating systems, philopatry and dispersal in birds and mammals. Animal Behaviour. 28: 1140-1162. Gros A., T. Hovestadt, and H.J. Poethke (2008). Evolution of sex-biased dispersal: The role of sex-specific dispersal costs, demographic stochasticity, and inbreeding. Ecological Modelling. 219: 226-233. Gros A., H.J. Poethke, and T. Hovestadt (2009). Sex-specific spatio-temporal variability in reproductive success promotes the evolution of sex-biased dispersal. Theoretical Population Biology. 76: 13-18. Gyllenberg M., E. Kisdi, and M. Utz (2008). Evolution of condition-dependent dispersal under kin competition. Journal of Mathematical Biology. 57: 285-307. Hammill E., R.G. Fitzjohn, and D.S. Srivastava (2015). Conspecific density modulates the effect of predation on dispersal rates. Oecologia. 178: 1149-1158. Hanski I., A. Peltonen, and L. Kaski (1991). Natal dispersal and social dominance in the common shrew Sorex araneus. Oikos. 62: 48-58. Harada T., and J.R. Spence (2000). Nymphal density and life histories of two water striders (Hemiptera: Gerridae). The Canadian Entomologist. 132: 353-363.

48

Hewett Ragheb E.L., and J.R. Walters (2011). Favouritism or intrabrood competition? Access to food and the benefits of philopatry for red-cockaded woodpeckers. Animal Behaviour. 82: 329-338. Hofmann S., and K. Henle (2006). Male reproductive success and intrasexual selection in the common lizard determined by DNA-microsatellites. Journal of Herpetology. 40: 1-6. Howard W.E. (1960). Innate and environmental dispersal of individual vertebrates. American Midland Naturalist. 63: 152-161. Ims R.A., and D.O. Hjermann (2001). Condition-dependent dispersal. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 203-216. Johns A., H. Gotoh, E.L. McCullough, D.J. Emlen, and L.C. Lavine (2014). Heightened condition-dependent growth of sexually selected weapons in the rhinoceros beetle, Trypoxylus dichotomus (Coleoptera: Scarabaeidae). Integrative and Compative Biology. 54: 614-621. Johnson B. (1965). Wing polymorphism in aphids II. Interaction between aphids. Entomological Experiments and Applications. 8: 49-64. Johnson M.L., and M.S. Gaines (1990). Evolution of dispersal: theoretical models and empirical tests using birds and mammals. Annual Review of Ecology and Systematics. 21: 449-480. Kendall B.E., O.N. Bjꬾrnstad, J. Bascompte, T.H. Keitt, and W.F. Fagan (2000). Dispersal, environmental correlations, and spatial synchrony in population dynamics. The American Naturalist. 155: 628-636. Kennedy P.L., and J.M. Ward (2003). Effects of experimental food supplementation on movements of juvenile northern goshawks (Accipiter gentilis atricapillus). Oecologia. 134: 284-291. Kuussaari M., M. Nieminen, and I. Hanski (1996). An experimental study of migration in the Glanville fritillary butterfly Melitaea cinxia. Journal of Animal Ecology. 65: 791-801. Lambin X., J. Aars, and S.B. Piertney (2001). Dispersal, intraspecific competition, kin competition and kin facilitation: a review of the empirical evidence. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 110-122.

49

Lawrence W.S. (1987). Dispersal: an alternative mating tactic conditional on sex ratio and body size. Behavioral Ecology and Sociobiology. 21: 367-373. Léna J.-P., J. Clobert, M. de Fraipont, J. Lecomte, and G. Guyot (1998). The relative influence of density and kinship on dispersal in the common lizard. Behavioral Ecology. 9: 500-507. Lindstrom T., G.P. Brown, S.A. Sisson, B.L. Phillips, and R. Shine (2013). Rapid shifts in dispersal behavior on an expanding range edge. Proceedings of the National Academy of Sciences. 110: 13452-13456. Lowe W.H., G.E. Likens, and B.J. Cosentino (2006). Self-organisation in streams: the relationship between movement behaviour and body condition in a headwater salamander. Freshwater Biology. 51: 2052-2062. Lowe W.H., and M.A. McPeek (2014). Is dispersal neutral? Trends in Ecology and Evolution. 29: 444-450. Lüdecke D. (2018). esc: Effect size computation for meta analysis. R package version 0.4.1. https://CRAN.R-project.org/package=esc Mabry K.E., E.L. Shelley, K.E. Davis, D.T. Blumstein, and D.H. Van Vuren (2013). Social mating system and sex-biased dispersal in mammals and birds: a phylogenetic analysis. PLoS One. 8: e57980. Massot M., and J. Clobert (2000). Processes at the origin of similarities in dispersal behaviour among siblings. Journal of Evolutionary Biology. 13: 707-719. Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock). Oxford University Press Oxford, pp. 3-18. McCauley S.J. (2010). Body size and social dominance influence breeding dispersal in male Pachydiplax longipennis (Odonata). Ecological Entomology. 35: 377-385. McCauley S.J., T. Brodin, and J. Hammond (2010). Foraging rates of larval dragonfly colonists are positively related to habitat isolation: results from a landscape-level experiment. The American Naturalist. 175: E66-E73. McDevitt A.D., M.K. Oliver, S.B. Piertney, P.A. Szafrańska, M. Konarzewski, and K. Zub (2013). Individual variation in dispersal associated with phenotype influences fine-scale genetic structure in weasels. Conservation Genetics. 14: 499-509.

50

Morris D.W. (1987). Spatial scale and the cost of density-dependent habitat selection. Evolutionary Ecology. 1: 379-388. Nachappa P., D.C. Margolies, J.R. Nechols, and T.J. Morgan (2009). Response of a complex foraging phenotype to artificial selection on its component traits. Evolutionary Ecology. 24: 631-655. Nilsson J.A., M. Akesson, and J.F. Nilsson (2009). Heritability of resting metabolic rate in a wild population of blue tits. Journal of Evolutionary Biology. 22: 1867-1874. O'Connor M.I., J.F. Bruno, S.D. Gaines, B.S. Halpern, S.E. Lester, B.P. Kinlan, and J.M. Weiss (2007). Temperature control of larval dispersal and the implications for marine ecology, evolution, and conservation. Proceedings of the National Academy of Sciences. 104: 1266-1271. O'Riain M.J., J.U.M. Jarvis, and C.G. Faulkes (1996). A dispersive morph in the naked mole-rat. Nature. 380: 619-621. Olsson M., and R. Shine (2003). Female-biased natal and breeding dispersal in an alpine lizard, Niceoscincus microlepidotus. Biological Journal of the Linnean Society. 79: 277-283. Pachkowski M., S.D. Côté, and M. Festa-Bianchet (2013). Spring-loaded reproduction: effects of body condition and population size on fertility in migratory caribou (Rangifer tarandus). Canadian Journal of Zoology. 91: 473-479. Pasinelli G., and J.R. Walters (2002). Social and environmental factors affect natal dispersal and philopatry of male red-cockaded woodpeckers. Ecology. 83: 2229-2239. Perrin N., and V. Mazalov (2000). Local competition, inbreeding, and the evolution of sex- biased dispersal. The American Naturalist. 155: 116-127. Peters J.L., A.J. Sutton, D.R. Jones, K.R. Abrams, and L. Rushton (2008). Contour-enhanced meta-analysis funnel plots help distinguish publication bias from other causes of asymmetry. Journal of Clinical Epidemiology. 61: 991-996. Pettorelli N., J.M. Gaillard, G. Van Laere, P. Duncan, P. Kjellander, O. Liberg, D. Delorme, and D. Maillard (2002). Variations in adult body mass in roe deer: the effects of population density at birth and of habitat quality. Proceedings of the Royal Society B: Biological Sciences. 269: 747-753. Pfenning B., H.J. Poethke, and T. Hovestadt (2007). Dealing with time constraints on development: the effect of food availability. Ecological Entomology. 32: 273-278.

51

Phillips B.L., G.P. Brown, J.K. Webb, and R. Shine (2006). Invasion and the evolution of speed in toads. Nature. 439: 803. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Roff D.A. (1984). The cost of being able to fly: a study of wing polymorphism in two species of crickets. Oecologia. 63: 30-37. Schamber J.L., D. Esler, and P.L. Flint (2009). Evaluating the validity of using unverified indices of body condition. Journal of Avian Biology. 40: 49-56. Selonen V., and I.K. Hanski (2010). Condition-dependent, phenotype-dependent and genetic- dependent factors in the natal dispersal of a solitary rodent. Journal of Animal Ecology. 79: 1093-1100. Selonen V., I.K. Hanski, and S. Mäkeläinen (2012). Predictors of long-distance dispersal in the Siberian flying squirrel. Evolutionary Ecology. 26: 1361-1369. Skalski G.T., and J.F. Gilliam (2000). Modeling diffusive spread in a heterogeneous population: a movement study with stream fish. Ecology. 81: 1685-1700. Sloggett J.J., and W.W. Weisser (2002). Parasitoids induce production of the dispersal morph of the pea aphid, Acyrthosiphon pisum. Oikos. 98: 323-333. Snider S.B., and J.F. Gilliam (2008). Movement ecology: size-specific behavioral response of an invasive snail to food availability. Ecology. 89: 1961-1971. Southwood T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of Animal Ecology. 46: 336-365. Spinks A.C., J.U.M. Jarvis, and N.C. Bennett (2000). Comparative patterns of philopatry and dispersal in two common mole-rat populations: implications for the evolution of mole-rat society. Journal of Animal Ecology. 69: 224-234. Stamps J.A. (2006). The silver spoon effect and habitat selection by natal dispersers. Ecology Letters. 9: 1179-1185. Stevens V.M., S. Whitmee, J.F. Le Galliard, J. Clobert, K. Bohning-Gaese, D. Bonte, M. Brandle, D. Matthias Dehling, C. Hof, A. Trochet, and M. Baguette (2014). A comparative analysis of dispersal syndromes in terrestrial and semi-terrestrial animals. Ecology Letters. 17: 1039-1052.

52

Sundstrom L. (1995). Dispersal polymorphism and physiological condition of males and females in the ant, Formica truncorum. Behavioral Ecology. 6: 132-139. Tarwater C.E., and S.R. Beissinger (2012). Dispersal polymorphisms from natal phenotype- environment interactions have carry-over effects on lifetime reproductive success of a tropical parrot. Ecology Letters. 15: 1218-1229. Tejedo M., R.D. Semlitsch, and H. Hotz (2000). Covariation of morphology and jumping performance in newly metamorphosed water frogs: effects of larval growth history. Copeia. 2: 448-458. Thompson D.J. (1991). Size-biased dispersal prior to breeding in a damselfly: conflicting evidence from a natural population. Oecologia. 87: 600-601. Thompson P.L., and A. Gonzalez (2017). Dispersal governs the reorganization of ecological networks under environmental change. Nature Ecology and Evolution. 1: 162. Tummers B. (2006). DataThief III. https://datathief.org/ Van Dyck H., and M. Baguette (2005). Dispersal behaviour in fragmented landscapes: Routine or special movements? Basic and Applied Ecology. 6: 535-545. Wahlstrōm L.K., and O. Liberg (1995). Patterns of dispersal and seasonal migration in roe deer (Capreolus capreolus). Journal of Zoology. 235: 455-467. Wild G., and P.D. Taylor (2004). Kin selection models for the co-evolution of the sex ratio and sex-specific dispersal. Evolutionary Ecology Research. 6: 481-502. Wilder S.M., D. Raubenheimer, S.J. Simpson, and K.P. Lee (2016). Moving beyond body condition indices as an estimate of fitness in ecological and evolutionary studies. Functional Ecology. 30: 108-115. Woodroffe R., D.W. Macdonald, and J. da Silva (1993). Dispersal and philopatry in the European badger, Meles meles. Journal of Zoology. 237: 227-239. Zera A. (1984). Differences in survivorship, development rate and fertility between the long- winged and wingless morphs of the waterstrider, Limnoporus canaliculatus. Evolution. 38: 1023-1032.

53

Chapter 3 The Joint Evolution of Density- and Body Condition-Dependent Dispersal

This chapter was conducted in collaboration with Justin Travis, Greta Bocedi, and Shannon McCauley. I conceived the ideas with SM. I designed the methodology with JT and GB. GB coded the model in C++. I analyzed the results and wrote the first draft of the article.

3.1 Abstract

The fitness consequences of dispersal are affected by multiple aspects of the environment, as well as by individual phenotype. Individuals must therefore integrate cues from multiple sources when making dispersal decisions. The importance of multiple causation on dispersal evolution and the consequences of dispersal for ecological and evolutionary dynamics has been stated previously, yet dispersal responses to multiple factors have rarely been investigated in empirical or theoretical studies. Here, we developed an individual based model to investigate the evolution of dispersal in response to the joint effects of population density and body condition. In our model, condition is a negative function of density, such that dispersal ability (which increases with condition) is negatively associated with dispersal motivation (which increases with density). Consistent with previous models, we found that dispersal evolved to be a steep positive function of both density and condition. A novel finding of our study was that the body condition threshold for dispersal decreased with increasing density. We also found that increasing the cost of dispersal decreased both emigration rate and dispersal mortality. Dispersal that depends on phenotype and environment will have consequences for metapopulation dynamics including gene flow and metapopulation stability. We suggest that future studies incorporate individual variation in dispersal ability and motivation, in order to better predict the movement rates of individuals and genes among demes in metapopulations.

3.2 Introduction

Understanding where individuals reproduce is a central theme in ecology and evolution. The distribution of individuals and their genes across the landscape influences biological dynamics including metapopulation persistence (Kuno 1981), local adaptation (Holt and Gomulkiewicz 54

1997, Rasanen and Hendry 2008), and community assembly (Doi et al. 2010, Lindstrom and Ostman 2011). Theory predicts that individuals should reproduce in patches which maximize fitness (Southwood 1977), and therefore they must balance the potential benefits of dispersing to a new patch against the costs associated with dispersal (Benard and McCauley 2008, Clobert et al. 2009). This imposes selection on individuals to be responsive to cues providing information about both the costs and benefits of dispersal (informed dispersal; Clobert et al. 2009), and integrate information from multiple sources to make optimal dispersal decisions (Matthysen 2012).

Despite evidence that multiple environmental factors and phenotypic traits influence dispersal (Bowler and Benton 2005, Clobert et al. 2009), most previous studies have focused on a single proximate cause of dispersal, and have ignored interactive effects. This represents a gap in our understanding of dispersal behaviour because interactive effects can influence which patches are sending out emigrants and/or the phenotypes of those emigrants, thereby altering metapopulation connectivity networks and the rate and direction of gene flow. For example, Hanski et al. (1991) found that in years with low population density, small shrews (Sorex araneus) were more likely to disperse than large shrews, but when densities were high, dispersal rates were higher and no longer size biased. Since small shrews are socially subordinate and have lower reproductive success (Hanski et al. 1991) this would result in low levels of gene flow in years or metapopulations with low population density. Despite the potentially wide-reaching consequences of interactive effects on dispersal, few theoretical models have investigated dispersal evolution in response to more than one factor. We therefore require studies that test whether dispersal is influenced by the interactive effects of multiple factors.

Here, we investigate how the interaction between environmental conditions and phenotype influences the evolution of dispersal. We chose to focus on density as the environmental driver of dispersal because multiple benefits are associated with moving away from dense natal patches, including avoidance of inbreeding and reduction of competition (with kin and/or non-kin). As a result, density is likely a strong driver of dispersal evolution in many species. Moreover, the evolution of density-dependent dispersal has been well-studied, which allows us to compare our model to previous models investigating the evolution of density-dependent dispersal in isolation from other factors. 55

Previous models have found that dispersal probability increases with population density above a certain threshold (Travis et al. 1999, Poethke and Hovestadt 2002, Kun and Scheuring 2006), because individuals in high density patches have low expected fitness, and therefore have high motivation to disperse. This is largely consistent with the empirical evidence. The observed relationship between density and dispersal is predominantly positive; individuals move away from dense patches in favour of low density patches (see Matthysen (2005) for a review). However, the opposite pattern (negative density-dependent dispersal) has also been observed (Kuussaari et al. 1996, Roland et al. 2000, Ims and Andreassen 2005, Hammill et al. 2015). The reasons for this inconsistency are still unknown (Matthysen 2005), but previous authors have suggested that it may be the result of other environmental or phenotypic variables interacting with density to produce variation in the observed level of density dependence in dispersal (Ims and Hjermann 2001). The small number of studies that have tested this hypothesis are supportive; they find that the effects of density on dispersal depend on other factors (predation risk: Baines et al. 2014, Hammill et al. 2015; habitat quality: Van Allen and Bhavsar 2014; body size: Hanski et al. 1991, Baines et al. in press).

The ability to disperse and bear the associated costs is influenced by phenotypes including body condition (i.e., the size of an individual’s energy reserves; Clobert et al. 2009). For example, high condition individuals have more energy to invest in energetically costly dispersal activities including locomotion (Cockbain 1961), and settlement (O'Riain et al. 1996, Bonte et al. 2011). Theoretical models predict that when high condition individuals incur lower costs or greater benefits from dispersal, positive condition-dependent dispersal evolves (Gyllenberg et al. 2008, Bonte and de la Pena 2009). Although many empirical studies have supported this prediction (e.g., Meylan et al 2002, Eraud et al. 2011, Baines et al. 2015), there are also a number of examples of negative and non-monotonic condition-dispersal relationships (McMahon and Tash 1988, Tarwater and Beissinger 2012, Moore and Whiteman 2016). The observed variance in the effect of body condition on dispersal suggests that interactive effects may play a role in condition-dependent dispersal patterns (Ims and Hjermann 2001).

In this paper, we develop an individual based model to investigate the evolution of dispersal in response to both density and condition. We varied the cost of dispersal, and asked whether this

56

altered the evolution of density- or condition-dependent dispersal, as well as the evolution of emigration rates and dispersal mortality.

3.3 The model 3.3.1 The landscape

In this IBM we model a diploid, sexual species that exists in a spatially explicit landscape composed of patches arranged in a 10 × 10 lattice with reflective boundaries. Each patch within the landscape has a finite amount of food resources, which is re-set every generation. Resource availability varies spatially among patches. There is no spatial autocorrelation in resource availability. The amount of food present in each patch in generation 0 is randomly selected from a uniform distribution over the interval [0, 100], and then varies across generations, with positive temporal autocorrelation. Temporal variability was modeled as in Bocedi et al. (2012) such that the quantity of resources in patch (x,y) at time t is given by:

푅(푥,푦,푡) = 푅̅(1 + 휀(푥,푦,푡)) [1] where 푅̅ is the mean resource availability of a given patch and ε represents environmental noise, with its value given by:

2 휀(푥,푡,푡+1) = 휅휀(푥,푦,푡) + 휔푡√1 − 휅 [2] where κ is the autocorrelation coefficient and ω is a random normal variable with mean 0 and standard deviation σ (Ruokolainen et al. 2009). For this study, σ was always 0.8 and κ was always 0.2.

The food resources present in each patch are divided into discrete parcels. The size of these parcels within each patch follows a uniform distribution over the interval [0.2, 0.6]. Juveniles born in each patch must compete for these parcels of resources. Each juvenile has an equal chance of obtaining each parcel.

57

3.3.2 Body condition

Individual body condition (ρi) is a sigmoid function of the amount of resources each individual obtains as a juvenile (r):

1 휌 = [3] 푖 1+푒−(푟−훽)∗훼 where α and β are constants representing the maximum slope and inflection point of the curve, respectively. Condition, ρi, is bounded between 0 and 1. In this study, α = 8, and β = 0.5. Juveniles that obtain no resource parcels die.

3.3.3 Dispersal

Individuals make dispersal decisions immediately after they mature. Emigration probability, d, of individual i in patch j was determined by a logistic function modified from Kun and Scheuring (2006). In Kun and Scheuring (2006), emigration probability was a logistic function of density alone. In the current model, emigration probability is a logistic function of both density and body condition:

1 푑 = [4] 푖,푗 −훼 [퐵 −훽 ]−훼 [휌 −훽 ]−훾퐵 휌 1+푒 퐷 푗 퐷 휌 푖 휌 푗 푖 where Bj is the density of patch j (calculated as the population size divided the by total amount of resources in that patch, at the beginning of that generation), and ρi is the body condition of individual i. The remaining parameters control the dispersal response to density and body condition: αD is the slope at the inflection point of the function of density on emigration probability, βD is the value of the inflection point of the function of density on emigration probability, αρ is the slope at the inflection point of the function of body condition on emigration probability, βρ is the value of the inflection point of the function of body condition on emigration probability, and γ represents the interactive effects of condition and density on dispersal.

Each trait is controlled by a single locus. Individuals inherit one allele from each parent at each locus, and offspring phenotype is the sum of the parental alleles (additive genetic model). In each generation, each allele has a probability of mutating (1 in 100) that is independent of the mutation probability of other alleles. The size of each mutation is drawn from a normal

58

distribution with mean zero and standard deviation 1 (αρ, αD) or 0.1 (βρ, βD, γ, d0). Dispersal is nearest-neighbour, meaning individuals can only move to one of the eight patches neighbouring their natal patch; they have an equal probability of moving to each neighbouring patch.

Dispersal imposes an energetic cost on individuals. We modeled this by reducing the body condition of dispersers by an amount, c. This is an absolute cost, and so is independent of initial condition, although the proportional cost of dispersal decreases with increasing condition. This cost could represent investment into the production of dispersal structures, the cost of moving between patches, and/or the cost of settlement such as the building of a burrow, web, or nest. Individuals that attempt dispersal when the energetic cost, c, is greater than the size of their energy reserves (condition, ρi) cannot reproduce in their new patch (i.e., are functionally dead). There is no additional mortality risk imposed on dispersers. We simulated three energetic costs (c = 0.05, 0.1, and 0.3).

3.3.4 Reproduction

The number of offspring produced by each female is selected from a Poisson distribution with a mean μi given by:

휇푖 = 휌푖퐹 [5] where F represents the mean fecundity of a female in high condition (ρi = 1). In this study, F was set to 8. The sex ratio of the offspring was 50:50.

Male mating success is weighted by condition such that the probability of a male siring each offspring produced in his patch is:

휌푖 푚푖 = 푛 [6] ∑푖=1 휌푖

푛 Where ∑푖=1 휌푖 is the sum of the body conditions of all males in the patch. The probability of siring an offspring is independent of the probability of siring other offspring in the patch. Dispersal therefore reduces reproductive success by lowering female fecundity, and the ability of males to compete for mates.

59

3.3.5 Order of events 1. Resource availability in each patch is determined by that patch’s time series. 2. Individuals are born. 3. Juveniles compete for resources. 4. Juveniles mature (all individuals in the metapopulation mature simultaneously). Body condition is determined at this point. 5. Adults disperse or remain philopatric. The body condition of dispersive individuals is reduced by the amount c. 6. Adults reproduce according to their post-dispersal body condition and then die.

3.3.6 Simulation experiments

The IBM was built in C++. Each simulation was initialized with a population of 2000 individuals uniformly distributed across the landscape. Individuals in initial populations were assigned trait values that were randomly selected from a normal distribution with mean = 0 and standard deviation = 1 for parameters αD, αρ, and γ, and mean = 0.5 and standard deviation 0.1 for βD, and

βρ. Simulations were run for 100,000 generations to reach evolutionarily stable strategies (Appendix B: Figure B1). Every combination of parameters was replicated 20 times. For each combination of parameters, we present the final evolved dispersal strategies as a function of body condition and population density, by averaging the values of the evolved dispersal traits across all 20 replicates.

3.3.7 Sensitivity analysis

The results shown are from simulations run using a 10 × 10 lattice. The results were the same when a 15 × 15 lattice was used. In the simulations presented, the amount of resources per patch ranged from 0 to 100. The results were consistent when resources ranged from 25 to 75. The results were also independent of resource parcel size, and initial trait values.

3.3.8 Realized dispersal patterns

Condition was a negative function of density in the model (Appendix B: Figure B2); as a result, some combinations of natal patch density and condition did not occur in the simulated metapopulations. Therefore, the realized dispersal patterns exhibited in the metapopulation (i.e.,

60

patterns of emigration status plotted against density or body condition for the whole metapopulation) may not perfectly reflect the evolved dispersal traits (i.e., the alleles determining whether each individual should disperse or not for a given combination of density and body condition). In other words, the metapopulation does not occupy the full parameter space of the model, and so some dispersal strategies will be realized and others will not. We estimated realized dispersal patterns by plotting the emigration status (emigrated or did not emigrate, a binary variable) of each individual in the final generation of our simulated metapopulations against body condition and natal patch density. We tested the significance of these relationships using generalized linear models (GLMs) with a binomial error structure. We used emigration status as the response variable. Natal patch density and body condition were used as predictors. Because some of the simulated metapopulations had realized dispersal patterns that appeared non-monotonic, we also included the squares of density and condition as predictors. These analyses were conducted in R v 3.3.3 (R Core Team 2017).

3.4 Results

3.4.1 The evolution of a condition- and density-dependent dispersal strategy

When dispersal is responsive to the joint effects of density and condition, the evolved dispersal strategy is a steep, positive threshold function of both factors (Figure 3.1). The condition threshold for dispersal decreases with increasing density (or, equivalently, the density threshold for dispersal decreases with increasing condition). Moreover, there is curvature in the threshold response such that i) there is no dispersal out of very low density patches, and ii) individuals do not disperse if they are in very low condition even when patch density is very high. Increasing the energetic cost of dispersal decreases overall emigration rates, but does not alter the shape of the response surface (Figure 3.1).

61

Figure 3.1. Evolved dispersal strategy in response to both natal patch density and condition when c = 0.05 (A), c = 0.1 (B), and c = 0.3 (C). The strategy displayed is calculated by taking the average of the dispersal traits in the final generation of each simulation across 20 replicates. Colour corresponds to probability of emigration; purple = low emigration probability, red = high emigration probability. 62

3.4.2 The evolution of emigration rate and dispersal mortality

Metapopulations evolved emigration rates that were proportional to the cost of dispersal; emigration probability decreased as cost of dispersal increased (Figure 3.2A). Dispersal mortality (i.e., the proportion of dispersers that died) also decreased with increasing costs of dispersal (Figure 3.2B).

Figure 3.2. A) The proportion ± 95% CIs of the metapopulation that emigrated in the last generation of the simulation as a function of the dispersal cost, c. B) The proportion ± 95% CIs of dispersers that died, i.e., had a body condition value, ρi, smaller than the cost of dispersal, c, in the last generation of the simulation as a function of the dispersal cost, c.

3.4.3 Realized dispersal patterns

Realized emigration probability was a negative function of natal patch density (Figure 3.3A, C, E). Emigration probability was a positive logistic function of body condition (Figure 3.3B, D, F). This function became hump-shaped when the cost of dispersal was high (c = 0.3). 63

Figure 3.3. Realized dispersal probability ± 95% CIs as a function of natal patch density (A, C, E) and body condition (B, D, F). The cost of dispersal, c, increases going from the top to the bottom panels. The log-odds of emigration was a quadratic function of density when c = 0.05 and c = 0.3 (p < 0.0001 for both). When c = 0.1, the log-odds of emigration was a linear function of density (p < 0.0001), and including the square of density did not improve fit (p = 0.40). The log- odds of emigration was a quadratic function of body condition (p < 0.0001 for all values of c). Data shown here are from a single representative replicate of each simulation run.

64

3.5 Discussion

Here, we build on previous studies by modeling the evolution of dispersal in response to the interactive effects of density and body condition. We found that when individuals use information about both density and condition, the evolved dispersal strategy is a steep, positive, logistic function of both variables, but there are interactive effects that are apparent in the shape of the dispersal response surface. The density threshold for dispersal decreases as condition increases (equally, the condition threshold for dispersal decreases as density increases). Moreover, there is curvature in the surface such that populations evolve a strategy where they do not disperse if they are in low body condition, even when density is very high, because they have insufficient dispersal capacity. The evolved strategy also dictates that individuals in very low density patches should not disperse, even if they have high body condition, because they have low motivation to do so; however, density values below this threshold rarely occur in our simulated metapopulations. Previous authors have discussed the roles of motivation and capacity on the evolution of dispersal behaviour, but this has not yet been modeled. Our results support the verbal reasoning of previous authors (Southwood 1977, Benard and McCauley 2008), who have argued that organisms disperse when they have both sufficient motivation and capacity, which will in turn depend on environmental context. Accounting for both motivation and capacity will resolve some inconsistencies observed in dispersal patterns. For example, failure to disperse out of high density patches may be due to constraints on capacity caused by low quality habitat, rather than motivation to remain in a high density patch. A framework which incorporates both motivation and capacity will also be necessary for modeling the consequences of dispersal. For example, models of range shifts in response to climate change will come to different conclusions depending on whether they assume decreasing climate suitability will i) increase dispersal probability because of increasing motivation to disperse away from low quality habitat, or ii) decrease dispersal probability because of decreasing capacity caused by low quality habitat. Our model is one of the first to consider both multiple causes of dispersal evolution, and multiple avenues by which environment and phenotype can act on dispersal.

Our results are similar to previous modeling efforts exploring the evolution of dispersal, but provide new insights into dispersal evolution in response to multiple selective pressures. The shape of the density-dependent dispersal strategy we find here is very similar to the results of

65

Kun and Scheuring (2006), who showed that dispersal evolves to be a steep, positive logistic function of density, under a wide range of conditions. Our results are also similar to those of Travis et al. (1999) who showed that dispersal evolves to be a positive function of density, and predicted a density threshold under which no dispersal occurs. Previous work has also examined condition-dependent dispersal and our findings are similar to this earlier work. For example, Gyllenberg et al. (2008) showed that dispersal evolves to be a steep, positive, logistic function of condition, when dispersal mortality decreases with increasing condition. Our results are also consistent with those of Bonte and de la Pena (2009) who predicted that dispersal should be a positive function of condition when there is temporal variability in habitat quality, and dispersal mortality is moderate to low. However, unlike our model, these previous studies did not consider multicausality in dispersal evolution. Our results therefore provide a novel understanding of how these forces interact, including how their influence on different aspects of dispersal (i.e., the propensity and ability to disperse) may shape realized dispersal patterns.

When we plot realized emigration probability against density and condition, we observe, as expected, emigration probability is a positive logistic function of condition. However, the pattern of density-dependent dispersal is unintuitive. Realized emigration probability was a negative function of density, even though the evolved dispersal strategy is a positive threshold function of density. This pattern arises because of the negative association between body condition and natal patch density. Few individuals in high density patches have body condition values above the required threshold. Individuals from high density patches are therefore unlikely to disperse, even though their expected reproductive output is lower in their current patch than the metapopulation average. There is some evidence for this kind of energetic constraint on dispersal strategies in nature. For example, Muraji et al. (1989) found that when the wing dimorphic insect, Microvelia douglasi, was reared at high densities, a greater proportion of individuals developed into winged adults. However, when juveniles were food limited, very few individuals developed into winged adults, and the effect of density on wing development disappeared (Muraji et al. 1989). This suggests that the effects of low food availability/high competition on phenotype limits the ability of individuals to display adaptive dispersal behaviour in response to environmental stressors. Unlike previous models that considered the evolution of dispersal in response to a single factor, our model can account for complex dispersal patterns such as those displayed by M. douglasi.

66

Our model results provide important insights on an open question in the dispersal literature: the observation that a substantial amount of variability exists in condition-dependent and density- dependent dispersal (Bowler and Benton 2005, Matthysen 2005), despite the fact that theoretical models generally predict that emigration should be a positive function of both density and condition (Travis et al. 1999, Kun and Scheuring 2006, Gyllenberg et al. 2008, Bonte and de la Pena 2009). Previous authors have suggested that negative density-dependent dispersal may be the result of the benefits of group living or Allee effects (Bowler and Benton 2005, Matthysen 2005), and that negative condition-dependent dispersal may be due to the effects of strong kin selection and/or elevated dispersal ability in low condition individuals (Gyllenberg et al. 2008, Bonte and de la Pena 2009). Here, we provide evidence for a more general hypothesis that can explain variation in both condition- and density-dependent dispersal: multiple factors interact to drive dispersal evolution, and differences in the importance of these factors, or in the association between them (e.g., the effect of natal patch density on body condition) will produce variation in observed dispersal patterns across metapopulations or species. In this case negative-density dependent dispersal is not an adaptive strategy, but results from a constraint: individuals in dense patches have motivation to disperse, but are limited by their low dispersal capacity.

In this paper, we show that the interaction between condition and density can influence the dispersal patterns observed in metapopulations. However, interactions among any factors that influence dispersal could also contribute to variation in dispersal patterns. This has been observed previously in empirical studies. For example, dispersal depends on the interaction between population density and predation risk (Baines et al. 2014, Hammill et al. 2015), and between density and body size (Hanski et al. 1991). Phenotype- and density-dependent dispersal have important consequences for ecological and evolutionary dynamics (French and Travis 2001, Best et al. 2007, Duckworth and Badyaev 2007, Amarasekare 2010, Burgess and Marshall 2011, Bolnick and Otto 2013, Cote et al. 2017). For example, dispersing mosquitofish (Gambusia affinis) had higher body condition, were more asocial (Cote et al. 2010) and had larger effects on prey communities than non-dispersers (Cote et al. 2017). Therefore, improving our understanding of the variation in dispersal patterns represents a substantial advance in ecology and evolutionary biology.

67

The number of individuals that emigrate from patches, and the number that successfully immigrate into new patches influences metapopulation connectivity, persistence, and gene flow (Gomulkiewicz 1999, Lecomte et al. 2004, Dey and Joshi 2013, Deere et al. 2017). Here, we found that emigration rates were negatively related to the energetic cost of dispersal. This is consistent with previous models (Gandon and Michalakis 2001), and with empirical data showing that increasing the costs of dispersal leads to the evolution of lower emigration rates (Schtickzelle et al. 2006). Counter-intuitively, we also found that dispersal mortality decreased as the cost of dispersal increased. This resulted from the fact that high dispersal costs produce strong selection on condition thresholds for dispersal. The condition threshold for dispersal always evolved to be higher than the cost of dispersal; however, selection on this threshold was highest when the cost of dispersal was high (c = 0.3) resulting in very low genetic variation for the value of the threshold, and few individuals making “mistakes” – dispersing when they had insufficient energy reserves. Empirical evidence suggests that dispersal mortality increases with increasing dispersal costs (e.g., Schtickzelle et al. 2006); however, our model prediction is specific to metapopulations that have reached an evolutionarily stable strategy that is adapted to the landscape and the cost of dispersal in that landscape. Whereas, previous studies have subjected a population to a range of dispersal costs without allowing them to evolve dispersal strategies. Future empirical research should explore the evolution of dispersal mortality, particularly in response to landscape alterations that increase the costs of dispersal such as habitat fragmentation.

Our model makes novel predictions about the consequences of increasing dispersal costs, because of the interplay between dispersal costs and the evolution of condition-dependent dispersal. High dispersal costs result in the evolution of lower emigration rates; however, it also results in higher condition thresholds for dispersal, meaning that the lower emigration rates are mostly due to fewer low condition individuals dispersing. Since individuals with low body condition also have low reproductive output, lower dispersal rates among these individuals should have smaller effects on connectivity and gene flow than would be expected if dispersers were a random sample of the population (Benard and McCauley 2008). Our model also predicts that dispersal mortality evolves to be lower in metapopulations that experience high dispersal costs because fewer individuals make the mistake of dispersing when they have insufficient

68

capacity. In metapopulations where dispersal is not an important source of mortality, it will still impose costs (e.g., energetic and opportunity costs; Bonte et al. 2012). These alternate costs may affect metapopulation dynamics, for example, by affecting population growth rates through fecundity reductions rather than mortality increases. However, most studies to date have assumed that the only cost of dispersal is mortality risk. Future studies modeling dispersal evolution should incorporate both condition thresholds, and more diverse dispersal costs, in order to make realistic predictions about the consequences of changing dispersal costs on metapopulation dynamics including gene flow and population growth.

Empirical evidence has demonstrated that organisms integrate information about multiple aspects of their environment (e.g., population density) and their phenotype (e.g., body condition) to make dispersal decisions. Yet, most theoretical studies model the evolution of dispersal in response to a single factor in isolation from the broader ecological context. This represents a substantial gap in our understanding of dispersal, and may explain why the predictions of dispersal models do not closely match empirical observations. In this study, we modeled the evolution of dispersal in response to the combined effects of both population density and body condition, and explored the consequences of the evolved dispersal strategies for overall emigration rates, and realized dispersal patterns. We found that dispersal evolved to be a threshold function of both natal patch density and body condition, but the two factors interacted such that the value of the threshold for one factor depended on the other factor. Moreover, we found that the association between body condition and natal patch density resulted in non- intuitive realized dispersal patterns: realized emigration probability was a negative function of natal patch density. We provide evidence for the hypothesis that interactions between the multiple factors that influence dispersal produces variation in phenotype- and context-dependent dispersal. This is an important step towards understanding metapopulation dynamics, and the consequences of dispersal for ecological and evolutionary processes including metapopulation persistence, coexistence of species with competitors and predators, and local adaptation.

3.6 References

Amarasekare P. (2010). Effect of non-random dispersal strategies on spatial coexistence mechanisms. Journal of Animal Ecology. 79: 282-293.

69

Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287. Baines C.B., S.J. McCauley, and L. Rowe (2015). Dispersal depends on body condition and predation risk in the semi-aquatic insect, Notonecta undulata. Ecology and Evolution. 5: 2307-2316. Benard M.F., and S.J. McCauley (2008). Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist. 171: 553-567. Best A.S., K. Johst, T. Munkemuller, and J.M.J. Travis (2007). Which species will successfully track climate change? The influence of intraspecific competition and density dependent dispersal on range shifting dynamics. Oikos. 116: 1531-1539. Bocedi G., J. Heinonen, and J.M. Travis (2012). Uncertainty and the role of information acquisition in the evolution of context-dependent emigration. The American Naturalist. 179: 606-620. Bolnick D.I., and S.P. Otto (2013). The magnitude of local adaptation under genotype-dependent dispersal. Ecology and Evolution. 3: 4722-4735. Bonte D., and E. de la Pena (2009). Evolution of body condition-dependent dispersal in metapopulations. Journal of Evolutionary Biology. 22: 1242-1251. Bonte D., N. De Meester, and E. Matthysen (2011). Selective integration advantages when transience is costly: immigration behaviour in an agrobiont spider. Animal Behaviour. 81: 837-841. Bonte D., H. Van Dyck, J.M. Bullock, A. Coulon, M. Delgado, M. Gibbs, V. Lehouck, E. Matthysen, K. Mustin, M. Saastamoinen, N. Schtickzelle, V.M. Stevens, S. Vandewoestijne, M. Baguette, K. Barton, T.G. Benton, A. Chaput-Bardy, J. Clobert, C. Dytham, T. Hovestadt, C.M. Meier, S.C. Palmer, C. Turlure, and J.M. Travis (2012). Costs of dispersal. Biological Reviews. 87: 290-312. Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Burgess S.C., and D.J. Marshall (2011). Are numbers enough? Colonizer phenotype and abundance interact to affect population dynamics. Journal of Animal Ecology. 80: 681- 687.

70

Clobert J., J.F. Le Galliard, J. Cote, S. Meylan, and M. Massot (2009). Informed dispersal, heterogeneity in animal dispersal syndromes and the dynamics of spatially structured populations. Ecology Letters. 12: 197-209. Cockbain A.J. (1961). Fuel utilization and duration of tethered flight in Aphis fabae Scop. Journal of Experimental Biology. 38: 163-174. Cote J., T. Brodin, S. Fogarty, and A. Sih (2017). Non-random dispersal mediates invader impacts on the invertebrate community. Journal of Animal Ecology. 86: 1298-1307. Cote J., S. Fogarty, K. Weinersmith, T. Brodin, and A. Sih (2010). Personality traits and dispersal tendency in the invasive mosquitofish (Gambusia affinis). Proceedings of the Royal Society B: Biological Sciences. 277: 1571-1579. Deere J.A., T. Coulson, S. Cubaynes, and I.M. Smallegange (2017). Unsuccessful dispersal affects life history characteristics of natal populations: The role of dispersal related variation in vital rates. Ecological Modelling. 366: 37-47. Dey S., and A. Joshi (2013). Effects of constant immigration on the dynamics and persistence of stable and unstable Drosophila populations. Scientific Reports. 3: 1405. Doi H., K.H. Chang, and S. Nakano (2010). Dispersal, connectivity, and local conditions determine zooplankton community composition in artificially connected ponds. Aquatic Biology. 10: 47-55. Duckworth R.A., and A.V. Badyaev (2007). Coupling of dispersal and aggression facilitates the rapid range expansion of a passerine bird. Proceedings of the National Academy of Sciences. 104: 15017-15022. Eraud C., A. Jacquet, and P. Legagneux (2011). Post-fledging movements, home range, and survival of juvenile Eurasian collared-doves in western France. The Condor. 113: 150- 158. French D.R., and J.M.J. Travis (2001). Density-dependent dispersal in host-parasitoid assemblages. Oikos. 95: 125-135. Gandon S., and Y. Michalakis (2001). Multiple causes of the evolution of dispersal. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 155-167.

71

Gomulkiewicz R. (1999). The effects of density dependence and immigration on local adaptation and niche evolution in a black-hole sink environment. Theoretical Population Biology. 55: 283-296. Gyllenberg M., E. Kisdi, and M. Utz (2008). Evolution of condition-dependent dispersal under kin competition. Journal of Mathematical Biology. 57: 285-307. Hammill E., R.G. Fitzjohn, and D.S. Srivastava (2015). Conspecific density modulates the effect of predation on dispersal rates. Oecologia. 178: 1149-1158. Hanski I., A. Peltonen, and L. Kaski (1991). Natal dispersal and social dominance in the common shrew Sorex araneus. Oikos. 62: 48-58. Holt R.D., and R. Gomulkiewicz (1997). How does immigration influence local adaptation? A reexamination of a familiar paradigm. The American Naturalist. 149: 563-572. Ims R.A., and H.P. Andreassen (2005). Density-dependent dispersal and spatial population dynamics. Proceedings of the Royal Society B: Biological Sciences. 272: 913-918. Ims R.A., and D.O. Hjermann (2001). Condition-dependent dispersal. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 203-216. Kun Á., and I. Scheuring (2006). The evolution of density-dependent dispersal in a noisy spatial population model. Oikos. 115: 308-320. Kuno E. (1981). Dispersal and the persistence of populations in unstable habitats: a theoretical note. Oecologia. 49: 123-126. Kuussaari M., M. Nieminen, and I. Hanski (1996). An experimental study of migration in the Glanville fritillary butterfly Melitaea cinxia. Journal of Animal Ecology. 65: 791-801. Lecomte J., K. Boudjemadi, F. Sarrazin, K. Cally, and J. Clobert (2004). Connectivity and homogenisation of population sizes: an experimental approach in Lacerta vivipara. Journal of Animal Ecology. 73: 179-189. Lindstrom E.S., and O. Ostman (2011). The importance of dispersal for bacterial community composition and functioning. PLoS One. 6: e25883. Matthysen E. (2005). Density-dependent dispersal in birds and mammals. Ecography. 28: 403- 416.

72

Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock). Oxford University Press Oxford, pp. 3-18. McMahon T.E., and J.C. Tash (1988). Experimental analysis of the role of emigration in population regulation of desert pupfish. Ecology. 69: 1871-1883. Meylan S., J. Belliure, J. Clobert, and M. de Fraipont (2002). Stress and body condition as prenatal and postnatal determinants of dispersal in the common lizard (Lacerta vivipara). Hormones and Behavior. 42: 319-326. Moore M.P., and H.H. Whiteman (2016). Natal philopatry varies with larval condition in salamanders. Behavioral Ecology and Sociobiology. 70: 1247-1255. Muraji M., T. Miura, and F. Nakasuji (1989). Phenological studies on the wing dimorphism of a semi-aquatic bug, Microvelia douglasi (Heteroptera: Veliidae). Research in Population Ecology. 31: 129-138. O'Riain M.J., J.U.M. Jarvis, and C.G. Faulkes (1996). A dispersive morph in the naked mole-rat. Nature. 380: 619-621. Poethke H.J., and T. Hovestadt (2002). Evolution of density- and patch-size-dependent dispersal rates. Proceedings of the Royal Society B: Biological Sciences. 269: 637-645. Rasanen K., and A.P. Hendry (2008). Disentangling interactions between adaptive divergence and gene flow when ecology drives diversification. Ecology Letters. 11: 624-636. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Roland J., N. Keyghobadi, and S. Fownes (2000). Alpine parnassius butterfly dispersal: effects of landscape and population size. Ecology. 81: 1642-1653. Ruokolainen L., A. Linden, V. V. Kaitala, and M.S. Fowler (2009). Ecological and evolutionary dynamics under coloured environmental variation. Trends in Ecology and Evolution. 24: 555–563. Schtickzelle N., G. Mennechez, and M. Baguette (2006). Dispersal depression with habitat fragmentation in the bog fritillary butterfly. Ecology. 87: 1057-1065. Southwood T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of Animal Ecology. 46: 336-365.

73

Tarwater C.E., and S.R. Beissinger (2012). Dispersal polymorphisms from natal phenotype- environment interactions have carry-over effects on lifetime reproductive success of a tropical parrot. Ecology Letters. 15: 1218-1229. Travis J.M.J., D.J. Murrell, and C. Dytham (1999). The evolution of density-dependent dispersal. Proceedings of the Royal Society B: Biological Sciences. 266: 1837-1842. Van Allen B.G., and P. Bhavsar (2014). Natal habitat effects drive density-dependent scaling of dispersal decisions. Oikos. 123: 699-704.

74

Chapter 4 Phenotype-by-Environment Interactions Influence Dispersal: Results From a Mark-Release-Recapture Study in an Insect

This chapter was conducted in collaboration with Shannon McCauley and Ilia Ferzoco. I conceived the ideas and designed the methodology with SM. All authors collected the data. I conducted the statistical analysis and wrote the first draft of the article. All authors contributed substantially to revisions.

This chapter is published as: Baines, C. B., I. M. Fercozo, and S. J. McCauley. Phenotype-by- environment interactions influence dispersal. In press at Journal of Animal Ecology.

4.1 Abstract

Numerous studies have demonstrated that dispersal is dependent on both disperser phenotype and the local environment. However, there is substantial variability in the observed strength and direction of phenotype- and environment-dependent dispersal. This has been hypothesized to be the result of interactive effects among the multiple phenotypic and environmental factors that influence dispersal. Here, our goal was to test the hypothesis that these interactions are responsible for generating variation in dispersal behaviour. We conducted a large, 2-year, mark- release-recapture study of the backswimmer, Notonecta undulata in an array of 36 semi-natural ponds. We measured the effects of multiple phenotypic (sex, body size) and environmental (population density, sex ratio) factors, on both dispersal probability and dispersal distance. We found support for the hypothesis that interactive effects influence dispersal and produce variability in phenotype- and environment-dependent dispersal: dispersal probability was dependent on the three-way interaction between sex, body mass, and population density. Small males displayed strong, positive density dependence in their dispersal behaviour, while large males and females overall did not respond strongly to density. Small backswimmers, regardless of sex, were more likely to disperse, but this effect was strongest at high population densities. Finally, the distance dispersed by backswimmers was a negative function of population density, a pattern which we hypothesize could be related to i) individuals from high and low density patches having different dispersal strategies, or ii) the effect of density on dispersal capacity. These results suggest that phenotype-by-environment interactions strongly influence dispersal. 75

Since phenotype- and environment-dependent dispersal has different consequences for ecological and evolutionary dynamics (e.g., metapopulation persistence and local adaptation) than random dispersal, interactive effects may have wide-reaching impacts on populations and communities. We therefore argue that greater investment should be put toward estimating the effects of multiple, interacting factors on dispersal, and determining whether similar interactive effects are acting across systems.

4.2 Introduction

Dispersal, the movement of organisms among habitat patches, which produces the potential for gene flow, is a central process in ecology and evolutionary biology. Numerous empirical studies have demonstrated that dispersal is influenced by environmental factors and phenotype (reviewed in Bowler and Benton 2005, Clobert et al. 2009). Yet, so far we have been unable to come to general conclusions about how environmental factors (including population density and sex ratio) and phenotypes (including body size and sex) influence dispersal, because the results of empirical studies are highly contradictory (Bowler and Benton 2005, Clobert et al. 2009, Matthysen 2012). Researchers studying dispersal have been aware of this variability for decades (Howard 1960, Greenwood 1980), however, its cause is still an open question in the dispersal literature (Ims and Hjermann 2001, Matthysen 2005, Mabry et al. 2013). Recently, interest in this topic has increased because of studies demonstrating that context- and phenotype-dependent dispersal impacts ecological and evolutionary dynamics such as local adaptation and interspecific interactions. For example, dispersing mosquitofish (Gambusia affinis) have a greater impact on prey populations than non-dispersers (Cote et al. 2017) as a result of their being larger and less social (Cote et al. 2010). Therefore, explaining variability in context and phenotype-dependent dispersal is vital both for clearing up contradictions in the dispersal literature, and for predicting the ecological and evolutionary consequences of dispersal.

Interactions among factors affecting dispersal may account for variability in dispersal observed across ecological contexts (Ims and Hjermann 2001, Matthysen 2012). A small number of studies support this hypothesis (Cote and Clobert 2007, Delattre et al. 2013, Marjamaki et al. 2013). For example, the effect of predation risk on dispersal depends on the intensity of local competition (Baines et al. 2014, Hammill et al. 2015), and in the tropical parrot (Forpus

76

passerinus) sex ratio influences the strength of body condition-dependent dispersal (Tarwater and Beissinger 2012). However, surprisingly little work has been invested into measuring multiple dispersal drivers simultaneously, and estimating the strength of their interactive effects on dispersal in natural populations. Because these factors may all contribute to the patterns of dispersal observed, including through their interactions, large-scale dispersal studies that measure multiple dispersal drivers are required in order to test the hypothesis that interactive effects generate variability in context- and phenotype-dependent dispersal.

Theory predicts that organisms should emigrate away from high density patches if they can expect to settle in less dense patches with lower competition (Travis et al. 1999, Kun and Scheuring 2006). Empirical evidence, however, is equivocal on this point. There are numerous examples of both positive and negative density-dependent dispersal (reviewed in Matthysen 2005). Negative density dependence is often attributed to the benefits of group living, Allee effects, or conspecific attraction (Stamps 2001, Bowler and Benton 2005, Matthysen 2005, Clobert et al. 2009). However, a non-mutually exclusive possibility is that the effect of density depends on the phenotypes of individuals within those populations. For example, the fitness of small individuals is often more sensitive to population density than the fitness of large individuals (Clutton-Brock et al. 1987, Festa-Bianchet et al. 1998). Therefore, small individuals would be expected to display positive density-dependent dispersal, whereas large individuals may exhibit weakly positive density-dependent or density-independent dispersal. Negative density-dependent dispersal may occur if individuals that develop in low quality habitats have low dispersal capacity, reducing their ability to escape those poor environments (Baines and McCauley 2018). In the same vein, variation in body size-dependent dispersal may be produced by interactions with density. For example, Hanski et al. (1991) found that in years when population density was low, dispersal rates in the common shrew (Sorex araneus) were low and biased toward small individuals, whereas when population density was high, dispersal rates were higher and biased (though non-significantly) towards large individuals. This suggests that the cost-benefit trade-off for dispersal depends on the interaction between body size and density. However, there is little empirical or theoretical evidence to support or refute this hypothesis.

The effects of density on dispersal may also interact with the effects of sex and sex ratio. Previous studies have demonstrated both male- and female-biased dispersal (e.g., Favre et al. 77

1997, Caudill 2003, Bowler and Benton 2009, Baines et al. 2017; and reviewed in Greenwood 1980, Mabry et al. 2013). Theory predicts that sex differences in dispersal may be fixed if one sex has greater spatiotemporal variation in fitness as the result of the mating system (e.g., males have greater variation in fitness in polygynous mating systems) or sex-dependent resource use (Greenwood 1980, Perrin and Mazalov 2000, Gros et al. 2009, Hovestadt et al. 2014). Fixed sex biases in dispersal may also result from sex differences in the costs of dispersal (Wild and Taylor 2004, Gros et al. 2008). Finally, sex-biases may be produced when one sex experiences greater levels of intrasexual competition than the other; in this case the direction of sex bias in dispersal is expected to vary with sex ratio (Hovestadt et al. 2014). Several studies support this hypothesis. For example, De Meester and Bonte (2010) found that males increased dispersal in response to increasingly male-biased sex ratios, and females increased dispersal in response to increasing female density. However, other studies have found no sex-specific effect of sex ratio on dispersal (e.g., Baines et al. 2017). This may be the result of interactions between sex, sex ratio, and total population density. In a theoretical study of sex-biased dispersal, Hovestadt et al. (2014) found that the magnitude of the effect of sex ratio on dispersal depended on total population density, with the effect of sex ratio greatest at intermediate densities. As a result of the fact that the total proportion of individuals that dispersed was close to zero at low total densities, and close to one at high total densities, the effect of sex ratio was maximized at intermediate total densities (Hovestadt et al. 2014). However, the role of interactions between sex and ecological context (i.e., sex ratio, population density) in determining dispersal rates has not been resolved.

Our aim in this study was to measure interactions among multiple dispersal drivers to test the hypothesis that these interactions are responsible for generating variation in context- and phenotype-dependent dispersal. We achieved this goal by taking advantage of a newly built array of 36 artificial ponds at the University of Toronto’s Koffler Scientific Reserve (KSR). We conducted a mark-release-recapture study of the backswimmer, Notonecta undulata, a semi- aquatic insect that had naturally colonized these ponds before the start of our study. We marked and recaptured backswimmers in multiple ponds across two years, and estimated both dispersal probability and dispersal distance. We measured multiple environmental (density, sex ratio) and phenotypic (sex, body size) factors, and examined whether they influenced dispersal singly, and in interaction with all other factors. We predicted that dispersal would depend on the interaction

78

between phenotype and environmental factors, namely i) density and body size, and ii) density, sex ratio and sex. Specifically, we predicted that small individuals would display strong positive density-dependent dispersal, while large individuals would display density independent dispersal, or potentially negative density-dependent dispersal if mate limitation reduces fitness in low density patches. A previous experimental study of this species of backswimmer found no evidence for an effect of sex ratio on dispersal at high population density (Baines et al. 2017). We predicted that in this study, an effect of sex ratio would only be visible at intermediate population densities.

4.3 Methods 4.3.1 Study system

The backswimmer, N. undulata, completes its life cycle entirely in the aquatic environment, but is flight capable in the adult stage (Hungerford 1934). More information on the natural history of N. undulata can be found in section 1.5 ‘Study system’.

We conducted a mark-release-recapture study of N. undulata in an array of artificial ponds at KSR. The array consists of 36 3m × 3m ponds build in nine sets of four (Figure 4.1; pond array described in Searcy et al. (2018)). The nine sets are laid out in a 3 x 3 grid across the landscape, with three north-south transects of three sets each. Each set is ~100 m away from neighbouring sets in the grid (Figure 4.1). Each pond slopes gradually deeper from a shallow end and then drops off quickly to a depth of approximately 54 - 90 cm. The artificial ponds were dug in 2014 and filled with water from a large, nearby pond. The water was filtered with a 30-μm filter which excluded macroinvertebrates but included phytoplankton. Zooplankton were stocked from a mix from nearby ponds which had been filtered to restrict stocking to only zooplankton. The artificial ponds were allowed to be colonized naturally by aquatic plants, macroinvertebrates, and amphibians. By 2016, the first year of our study, all 36 ponds had established populations of N. undulata. The majority of colonists likely originated from the closest natural ponds at KSR (Barn and Gazebo ponds, approximately 100 m from the artificial pond array (Figure 4.1), and Dufferin pond, approximately 1000 m from the pond array). All of these ponds are within the maximum observed dispersal distance for N. undulata.

79

Figure 4.1. A) Satellite view of the artificial pond array at the University of Toronto’s Koffler Scientific Reserve, King, Ontario. Numbered circles indicate the locations of the nine sets of four ponds; each set is designated a number which indicates its position in the array. B) Diagram of the artificial pond array and survey schedule of the mark-release recapture study in 2017. Ponds are arranged in nine sets of four in a 3 × 3 grid. Sets are arranged in three transects (west, middle, and east). Transects are approximately 100 m apart. Within each transect, sets are approximately 100, 200, and 300 metres away from two pre-existing source ponds (gazebo pond and barn pond). Surveys were conducted on a two week rotation. Backswimmer collection: we collected unmarked backswimmers and recorded previously marked individuals from eight ponds (indicated in gray). Searching for marked backswimmers: We recaptured marked backswimmers in the remaining ponds (indicated in black). Note that the diagram in (B) is not shown to scale.

4.3.2 Collection methods and survey design

We conducted a mark-release-recapture study in the artificial pond array in two consecutive years. Backswimmers were collected using a 2.4 m × 1.2 m seine with 5 mm mesh. We

80

developed a standard seining protocol which involved dragging the seine through each pond four consecutive times, to standardize collection effort across dates and researchers.

4.3.2.1 Backswimmer collection

In both years, we collected and marked adult N. undulata from a subset of ponds in the array (Figure 4.1, Appendix C: Figure C1) every month from 18 Mar to 18 Sep in 2016 and every two weeks from 4 May to 10 Aug in 2017. We collected from only a subset of ponds because we did not have enough person hours to make collecting and measuring individuals from all ponds feasible.

We brought the backswimmers into the lab at KSR. We selected a subset of individuals from each pond: up to 20 individuals per pond in 2016, and up to 80 per pond in 2017 (the increase in the number of individuals selected was due an increased number of person hours in 2017). On this subset of individuals, we measured body size (body width to the nearest 0.01 mm in 2016 and body mass to the nearest milligram in 2017), and recorded the date (hereafter, date of original capture), the pond they were captured in (hereafter, site of original capture), their sex, and whether they were infected with Hydrachnidia mites. Finally, we used a Sharpie permanent marker to write a unique four-digit ID on the hemelytra of each individual. In 2017, we counted the total number of backswimmers collected from each pond (if we had more than 80 backswimmers from any pond, we counted those surplus individuals), in order to generate an estimate of the density of each “collection” pond. In 2016, we did not count the total number of individuals, so we do not have density estimates for any ponds in this year. We then returned the backswimmers to the same ponds from which they were collected within 30 hours of the time they were captured. If, on a collection day, we captured backswimmers that had already been marked (i.e., recaptured marked individuals) we recorded the date and the pond in which it was found.

4.3.2.2 Searching for marked backswimmers

In both years we surveyed the ponds for marked individuals at regular intervals. We did this by performing the standardized seining protocol, and sorting the backswimmers we collected for marked individuals. When a marked individual was found (i.e., “re-captured”), we recorded the date of each recapture, and the pond in which it was found. In 2016, we did this every month, 81

approximately two weeks after each collection day, and in 2017, we did this every two weeks, approximately one week after each collection day, plus one additional search day on 04 Oct 2017. We did not count individuals on “search” days, so we do not have density estimates for every pond in which marked backswimmers were recaptured.

In 2016, we did not have enough available person-hours to make it feasible to survey every pond in the artificial pond array for marked backswimmers. We therefore focused our search efforts in the ponds from which we collected backswimmers to be marked (the ponds in sets 1, 3, 7, and 9; Appendix C: Figure C1). We did this so that we were able to observe both short distance dispersal events (movement between ponds within the same set) and long-distance dispersal events (movement between ponds in different sets). This also gave us a high probability of recapturing individuals who did not disperse. In 2017, we had more person hours, and so searched for marked backswimmers in all ponds in the artificial pond array, excepting the eight ponds from which backswimmers were collected to be marked; on collection days, marked individuals in the collection ponds were captured and recorded (Figure 4.1). Not searching in all ponds in 2016 likely reduced our estimates of dispersal rates in that year; however, we still found a reasonable number of dispersers when compared to 2017 (see section 4.4 - Results).

4.3.3 Classifying dispersers

We classified individuals we recaptured as “residents” or “dispersers”. Residents were defined as individuals that were only recaptured in the same pond in which they were originally captured. Dispersers were defined as individuals that were recaptured in a pond different from the site of original capture. Individuals that were recaptured multiple times were classified as residents if they remained in their original capture site, and as dispersers if they were observed in a new pond on any recapture date. Only a single individual in our dataset had more than one dispersal event (i.e., moved from pond A to pond B to pond C). We categorized this individual as a disperser, and used information from its first dispersal event only. Removing this individual from the analysis did not change our conclusions. Note that some backswimmers may have dispersed before they were initially captured and marked.

82

4.3.4 Statistical analysis

All analyses were conducted in R v 3.4.3 (R Core Team 2017). We tested for sex differences in body size in 2016 by conducting a linear mixed model (LMM) using the ‘lme4’ package in R (Bates et al. 2015). We used body width (a proxy for body size which can be reliably measured) as the response, sex as a fixed effect, and the site of original capture as a random effect. We tested for the effects of sex and the population density of the original capture site on body mass in 2017 by conducting an LMM. We used body mass as the response, and sex and the density of the original capture site as fixed effects. Site of original capture was included as a random effect.

To test for spatial autocorrelation in population density in the artificial pond array, we conducted Moran’s I tests separately for each date on which population density was measured (every “collection date”). We used the Euclidean distance between each pair of ponds to estimate the distance matrix.

4.3.4.1 Dispersal

In the two years of this study, 451 (15%) of individuals we collected were infected with mites. Since mites generally cluster on the dorsum under the hemelytra, often damaging the wings of the backswimmers (C. Baines, personal observation), and potentially causing other damage that reduces flight ability, we excluded individuals with mite infections from all analyses of dispersal patterns.

4.3.4.1.1 Dispersal probability

To investigate whether backswimmers exhibit sex biases in dispersal, we conducted a generalized linear mixed model (GLMM) with a binomial error structure and a logit link using the ‘lme4’ package in R. We used dispersal status (resident (0) or disperser (1)) as the binary response variable. We included sex and date of original capture as fixed predictors, and original capture site as a random effect. We included the date of original capture in the model because individuals marked on different dates were tracked for different amounts of time; individuals that had overwintered were tracked from the date they were marked until their death in the summer, while newly emerged individuals were tracked from the date they were marked until the last day in each year that we searched for marked individuals (4 Oct in both years). The site of original

83

capture was included as a random effect to account for differences in dispersal rates between ponds caused by unmeasured variables. Data from 2016 and 2017 were pooled for this analysis to maximize power.

We used a GLMM with a binomial error structure to test whether body size, sex, and sex ratio and their interactions affected dispersal probability in 2016. We did not test for an effect of population density, because density was not estimated in 2016. We again used dispersal status of each individual as the response variable. We included body width, sex, sex ratio, and their two- and three-way interactions as fixed effects. Date of original capture was also included as a fixed effect. Original capture site was included as a random effect. Nonsignificant interaction terms (p > 0.05) were removed before evaluating the significance of lower order terms.

We tested for an effect of backswimmer density, sex ratio, sex, and body mass on dispersal probability in 2017 using a GLMM with a binomial error structure. The response variable was the dispersal status of each individual. We first fit a full model to estimate the effects of all variables of interest and their interactions:

Model 1: DISP ~ DENS + SR + BM + SEX + DENS:SR + DENS:BM + DENS:SEX + SR:BM + SR:SEX + BM:SEX + DENS:SR:BM + DENS:SR:SEX + DENS:BM:SEX + SR:BM:SEX + date of original capture + original capture site (random) where DISP is dispersal status, DENS is density of original capture site, SR is sex ratio of original capture site, and BM is body mass. We excluded observations in which density (number of backswimmers captured in four seines) was less than 10, in order to ensure our estimate of sex ratio was reliable. Nonsignificant interaction terms (p > 0.05) were removed before evaluating the significance of lower order terms.

Neither the main nor the interactive effects of sex ratio were significant (all p > 0.17). We therefore fit a second model which excluded sex ratio and its two- and three-way interactions, in order to estimate the effects of density of the original capture site, sex, body mass, and date of original capture on dispersal:

Model 2: DISP ~ DENS + BM + SEX + DENS:BM + DENS:SEX + BM:SEX + DENS:BM:SEX + date of original capture + original capture site (random). 84

4.3.4.1.2 Dispersal distance

For each individual that moved between ponds, we also measured dispersal distance, the Euclidean distance between the original capture site and the site of recapture. Dispersal distance was not normally distributed. Therefore, to investigate whether there were sex differences in dispersal distance, we conducted a Mann-Whitney U test using dispersal distance as the response, and sex as the predictor. We combined data from 2016 and 2017 for this analysis to maximize power.

We used path analysis, a type of structural equation modeling, to investigate the effects of environment (i.e., density) and body mass on dispersal distance in 2017. We hypothesized that population density and sex would influence body mass, and population density and body mass would influence dispersal distance (Fig 5A). Using the ‘sem’ function in the ‘lavaan’ package in R (Rosseel 2012), we estimated the coefficients of the hypothesized model. Our dataset violated the assumption of multivariate normality of structural equation modeling. To deal with this, we used the robust estimator “MLM” in the ‘sem’ function. “MLM” is maximum likelihood estimation with robust standard errors and a Satorra-Bentler scaled test statistic. The resulting model satisfied a goodness of fit test (χ2 = 1.73, df = 1, p = 0.19). Only data from 2017 were used in this analysis, because we did not have data on body mass or population density from 2016.

4.4 Results

In 2016, we marked 1354 individuals and we recaptured 440 (32%) of these. In 2017, we marked 1139 individuals and we recaptured 355 (31%) of these. In 2016, 80 of the recaptured individuals (18%) were classified as dispersers, and in 2017, 58 individuals (16%) were classified as dispersers. The probability of recapture was independent of sex, population density at original capture, and body width, but was a positive function of body mass (Appendix C: Figure C2).

2 Males were smaller than females (body width in 2016: χ 1 = 380.76, p < 0.0001, Appendix C: 2 Figure C3; body mass in 2017: χ 1 = 193.66, p < 0.0001, Appendix C: Figure C4). Body mass 2 overall was a negative function of backswimmer population density (χ 1 = 76.04, p < 0.0001; Appendix C: Figure C4); however, there was more variation in body mass within ponds than between ponds of different densities (Appendix C: Figure C4). There was no spatial

85

autocorrelation in density in 2017, except for one date when there was positive autocorrelation (Appendix C: Figure C5, Table C1).

Figure 4.2. Probability of dispersal ± 95% confidence intervals in 2016 (A) and 2017 (B) as a function of sex ratio and sex. Dispersal was not a significant function of sex or sex ratio

(proportion males) in the original capture site. 2016: nmale = 193, nfemale = 242, 2017: nmale = 178, nfemale = 162.

2 Males and females had equal dispersal probability overall (pooled 2016 and 2017 data: χ 1 = 0.052, p = 0.82; Figure 4.2). Sex ratio did not influence dispersal probability in males or females in either year of the study, and did not interact with sex, body size/mass, or density (2016: Figure 4.2A, Table 4.1, 2017: Figure 4.2B, Appendix C: Table C2). The probability that backswimmers dispersed from their original capture site in 2016 did not depend on body width (Table 4.1, Appendix C: Figure C6).

Dispersal probability in 2017 depended on the three-way interaction between population density, body mass, and sex (Table 4.2, Figure 4.3). Dispersal probability decreased with increasing body mass, and this effect was strongest at high population densities, for both males and females (Figure 4.3, Appendix C: Figure C7). Dispersal probability was a strong, positive function of population density for males with low body mass (Figure 4.3A, Appendix C: Figure C8). However, males with medium to high body mass and females across the range of body mass 86

values showed a weakly positive or no response to population density (Figure 4.3, Appendix C: Figure C8).

Table 4.1. Results of the effects of sex, sex ratio, body width (width), and date of original capture on the probability of dispersal in 2016. s.e. represents the standard error of the estimate. Parameter Estimate s.e. χ2 df p sex 0.312 0.322 0.914 1 0.339 sex ratio 0.783 1.82 0.180 1 0.672 width 0.613 0.972 0.384 1 0.536 sex × sex ratio 0.873 3.027 0.0718 1 0.789 sex × width -1.301 1.852 0.507 1 0.477 sex ratio × width 1.826 9.443 0.0335 1 0.855 sex × sex ratio × width 6.091 15.994 0.127 1 0.721 date of original capture † † 9.564 6 0.144 †Estimates and standard errors provided in Table C3.

Table 4.2. Results of model 2 showing the effects of date of original capture, and the three-way interaction between density of the original capture site (density), sex, and body mass (mass) on the probability of dispersal in 2017. s.e. represents the standard error of the estimate. Parameter Estimate s.e. χ2 df p density × sex × mass -0.00480 0.000918 4.771 1 0.0290* date of original capture † † 13.262 6 0.0388* †Estimates and standard errors provided in Table C4.

The probability of dispersal did not depend on date of original capture in 2016 (Table 4.1, Appendix C: Figure C9A). However, in 2017, the probability of dispersal was a unimodal function of the date of original capture, peaking in early summer (29 Jun 2017; Table 4.2, Appendix C: Figure C9B).

87

Figure 4.3. Probability of dispersing from the original capture site in 2017, as a function of body mass and backswimmer density in the original capture site for males (A) and females (B). Colours correspond to dispersal probability (green = low dispersal probability, red = high dispersal probability). Note the density axis increases as you move left/out of the page. Data shown are from 2017 only. nmale = 178, nfemale = 162.

Dispersal distances ranged the entire spectrum of possible distances (minimum and maximum distances between ponds in the artificial array were 5 and 305 m, respectively; Figure 4.4A). The dispersal kernel peaked at short distances, and then decreased monotonically (Figure 4.4A). Median dispersal distances did not differ between males and females (W = 2648, p = 0.41; Appendix C: Figure C10). The path analysis model explained 11% and 8% of the variation in dispersal distance and body mass, respectively, in 2017 (Figure 4.5B). When analyzing dispersers only, body mass was not a function of density at the original capture site (Figure 4.5B). Individuals from low density patches dispersed longer distances than those from high density patches (Figure 4.4B, Figure 4.5B), but dispersal distance was unrelated to body mass (Figure 4.4C, Figure 4.5B).

88

Figure 4.4. A) Histogram of backswimmer dispersal distances. The dotted line indicates the median dispersal distance. Note that data shown are only from individuals who dispersed. Data from 2016 and 2017 are pooled. n = 138. B, C) Dispersal distance as a function of B) backswimmer density (p < 0.01), and C) body mass (p > 0.05). Regression curves show mean dispersal distance ± 95% confidence intervals, and were generated using loess local regression. Data shown are from 2017 only. n = 58.

89

Figure 4.5. A) Path diagram showing the hypothesized relationships between population density of the original capture site (DENS), sex, body mass (MASS), and dispersal distance (DIST). ‘p’ represents a partial regression coefficient. ‘e’ represents residual variances. B) Path diagram showing the estimated coefficients of the paths between DENS, SEX, MASS, and DIST. Values associated with straight arrows provide standardized regression coefficients with 95% confidence intervals. Values in circular arrows represent the standardized residual variance of each factor. Asterisks indicate significance of the z-test for each standardized coefficient (* < 0.05, ** < 0.01). n = 58. Unstandardized coefficients are provided in Table C5.

4.5 Discussion

We found that the interaction between population density and body mass influenced dispersal, but this was sex-specific. Small males had strong, positive density dependence in their dispersal behaviour, while large males and females across the range of body mass values had weakly positive or no response to density. Small individuals had higher dispersal probability, overall. This finding is consistent with the fact that small individuals were less likely to be recaptured, as dispersive individuals may leave the study area and/or suffer mortality during dispersal, thereby

90

reducing the probability they are recaptured. Interestingly, the effect of body mass on dispersal was strongest when population density was high. This result supports the hypothesis that variability in the strength and direction of phenotype- and environment-dependent dispersal is due to interactions among the multiple factors influencing dispersal. Observations of interactive effects between body size and density on dispersal are rare (see Hanski et al. (1991) for an exception), but based on theory, we predict this effect is common in natural systems. Theory predicts that individuals should disperse when their expected fitness if they disperse (including the costs involved with dispersal; Bonte et al. 2012) is greater than their expected fitness if they remain philopatric (Southwood 1977). Since the strength of density dependence in fitness often depends on body size (e.g., Clutton-Brock et al. 1987; Festa-Bianchet et al. 1998), the cost- benefit ratio of dispersal should also depend on the interaction between body size and density. This would tend to result in high dispersal rates among small individuals in high density patches, a pattern we observed among males in our study. The pattern in females also trended in this direction (dispersal rates were high for small females and females in high density ponds), but the interaction between body size and density was smaller among females.

Although males and females differed in their responses to body mass and population density, they had equal dispersal probabilities overall in this study. This result contradicts a previous study of this species, which found male-biased dispersal (Baines et al. 2017). Theory predicts that sex biases in dispersal are caused by sex differences in the cost to benefit ratio of dispersal (Gros et al. 2008, Gros et al. 2009). These cost to benefit ratios may vary across seasons. For example, males may be incentivized to disperse during the prime breeding season, in order to maximize their number of mating partners, while females may reserve resources for egg production rather than investing in dispersal. This would result in sex biases in dispersal being present only during prime breeding season; however, our data do not adequately cover the peak of the breeding season for this species (approx. Mar – Apr), and therefore we may not have had sufficient power to detect sex biases in dispersal in our data. Sex biases may also vary with another factor (e.g., predation risk, food availability) that may have differed between the current study and the previous study, but were not measured here. Future research should focus on the causes of sex biases, and the factors that potentially interact with sex to produce variation in the magnitude of sex-biased dispersal.

91

In the current study, sex ratio had no effect on dispersal probability. This is consistent with a previous study of the same species, which also found no effect of sex ratio on dispersal (Baines et al. 2017). This result runs counter to the results of studies of other taxa, which have found either increased dispersal in response to increased levels of intrasexual competition (as the result of changes in sex ratio; Lawrence 1987, De Meester and Bonte 2010), or increased dispersal of both sexes in response to male-biased sex ratios, as the result of male aggression (Odendaal et al. 1989, Trochet et al. 2013). Theory predicts that the effects of sex ratio on dispersal may depend on total population density (Hovestadt et al. 2014). We found no evidence that sex ratio interacted with density to influence dispersal. Our results therefore suggest that sex ratio is not an important factor influencing dispersal in this species. We previously hypothesized that although this species is sexually dimorphic with respect to size and colouration (Appendix C: Figures C3 and C4, Baines et al. 2015, C. Baines unpublished data), these differences are small and may not be apparent to backswimmers themselves (Baines et al. 2017). Backswimmers may therefore not be able to assess the sex ratio of the populations they inhabit. This would suggest that the lack of an effect of sex ratio may be limited to this species (or higher taxonomic grouping, e.g., Notonectidae). Alternatively, sex ratio may influence dispersal only when it is extremely skewed, or only in particular time periods (e.g., the beginning of the mating season), situations for which we are lacking data. Future studies should investigate the interactive effects of density, sex ratio, and sex on dispersal since Notonecta may be idiosyncratic in its lack of a response to sex ratio.

The dispersal kernel for N. undulata peaks at short distances (approx. 0 – 50 m) and then decreases monotonically to the extent of our study site (305 m). We are likely missing dispersal events at distances greater than the maximum extent of our study site (> 305 m) because the maximum observed dispersal distance for this species is ~1000 m (C. Baines, unpublished data). However, we were prevented from tracking individuals who dispersed further than 305 m by the physical arrangement of the artificial pond array. Moreover, because the observed dispersal kernel is decreasing monotonically, we believe that our study has captured the greater part of the dispersal kernel of this species; of the 138 dispersers we observed over the two years of our study, only 5 individuals (< 4%) dispersed further than 300 m away from the site from which they were originally captured.

92

The dispersal kernel we observed here for N. undulata is similar to those published for other animals (e.g., Paradis et al. 1998, Byrne et al. 2014); however, it differed from the hump-shaped kernel observed for a related species, N. irrorata (McCauley et al. 2009). McCauley et al. (2009) suggested that the hump-shaped kernel displayed by N. irrorata may due to i) time gaps (“refractory periods”) between the initiation of dispersal and the time at which individuals are responsive to stimuli indicating suitable habitat patches, ii) a behavioural strategy to increase the probability of immigrating into a high quality patch in landscapes with positive spatial autocorrelation in habitat quality, or iii) if backswimmers use stepping stones as they disperse, the peak of the hump may correspond to the typical distance of one “leg” of a dispersal event. We suggest the difference in the shape of the dispersal kernels of N. undulata and N. irrorata may be due to different refractory periods, different behavioural strategies for dispersal (e.g., because of variation in the magnitude of spatial autocorrelation in habitat quality across study sites), or differences in the distance of a typical dispersal “leg”.

The distance travelled by backswimmers that we recaptured in our study was negatively related to the population density of the pond from which they emigrated. There are two non-mutually exclusive explanations for this result. The first explanation is that dispersal distance may be determined by the factor triggering dispersal. For example, males searching for mates may stop in multiple ponds and ultimately travel longer distances before settlement than individuals emigrating away from high density patches, who may only need to reach the next pond to increase fitness (since population density within ponds does not exhibit positive spatial autocorrelation). A similar hypothesis has been proposed by Clobert et al. (2009). The second explanation is that density influences dispersal distance indirectly through effects on dispersal capacity. Body mass is often positively related to dispersal capacity (Beck and Congdon 2000, Phillips et al. 2006), and in our system body mass is a negative function of population density. Therefore, smaller individuals from high density patches may not be able to travel as far as larger individuals from low density patches. This hypothesis was not supported by our results. We found that dispersers typically had low body mass regardless of the density of their original pond, meaning dispersers from high density patches were not smaller than dispersers from low density patches. Moreover, our path analysis found no relationship between body mass and dispersal distance. However, our power for the path analysis was low, and therefore could be

93

missing real effects. We therefore cannot make firm conclusions about how body mass influences dispersal distance; the potential for density to influence dispersal distance through effects on body mass is a hypothesis that should be tested further.

Dispersal distance being a function of phenotype and/or environment violates the assumptions of most theoretical models of dispersal, indicating that the consequences of dispersal in real systems may differ from theoretical expectations (Benard and McCauley 2008). For example, since body mass influences reproductive success (Festa-Bianchet et al. 1998, Pachkowski et al. 2013), mass- dependent dispersal distances may result in greater long distance gene flow than would be expected assuming dispersal kernels were independent of body mass. Incorporating interactive effects on dispersal will be especially important for models designed to predict real world phenomena such as range shifts of species due to climate change (Bonte and Dahirel 2017). Published models already incorporate the effects of factors such as population density and interspecific interactions (Best et al. 2007, Thompson and Gonzalez 2017), however models that incorporate the interactive effects of environment and phenotype are rare.

Overall, our study joins previous work in providing empirical support for the hypothesis that interactions among the multiple phenotypic and environmental factors influencing dispersal are responsible, at least in part, for generating the observed variability in the strength and direction of context- and phenotype-dependent dispersal. Future empirical research on dispersal should focus on estimating the interactive effects of multiple factors at each stage of dispersal, to determine if these interactions are having similar effects across species or populations. If so, incorporating interactive effects may unify the seemingly contradictory results of dispersal studies, and allow researchers to make general conclusions about context- and phenotype- dependent dispersal patterns. This, in turn, would improve our ability to predict the consequences of dispersal. Several recent studies have demonstrated that the consequences of dispersal for ecological and evolutionary processes including population growth and local adaptation depend on the phenotypic and environmental causes and correlates of dispersal. For example, Burgess and Marshall (2011) found that the phenotypes of colonizers (in this case, body condition) influenced the population growth rates of newly established populations. In a theoretical study, Bolnick and Otto (2013) found that contrary to the expectation that gene flow reduces the rate of local adaptation, dispersal that is non-random with respect to genotype (such that individuals 94

preferentially move from patches where they have low fitness to patches where they have high fitness) can increase the rate of local adaptation. However, it is difficult to apply this to generate predictions about the consequences of dispersal in natural systems because the empirical evidence for phenotype- and environment-dependent dispersal is highly variable. Unifying these contradictory results would improve the ability of researchers to incorporate phenotype- and environment-dependence into dispersal models.

4.6 References

Baines C.B., I.M. Ferzoco, and S.J. McCauley (2017). Sex-biased dispersal is independent of sex ratio in a semiaquatic insect. Behavioral Ecology and Sociobiology. 71: 119. Baines C.B., and S.J. McCauley (2018). Natal habitat conditions have carryover effects on dispersal capacity and behavior. Ecosphere. 9: e02465. Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287. Baines C.B., S.J. McCauley, and L. Rowe (2015). Dispersal depends on body condition and predation risk in the semi-aquatic insect, Notonecta undulata. Ecology and Evolution. 5: 2307-2316. Bates D., M. Maechler, B. Bolker, and S. Walker (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 67: 1-48. Beck C.W., and J.D. Congdon (2000). Effects of age and size at metamorphosis on performance and metabolic rates of Southern Toad, Bufo terrestris, metamorphs. Functional Ecology. 14: 32-38. Benard M.F., and S.J. McCauley (2008). Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist. 171: 553-567. Best A.S., K. Johst, T. Munkemuller, and J.M.J. Travis (2007). Which species will successfully track climate change? The influence of intraspecific competition and density dependent dispersal on range shifting dynamics. Oikos. 116: 1531-1539. Bolnick D.I., and S.P. Otto (2013). The magnitude of local adaptation under genotype-dependent dispersal. Ecology and Evolution. 3: 4722-4735. Bonte D., and M. Dahirel (2017). Dispersal: a central and independent trait in life history. Oikos. 126: 472-479.

95

Bonte D., H. Van Dyck, J.M. Bullock, A. Coulon, M. Delgado, M. Gibbs, V. Lehouck, E. Matthysen, K. Mustin, M. Saastamoinen, N. Schtickzelle, V.M. Stevens, S. Vandewoestijne, M. Baguette, K. Barton, T.G. Benton, A. Chaput-Bardy, J. Clobert, C. Dytham, T. Hovestadt, C.M. Meier, S.C. Palmer, C. Turlure, and J.M. Travis (2012). Costs of dispersal. Biological Reviews. 87: 290-312. Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Bowler D.E., and T.G. Benton (2009). Variation in dispersal mortality and dispersal propensity among individuals: the effects of age, sex and resource availability. Journal of Animal Ecology. 78: 1234-1241. Burgess S.C., and D.J. Marshall (2011). Are numbers enough? Colonizer phenotype and abundance interact to affect population dynamics. Journal of Animal Ecology. 80: 681- 687. Byrne A.W., J.L. Quinn, J.J. O'Keeffe, S. Green, D.P. Sleeman, S.W. Martin, and J. Davenport (2014). Large-scale movements in European badgers: has the tail of the movement kernel been underestimated? Journal of Animal Ecology. 83: 991-1001. Caudill C.C. (2003). Measuring dispersal in a metapopulation using stable isotope enrichment: high rates of sex-biased dispersal between patches in a mayfly metapopulation. Oikos. 101: 624-630. Clobert J., J.F. Le Galliard, J. Cote, S. Meylan, and M. Massot (2009). Informed dispersal, heterogeneity in animal dispersal syndromes and the dynamics of spatially structured populations. Ecology Letters. 12: 197-209. Clutton-Brock T.H., M. Major, S.D. Albon, and F.E. Guinness (1987). Early development and population dynamics in red deer. I. Density-dependent effects on juvenile survival. Journal of Animal Ecology. 56: 53-67. Cote J., T. Brodin, S. Fogarty, and A. Sih (2017). Non-random dispersal mediates invader impacts on the invertebrate community. Journal of Animal Ecology. 86: 1298-1307. Cote J., and J. Clobert (2007). Social information and emigration: lessons from immigrants. Ecology Letters. 10: 411-417.

96

Cote J., S. Fogarty, K. Weinersmith, T. Brodin, and A. Sih (2010). Personality traits and dispersal tendency in the invasive mosquitofish (Gambusia affinis). Proceedings of the Royal Society B: Biological Sciences. 277: 1571-1579. De Meester N., and D. Bonte (2010). Information use and density-dependent emigration in an agrobiont spider. Behavioral Ecology. 21: 992-998. Delattre T., M. Baguette, F. Burel, V.M. Stevens, H. Quénol, and P. Vernon (2013). Interactive effects of landscape and weather on dispersal. Oikos. 122: 1576-1585. Favre L., F. Balloux, J. Goudet, and N. Perrin (1997). Female-biased dispersal in the monogamous mammal Crocidura russula: evidence from field data and microsatellite patterns. Proceedings of the Royal Society B: Biological Sciences. 264: 127-132. Festa-Bianchet M., J.-M. Gaillard, and J.T. Jorgenson (1998). Mass- and density-dependent reproductive success and reproductive costs in a capital breeder. The American Naturalist. 152: 367-379. Greenwood P.J. (1980). Mating systems, philopatry and dispersal in birds and mammals. Animal Behaviour. 28: 1140-1162. Gros A., T. Hovestadt, and H.J. Poethke (2008). Evolution of sex-biased dispersal: The role of sex-specific dispersal costs, demographic stochasticity, and inbreeding. Ecological Modelling. 219: 226-233. Gros A., H.J. Poethke, and T. Hovestadt (2009). Sex-specific spatio-temporal variability in reproductive success promotes the evolution of sex-biased dispersal. Theoretical Population Biology. 76: 13-18. Hammill E., R.G. Fitzjohn, and D.S. Srivastava (2015). Conspecific density modulates the effect of predation on dispersal rates. Oecologia. 178: 1149-1158. Hanski I., A. Peltonen, and L. Kaski (1991). Natal dispersal and social dominance in the common shrew Sorex araneus. Oikos. 62: 48-58. Hovestadt T., O. Mitesser, and H.J. Poethke (2014). Gender-specific emigration decisions sensitive to local male and female density. The American Naturalist. 184: 38-51. Howard W.E. (1960). Innate and environmental dispersal of individual vertebrates. American Midland Naturalist. 63: 152-161. Hungerford H.B. (1934). The genus Notonecta of the world (Notonectidae-Hemiptera). University of Kansas Science Bulletin. 21: 5-195.

97

Ims R.A., and D.O. Hjermann (2001). Condition-dependent dispersal. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 203-216. Kun Á., and I. Scheuring (2006). The evolution of density-dependent dispersal in a noisy spatial population model. Oikos. 115: 308-320. Lawrence W.S. (1987). Dispersal: an alternative mating tactic conditional on sex ratio and body size. Behavioral Ecology and Sociobiology. 21: 367-373. Mabry K.E., E.L. Shelley, K.E. Davis, D.T. Blumstein, and D.H. Van Vuren (2013). Social mating system and sex-biased dispersal in mammals and birds: a phylogenetic analysis. PLoS One. 8: e57980. Marjamaki P.H., A.L. Contasti, T.N. Coulson, and P.D. McLoughlin (2013). Local density and group size interacts with age and sex to determine direction and rate of social dispersal in a polygynous mammal. Ecology and Evolution. 3: 3073-3082. Matthysen E. (2005). Density-dependent dispersal in birds and mammals. Ecography. 28: 403- 416. Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock). Oxford University Press Oxford, pp. 3-18. McCauley S.J., C.J. Davis, J. Nystrom, and E.E. Werner (2009). A hump-shaped relationship between isolation and abundance of Notonecta irrorata colonists in aquatic mesocosms. Ecology. 90: 2635-2641. Odendaal F.J., P. Turchin, and F.R. Stermitz (1989). Influence of host-plant density and male harassment on the distribution of female Euphydryas anicia (Nymphalidae). Oecologia. 78: 283-288. Pachkowski M., S.D. Côté, and M. Festa-Bianchet (2013). Spring-loaded reproduction: effects of body condition and population size on fertility in migratory caribou (Rangifer tarandus). Canadian Journal of Zoology. 91: 473-479. Paradis E., S.R. Baillie, W.J. Sutherland, and R.D. Gregory (1998). Patterns of natal and breeding dispersal in birds. Journal of Animal Ecology. 67: 518-536. Perrin N., and V. Mazalov (2000). Local competition, inbreeding, and the evolution of sex- biased dispersal. The American Naturalist. 155: 116-127.

98

Phillips B.L., G.P. Brown, J.K. Webb, and R. Shine (2006). Invasion and the evolution of speed in toads. Nature. 439: 803. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Rosseel Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software. 48: 1-36. Searcy C.A., B. Gilbert, M. Krkosek, L. Rowe, and S.J. McCauley (2018). Positive correlation between dispersal and body size in green frogs (Rana clamitans) naturally colonizing an experimental landscape. Canadian Journal of Zoology. 96: 1378-1384. Southwood T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of Animal Ecology. 46: 336-365. Stamps J. (2001). Habitat selection by dispersers: Integrating proximate and ultimate approaches. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 230-242. Tarwater C.E., and S.R. Beissinger (2012). Dispersal polymorphisms from natal phenotype- environment interactions have carry-over effects on lifetime reproductive success of a tropical parrot. Ecology Letters. 15: 1218-1229. Thompson P.L., and A. Gonzalez (2017). Dispersal governs the reorganization of ecological networks under environmental change. Nature Ecology and Evolution. 1: 162. Travis J.M.J., D.J. Murrell, and C. Dytham (1999). The evolution of density-dependent dispersal. Proceedings of the Royal Society B: Biological Sciences. 266: 1837-1842. Trochet A., D. Legrand, N. Larranaga, S. Ducatez, O. Calvez, J. Cote, J. Clobert, and M. Baguette (2013). Population sex ratio and dispersal in experimental, two-patch metapopulations of butterflies. Journal of Animal Ecology. 82: 946-955. Wild G., and P.D. Taylor (2004). Kin selection models for the co-evolution of the sex ratio and sex-specific dispersal. Evolutionary Ecology Research. 6: 481-502.

99

Chapter 5 Natal Habitat Conditions Have Carryover Effects on Dispersal Capacity and Behaviour

This chapter was conducted in collaboration with Shannon McCauley. Both authors conceived the ideas and designed the methodology. I collected the data, conducted the statistical analysis and wrote the first draft of the article. Both authors contributed substantially to revisions.

This chapter is published as: Baines, C. B., and S. J. McCauley. (2018). Natal habitat conditions have carryover effects on dispersal capacity and behavior. Ecosphere. 9: e02465.

5.1 Abstract

Local habitat quality influences dispersal, and thereby affects regional dynamics including metapopulation persistence and speciation. However, most previous studies have not been able to determine whether dispersal is more strongly affected by habitat quality experienced at the dispersal stage, or carryover effects of habitat quality from previous life stages. Strong carryover effects will cause dispersal to be temporally disconnected from its drivers, altering the impact of dispersal on metapopulation dynamics, and potentially complicating empirical estimates of context-dependent dispersal. Here, we used a fully factorial mesocosm experiment to assess how habitat quality experienced during development and at adulthood affected emigration in adult backswimmers (Notonecta undulata). We found strong carryover effects of natal habitat quality on dispersal; individuals from high quality natal environments had higher emigration rates than individuals from low quality natal environments. However, emigration did not depend on adult habitat quality. This suggests that conditions experienced during development can outweigh the effects of habitat quality at later life stages, resulting in time lags between environmental triggers and the onset of dispersal behavior. If there are critical life stages at which dispersal rates are determined, habitat quality at those stages may have outsized impacts on biological dynamics in spatiotemporally variable landscapes.

5.2 Introduction

Understanding the causes of dispersal behaviour and how dispersal varies among patches in the landscape is vital because dispersal shapes multiple aspects of the ecology and evolution of 100

species. Dispersal among habitat patches influences metapopulation persistence (Kuno 1981), local adaptation (Holt and Gomulkiewicz 1997, Rasanen and Hendry 2008), and interspecific interactions including the coexistence of species with competitors, predators, and parasites (Sabelis and Diekmann 1988, Comins and Hassell 1996). The habitat conditions organisms experience influence the costs and benefits associated with dispersal, and thus affect the number and phenotypes of emigrants and settlers (Bowler and Benton 2005, Clobert et al. 2012). One way in which habitat conditions can influence dispersal is via carryover effects, whereby conditions experienced during development affect the capacity and motivation for dispersal in a subsequent life stage (Benard and McCauley 2008, O'Connor et al. 2014). For example, when the potato tuberworm (Phthorimaea operculella) is reared on a low quality host plant, they are more likely to disperse as adults, and dispersers are more likely to have traits that reduce the cost of flight including low wing loading and high lipid content (Coll and Yuval 2004). This apparent increased investment in dispersal related traits may facilitate escape from low quality habitats.

The effects of natal habitat conditions on dispersal will depend on ecological context, and therefore may differ among systems (Benard and McCauley 2008). Individuals should be motivated to disperse away from low quality habitats. This may promote investment into dispersal capacity (e.g., wings, lipid reserves), and result in higher emigration from low quality habitats (Harada et al. 1997, Coll and Yuval 2004, Dmitriew et al. 2009). However, low quality habitats may also constrain investment into dispersal capacity, resulting in low dispersal rates even if motivation to disperse is high (Lens and Dhondt 1994, Harada and Spence 2000, Chelgren et al. 2006, O’Sullivan et al. 2014). These opposing effects may occur at different points along a continuum of natal habitat quality; empirical evidence suggests that organisms increase investment in dispersal capacity as natal habitat quality decreases (Johnson 1965, Harada et al. 1997, Dmitriew et al. 2009), but very low quality habitats may produce individuals that do not have sufficient resources to develop dispersal traits. For example, Muraji et al. (1989) showed that when the wing dimorphic, semi-aquatic insect, Microvelia douglasi, is reared at high densities, they are more likely to develop into winged adults. However, when they are food limited during development, very few individuals produce wings and the effect of density on wing development disappears. Negative correlations between dispersal motivation and dispersal

101

capacity along a continuum of natal habitat quality may often result in non-linear relationships between habitat quality and dispersal (Benard and McCauley 2008, Gyllenberg et al. 2008).

Heterogeneous habitat quality leading to asymmetry in the number or phenotypes of emigrants and immigrants among patches has wide reaching effects on ecological dynamics. For example, Vuilleumier and Possingham (2006) found that when habitat patches have asymmetrical colonization rates, the risk of metapopulation extinction increases. Moreover, a larger number of connected patches are required for the metapopulation to be viable when dispersal rates are asymmetric (Vuilleumier and Possingham 2006). The effects of habitat quality on the phenotypes of dispersers can also influence ecological dynamics. For example, Van Allen and Rudolf (2013) found that mealworm beetles (Tribolium castaneum) that were reared in high quality habitats had higher population growth rates in their colonized patches than beetles that were reared in low quality habitats. Further, patches established with beetles from high quality habitats had higher carrying capacities, and this effect lasted for multiple generations after the initial colonization event (Van Allen and Rudolf 2013). Hence, understanding the effect of habitat quality on dispersal will provide novel insights into how dispersal affects population dynamics and stability.

One aspect of this question that has historically been overlooked is the timing of the dispersal response to changes in habitat quality. If conditions experienced at one life stage often have carryover effects on dispersal in subsequent life stages, then the factors shaping dispersal may be temporally disconnected from the dispersal period itself. This has important implications for theoretical dispersal models; time lags between changes in habitat quality and the resulting dispersal behaviour could fundamentally change the predicted consequences of dispersal for ecological processes including metapopulation persistence and synchrony (Benard and McCauley 2008). Temporal disconnects can also cause difficulty for studies attempting to empirically estimate the effects of environmental factors on dispersal, and may result in the overestimation of the variance in effect sizes across studies. Yet, very few studies have considered carryover effects on dispersal (Benard and McCauley 2008) and, to our knowledge, only one study has attempted to disentangle the carryover effects of past natal habitat from the effects of the habitat experienced at the dispersal stage. Van Allen and Bhavsar (2014) found that both natal and current habitat conditions influenced emigration rates, as well as the strength of 102

density dependence in the dispersal response. This single study suggests that carryover effects of habitat quality at early life stages influences dispersal in combination with current habitat quality. We therefore require additional studies that measure the dispersal response to habitat conditions experienced at multiple life stages, and determine whether carryover effects are a common phenomenon influencing dispersal across taxonomic groups.

Carryover effects may produce time lags between the trigger of dispersal and the dispersal behaviour itself, but we also expect that habitat quality experienced at different life stages may have different types of effects on dispersal capacity and motivation (De Meester and Bonte 2010). For example, natal conditions determine structural body size (Chelgren et al. 2006), which is often an important factor influencing dispersal capacity in intra- and inter-specific comparisons (Beck and Congdon 2000, O’Sullivan et al. 2014, Stevens et al. 2014). In taxa in which structural body size and morphology (e.g., wing presence or size) is fixed at adulthood, conditions experienced after maturity cannot influence dispersal through effects on morphology. Habitat quality experienced during adulthood likely influences more labile aspects of dispersal capacity (e.g., lipid reserves; Baines et al. 2015), but its impacts on dispersal motivation are probably more important than effects on capacity at this life stage. Finally, changes in habitat quality through time or across life stages may influence dispersal propensity. For example, experience in the natal habitat may influence the perception of habitat quality (and therefore dispersal motivation) later in life (Stamps et al. 2009). Understanding how individual phenotypic variation related to natal habitat conditions (e.g., variation in body size) interacts with habitat conditions experienced at the dispersal stage is important for understanding dispersal in spatiotemporally variable environments.

In this study, we tested whether natal habitat conditions interacted with adult habitat quality to determine emigration rates in adults using a full factorial experiment on the backswimmer, Notonecta undulata. Furthermore, we measured the effect of natal habitat quality on dispersal capacity (body size, lipid content, protein content, and wing morphology), to test whether individual variation in these traits could explain variation in dispersal behaviour. We predicted that emigration rates of adults would increase with decreasing habitat quality, but that the expression of this dispersal behaviour would depend on natal habitat quality. Specifically, we predicted that emigration out of high quality adult patches should be low overall, and not 103

dependent on natal patch quality. Whereas, motivation to disperse out of low quality adult patches should be high, and dispersal probability should increase with increasing dispersal capacity, which we predicted would be a positive function of natal habitat quality. Therefore, we expected mean emigration rates in the treatment groups to rank in the following order (in order from highest to lowest emigration rate): 1) high natal/ low adult habitat quality, 2) low natal/low adult habitat quality, 3) high natal/high adult habitat quality, and 4) low natal/high adult habitat quality. In addition, we predicted that within a given treatment, large individuals should have higher emigration probabilities because they have higher dispersal capacity than small individuals, and so will be more likely to attempt dispersal.

5.3 Methods

5.3.1 Study system

The backswimmer N. undulata is a generalist carnivore which eats a variety of aquatic prey (including zooplankton and Diptera larvae), but also scavenges for animals trapped on the water surface (Hungerford 1934). Notonecta complete their entire life cycle in the aquatic environment. There are five juvenile instars which are wingless and restricted to the body of water in which they are oviposited (Hungerford 1934). Adult backswimmers have wings and can disperse by flight between water bodies. More information on the natural history of N. undulata can be found in section 1.5 ‘Study system’.

5.3.2 Manipulating natal habitat quality

We collected third and fourth instar N. undulata from two fishless ponds at the Koffler Scientific Reserve (KSR) from 11 – 17 Jul 2017. On the day they were collected, they were placed individually in plastic cups (diameter: 11 cm, height: 9 cm) filled with ~250 mL of dechlorinated well water. The cups were placed on shelves in the laboratory at KSR (mean water temperature = 19° C).

We randomly assigned backswimmers to one of two diet treatments: high natal quality (306 juveniles) or low natal quality (432 juveniles). Juveniles in the high quality treatment were fed a mixture of zooplankton and mosquito larvae every day (approximately 7 mosquito larvae and 52 zooplankton daily). This amount of food corresponds to approximately the maximum amount

104

they can eat in a single day. Juveniles in the low quality treatment were fed the same food mixture two of every three days, but were starved the third day. Similar types of diet regimes have been used previously to manipulate food availability in heteropterans (e.g., Muraji et al. 1989). This protocol ensured that individuals in the low quality treatment were receiving less food, and experiencing periods in which they perceived food availability to be low. This feeding regime may also be similar to conditions backswimmers experience in natural ponds: long periods with no or low food availability, followed by a large meal.

We monitored all cups daily and recorded the date of death of any individuals that died, and the date that each individual molted to the adult stage. Once most ( > 95%) of the individuals had matured, we ended the diet manipulation. On 31 Aug and 1 Sep 2017, we recorded the sex of each adult, and measured fresh adult body mass to the nearest milligram (VWR 403B scale). We randomly selected 27 individuals from the high quality natal treatment and 16 individuals from the low quality natal treatment, and preserved them in 70% ethanol, in order to measure the effects of the juvenile diet treatment on dispersal traits (body condition, body size, wing morphology).

5.3.3 Dispersal experiment

We measured emigration probability using a mesocosm experiment in the field at KSR. We marked each backswimmer by writing a unique four-digit number on their hemelytra using a Sharpie permanent marker, in order to track individual dispersal behaviour. On 1 Sep 2017, we placed 10 backswimmers from the high quality natal treatment or 10 from the low quality natal treatment in each of 18 tanks (19L) filled with dechlorinated well water. Backswimmers from the high and low quality natal treatments were separated into different tanks to minimize cannibalism. Backswimmers exhibit size-structured intraguild predation (Sih 1982; IM Ferzoco, personal communication), and individuals from the low quality natal treatment were smaller than individuals from the high quality natal treatment.

Backswimmers were randomly chosen from the natal habitat treatments and assigned to adult tanks. Each tank had structures made out of fiberglass mesh and plastic ribbon to provide refuge. We randomly assigned each tank to one of three adult habitat quality treatments (low, medium, or high food availability). Low, medium, and high adult treatments received food (zooplankton) 105

at a ratio of 1:3:9 units, respectively, per day (1 food unit was approximately 700 zooplankton). Therefore, the density of prey (no. prey per unit volume of water) in the low quality adult treatment was lower than prey density in the juvenile diet, and the density of prey in the high quality adult treatment was greater than that in the juvenile diet. We covered the tanks for two days to allow backswimmers to acclimate to their new environments. On 3 Sep 2017, we randomly selected one backswimmer from each tank, and preserved it in 70% ethanol in order to estimate the effects of the adult habitat quality treatments on body condition. We took this sample before any dispersal occurred so that we could estimate the effects of adult habitat quality treatment on a random sample of backswimmers, without the potentially confounding effects of mass-dependent dispersal.

On 3 Sep 2017, we removed the covers from the mesocosm tanks and allowed backswimmers to disperse. Every three days from 6 Sep to 15 Sep 2017, we recorded the ID numbers of each individual remaining in the tanks, as well as the ID numbers of dead individuals. Backswimmers that were missing from the tanks were assumed to have dispersed by flight (we did not attempt to recapture dispersers, and no backswimmers flew from one mesocosm into another). Dispersers can be distinguished from cannibalized individuals because backswimmers are piercing, sucking predators that leave the exoskeletons of their prey whole, so we were able to read the ID numbers of cannibalized individuals. At the end of the experiment, all remaining backswimmers were preserved in 70% ethanol.

5.3.4 Measurement of dispersal capacity

To estimate the effects of the natal treatment on dispersal capacity, we measured several traits in individuals that were preserved immediately after the juvenile diet manipulation was terminated: body size, dry body mass, dry lipid mass, dry protein mass, and wing area. We took digital photographs of each individual, and measured pronotum width as a proxy for body size (C. Baines, unpublished data) using ImageJ (Schneider et al. 2012). We also photographed the right wing of each backswimmer, after removing it from the rest of the body. We measured the area of each wing using ImageJ. We then measured dry body mass, dry lipid mass, and dry protein mass using the same methods as Baines et al. (2015). In brief, we dehydrated the backswimmers to a constant mass in a drying oven and weighed them to the nearest 0.01 mg using a Mettler AE240

106

scale. We used chloroform redux to dissolve triglyceride fat, and then dried and weighed the backswimmers again. Finally, we submerged the backswimmers in 0.2 mol/L potassium hydroxide to dissolve protein, and then dried and weighed them a final time. Dry lipid mass was calculated as the difference between dry body mass and dry fatless mass, and dry protein mass was calculated as the difference between dry fatless mass and dry fatless, proteinless mass. The same person took all of the measurements of mass, body size, and wing area, and was blind to the sex and treatment of the backswimmers.

We estimated the effect of the adult habitat quality treatment on body condition by measuring the dry body mass, dry lipid mass, and dry protein mass of individuals preserved on 3 Sep 2017 (after the acclimation period in the adult tanks but before the dispersal period) and a random subset of those remaining on 15 Sep 2017 (after the dispersal period). These traits were measured as described above.

5.3.5 Statistical analysis

To test whether natal habitat quality influenced structural body size we used an ANOVA on the "pre" (individuals preserved immediately after the juvenile diet manipulation) samples. We used body width as the response, and natal habitat quality, sex, and their interaction as predictors.

To test whether natal habitat quality influenced body mass we used an ANOVA on the “pre” samples. We used dry body mass as the response, and natal habitat quality, sex, and their interaction as predictors. Using fresh mass as the response variable gave similar results (data not shown).

We tested whether body mass changed while backswimmers were in the adult tanks using an ANOVA with dry body mass as the response variable, and natal habitat quality, sex, date of preservation, and the two-way interactions between natal habitat quality and date of preservation, and sex and date of preservation as predictors. There was a significant interaction between natal habitat quality and the date of preservation on dry body mass, so we performed post hoc tests to examine whether the high and low natal quality treatments both increased in body mass while in the adult tanks. We used a Bonferroni corrected α = 0.025 to account for multiple comparisons.

107

We tested whether the quality of the adult habitat influenced dry body mass by conducting a linear mixed model (LMM) on the “post” samples (individuals who were used in the dispersal experiment, but did not disperse, and were preserved at the end of the experiment). We used dry body mass as the response variable, and adult habitat quality, natal habitat quality, sex, and all two-way interactions as fixed predictors. Mesocosm tank was used as a random effect. The LMM was built using the ‘lme4’ package in R (Bates et al. 2015).

To test whether wing area depended on natal habitat quality, we conducted an ANCOVA on the "pre" samples. We used wing area as the response, and natal habitat quality, fresh body mass, and their interaction as predictors. We used fresh mass as a covariate when analyzing wing area rather than dry mass because fresh mass is more relevant to their actual flight ability.

To test the effects of natal and adult habitat quality on dispersal probability, we built a generalized linear mixed model (GLMM) with a binomial error structure and a logit link using the ‘lme4’ package in R (Bates et al. 2015). We used dispersal status at the end of the experiment (dispersed or did not disperse) as the response variable. We included fresh body mass, sex, natal habitat quality, adult habitat quality, and the interaction between natal and adult habitat quality as fixed predictors. We did not perform a full interaction model because sample sizes were insufficient to handle a large number of predictors. We included the identity of the adult (dispersal) tank and adult age (days since adult emergence) as random effects. Age was included in the model because insects may be unable to fly for several days after emergence (Johnson 1969), and so time since emergence may explain some variation in dispersal probability. Individuals that died during the dispersal experiment (37 in total) were excluded from the analysis. Therefore, our analysis included 125 individuals, 20 of which were dispersers.

The significance of all terms was evaluated using type II sum of squares. Nonsignificant interaction terms (p > 0.05) were removed before evaluating the significance of main effects. All analyses were conducted in R v 3.3.3 (R Core Team 2017).

108

5.4 Results 5.4.1 Effects of habitat quality on survival, development rate, body size, and body composition

The natal habitat quality treatment had large effects on backswimmer survival and development. 2 Individuals from the high quality natal treatment had higher survival (χ 1 = 165.2, p < 0.0001; 2 Appendix D: Figure D1), and developed faster (χ 1 = 162.2, p < 0.0001; Appendix D: Figure D2) than individuals from the low quality natal treatment.

Figure 5.1. Body width (mm) of female and male backswimmers in both natal habitat quality treatments. Results shown are from the "pre" samples (animals that were preserved immediately after the juvenile diet manipulation).

Immediately after the juvenile diet manipulation, backswimmers reared in the high quality natal environment were larger (F1,37 = 9.96, p = 0.003; Figure 5.1), and heavier (F1,37 = 47.16, p < 0.0001; Figure 5.2A) than individuals in the low quality natal treatment. Total wing area increased with fresh body mass (F1,18 = 44.00, p < 0.0001; Figure 5.3), but after controlling for body mass, natal habitat quality had no effect on wing area (F1,18 = 1.30, p = 0.27; Figure 5.3).

109

Figure 5.2. Dry body mass (mg) as a function of natal habitat quality, adult habitat quality, and the time at which the backswimmers were preserved. A) “pre” indicates animals that were preserved immediately after the juvenile diet manipulation. B) “mid” indicates animals that were preserved after the acclimation period in the adult tanks, but before the dispersal experiment. C) “post” indicates animals that were preserved after the dispersal experiment.

There was an interaction between timing of preservation and natal treatment on body mass (F2,111 = 5.42, p = 0.0057; Figure 5.2A,C). Post-hoc tests revealed that backswimmers from the low quality natal treatment gained weight during the dispersal experiment, but backswimmers from the high quality natal treatment did not (low natal: F2,52 = 11.43, p = 0.0001; high natal: F2,60 = 0.39, p = 0.68; Figure 5.2A,C). As a result, by the end of the experiment, the difference in body 2 mass between the high and low natal habitat treatments was no longer significant (χ 1 = 2.64, p = 0.10; Figure 5.2C). Body mass was a positive function of adult habitat quality at the end of the 2 dispersal experiment (χ 2 = 9.23, p = 0.0099; Figure 5.2C).

110

Figure 5.3. Wing area (mm2) as a function of fresh body mass (mg) and natal habitat quality. All data shown are from "pre" samples (animals that were preserved immediately after the juvenile diet manipulation). Lines and bands represent mean wing area ± 95% confidence intervals as a function of fresh body mass and natal habitat quality.

Lipid mass and protein mass were both positively related to total body mass (Appendix D). After controlling for differences in total body mass, the effects of natal and adult habitat quality on lipid and protein mass were minimal. Further details on the statistical analysis and results for body composition can be found in Appendix D.

5.4.2 Dispersal

Backswimmers from the high quality natal habitat were more likely to disperse than those from 2 the low quality natal habitat (χ 1 = 6.89, p = 0.008; Figure 5.4, Figure 5.5). The odds ratio for dispersal for individuals in the high natal habitat quality treatment compared to individuals in the low natal treatment was 5.78. Overall, the probability of dispersal increased with body mass 2 (fresh body mass: χ 1 = 3.91, p = 0.048; Figure 5.4). There was no main effect of adult habitat 2 quality on dispersal (χ 2 = 0.0051, p = 0.997; Figure 5.5), and no interaction between natal 2 habitat quality and adult habitat quality (χ 2 = 1.69, p = 0.43; Figure 5.5). Males and females did 2 not significantly differ in dispersal probability (χ 1 = 3.22, p = 0.073; Appendix D: Figure D7).

111

Figure 5.4. Dispersal status at the end of the experiment for each individual backswimmer, as a function of fresh body mass, and natal habitat quality. Each point represents an individual. Note that individuals either dispersed (y = 1) or did not disperse (y = 0), but points are jittered with respect to the y axis to improve visibility. Lines and bands represent the probability of dispersal ± 95% confidence intervals as a function of fresh body mass and natal habitat quality. Regression lines were generated using a generalized linear model with a binomial error structure.

112

Figure 5.5. Proportion of backswimmers dispersed ± 95% confidence intervals for each natal habitat quality × adult habitat quality treatment.

5.5 Discussion

Natal habitat quality had strong effects on dispersal capacity and dispersal probability; backswimmers in the high quality natal treatment were larger, heavier, and more likely to disperse (the odds of dispersal were more than five times higher for individuals from the high natal habitat quality treatment than for those from the low natal habitat quality treatment). This suggests that low quality natal habitat constrains investment in dispersal capacity, and thereby reduces dispersal rates. Our results are congruent with previous studies that have found that small body size resulting from low food availability during development can decrease dispersal probability (Chelgren et al. 2006, Van Allen and Bhavsar 2014), or lead to delayed dispersal (Nunes and Holekamp 1996). Reduced or delayed dispersal can negatively affect fitness. For example, Spear et al. (1998) found that philopatric western gulls (Larus occidentalis) experienced higher fitness costs (lower survival in males, and lower reproductive success in females) than dispersive gulls, and Lens and Dhondt (1994) found that late dispersers were less likely to settle in high quality habitats than early dispersers.

Our study suggests that the developmental environment is an important determinant of dispersal rates in backswimmers. We found that low quality habitat patches produce fewer dispersers, and

113

therefore likely contribute less to demographic and genetic connectivity than high quality habitat patches. This may have important consequences for ecological and evolutionary dynamics. For example, asymmetries in the number of dispersers each patch contributes may increase the probability of metapopulation extinction (Vuilleumier and Possingham 2006). If individuals in low quality patches are unable to disperse because of their low dispersal capacity, this may also lead to increased extinction of lineages as the result of being “trapped” in low quality patches subject to stochastic extinction. Asymmetries in dispersal rates may also influence local adaptation. For example, in spatially heterogeneous landscapes, large numbers of dispersers from high quality habitat types may impede the rate of adaptation to low quality habitat types (Rasanen and Hendry 2008).

Contrary to our prediction, adult habitat treatment did not influence dispersal probability, and there was no interaction between natal and adult habitat treatments. This is surprising given that previous studies have observed that habitat quality at the dispersal stage influences dispersal probability (Baines et al. 2014, Van Allen and Bhavsar 2014), and that the adult habitat treatment in our study had a strong enough influence on body mass that effects were measurable after only two weeks (the duration of the time backswimmers were in the adult tanks). There are several potential explanations for the lack of an effect of adult habitat quality on dispersal. It is possible that backswimmers perceived the adult mesocosms to be poor quality regardless of adult habitat treatment because the mesocosms were smaller and had lower prey diversity than typical backswimmer habitats. In this case, they may all have been motivated to disperse, but some (mostly from the low quality natal habitats) were prevented from dispersing by insufficient dispersal capacity. Since the backswimmers from the low quality natal treatment gained weight (and lipid mass; Appendix D) during the dispersal experiment, but still had low dispersal rates, the constraint on dispersal capacity could be due to small structural body sizes, or improperly developed flight musculature. Another possible explanation is that two weeks of low food availability may not trigger dispersal in backswimmers. Backswimmers scavenge for dead arthropods on the surface of the water, and may go long periods with low food availability, followed by a large scavenged meal. Therefore, dispersal in backswimmers may not be triggered unless prey densities are low for a longer period of time than the duration of this experiment.

114

Finally, food availability may influence dispersal probability, but not within the range of prey densities tested in this experiment.

If natal habitat quality tends to have large effects on dispersal, then temporally variable landscapes may exhibit only weak relationships between dispersal rates and concurrent habitat quality. This may help to explain some observations in the empirical literature. For example, dispersal generally exhibits positive density dependence (Johnson 1965, De Meester and Bonte 2010, Nowicki and Vrabec 2011, Baines et al. 2014), but several studies have reported no relationship between dispersal and density (Arcese 1989, Keppie and Towers 1992, Gaillard et al. 2008). These null relationships could be due to time lags in dispersal responses to natal population density. The effects of spatiotemporal variability on dispersal evolution have been well-studied (Roff 1975, McPeek and Holt 1992, Mathias et al. 2001, Leturque and Rousset 2002, Kun and Scheuring 2006), but the carryover effects of natal habitat conditions on dispersal in subsequent life stages have received less attention (Benard and McCauley 2008). Modeling carryover effects in spatiotemporally variable landscapes would be a useful line of future research that may provide new insights into dispersal behaviour and explain dispersal patterns that currently remain poorly understood.

In our study, dispersal probability was a positive function of body mass. This was evident from the fact that backswimmers from the high quality natal treatment were heavier, and had higher dispersal rates. Additional evidence of a positive body mass-dispersal association comes from the observation that after controlling for natal and adult habitat treatments, individuals with high body mass were more likely to disperse than small individuals. This positive body mass- dependent dispersal is consistent with previous studies (Holekamp 1986, Wahlstrōm and Liberg 1995, Debeffe et al. 2012, Selonen et al. 2012). These results indicate than in backswimmers, dispersers are expected to be disproportionately sampled from high quality patches, and are expected to have higher body mass than the metapopulation average. Since body size is positively correlated with reproductive success in insects (Honěk 1993, Wiklund and Kaitala 1995), positive mass-dependent dispersal may have consequences for metapopulation dynamics. For example, heavy dispersers may contribute more to rescue effects (Brown and Kodric-Brown 1977) and gene flow than would be expected assuming that dispersers were a random sample of

115

the metapopulation. This effect may be amplified if large dispersers have higher immigration success (Bonte et al. 2011, Remy et al. 2011, O’Sullivan et al. 2014).

Spatial variation in habitat quality that results in asymmetries in dispersal rates between habitat patches can influence ecological and evolutionary processes including metapopulation persistence, and local adaptation. Here, we show that carryover effects of habitat quality experienced during juvenile development influenced dispersal capacity and probability at the adult stage. However, although conditions experienced at the adult stage may have influenced dispersal capacity through effects on body mass, adult habitat quality did not influence dispersal probability. Moreover, we found that dispersal probability was a positive function of body mass. These results indicate that high quality natal patches contribute more to demographic and genetic connectivity than low quality natal patches. Variation among patches in resource availability during juvenile development and its effects on traits such as body size is likely to be an important determinant of dispersal rates, and therefore influence ecological and evolutionary dynamics in metapopulations.

5.6 References

Arcese P. (1989). Intrasexual competition, mating system and natal dispersal in song sparrows. Animal Behaviour. 38: 958-979. Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287. Baines C.B., S.J. McCauley, and L. Rowe (2015). Dispersal depends on body condition and predation risk in the semi-aquatic insect, Notonecta undulata. Ecology and Evolution. 5: 2307-2316. Bates D., M. Maechler, B. Bolker, and S. Walker (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 67: 1-48. Beck C.W., and J.D. Congdon (2000). Effects of age and size at metamorphosis on performance and metabolic rates of Southern Toad, Bufo terrestris, metamorphs. Functional Ecology. 14: 32-38. Benard M.F., and S.J. McCauley (2008). Integrating across life-history stages: consequences of natal habitat effects on dispersal. The American Naturalist. 171: 553-567.

116

Bonte D., N. De Meester, and E. Matthysen (2011). Selective integration advantages when transience is costly: immigration behaviour in an agrobiont spider. Animal Behaviour. 81: 837-841. Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Brown J.H., and A. Kodric-Brown (1977). Turnover rates in insular biogeography: effect of immigration on extinction. Ecology. 58: 445-449. Chelgren N.D., D.K. Rosenberg, S.S. Heppell, and A.I. Gitelman (2006). Carryover aquatic effects on survival of metamorphic frogs during pond emigration. Ecological Applications. 16: 250-261. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock (eds.) (2012). Dispersal ecology and evolution. Oxford University Press, Oxford. Coll M., and B. Yuval (2004). Larval food plants affect flight and reproduction in an oligophagous insect herbivore. Environmental Entomology. 33: 1471-1476. Comins H.N., and M.P. Hassell (1996). Persistence of multispecies host-parasitoid interactions in spatially distributed models with local dispersal. Journal of Theoretical Biology. 183: 19-28. De Meester N., and D. Bonte (2010). Information use and density-dependent emigration in an agrobiont spider. Behavioral Ecology. 21: 992-998. Debeffe L., N. Morellet, B. Cargnelutti, B. Lourtet, R. Bon, J.M. Gaillard, and A.J. Mark Hewison (2012). Condition-dependent natal dispersal in a large herbivore: heavier animals show a greater propensity to disperse and travel further. Journal of Animal Ecology. 81: 1327-1337. Dmitriew C., J. Carroll, and L. Rowe (2009). Effects of early growth conditions on body composition, allometry, and survival in the ladybird beetle Harmonia axyridis. Canadian Journal of Zoology. 87: 175-182. Gaillard J.M., A.J. Hewison, P. Kjellander, N. Pettorelli, C. Bonenfant, B. Van Moorter, O. Liberg, H. Andren, G. Van Laere, F. Klein, J.M. Angibault, A. Coulon, and C. Vanpe (2008). Population density and sex do not influence fine-scale natal dispersal in roe deer. Proceedings of the Royal Society B: Biological Sciences. 275: 2025-2030.

117

Gyllenberg M., E. Kisdi, and M. Utz (2008). Evolution of condition-dependent dispersal under kin competition. Journal of Mathematical Biology. 57: 285-307. Harada T., and J.R. Spence (2000). Nymphal density and life histories of two water striders (Hemiptera: Gerridae). The Canadian Entomologist. 132: 353-363. Harada T., R. Tabuchi, and J. Koura (1997). Migratory syndrome in the water strider Aquarius paludum (Heteroptera: Gerridae) reared in high versus low nymphal densities. European Journal of Entomology. 94: 445-452. Holekamp K.E. (1986). Proximal causes of natal dispersal in Belding's ground squirrels (Spermophilus beldingi). Ecological Monographs. 56: 365-391. Holt R.D., and R. Gomulkiewicz (1997). How does immigration influence local adaptation? A reexamination of a familiar paradigm. The American Naturalist. 149: 563-572. Honěk A. (1993). Intraspecific variation in body size and fecundity in insects: a general relationship. Oikos. 66: 483-492. Hungerford H.B. (1934). The genus Notonecta of the world (Notonectidae-Hemiptera). University of Kansas Science Bulletin. 21: 5-195. Johnson B. (1965). Wing polymorphism in aphids II. Interaction between aphids. Entomological Experiments and Applications. 8: 49-64. Johnson C.G. (1969). Migration and dispersal of insects by flight. Methuen & Co. Ltd, London. Keppie D.M., and J. Towers (1992). A test on social behavior as a cause of dispersal of spruce grouse. Behavioral Ecology and Sociobiology. 30: 343-346. Kun Á., and I. Scheuring (2006). The evolution of density-dependent dispersal in a noisy spatial population model. Oikos. 115: 308-320. Kuno E. (1981). Dispersal and the persistence of populations in unstable habitats: a theoretical note. Oecologia. 49: 123-126. Lens L., and A.A. Dhondt (1994). Effects of habitat fragmentation on the timing of crested tit Parus cristatus natal dispersal. The Ibis. 136: 147-152. Leturque H., and F. Rousset (2002). Dispersal, kin competition, and the ideal free distribution in a spatially heterogeneous population. Theoretical Population Biology. 62: 169-180. Mathias A., E. Kisdi, and I. Olivieri (2001). Divergent evolution of dispersal in a heterogeneous landscape. Evolution. 55: 246-259.

118

McPeek M.A., and R.D. Holt (1992). The evolution of dispersal in spatially and temporally varying environments. The American Naturalist. 140: 1010-1027. Muraji M., T. Miura, and F. Nakasuji (1989). Phenological studies on the wing dimorphism of a semi-aquatic bug, Microvelia douglasi (Heteroptera: Veliidae). Research in Population Ecology. 31: 129-138. Nowicki P., and V. Vrabec (2011). Evidence for positive density-dependent emigration in butterfly metapopulations. Oecologia. 167: 657-665. Nunes S., and K.E. Holekamp (1996). Mass and fat influence the timing of natal dispersal in Belding's ground squirrels. Journal of Mammalogy. 77: 807-817. O'Connor C.M., D.R. Norris, G.T. Crossin, and S.J. Cooke (2014). Biological carryover effects: linking common concepts and mechanisms in ecology and evolution. Ecosphere. 5: 28. O’Sullivan D., T.G. Benton, and T.C. Cameron (2014). Inter-patch movement in an experimental system: the effects of life history and the environment. Oikos. 123: 623-629. Rasanen K., and A.P. Hendry (2008). Disentangling interactions between adaptive divergence and gene flow when ecology drives diversification. Ecology Letters. 11: 624-636. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Remy A., J.F. Le Galliard, G. Gundersen, H. Steen, and H.P. Andreassen (2011). Effects of individual condition and habitat quality on natal dispersal behaviour in a small rodent. Journal of Animal Ecology. 80: 929-937. Roff D.A. (1975). Population stability and the evolution of dispersal in a heterogeneous environment. Oecologia. 19: 217-237. Sabelis M.W., and O. Diekmann (1988). Overall population stability despite local extinction: the stabilizing influence of prey dispersal from predator-invaded patches. Theoretical Population Biology. 34: 169-176. Schneider C.A., W.S. Rasband, and K.W. Eliceiri (2012). NIH Image to ImageJ: 25 years of image analysis. Nature Methods. 9: 671-675. Selonen V., I.K. Hanski, and S. Mäkeläinen (2012). Predictors of long-distance dispersal in the Siberian flying squirrel. Evolutionary Ecology. 26: 1361-1369. Sih A. (1982). Foraging strategies and the avoidance of predation by an aquatic insect, Notonecta hoffmanni. Ecology. 63: 786-796.

119

Spear L.B., P. Pyle, and N. Nur (1998). Natal dispersal in the western gull: proximal factors and fitness consequences. Journal of Animal Ecology. 67: 165-179. Stamps J.A., V.V. Krishnan, and N.H. Willits (2009). How different types of natal experience affect habitat preference. The American Naturalist. 174: 623-630. Stevens V.M., S. Whitmee, J.F. Le Galliard, J. Clobert, K. Bohning-Gaese, D. Bonte, M. Brandle, D. Matthias Dehling, C. Hof, A. Trochet, and M. Baguette (2014). A comparative analysis of dispersal syndromes in terrestrial and semi-terrestrial animals. Ecology Letters. 17: 1039-1052. Van Allen B.G., and P. Bhavsar (2014). Natal habitat effects drive density-dependent scaling of dispersal decisions. Oikos. 123: 699-704. Van Allen B.G., and V.H.W. Rudolf (2013). Ghosts of habitats past: environmental carry-over effects drive population dynamics in novel habitat. The American Naturalist. 181: 596- 608. Vuilleumier S., and H.P. Possingham (2006). Does colonization asymmetry matter in metapopulations? Proceedings of the Royal Society B: Biological Sciences. 273: 1637- 1642. Wahlstrōm L.K., and O. Liberg (1995). Patterns of dispersal and seasonal migration in roe deer (Capreolus capreolus). Journal of Zoology. 235: 455-467. Wiklund C., and A. Kaitala (1995). Sexual selection for large male size in a polyandrous butterfly: the effect of body size on male versus female reproductive success in Pieris napi. Behavioral Ecology. 6: 6-13.

120

Chapter 6 The Effects of Parasites on Host Dispersal

Part 1 of this chapter was conducted in collaboration with Shannon McCauley and Salma Diab. SM and I conceived the ideas and designed the methodology. SD collected the data. I conducted the statistical analysis and wrote the first draft of the article. All authors contributed substantially to revisions.

Part 2 of this chapter was conducted in collaboration with Shannon McCauley. SM and I conceived the ideas, and designed the methodology. Both authors collected the data. I conducted the statistical analysis and wrote the first draft of the article. Both authors contributed substantially to revisions.

6.1 Part 1: Parasitism Risk Increases Host Dispersal Propensity 6.1.1 Abstract

Models of host-parasite interactions generally assume that host dispersal rate is independent of parasite abundance. However, dispersing away from patches with high parasite density may commonly be used as an effective way to avoid infection. Organisms should be incentivized to disperse away from parasites when parasites impose a fitness cost that is variable in space, such that dispersing hosts are able to move from patches with high parasite density to patches with low parasite density. In this study, we conducted a mesocosm experiment to test whether an insect (the backswimmer Notonecta undulata) increases dispersal propensity when exposed to two different types of cues of ectoparasitic Hydrachnidia mites: direct (visual, olfactory, and/or tactile) cues indicating the presence of parasites in the environment, and cues from parasitized conspecifics. We found that direct cues of parasite presence increased host dispersal rates, but cues from parasitized conspecifics did not. These results indicate that host movement rates are influenced by parasitism risk, which violates the assumption of independent host dispersal made in most theoretical models of host-parasite interactions. Parasite-induced dispersal will influence community dynamics including metacommunity viability, and the evolution of virulence and parasite resistance genes.

121

6.1.2 Introduction

Organisms employ a variety of defence mechanisms to minimize the fitness impacts of parasitism (reviewed in Hart 1990, Carton et al. 2005, Kutzer and Armitage 2016). Some of these mechanisms, including removal of parasites through grooming, immunological resistance, and tolerance, have been well-studied. However, the use of dispersal as a strategy to avoid parasites has received less attention. Dispersal in many organisms is a plastic behaviour (or suite of behaviours; Clobert et al. 2009) that can be deployed depending on the ratio of associated costs and benefits (Southwood 1977, Bonte et al. 2014). Costs of dispersal include heightened mortality risk and energy expenditure while dispersing, and reduced time available for activities such as foraging or mate-searching (Bonte et al. 2012). There are several potential benefits of dispersal (reviewed in Clobert et al. 2001), but here we focus on the opportunity to access habitats with lower parasitism risk. High parasitism risk should trigger dispersal if parasites impose a fitness cost that is larger than the costs of dispersal, and there is spatial variation in parasite abundance, providing the opportunity for dispersers to locate refuges with lower parasite abundance (Sloggett and Weisser 2002).

Most theoretical models of host-parasite interactions assume that host dispersal rate is independent of parasite abundance (Hassell et al. 1991, Comins and Hassell 1996, White et al. 1996, Fenton et al. 2002). However, if parasitism risk induces dispersal, this may alter community dynamics in host-parasite assemblages. For example, parasite-induced dispersal will decrease host-parasite encounter rates, resulting in decreased parasite prevalence in metacommunities. Since parasitic infection reduces host survival and reproductive success (Smith 1989, Lehmann 1993, Ebert et al. 2000), decreases in parasite prevalence are expected to lead to larger host population sizes and greater metapopulation viability. Increased host dispersal will also influence coevolution in host-parasite interactions. For example, Gandon et al. (1996) found that the relative dispersal rates of a host and its parasite influences the evolution of virulence and resistance genes. They found that when parasites had higher dispersal rates than hosts, hosts were more likely to be resistant to allopatric parasites than sympatric parasites, and when hosts had higher dispersal rates, they were more likely to be resistant to sympatric parasites (Gandon et al. 1996).

122

Previous research has demonstrated that parasites alter movement rates (Debeffe et al. 2014, Terui et al. 2017). For example, nestling cliff swallows (Petrochelidon pyrrhonota) infested with ecotoparasites are less likely to return to their natal sites to breed (Brown and Brown 1992, Brown et al. 2017). However, most previous studies have tested the effects of parasite infection on movement, and have not tested the hypothesis that parasitism risk influences dispersal behaviour. Infected individuals generally have lower dispersal ability and higher dispersal mortality than healthy individuals (e.g., Goater et al. 1993, Bradley and Altizer 2005), which may constrain movement. Moreover, individuals that are already infected by parasites may not be able to increase their fitness by dispersing away from infested habitat patches. Therefore, the effects of parasitic infection on dispersal are likely to differ from the effects of parasitism risk. However, we still are lacking experimental evidence for the effects of parasitism risk on dispersal.

The cues that organisms use to detect parasites before they are infected are similar to cues that organisms use to detect predators. These may include direct visual and olfactory cues of predator/parasite presence, as well as alarm chemicals from injured conspecifics, and chemical by-products of consumed conspecifics (Petranka et al. 1987, Schoeppner and Relyea 2009, Hettyey et al. 2012). For example, Bufo americanus tadpoles avoid chemical cues of free- swimming trematode parasites (Rohr et al. 2009), and Rana catesbeiana tadpoles avoid chemical cues of conspecifics infected with the pathogen Candida humicola (Kiesecker et al. 1999). Tactile cues from parasites can also induce avoidance behaviours. For example, infection by ticks increases grooming activity in impala (Aepyceros melampus; Mooring et al. 1996). These non-consumptive (or non-infective) effects can greatly reduce fitness (Giorgi et al. 2001, Horn and Luong 2018). Increased dispersal rates of hosts in response to parasites may be an important non-consumptive effect on host fitness, because dispersal can be associated with substantial costs (Bonte et al. 2012). Therefore, it is important to understand how cues of parasitism risk influence dispersal rates in hosts.

In this study, we tested the hypothesis that parasitism risk influences the dispersal propensity of healthy (i.e., uninfected) individuals. We tested whether individuals respond to direct (visual, olfactory, and/or tactile) cues of parasites, or cues from infected conspecifics. We conducted a mesocosm experiment in which we exposed the backswimmer, Notonecta undulata, to cues from 123

ectoparasitic Hydrachnidia mites. We manipulated cues indicating parasitism risk in two ways: free-swimming mites were either present or not present in mesocosms, and a backswimmer caged in the mesocosm was either infected or not infected with mites. We predicted that both direct cues of mite presence and cues from infected conspecifics would increase dispersal probability in uninfected individuals.

6.1.3 Methods

6.1.3.1 Study system

The backswimmer, N. undulata, is commonly infected with Hydrachnidia mites. Larval Hydrachnidia are parasitic on insects, while the adults are free-living predators (Di Sabatino et al. 2000). More information on the natural history of N. undulata and Hydrachnidia can be found in section 1.5 ‘Study system’.

6.1.3.2 Experimental design

To test whether parasitism risk influenced dispersal behaviour in uninfected backswimmers, we conducted an aquatic mesocosm experiment in which we exposed backswimmers to two different types of mite cues and measured emigration rates. The first cue type we tested was the presence or absence of free-swimming mites in the water (hereafter “water treatment”). The second cue type was the presence or absence of backswimmers that were infected with mites (hereafter “caged backswimmer treatment”). We crossed these treatments in a fully factorial 2×2 design. Treatments were randomly assigned to mesocosm tanks. There were five replicate mesocosms per treatment.

We set up 20 19-L mesocosm tanks in an outdoor space at KSR. Each tank had a cylindrical fibreglass screening cage (diameter = 10 cm) spanning the height of the tank and extending above the surface of the water. In half of these cages we placed one backswimmer infected with mites. We matched these backswimmers for mite loads to minimize variation in the intensity of the cue. We placed an uninfected backswimmer in the other half of the cages as a control. The cages were covered to prevent the backswimmers inside from dispersing or escaping into the mesocosm.

124

We crossed the caged backswimmer treatment with a treatment manipulating the presence of mites and mite kairomones in the water. We filled mesocosm tanks with water from six ponds. These ponds are part of an experimental pond array that was built in 2014. The ponds are uniform in size, depth, temperature, nutrient availability, and community composition (including the density of predators that eat backswimmers; Searcy et al. 2018; C. Searcy, B. Gilbert, M. Krkosek, L. Rowe, and S. McCauley, unpublished data), but differ in the incidence of mite infection. We measured infection incidence by taking a sample of backswimmers (~50 individuals) from each pond, and recording whether they were infected with mites. We chose three ponds that were free of mites (ponds in which 0% of backswimmers sampled were infected with mites), and three with high mite infection rates (>30% of backswimmers infected within each pond). Half of the mesocosm tanks were filled with the water collected from the three mite- free ponds. The other ten mesocosm tanks were filled with the water collected from the three ponds with high mite density. The water from the three mite ponds was mixed together before being added to the mesocosms, to reduce variability in the strength of the mite cue, and the same was done with the water from the three mite-free ponds. All water was filtered before being added to the tanks to remove macroinvertebrates, but allow free-swimming mites through. We used water from natural ponds rather than using dechlorinated water and manipulating mite density because infective stage larval mites are small and difficult to isolate. This design also ensured that the density of infective stage larval mites was within the naturally observed range of densities.

On 16 Jul 2018 we collected uninfected N. undulata from three mite-free ponds at KSR. We used only uninfected backswimmers because we wanted to test the effects of parasitism risk, rather than infection, on host dispersal rates. On 17 Jul 2018 we placed 16 uninfected backswimmers in each mesocosm, for a total of 320 individuals. Individuals were assigned to mesocosms randomly. We covered the mesocosms for one day to prevent immediate dispersal in response to the stress of being handled, and to allow backswimmers to acclimate to their environments. We then allowed the backswimmers to disperse for seven days, starting on 18 Jul. On each day, the covers were removed only during the hours 1100 to 1800, to prevent animals (e.g., mice) from disturbing them overnight. Covering them overnight should not have altered backswimmer dispersal behaviour, because they do not fly under low light conditions. At the end of the

125

dispersal period on 24 July, we counted the total number of backswimmers remaining in each mesocosm (including dead individuals). Backswimmers that were no longer present in the mesocosms were assumed to have dispersed. Food for backswimmers (zooplankton and mosquito larvae) were added to the mesocosms every other day.

6.1.3.3 Statistical analysis

To test whether parasitism risk influenced dispersal probability in uninfected backswimmers, we used a generalized linear mixed model with a binomial error structure using the ‘lme4’ package (Bates et al. 2015) in R (R Core Team 2017). We used dispersal status (emigrated or did not emigrate) as the binary response variable. We used caged backswimmer treatment and water treatment as fixed effects and mesocosm tank as a random effect. Individuals that died during the experiment were excluded from the analysis. One of the cages in the mesocosm detached during the experiment, and so this mesocosm was excluded from further analysis. As a result, the contaminated water/uninfected caged backswimmer treatment was represented by only four mesocosms. Significance was evaluated using type II sum of squares. The nonsignificant interaction term (p > 0.05) was removed before evaluating the significance of the main effects. This analysis was conducted in R v 3.4.3 (R Core Team 2017).

6.1.4 Results

Parasitism risk increased dispersal in uninfected backswimmers, but only when the cue was the presence of mites in the water (Table 6.1, Figure 6.1). The probability of backswimmers dispersing out of mesocosms with water contaminated with mites was 25% greater than the probability of backswimmers dispersing out of mesocosms which did not have mites present in the water. The presence of conspecifics infected with mites did not increase dispersal probability, and did not alter the effect of the water treatment on dispersal (Table 6.1, Figure 6.1).

126

Figure 6.1. Proportion of backswimmers that dispersed ± 95% confidence intervals as a function of water treatment (control: water from mite-free ponds; contaminated: water from ponds with mites) and caged backswimmer treatment (uninfected: the backswimmer within the cage was not infected with mites; infected: the backswimmer within the cage was infected with mites). n = 297.

Table 6.1. Results of the GLMM of the effects of the water treatment, the caged backswimmer treatment, and their interaction on backswimmer dispersal. Effect Estimate s.e. χ2 df p water treatment 0.613 0.247 6.27 1 0.012* cage treatment -0.279 0.245 1.30 1 0.254 cage × water -0.194 0.494 0.15 1 0.695

6.1.5 Discussion

Parasitism risk is hypothesized to induce dispersal in hosts when the fitness costs of parasitism are high, and there is variation in parasite density across space (Sloggett and Weisser 2002). In the backswimmer-Hydrachnidia system, mites impose a fitness cost (Lanciani 1982), and

127

spatiotemporal variation in the incidence of mite infection is high (Appendix E: Figure E2). We therefore predicted that backswimmers would increase dispersal probability in the presence of mites. Our results support this prediction: backswimmers were more likely to disperse away from environments with ectoparasitic mites than those without mites. To our knowledge, this is the first evidence demonstrating that uninfected hosts increase dispersal rates (i.e., movement among populations) when exposed to parasitism risk. The probability of dispersing away from an environment with mites was 25% greater than the probability of dispersing away from an environment without mites. This is likely to be a meaningful effect because even small differences in dispersal rate can have large impacts on dynamics such as metacommunity viability, and local adaptation (Gandon et al. 1996, Hess 1996, White et al. 1996).

In our study, backswimmers were more likely to disperse when they were in water from ponds that contained mites, but the presence of infected conspecifics did not influence dispersal propensity in healthy backswimmers. One explanation for the lack of a response to this treatment was that the cue was too weak. The density of infected conspecifics in our experiment was on the low end of the naturally observed range (C. Baines, personal observation). If backswimmers have a threat sensitive response to parasites as they do to predators (McCauley and Rowe 2010), they may not increase dispersal propensity until the density of infected conspecifics is higher than it was in our experiment. We cannot rule out the possibility that the water from mite ponds contained both mites and chemical cues of infected conspecifics, but the lack of a main or interactive effect of the caged backswimmer treatment indicates that these cues are not important alone or in combination with the direct effect of mites in the water.

Previous studies have found that hosts can sense their parasites before infection (Höller et al. 1994, Sloggett and Weisser 2002, Horn and Luong 2018), but few studies have investigated the source of the cues that hosts use to estimate parasitism risk (see Höller et al. (1994) for an exception). Future studies should investigate the specific cues (visual, olfactory, or tactile including attempts of mites to attach to their hosts) that hosts are using to identify parasites. The type of cue used may influence the probability that hosts are able to respond to parasites before infection. For example, if hosts are not able to detect parasites until they are attacked, more hosts are likely to be infected before they are able to disperse than if hosts are able to detect the parasites from a long distance (e.g., through olfactory cues in aquatic environments). 128

The effects of parasites on the spatial distribution of their hosts in the absence of infection are analogous to “non-consumptive” (Preisser et al. 2005) and “remote control” (Orrock et al. 2010) effects in predator-prey interactions that influence metacommunity viability, local adaptation, and trophic cascades (Preisser et al. 2005, Orrock et al. 2010, Horn and Luong 2018). For example, if hosts are able to disperse away from parasites before infection and move to parasite- free habitat patches, global infection prevalence will be lower than if susceptible hosts remain in patches with high parasite density, resulting in changes in host and parasite abundance and distribution. Host dispersal will also influence spatial patterning in the evolution of resistance and virulence genes (Gandon et al. 1996). Finally, parasite-induced dispersal may influence interspecific interactions. For example, if multiple competitor species differ in their susceptibility and/or dispersal response to parasites, this may affect whether they can coexist locally and regionally. Using a theoretical coexistence model, Comins and Hassell (1996) found that coexistence of two competing hosts can be maintained if the two hosts differ substantially in dispersal rate, and dispersal rate trades-off with susceptibility to parasitism. Therefore, host dispersal rates that change in response to parasites will impact host-parasite and competitive interactions.

Many theoretical studies of host-parasite interactions are spatially explicit, and have investigated the effects of both host and parasite dispersal on metacommunity viability, species distributions, and the evolution of virulence (Gandon et al. 1996, Hess 1996, White et al. 1996). However, most previous studies do not consider the effect of parasitism risk on dispersal propensity in the host. This represents a major gap in the literature on host-parasite interactions because differences in the relative dispersal rates of hosts and parasites influence metacommunity dynamics. For example, French and Travis (2001) conducted an experiment on a host-parasite interaction between a beetle (Callosobruchus chinensis) and a parasitoid wasp (Anisopteromalus calandrae). They did not find evidence that the parasite induced dispersal in its beetle host, but they did demonstrate that parasite dispersal was sensitive to the relative densities of the host and parasite (French and Travis 2001). They used these experimental data to inform a theoretical model, which found that “community-dependent dispersal” (dispersal dependent on the densities of other species, in this case, sensitivity of the parasite to host density) slowed down the rate of invasion compared to cases where parasite dispersal was independent of host density (French and

129

Travis 2001). Future studies should incorporate both incentives for hosts to disperse away from parasites, and the potentially negative effects of parasitic infection on dispersal ability (Goater et al. 1993, Bradley and Altizer 2005).

Many aspects of host-parasite interactions have been well-studied, but the effect of parasitism risk on host dispersal has been under-explored. Here, we found evidence that cues from ectoparasitic mites increase dispersal probability in uninfected hosts. Parasite-induced dispersal has the potential to influence community dynamics including metacommunity viability, and the evolution of virulence and resistance traits. Future empirical studies should estimate the effect of parasitism risk on host dispersal in other systems, and relate this to both the cost of parasitism for host fitness and the amount of spatiotemporal variability in parasite density in the landscape. Theoretical studies of host-parasite interactions should incorporate host dispersal strategies that are dependent on parasitism risk, as well as the potentially negative effects of direct parasitism on host dispersal ability (Goater et al. 1993, Bradley and Altizer 2005), to determine whether “community-dependent” dispersal strategies influence the consequences of host-parasite interactions.

6.2 Part 2: Parasite Infection Eliminates Density-Dependent Host Dispersal 6.2.1 Abstract

By causing anatomical damage and draining resources, parasites reduce dispersal capacity, which may prevent their hosts from exhibiting adaptive plasticity in dispersal in response to stressors such as high population density. Dispersal inhibition has been overlooked, but could represent a large, though indirect, cost of parasitism for hosts. Here, we conducted an experiment on ectoparasitic Hydrachnidia mites and an insect host (the backswimmer Notonecta undulata), to test the hypothesis that parasites eliminate density-dependent host dispersal. We experimentally confirmed that mite infection severely reduces backswimmer dispersal ability. We also conducted a large mark-release-recapture study which demonstrated that mites eliminate positive density-dependent host dispersal. Elimination of density dependent dispersal will affect the entire community by altering spatial patterns of abundance, increasing spatial variance in the intensity of intraspecific competition, and reducing gene flow in both the host and parasite.

130

6.2.2 Introduction

Plastic dispersal behavior allows organisms to maximize fitness in spatiotemporally heterogeneous landscapes (Bonte et al. 2014). Organisms emigrate out of patches where they have low fitness prospects due to stressors including high levels of intraspecific competition (Nowicki and Vrabec 2011, Baines et al. 2014), aggression from conspecifics (Odendaal et al. 1989, Trochet et al. 2013), or predation risk (Cronin et al. 2004, McCauley and Rowe 2010). Parasites are expected to have complex effects on dispersal. High within-patch parasite density creates an incentive for dispersal in host organisms (Boulinier et al. 2001) because parasites reduce survival and reproductive success (Smith 1989, Lehmann 1993, Ebert et al. 2000). However, by directly damaging host anatomy or draining resources, parasites can reduce locomotory performance (Goater et al. 1993, Bradley and Altizer 2005), limiting the ability of their hosts to disperse. Therefore, parasites may prevent hosts from displaying adaptive dispersal behavior in response to other environmental stressors (e.g., competition and predation risk).

If parasites limit their host’s ability to disperse away from low quality patches, this would result in fitness costs to hosts above and beyond the known costs of parasites which include reductions in health and survival, diversion of resources for parasite avoidance and resistance (e.g., grooming, immunity), and lowered fecundity (Lehmann 1993). Parasitized hosts would have to bear the additional fitness costs of remaining in otherwise low quality environments, for example, environments with high levels of competition or predation. This effect should scale up to the population and community level, causing increased extinction risk in both hosts and parasites (Brown and Kodric-Brown 1977, Hess 1996, Ebert et al. 2000), and potentially reducing population growth rates and altering interactions with competitors and predators. For example, a theoretical model developed by Comins and Hassell (1996) demonstrated that coexistence in communities with three species (two competing hosts with a shared parasite, or one host with two competing parasites) depends on the dispersal rates of the constituent species.

Theory predicts that dispersal should be positively density-dependent as a mechanism to minimize competition (Travis et al. 1999, Kun and Scheuring 2006). However, empirical studies have observed the full possible spectrum of effects of density on dispersal including positive density dependence, negative density dependence, and density independence. Several hypotheses

131

have been proposed to explain this variation, including the hypothesis that environmental and phenotypic factors interact with density to influence dispersal, resulting in contradictory results across studies (Ims and Hjermann 2001, Matthysen 2012). If parasites are changing the cost- benefit ratio for dispersal (Terui et al. 2017) or limiting the dispersal capacity of their hosts (Goater et al. 1993), then variation in parasite loads could account for some of the variation observed in density-dependent dispersal across studies. Evidence that host and parasite density interact to influence dispersal would help to resolve the question of variability in density- dependent dispersal that is still under contention in the literature (Matthysen 2005).

In this study, we tested the hypothesis that parasites alter the strength of density dependence in the dispersal of their hosts. We have previously documented positive density-dependent dispersal in the host species used in this study, the backswimmer Notonecta undulata (Baines et al. 2014, chapter 4 and published as Baines et al. in press). Here, we assessed how density-dependent dispersal in backswimmers is modified by ectoparasitic Hydrachnidia mites. We conducted both an experiment and a mark-release-recapture (MRR) study monitoring natural backswimmer dispersal behaviour. In the experiment, we examined the impact of mite infection on dispersal ability, while in the MRR study, we tested whether density dependence in dispersal was influenced by mite infection, and parasitism risk. We predicted that parasitism risk would increase dispersal probability in healthy backswimmers, and act synergistically with population density to raise dispersal rates. Because Hydrachnidia mites drain resources and damage host bodies (Di Sabatino et al. 2000), we predicted that mite infection would decrease dispersal capacity, resulting in lower dispersal rates in infected backswimmers. By extension, we predicted that if infected backswimmers are unable to disperse, mite infection would reduce or eliminate positive density dependence in backswimmer dispersal.

6.2.3 Methods

6.2.3.1 Study system

Information on the natural history of N. undulata and Hydrachnidia can be found in section 1.5 ‘Study system’.

132

6.2.3.2 Dispersal ability assay

We conducted a simple assay to test whether mite infection reduces backswimmer dispersal ability. On 17 Jul 2018 we collected 119 N. undulata from three ponds at the Koffler Scientific Reserve (KSR). We recorded whether each individual was infected with mites. We then placed each individual in a cup filled with ~250 mL of dechlorinated well water and left the cups outside from 13:00 to 18:00 to allow backswimmers to fly out of their cups. At 18:00, we recorded whether individuals were still present or had dispersed. Pilot data showed that being held in these conditions (i.e., in a small volume of water) induces the majority of healthy backswimmers to fly within 20 minutes. Therefore, this assay provides a rough measure of dispersal ability; lower dispersal rates in infected individuals relative to healthy controls would indicate reduced dispersal ability.

6.2.3.3 Statistical analysis of dispersal ability assay

To test whether mite infection altered dispersal ability we used a generalized linear mixed model (GLMM) with a binomial error structure using the ‘lme4’ package in R (Bates et al. 2015). We used dispersal status (emigrated or did not emigrate) as the binary response variable. Individual infection status (infected or not) was included as a fixed effect, and source pond was included as a random effect. This analysis was conducted in R v 3.4.3 (R Core Team 2017).

6.2.3.4 Mark-release-recapture study

We conducted a mark-release-recapture study of N. undulata in an array of semi-natural ponds at KSR. The methods of the mark-release-recapture study are described in detail in Chapter 4 (sections 4.3.2 and 4.3.3). Briefly, the pond array consists of 36 artificial ponds (3m × 3m) built in nine sets of four. The ponds were filled in 2014 with water from a large, nearby pond filtered to 30-μm. Backswimmers and Hydrachnidia naturally colonized the pond array. All 36 ponds had established populations of N. undulata for the entire length of our study. This array of ponds was ideal for studying how parasites influence backswimmer dispersal because of their relative uniformity. The ponds are identical in size, depth, and age (Searcy et al. 2018). They also had identical starting conditions (e.g., they received the same inoculum of zooplankton immediately after being filled; Searcy et al. 2018), except that half of the ponds received a pulse of added nutrients once, soon after they were filled. However, by 2015, there was no measurable effect of 133

nutrient pulse treatment on chlorophyll levels, and the nutrient addition treatment was unrelated to mite incidence, backswimmer density, or backswimmer dispersal probability (C. Baines, S. McCauley, and C. Searcy, unpublished data). Therefore, nutrient pulse treatment was not included in further analyses. The ponds are also similar in community composition (S. McCauley and C. Searcy, unpublished data). There are no fish in these ponds and the density of other predators is very low.

We conducted the mark-release-recapture study in 2016 and 2017. For this paper, we use data from 2017 only, the year in which we took reliable estimates of backswimmer population density (Chapter 4). We collected adult backswimmers using a protocol which standardized collection effort across dates and researchers. We captured and recaptured backswimmers using seine nets on alternate weeks from 4 May 2017 to 4 Oct 2017. Every two weeks, we collected N. undulata from eight different ponds. We marked up to 80 backswimmers from each pond per day with a unique ID on the hemelytra using a Sharpie permanent marker. We also recorded whether each backswimmer was infected by mites or not (a binary variable). We counted surplus backswimmers from each of these ponds, in order to estimate population density. We also recorded the location of previously marked individuals in these eight ponds. On alternate weeks, we sampled the remaining 28 ponds using the standard sampling method, and recorded the location of all marked individuals.

We split the eight ponds from which we collected backswimmers into two categories: ponds with low incidence of mite infection, and ponds with high incidence of mite infection (Appendix E; Figure E1). Since mite infection rates (proportion of backswimmer infected with mites) fluctuated across the season, we categorized them based on their maximum mite levels: ponds categorized as having high mite levels had maximum mite infection rates > 45%, and ponds categorized as having low mite levels had maximum mite infection rates < 30%. The average mite infection rates for ponds with low and high mite levels were 3% and 22%, respectively.

In 2017, we marked 1516 N. undulata, and 342 (23%) of these were infected with mites. Individuals with mites were slightly more likely to be recaptured (33% and 39% recapture rate for uninfected and infected individuals, respectively). In total, we recaptured 390 uninfected (117 and 273 from mite-free and mite-infected ponds, respectively) and 134 infected individuals.

134

6.2.3.5 Statistical analysis of mark-release-recapture study

To test whether mite infection altered the relationship between dispersal and population density, we used a GLMM with a binomial error structure using the ‘lme4’ package in R (Bates et al. 2015). Dispersal status (dispersed or did not disperse over the course of the study) was used as the binary response variable. We classified each individual based on its “mite status”: 0 = uninfected individuals in ponds with low mite levels, 1 = uninfected individuals in ponds with high mite levels, 2 = infected individuals in ponds with high mite levels. The number of infected backswimmers in ponds with low mite levels was too small to include in the analysis. Visual inspection of the data indicated non-monotonicity in the relationship between density and dispersal. We therefore used the ‘poly’ function in R to build a model with a 2nd degree polynomial of the population density of the original capture site as a fixed effect. We also included mite status, and the interaction between mite status and backswimmer density as fixed effects. We included the identity of the original capture site as a random effect. We also included the date of original capture as a random effect because individuals marked on different dates were tracked for different amounts of time.

The interaction between mite status and the 2nd degree polynomial of population density was significant. We therefore built additional GLMMs to test the effect of density separately in each mite status category. For each model, we used the 2nd degree polynomial of population density as a fixed effect, and the original capture site and date of original capture as random effects.

Nonsignificant interaction terms (p > 0.05) were removed before evaluating the significance of main effects. This analysis was conducted in R v 3.4.3 (R Core Team 2017).

6.2.4 Results

6.2.4.1 The effect of mite infection on host dispersal ability

Infection from Hydrachnidia mites strongly reduced dispersal probability (log-odds ± s.e.: 2.07 ± 2 0.43; χ 3 = 26.00, p < 0.0001; Figure 6.2).

135

Figure 6.2. The proportion ± 95% CIs of individuals that dispersed as a function of their infection status. Sample sizes are indicated above the bars.

6.2.4.2 The effect of mites on density-dependent host dispersal

The effect of density on dispersal probability depended on whether mites were present in the 2 ponds (density × mite status: χ 2 = 11.20, p = 0.0037; Figure 6.3A,B). In ponds without mites, dispersal probability was a quadratic function of density, with an overall trend toward positive density-dependent dispersal (Table 6.2, Figure 6.3A). Whereas, both uninfected and infected individuals in ponds with mites had density independent dispersal (Table 6.2, Figure 6.3B).

136

Figure 6.3. Probability of dispersal ± 95% confidence intervals as a quadratic function of backswimmer density for backswimmers in each mite status category: A) individuals in ponds with low mite levels, and B) uninfected and infected individuals in ponds with high mite levels. Sample sizes were 117, 139, and 134, respectively, for individuals from ponds with low mite incidence, uninfected individuals from ponds with high mite incidence, and infected individuals from ponds with high mite incidence.

Table 6.2. Summary table of the results of the GLMMs testing whether dispersal is a quadratic and linear function of backswimmer population density in the three mite status categories. Mite status Density effect Estimate s.e. χ2 df p ponds with low mite levels quadratic 15.60 7.86 4.13 1 0.042* linear 21.58 5.83 14.11 1 0.0002** uninfected individuals in quadratic -1.53 7.20 0.045 1 0.83 ponds with high mite levels linear -5.17 6.80 0.53 1 0.47 infected individuals in ponds quadratic -13.40 10.13 2.53 1 0.11 with high mite levels linear -19.43 10.15 2.38 1 0.12

137

6.2.5 Discussion

We found strong evidence for both our predictions. First, we found that mite infection strongly reduces backswimmer dispersal ability. We did not manipulate parasite load in the dispersal ability assay, so differences in dispersal ability could be due to individual traits that cause increased vulnerability to parasitism and decreased dispersal ability. However, we found no evidence for phenotypic differences between unparasitized and parasitized individuals (no differences in sex ratio or body size; Appendix E: Section 2), so this is unlikely to account for the large differences in dispersal ability. The reduction of dispersal ability in infected individuals is consistent with previous studies demonstrating that mite infection reduces locomotory performance (Goater et al. 1993, Bradley and Altizer 2005). Second, we found that infection by mites eliminates the density dependence in dispersal behavior that is exhibited by healthy backswimmers. To our knowledge, this is the first evidence of parasites eliminating density- dependent host dispersal.

Parasite-induced constraints on dispersal will have multiple effects on host populations. If infected individuals are prevented from leaving low quality habitats (i.e., patches with high population density), this will reduce fitness through multiple avenues: i) the direct effect of “trapping” individuals in a low quality patch in which their expectation of survival or reproductive success is low (e.g., Forrester 1995), ii) “trapped” individuals will be unable to spread their offspring among multiple patches to maximize expected fitness in spatiotemporally variable environments (i.e., risk spreading; den Boer 1968), and iii) “trapped” individuals may be forced to accept higher rates of inbreeding and kin competition (Hamilton and May 1977, Gandon 1999). Effects i and ii will be amplified if there is positive temporal autocorrelation in parasite abundance such that the offspring of infected individuals will also be exposed to high parasite densities. Therefore, the negative effect of the parasite on host fitness will go above and beyond the direct effects of parasitic infection on host survival and fecundity. This leads to the prediction that selection on host organisms for traits that increase tolerance or resistance to parasites will be stronger when the parasite decreases the dispersal capacity of their host. Exploring the selective effects of dispersal inhibition by parasites would be an interesting line of future research.

138

The fact that dispersal probability among uninfected individuals in ponds with mites was relative low is surprising since a previous study demonstrated that backswimmers disperse away from patches with high parasite densities (Chapter 6: Part 1). This effect is unlikely to be due to inherent differences between ponds with low and high mite incidence because ponds were uniform with respect to factors that may influence dispersal including pond size, pond depth, water temperature, chlorophyll levels, and community composition. We speculate that some of the individuals in ponds with mites that were classified as uninfected had sub-detectable infections. When mites are large, they displace the hemelytra of their backswimmer hosts and are easily visible. Small mites are visible through the hemelytra because of their bright red colour. However, it is possible that very small mites were not detectable through the hemelytra, and we did not move the hemelytra to search for mites because the hemelytra and wings are delicate and we did not want to risk damage. Small mites, however, may still reduce dispersal probability of their hosts, and would have grown and caused increasing amounts of damage.

Alternatively, uninfected individuals may be choosing to stay in ponds with mites. Backswimmers are most susceptible to mite infection during and immediately after moulting (Smith 1988, Di Sabatino et al. 2000), therefore the risk of parasitism is low after this period. If there are benefits to remaining in a pond with mites, this may counter the benefits of dispersing away from ponds with high levels of competition. For example, mite infection can drastically reduce swimming ability (C. Baines, personal observation) which likely reduces the ability of infected individuals to compete for shared food resources. Therefore, the intensity of competition for food in ponds with mites is likely lower than competition experienced in mite-free ponds with similar population density. Moreover, because infected individuals have reduced swimming ability, they may be more vulnerable to cannibalism, providing an additional (high-quality) food source to uninfected individuals who remain in ponds with mites.

From the point of view of the parasite, reducing the dispersal ability of the host is likely not a positive consequence of infection. Many parasites, including the parasitic mites studied here (Di Sabatino et al. 2000), have very limited dispersal ability and use their hosts as dispersal vectors (Boulinier et al. 2001). In these cases, low host dispersal rates can lead to decreased rates of spread and abundance in the parasite (Lopez et al. 2005), and increased extinction probability (Hess 1996). It has been suggested that selection may act on parasites to manipulate their hosts’ 139

dispersal rate to be optimal for the parasite (Boulinier et al. 2001, Lion et al. 2006). This is an interesting possibility, but our results suggest that this may not be feasible in host-parasite systems in which the parasite reduces host dispersal ability (e.g., Goater et al. 1993, this study). In this case, any selection on dispersal traits (e.g., take-off propensity) would be constrained by the direct negative effects of parasite infection on hosts. However, the consequences of parasites inhibiting dispersal in their hosts will depend on the degree of host specialization, and whether parasites have consistently negative impacts on the dispersal of multiple host species. For example, Hydrachnidia mites cause substantial reductions in Notonecta flight ability because they cluster under the hemelytra, damaging the hemelytra and wings. In other host taxa however, they attach to other parts of the body (e.g., the ventral surface of the thorax or abdomen, the base of the forelegs; Efford 1963, Mitchell 1968), which may cause less severe reductions in flight ability. Therefore, the inhibition of dispersal in Notonecta caused by Hydrachnidia mites may not have strongly negative consequences for Hydrachnidia populations, if they are able to use other insect taxa as dispersal vectors. Future investigations of parasite manipulation of host dispersal should consider both the degree of host specialization of the parasite, and the effect of the parasite on the dispersal ability of their hosts.

This study provides evidence that parasitic infection influences the strength of density dependence in dispersal, which adds to the results of previous studies demonstrating interactive effects between density and other factors including predation risk (Baines et al. 2014, Hammill et al. 2015), and body size (Hanski et al. 1991, Chapter 4). Taken together, these studies provide strong evidence that interactive effects contribute to the variability observed in the strength and direction of density-dependent dispersal across empirical studies. In host-parasite assemblages in which parasitic infection does not reduce the movement capacity of hosts (e.g., Taggart et al. 2018), parasites may still alter density-dependent dispersal by altering the cost-benefit ratio of dispersal across a density gradient. For example, parasite presence may exaggerate the negative effects of density on fitness. If active individuals are more likely to be parasitized (analogous to activity-risk trade-offs in predator-prey interactions), then the presence of parasites may force hosts to decrease foraging rates in already resource limited environments, resulting in synergistic effects between density and parasite presence on dispersal (similar effects between density and

140

predator presence have been found to influence dispersal; Poethke et al. 2010, Baines et al. 2014).

Density-dependent dispersal influences ecological and evolutionary dynamics differently than random dispersal. For example, positive density-dependent dispersal is a mechanism for population regulation, and thereby minimizes variation in density across habitat patches (Denno and Peterson 1995, Ruxton and Rohani 1998). Without emigration, patches may be more prone to extreme population size fluctuations, where patches greatly exceed their carrying capacities and then crash (Denno and Peterson 1995). Moreover, positive density-dependent dispersal increases metapopulation viability by promoting the flow of dispersers from high density to low density patches (Amarasekare 1998). Density-dependent dispersal also influences local adaptation. For example, when patches have low density as a result of a poor match between population traits and the local environment, positive density-dependent dispersal that results in high levels of gene flow from high density into low density patches may impede the rate of local adaptation in low density patches (Rasanen and Hendry 2008). Therefore, parasites that prevent the expression of positive-dependent dispersal will have cascading effects on the ecological and evolutionary dynamics of their hosts.

There is a wealth of information on host-parasite interactions, but the impact of parasites on host dispersal propensity and ability has been understudied. Here, we found that ectoparasitic mites reduce dispersal ability in their insect hosts, and we provide the first example of a parasite limiting the ability of their host to display adaptive dispersal behavior in response to another biotic stressor. This study adds to a growing body of evidence that interactions between density and environmental and phenotypic factors contribute to variation in density-dependent dispersal across studies. Reductions in dispersal rate and elimination of density-dependent dispersal caused by parasitic infection will have a variety of effects on the ecological and evolutionary dynamics of whole communities. Our results generate further questions that should be investigated including i) how do parasitic traits that prevent or promote host dispersal evolve? ii) what are the consequences of parasites limiting host dispersal for the population dynamics of hosts and parasites? and iii) what are the consequences of parasites limiting host dispersal for the evolution of host traits including immunological resistance to parasites?

141

6.3 References

Amarasekare P. (1998). Interactions between local dynamics and dispersal: insights from single species models. Theoretical Population Biology. 53: 44-59. Baines C.B., S.J. McCauley, and L. Rowe (2014). The interactive effects of competition and predation risk on dispersal in an insect. Biology Letters. 10: 20140287. Bates D., M. Maechler, B. Bolker, and S. Walker (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 67: 1-48. Bonte D., A. De Roissart, N. Wybouw, and T. Van Leeuwen (2014). Fitness maximization by dispersal: evidence from an invasion experiment. Ecology. 95: 3104-3111. Bonte D., H. Van Dyck, J.M. Bullock, A. Coulon, M. Delgado, M. Gibbs, V. Lehouck, E. Matthysen, K. Mustin, M. Saastamoinen, N. Schtickzelle, V.M. Stevens, S. Vandewoestijne, M. Baguette, K. Barton, T.G. Benton, A. Chaput-Bardy, J. Clobert, C. Dytham, T. Hovestadt, C.M. Meier, S.C. Palmer, C. Turlure, and J.M. Travis (2012). Costs of dispersal. Biological Reviews. 87: 290-312. Boulinier T., K.D. McCoy, and G. Sorci (2001). Dispersal and parasitism. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 169-179. Bradley C.A., and S. Altizer (2005). Parasites hinder monarch butterfly flight: implications for disease spread in migratory hosts. Ecology Letters. 8: 290-300. Brown C.R., and M.B. Brown (1992). Ectoparasitism as a cause of natal dispersal in cliff swallows. Ecology. 73: 1718-1723. Brown C.R., E.A. Roche, and M.B. Brown (2017). Why come back home? Breeding-site fidelity varies with group size and parasite load in a colonial bird. Animal Behaviour. 132: 167- 180. Brown J.H., and A. Kodric-Brown (1977). Turnover rates in insular biogeography: effect of immigration on extinction. Ecology. 58: 445-449. Carton Y., A.J. Nappi, and M. Poirie (2005). Genetics of anti-parasite resistance in invertebrates. Developmental and Comparative Immunology. 29: 9-32. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols (eds.) (2001). Dispersal. Oxford University Press, Oxford.

142

Clobert J., J.F. Le Galliard, J. Cote, S. Meylan, and M. Massot (2009). Informed dispersal, heterogeneity in animal dispersal syndromes and the dynamics of spatially structured populations. Ecology Letters. 12: 197-209. Comins H.N., and M.P. Hassell (1996). Persistence of multispecies host-parasitoid interactions in spatially distributed models with local dispersal. Journal of Theoretical Biology. 183: 19-28. Cronin J.T., K.J. Haynes, and F. Dillemuth (2004). Spider effects on planthopper mortality, dispersal, and spatial population dynamics. Ecology. 85: 2134-2143. Debeffe L., N. Morellet, H. Verheyden-Tixier, H. Hoste, J.-M. Gaillard, B. Cargnelutti, D. Picot, J. Sevila, and A.J.M. Hewison (2014). Parasite abundance contributes to condition- dependent dispersal in a wild population of large herbivore. Oikos. 123: 1121-1125. den Boer P.J. (1968). Spreading of risk and stabilization of animal numbers. Acta Biotheoretica. 18: 165-194. Denno R.F., and M.A. Peterson (1995). Density-dependent dispersal and its consequences for population dynamics. In: Population Dynamics: New Approaches and Synthesis (eds. Cappuccino N., and P.W. Price). Academic Press, Inc. San Diego, California, pp. 113- 130. Di Sabatino A., R. Gerecke, and P. Martin (2000). The biology and ecology of lotic water mites (Hydrachnidia). Freshwater Biology. 44: 47-62. Ebert D., M. Lipsitch, and K.L. Mangin (2000). The effect of parasites on host population density and extinction: experimental epidemiology with Daphnia and six microparasites. The American Naturalist. 156: 459-477. Efford I.E. (1963). The parasitic ecology of some watermites. Journal of Animal Ecology. 32: 141-156. Fenton A., J.P. Fairbairn, R. Norman, and P.J. Hudson (2002). Parasite transmission: reconciling theory and reality. Journal of Animal Ecology. 71: 893-905. Forrester G.E. (1995). Strong density-dependent survival and recruitment regulate the abundance of a coral reef fish. Oecologia. 103: 275-282. French D.R., and J.M.J. Travis (2001). Density-dependent dispersal in host-parasitoid assemblages. Oikos. 95: 125-135.

143

Gandon S. (1999). Kin competition, the cost of inbreeding and the evolution of dispersal. Journal of Theoretical Biology. 200: 345-364. Gandon S., Y. Capowiez, Y. Dubois, Y. Michalakis, and I. Olivieri (1996). Local adaptation and gene-for-gene coevolution in a metapopulation model. Proceedings of the Royal Society B: Biological Sciences. 263: 1003-1009. Giorgi M.S., R. Arlettaz, P. Christe, and P. Vogel (2001). The energetic grooming costs imposed by a parasitic mite (Spinturnix myoti) upon its bat host (Myotis myotis). Proceedings of the Royal Society B: Biological Sciences. 268: 2071-2075. Goater C.P., R.D. Semlitsch, and M.V. Bernasconi (1993). Effects of body size and parasite infection on the locomotory performance of juvenile toads, Bufo bufo. Oikos. 66: 129- 136. Hamilton W.D., and R.M. May (1977). Dispersal in stable habitats. Nature. 269: 578-581. Hammill E., R.G. Fitzjohn, and D.S. Srivastava (2015). Conspecific density modulates the effect of predation on dispersal rates. Oecologia. 178: 1149-1158. Hanski I., A. Peltonen, and L. Kaski (1991). Natal dispersal and social dominance in the common shrew Sorex araneus. Oikos. 62: 48-58. Hart B.L. (1990). Behavioral adaptations to pathogens and parasites: five strategies. Neuroscience and Biobehavioral Reviews. 14: 273-294. Hassell M.P., R.M. May, S.W. Pacala, and P.L. Chesson (1991). The persistence of host- parasitoid associations in patchy environments. I. A general criterion. The American Naturalist. 138: 568-583. Hess G. (1996). Disease in metapopulation models: implications for conservation. Ecology. 77: 1617-1632. Hettyey A., F. Rӧlli, N. Thürlimann, A.-C. Zürcher, and J. Van Buskirk (2012). Visual cues contribute to predator detection in anuran larvae. Biological Journal of the Linnean Society. 106: 820-827. Höller C., S.G. Micha, S. Schulz, W. Francke, and J.A. Pickett (1994). Enemy-induced dispersal in a parasitic wasp. Experientia. 50: 182-185. Horn C.J., and L.T. Luong (2018). Proximity to parasites reduces host fitness independent of infection in a Drosophila-Macrocheles system. Parasitology. 145: 1564-1569.

144

Ims R.A., and D.O. Hjermann (2001). Condition-dependent dispersal. In: Dispersal (eds. Clobert J., E. Danchin, A.A. Dhondt, and J.D. Nichols). Oxford University Press Oxford, pp. 203-216. Kiesecker J.M., D.K. Skelly, K.H. Beard, and E. Preisser (1999). Behavioral reduction of infection risk. Proceedings of the National Academy of Sciences. 96: 9165-9168. Kun Á., and I. Scheuring (2006). The evolution of density-dependent dispersal in a noisy spatial population model. Oikos. 115: 308-320. Kutzer M.A., and S.A. Armitage (2016). Maximising fitness in the face of parasites: a review of host tolerance. Zoology. 119: 281-289. Lanciani C.A. (1982). Parasite-mediated reductions in the survival and reproduction of the backswimmer Buenoa scimitra (Hemiptera: Notonectidae). Parasitology. 85: 593-603. Lehmann T. (1993). Ectoparasites: direct impact on host fitness. Parasitology Today. 9: 8-13. Lion S., M. van Baalen, and W.G. Wilson (2006). The evolution of parasite manipulation of host dispersal. Proceedings of the Royal Society B: Biological Sciences. 273: 1063-1071. Lopez J.E., L.P. Gallinot, and M.J. Wade (2005). Spread of parasites in metapopulations: an experimental study of the effects of host migration rate and local host population size. Parasitology. 130: 323-332. Matthysen E. (2005). Density-dependent dispersal in birds and mammals. Ecography. 28: 403- 416. Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J., M. Baguette, T.G. Benton, and J.M. Bullock). Oxford University Press Oxford, pp. 3-18. McCauley S.J., and L. Rowe (2010). Notonecta exhibit threat-sensitive, predator-induced dispersal. Biology Letters. 6: 449-452. Mitchell R. (1968). Site selection by larval water mites parasitic on the damselfly Coercion hieroglyphicum Brauer. Ecology. 49: 40-47. Mooring M. (1996). Grooming in impala: Role of oral grooming in removal of ticks and effects of ticks in increasing grooming rate. Physiology & Behavior. 59: 965-971. Nowicki P., and V. Vrabec (2011). Evidence for positive density-dependent emigration in butterfly metapopulations. Oecologia. 167: 657-665.

145

Odendaal F.J., P. Turchin, and F.R. Stermitz (1989). Influence of host-plant density and male harassment on the distribution of female Euphydryas anicia (Nymphalidae). Oecologia. 78: 283-288. Orrock J.L., L.M. Dill, A. Sih, J.H. Grabowski, S.D. Peacor, B.L. Peckarsky, E.L. Preisser, J.R. Vonesh, and E.E. Werner (2010). Predator effects in predator-free space: the remote effects of predators on prey. The Open Ecology Journal. 3: 22-30. Petranka J.W., L.B. Kats, and A. Sih (1987). Predator-prey interactions among fish and larval amphibians: use of chemical cues to detect predatory fish. Animal Behaviour. 35: 420- 425. Poethke H.J., W.W. Weisser, and T. Hovestadt (2010). Predator-induced dispersal and the evolution of conditional dispersal in correlated environments. The American Naturalist. 175: 577-586. Preisser E.L., D.I. Bolnick, and M.F. Benard (2005). Scared to death? The effects of intimidation and consumption in predator-prey interactions. Ecology. 86: 501-509. Rasanen K., and A.P. Hendry (2008). Disentangling interactions between adaptive divergence and gene flow when ecology drives diversification. Ecology Letters. 11: 624-636. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Rohr J.R., A. Swan, T.R. Raffel, and P.J. Hudson (2009). Parasites, info-disruption, and the ecology of fear. Oecologia. 159: 447-454. Ruxton G.D., and P. Rohani (1998). Fitness-dependent dispersal in metapopulations and its consequences for persistence and synchrony. Journal of Animal Ecology. 67: 530-539. Schoeppner N.M., and R.A. Relyea (2009). Interpreting the smells of predation: how alarm cues and kairomones induce different prey defences. Functional Ecology. 23: 1114-1121. Searcy C.A., B. Gilbert, M. Krkosek, L. Rowe, and S.J. McCauley (2018). Positive correlation between dispersal and body size in green frogs (Rana clamitans) naturally colonizing an experimental landscape. Canadian Journal of Zoology. 96: 1378-1384. Sloggett J.J., and W.W. Weisser (2002). Parasitoids induce production of the dispersal morph of the pea aphid, Acyrthosiphon pisum. Oikos. 98: 323-333. Smith B.P. (1988). Host-parasite interactions and impact of larval water mites on insects. Annual Review of Entomology. 33: 487-507.

146

Smith B.P. (1989). Impact of parasitism by larval Limnochares aquatica (Acari: Hydrachnidia; Limnocharidae) on juvenile Gerris comatus, Gerris alacris, and Gerris buenoi (Insecta: Hemiptera; Gerridae). Canadian Journal of Zoology. 67: 2238-2243. Southwood T.R.E. (1977). Habitat, the templet for ecological strategies? Journal of Animal Ecology. 46: 336-365. Taggart P.L., S.T. Leu, O. Spiegel, S.S. Godfrey, A. Sih, and C.M. Bull (2018). Endure your parasites: sleepy lizard movement is not affected by their ectoparasites. Canadian Journal of Zoology: doi: 10.1139/cjz-2017-0352. Terui A., K. Ooue, H. Urabe, and F. Nakamura (2017). Parasite infection induces size-dependent host dispersal: consequences for parasite persistence. Proceedings of the Royal Society B: Biological Sciences. 284: 20171491. Travis J.M.J., D.J. Murrell, and C. Dytham (1999). The evolution of density-dependent dispersal. Proceedings of the Royal Society B: Biological Sciences. 266: 1837-1842. Trochet A., D. Legrand, N. Larranaga, S. Ducatez, O. Calvez, J. Cote, J. Clobert, and M. Baguette (2013). Population sex ratio and dispersal in experimental, two-patch metapopulations of butterflies. Journal of Animal Ecology. 82: 946-955. White A., M. Begon, and R.G. Bowers (1996). Host-pathogen systems in a spatially patchy environment. Proceedings of the Royal Society B: Biological Sciences. 263: 325-332.

147

Chapter 7 General Conclusion

7.1 Concluding remarks

Dispersal is a complex behaviour that is influenced by many aspects of the environment and individual phenotype (Matthysen 2012). However, studies have varied widely in their estimation of the magnitude and even direction of the relationships between dispersal and factors including population density, sex ratio, body size, body condition, and sex (Bowler and Benton 2005). The multiple, interacting causes of dispersal hypothesis proposes that interactions between factors (either environmental or phenotypic) produce the observed variability in dispersal patterns. The main goal of this thesis was to test this hypothesis by determining how environmental and phenotypic factors interact in their effects on dispersal, and the extent to which they account for the observed variation. I accomplished this goal using three approaches: 1) quantitative review, 2) individual based modeling, and 3) empirical investigations.

In chapter 2, I conducted a meta-analysis to estimate the effect of body size and body condition on dispersal. I found that large individuals disperse longer distances, but there was no consistent effect of body size or condition on emigration or immigration behaviour. Within each dispersal stage I observed high levels of heterogeneity across studies. I tested whether this heterogeneity could be explained by taxonomic group or spatial scale, but found no support for these predictions. I also reviewed the evidence for the “multiple interacting causes of dispersal” hypothesis, and found preliminary support for the prediction that interactive effects alter dispersal patterns. In chapter 3, I developed an individual based model of the evolution of dispersal in response to the joint effects of body condition and population density. I found that dispersal evolved to be a positive threshold function of density, and the value of this threshold depended on condition. Increasing the cost of dispersal did not change the evolved dispersal strategy, but did decrease emigration rates and dispersal mortality. Interestingly, realized dispersal patterns did not match evolved dispersal strategies, because density constrains the distribution of body condition values within patches. In chapter 4, I conducted a mark-release- recapture study of the backswimmer, Notonecta undulata. I found that small males had high dispersal probability, but only when they were from high density patches. Dispersal probability

148

in females decreased with body mass and increased with density, but these effects were additive. Moreover, individuals from high density populations dispersed shorter distances than individuals from low density populations. In chapter 5, I conducted an experiment to investigate the effects of temporal variability in habitat quality on dispersal. I found that backswimmers that developed in low quality patches had lower dispersal probability as adults than those that developed in high quality patches, but the quality of the habitat at the adult stage had no effect on dispersal. Finally, in chapter 6, I performed a series of studies investigating the effects of ectoparasitic mites on the dispersal behaviour of backswimmers. I found that risk of parasitism induces dispersal in healthy, uninfected backswimmers. However, backswimmers that are infected by mites have low dispersal ability. This effect impacts dispersal in natural environments: mites eliminate density dependence in backswimmer dispersal, likely because infected backswimmers are unable to disperse away from low quality habitats.

My work provides strong support for the multiple, interacting causes of dispersal hypothesis. In particular, I provide insight into the reasons for the observed variation in body size or body condition-dependent dispersal, and density-dependent dispersal. Variation in size- and condition- dependent dispersal has previously been poorly explained. I have quantified this variation, and investigated its sources (chapter 2). While some variation is produced by differences in the effects of body size/condition on different dispersal stages, there is still substantial heterogeneity in this relationship that cannot be explained. My work suggests that variation in size or condition-dependent dispersal may be due to interactions with density (chapters 3 and 4). Theory predicts that dispersal should be a positive function of density. Previous authors have hypothesized that deviations from this prediction in empirical studies may be due to Allee effects, and/or the benefits of group living. I add an alternative hypothesis to this list: organisms may not have the capacity to disperse in response to high density because of low dispersal ability caused by low habitat quality (chapters 3 and 6). Poor habitat conditions including low food availability and high parasite abundance increase dispersal motivation, yet low food availability and parasitic infection may also reduce dispersal capacity (chapters 5 and 6), resulting in variation in dispersal responses to habitat quality across contexts. The opposing effects of habitat quality on dispersal motivation and dispersal capacity are likely responsible for much of the observed variability in dispersal patterns. My results highlight the importance of considering two

149

facets of dispersal, motivation and capacity, separately for understanding overall dispersal behaviour (chapters 2, 3, 5, and 6).

My results alter our understanding of the consequences of dispersal for ecological and evolutionary dynamics. Differences between patches in dispersal rates influence ecological and evolutionary dynamics including metapopulation persistence (Vuilleumier and Possingham 2006) and metapopulation genetic structure (McDevitt et al. 2013). Density-dependent dispersal in particular influences population regulation. Positive density dependence is hypothesized to move individuals from crowded to underutilized patches, reducing spatial variation in fitness (Travis et al. 1999). My results suggest that density dependence in dispersal can vary across ecological contexts (chapters 3 and 6), but the impact of this variation for dynamics including population regulation and metapopulation persistence are unknown. Variation in density- dependent dispersal will have additional implications when that variation is caused by the presence of another species. I found that positive density-dependent dispersal in the backswimmer N. undulata is eliminated in patches containing ectoparasitic Hydrachnidia mites. Mites therefore disrupt population regulation processes in backswimmers, and increase spatial variation in fitness. Effects of mites on the dispersal of their hosts may also influence the evolution of virulence and resistance genes (Gandon et al. 1996). By “trapping” backswimmers in parasitized populations, parasitic mites reduce gene flow, and impose strong selection on resistance. Although the importance of dispersal for host-parasite coevolution has been recognized (Gandon et al. 1996), there are still many open questions regarding the correlated evolution of dispersal with virulence and resistance traits. My results highlight the feedback loop linking dispersal to host-parasite coevolution and the relative abundance of hosts and parasites.

Disperser phenotypes (including body size, sex, aggressiveness, etc.) will determine their effects on the patches they colonize (Duckworth and Badyaev 2007, Zajitschek et al. 2009, Burgess and Marshall 2011). For example, the body condition of marine bryozoan dispersers influences the growth rates of the populations they colonize (Burgess and Marshall 2011). My work suggests that dispersers are not consistently smaller or larger than non-dispersers, even within a species, but that the phenotypes of dispersers depend on the environmental conditions of the patches from which they emigrate (chapters 2 – 5). How this influences the phenotypic distribution of the disperser pool through space and time, and what effects this will have on ecological and 150

evolutionary dynamics including metapopulation persistence, metapopulation genetic structure, and local adaptation are critical questions that need to be addressed. I conclude that interactive effects are important determinants of dispersal patterns, and that future investigations of the causes and consequences of dispersal should account for the fact that dispersal rates and the phenotypic distribution of dispersers can vary widely across ecological contexts.

7.2 References

Bowler D.E., and T.G. Benton (2005). Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews. 80: 205-225. Burgess S.C., and D.J. Marshall (2011). Are numbers enough? Colonizer phenotype and abundance interact to affect population dynamics. Journal of Animal Ecology. 80: 681- 687. Duckworth R.A., and A.V. Badyaev (2007). Coupling of dispersal and aggression facilitates the rapid range expansion of a passerine bird. Proceedings of the National Academy of Sciences. 104: 15017-15022. Gandon S., Y. Capowiez, Y. Dubois, Y. Michalakis, and I. Olivieri (1996). Local adaptation and gene-for-gene coevolution in a metapopulation model. Proceedings of the Royal Society B: Biological Sciences. 263: 1003-1009. Matthysen E. (2012). Multicausality of dispersal: a review. In: Dispersal Ecology and Evolution (eds. Clobert J, M Baguette, TG Benton, and JM Bullock). Oxford University Press Oxford, pp. 3-18. McDevitt A.D., M.K. Oliver, S.B. Piertney, P.A. Szafrańska, M. Konarzewski, and K. Zub (2013). Individual variation in dispersal associated with phenotype influences fine-scale genetic structure in weasels. Conservation Genetics. 14: 499-509. Travis J.M.J., D.J. Murrell, and C. Dytham (1999). The evolution of density-dependent dispersal. Proceedings of the Royal Society B: Biological Sciences. 266: 1837-1842. Vuilleumier S., and H.P. Possingham (2006). Does colonization asymmetry matter in metapopulations? Proceedings of the Royal Society B: Biological Sciences. 273: 1637- 1642. Zajitschek S.R., F. Zajitschek, and R.C. Brooks (2009). Demographic costs of inbreeding revealed by sex-specific genetic rescue effects. BMC Evolutionary Biology. 9: 289.

151

Appendix A: Supplementary material for chapter 2

Table A1. Formulae for the body condition indices used by studies in the meta-analysis database. Reference Study group BCI formula Tarwater and Dispersal status LR Residuals from OLS regression: body mass ~ age Beissinger 2012 + wing chord + sex + population + year Lowe et al. 2006 Dispersal distance Residuals from OLS regression: log(mass) ~ log(SVL†) Olsson and Shine Dispersal distance Residuals from OLS regression: body mass ~ 2003 SVL Massot and Clobert Disperser-resident Residuals from OLS regression: body mass ~ 2000 comparison SVL Dubois et al. 2010 Disperser-resident Residuals from OLS regression: log(body mass) comparison ~ log(right elytron length*) Coll and Yuval Disperser-resident Lipid content (fat mass per unit body mass) 2004 comparison †SVL = snout-vent length *elytron length is a proxy for body size in this species

Table A2. Results of three-level meta-analytic mixed effects models for each predictor variable tested, for the disperser-resident comparison group. Model predictor χ2 df p dispersal stage 0.58 2 0.75 spatial scale 0.12 1 0.73 phenotype 4.69 3 0.2 sex 1.47 2 0.48 taxon 7.39 4 0.12

152

Table A3. Results of three-level meta-analytic mixed effects models for each predictor variable tested, for the dispersal distance group. Model predictor χ2 df p spatial scale 2.14 1 0.14 phenotype 4.78 3 0.19 sex 0.16 2 0.92 taxon 7.46 4 0.11

Figure A1. Effect size estimates (Hedge’s g) ± 95% confidence intervals as a function of taxonomic group for the disperser-resident comparison dataset. Point size is inversely proportional to the standard error of each category. ‘Studies’ indicates the number of studies represented in each category, and ‘Results’ indicates the number of results in each category.

153

Figure A2. Effect sizes (Fisher’s Z) as a function of A) spatial scale, B) phenotype metric, and C) sex for studies in the dispersal distance dataset. Data are effect size estimates ± 95% confidence intervals derived from three-level meta-analytic regression models. The dashed vertical line indicates an effect size of zero. Abbreviations: BCI = body condition index, DS = dispersal structure size, ‘Studies’ indicates the number of studies represented in each category, and ‘Results’ indicates the number of results in each category. The number of studies in each category does not sum to the total in (B) and (C) because some studies reported effect sizes for multiple categories. Note that the x-axis scale varies across panels. Point size is inversely proportional to the standard error of each category.

154

Figure A3. Effect size estimates (Fisher’s Z) ± 95% confidence intervals as a function of taxonomic group for the dispersal distance dataset. Point size is inversely proportional to the standard error of each category. ‘Studies’ indicates the number of studies represented in each category, and ‘Results’ indicates the number of results in each category.

155

Figure A4. Effect size estimates (Log-odds) ± 95% confidence intervals for each result in the dispersal status LR dataset. The overall effect could not be estimated because of the small number of studies. Point size is inversely proportional to the standard error of each result.

156

Figure A5. Funnel plot of effect sizes in the disperser-resident comparison group. Contour lines indicate significance at the 0.05 and 0.01 levels. Median study-level power = 5.1%.

Figure A6. Funnel plot of effect sizes in the dispersal distance group. Contour lines indicate significance at the 0.05 and 0.01 levels. Median study-level power = 6.5%.

157

References

Baines C.B., and S.J. McCauley (2018). Natal habitat conditions have carryover effects on dispersal capacity and behavior. Ecosphere. 9: e02465. Coll M., and B. Yuval (2004). Larval food plants affect flight and reproduction in an oligophagous insect herbivore. Environmental Entomology. 33: 1471-1476. Dubois G.F., P.J. Le Gouar, Y.R. Delettre, H. Brustel, and P. Vernon (2010). Sex-biased and body condition dependent dispersal capacity in the endangered saproxylic beetle Osmoderma eremita (Coleoptera: Cetoniidae). Journal of Insect Conservation. 14: 679- 687. Loe L.E., A. Mysterud, V. Veiberg, and R. Langvatn (2010). No evidence of juvenile body mass affecting dispersal in male red deer. Journal of Zoology. 280: 84-91. Lowe W.H., G.E. Likens, and B.J. Cosentino (2006). Self-organisation in streams: the relationship between movement behaviour and body condition in a headwater salamander. Freshwater Biology. 51: 2052-2062. Massot M., and J. Clobert (2000). Processes at the origin of similarities in dispersal behaviour among siblings. Journal of Evolutionary Biology. 13: 707-719. Olsson M., and R. Shine (2003). Female-biased natal and breeding dispersal in an alpine lizard, Niceoscincus microlepidotus. Biological Journal of the Linnean Society. 79: 277-283. Remy A., J.F. Le Galliard, G. Gundersen, H. Steen, and H.P. Andreassen (2011). Effects of individual condition and habitat quality on natal dispersal behaviour in a small rodent. Journal of Animal Ecology. 80: 929-937. Tarwater C.E., and S.R. Beissinger (2012). Dispersal polymorphisms from natal phenotype- environment interactions have carry-over effects on lifetime reproductive success of a tropical parrot. Ecology Letters. 15: 1218-1229. van der Jeugd H.P. (2001). Large barnacle goose males can overcome the social costs of natal dispersal. Behavioral Ecology. 12: 275-282.

158

Appendix B: Supplementary material for chapter 3

Figure B1. Boxplots of the change in trait values for αρ (A), βρ (B), αD (C), βD (D), and γ (E) from generation 0 to 100,000 in the simulations. Data from all 20 replicates of the model with c = 0.1 are shown.

159

Figure B2. Individual body condition at maturity (before dispersal) as a function of natal patch density. The dotted line shows the mean body condition; it was generated using LOESS local regression. Data from all 20 replicates of the model with c = 0.1 are shown.

160

Appendix C: Supplementary material for chapter 4

Figure C1. Diagram of the survey schedule of the mark-release recapture study in 2016. Numbers above pond sets indicate their designation (i.e., their location in the array). Surveys were conducted on a two month rotation. Week 1: We collected up to 20 unmarked backswimmers and recorded marked individuals in each light gray pond. Week 3: We recaptured marked backswimmers in the black outlined ponds. Week 5: We collected up to 20 unmarked backswimmers and recorded marked individuals in each dark gray pond. Week 7: We recaptured marked backswimmers in the black outlined ponds. We then repeated the process starting with Week 1.

161

Figure C2. Probability of being recaptured at least once ± 95% confidence intervals. A) Recapture probability was independent of density (GLMM with backswimmer density as a fixed 2 effect: χ 1 = 3.19, p = 0.074). Each panel in (A) presents data from a single pond, with the pond number designation labeled above. Data are from 2017 only, the year in which we collected information about backswimmer density. B) Recapture probability did not differ between the 2 sexes (GLMM with sex as a fixed effect: χ 1 = 0.053, p = 0.82). C) Recapture probability was 2 independent of body width (GLMM with body width and sex as fixed effects: χ 1 (body width) = 2.07, p = 0.15). Data are from 2016 only. D) Recapture probability was a positive function of 2 body mass (GLMM with body mass and sex as fixed effects: χ 1 (body mass) = 13.30, p = 0.0003). Data are from 2017 only. All panels: GLMMs with binomial error structures and logit links were used to test the effect of environment and phenotype on recapture probability. We used the binary variable “recapture status” (recaptured at least once (1), or never recaptured (0)) as the response. We included date and site of original capture as random effects. Fixed effects are as described above for each panel. N(2016) = 1354, N(2017) = 1139.

162

Figure C3. Mean backswimmer body width (mm) ± 95% confidence intervals for females and males. Data from 2016 only. N = 1354.

Figure C4. Body mass (mg) as a function of backswimmer population density and sex. Regression lines show means ± 95% confidence intervals. Data from 2017 only. N = 1123.

163

Figure C5. Diagram showing the spatial arrangement of the ponds for which we estimated backswimmer population density for eight dates in 2017. The shade of the pond corresponds to the backswimmer population density of that pond on that date. Gray lines separate pond sets. Size of ponds and distance between ponds are not to scale. Dates are given as year-month-day.

164

Table C1. Results of Moran’s I test for spatial autocorrelation in backswimmer population density for each date in 2017 on which density was estimated. Positive values indicate that ponds close together have more similar density values than ponds that are far apart. Negative values indicate that ponds that are close together have less similar density values than ponds that are far apart. Values close to zero indicate no spatial autocorrelation in density. * indicates a significant value of Moran’s I (p < 0.05). Dates are given as year-month-day.

Date Moran's I p-value 2017-05-04 0.800 0.019* 2017-05-18 0.006 0.312 2017-06-01 -0.339 0.429 2017-06-16 -0.598 0.271 2017-06-29 -0.557 0.351 2017-07-12 -0.738 0.151 2017-07-27 -0.047 0.813 2017-08-10 -0.153 0.924

165

Table C2. Results of model 1 showing the effects of sex ratio of the original capture site (sex ratio), density of the original capture site (density), sex, body mass (mass), and date of original capture on the probability of dispersal in 2017. s.e. represents the standard error of the estimate. N = 435.

Parameter Estimate s.e. χ2 df p sex ratio 1.24 3.47 0.124 1 0.725 sex ratio × density 0.00848 0.152 0.0039 1 0.950 sex ratio × sex 2.36 5.47 0.180 1 0.671 sex ratio × mass 0.0945 0.331 0.0709 1 0.790 sex ratio × density × sex 0.175 0.260 0.456 1 0.499 sex ratio × density × mass -0.000520 0.00274 0.0004 1 0.985 sex ratio × sex × mass 0.846 0.580 1.834 1 0.176 density × sex × mass -0.00488 0.000958 4.880 1 0.0272* date of original capture 12.914 6 0.0444*

Figure C6. Probability of dispersing from the original capture site ± 95% confidence intervals, as a function of body width and sex. Data are from 2016 only. N = 440.

166

Figure C7. Probability of dispersal ± 95% confidence intervals as a function of backswimmer body mass for males and females that were originally captured in ponds with low population density (<50 individuals per pond), medium density (≥ 50 and < 60 individuals per pond), and high density (≥ 60 individuals per pond). Each point represents a single individual. Density categories were arbitrary, and were designed to equalize sample sizes across groups. Density was measured as the number of individuals in four standardized seines of a pond. Note that individuals either dispersed (1) or did not disperse (0), but points are jittered with respect to the y-axis to improve visibility. nmale = 178, nfemale = 162.

167

Figure C8. Probability of dispersal ± 95% confidence intervals as a function of the population density of the original capture site for males and females with low (females: <60 mg; males: < 54 mg), medium (females: ≥ 60 and < 70 mg; males: ≥ 54 and < 60 mg), and high (females: > 70 mg; males: > 60 mg) body mass. Each point represents a single individual. Mass categories were arbitrary, and were designed to equalize sample sizes across groups. Note that individuals either dispersed (1) or did not disperse (0), but points are jittered with respect to the y-axis to improve visibility. nmale = 178, nfemale = 162.

168

Figure C9. The proportion ± 95% confidence intervals of recaptured individuals that dispersed, as a function of the date each individual was originally captured. Values above bars indicate the number of individuals marked on each date that were recaptured at least once. Dates are given as year-month-day.

169

Table C3. Estimate and standard error (s.e.) of the difference in dispersal probability for individuals that were originally captured on different dates in 2016. Dates are given as year- month-day.

Parameter Estimate s.e. 2016-03-18 – 2016-04-18 0.408 0.750 2016-03-18 – 2016-05-16 0.7937 0.768 2016-03-18 – 2016-06-22 -0.659 0.899 2016-03-18 – 2016-07-18 -0.968 0.622 2016-03-18 – 2016-08-15 -0.449 0.642 2016-03-18 – 2016-09-18 -0.183 0.596 2016-04-18 – 2016-05-16 0.385 0.795 2016-04-18 – 2016-06-22 -1.066 0.728 2016-04-18 – 2016-07-18 -1.37 0.687 2016-04-18 – 2016-08-15 -0.857 0.485 2016-04-18 – 2016-09-18 -0.591 0.702 2016-05-16 – 2016-06-22 -1.451 0.982 2016-05-16 – 2016-07-18 -1.759 0.790 2016-05-16 – 2016-08-15 -1.24 0.815 2016-05-16 – 2016-09-18 -0.977 0.738 2016-06-22 – 2016-07-18 -0.314 0.87 2016-06-22 – 2016-08-15 0.204 0.715 2016-06-22 – 2016-09-18 0.467 0.884 2016-07-18 – 2016-08-15 0.517 0.661 2016-07-18 – 2016-09-18 0.783 0.587 2016-08-15 – 2016-09-18 0.264 0.675

170

Table C4. Estimate and standard error (s.e.) of the difference in dispersal probability for individuals that were originally captured on different dates in 2017. Results are from model 2. Dates are given as year-month-day.

Parameter Estimate s.e. 2017-05-04 – 2017-05-18 -0.597 0.817 2017-05-04 – 2017-06-01 2.093 1.621 2017-05-04 – 2017-06-29 -0.119 0.705 2017-05-04 – 2017-07-12 -0.888 0.536 2017-05-04 – 2017-07-27 -1.542 0.647 2017-05-04 – 2017-08-10 -1.991 1.169 2017-05-18 – 2017-06-01 2.690 1.587 2017-05-18 – 2017-06-29 0.479 0.867 2017-05-18 – 2017-07-12 -0.290 0.786 2017-05-18 – 2017-07-27 -0.945 0.800 2017-05-18 – 2017-08-10 -1.395 1.217 2017-06-01 – 2017-06-29 -2.196 1.715 2017-06-01 – 2017-07-12 -2.964 1.667 2017-06-01 – 2017-07-27 -3.621 1.626 2017-06-01 – 2017-08-10 -4.074 1.844 2017-06-29 – 2017-07-12 -0.769 0.572 2017-06-29 – 2017-07-27 -1.423 0.705 2017-06-29 – 2017-08-10 -1.873 1.183 2017-07-12 – 2017-07-27 -0.651 0.574 2017-07-12 – 2017-08-10 -1.096 1.138 2017-07-27 – 2017-08-10 -0.448 1.181

171

Figure C10. Boxplot of dispersal distance as a function of sex and year. n(2016) = 82, n(2017) = 58.

Table C5. Unstandardized partial regression coefficients, and results of the z-tests from the path analysis of the effects of body mass and density on dispersal distance, and density and sex on body mass in 2017.

Response Predictor Estimate s.e. z p 95% CI dispersal distance body mass 0.93 1.50 0.62 0.54 -2.02, 3.87 dispersal distance density -1.33 0.50 -2.68 0.0073 -2.31, -0.36 body mass density -0.055 0.043 -1.29 0.20 -0.14, 0.029 body mass sex -4.38 2.30 -1.91 0.056 -8.88, 0.12

172

Appendix D: Supplementary material for chapter 5 Statistical analysis

We tested whether natal habitat quality influenced survival and development time using survival analysis. Using the 'coxme' package (Therneau 2018) in R, we fit a mixed effects Cox model using survival and adult emergence as the response variables in separate analyses. For both analyses, we included natal habitat quality, and instar at the beginning of the experiment as fixed effects, and the pond of origin as a random effect.

We tested whether natal habitat quality influenced dry lipid mass using an ANCOVA on the "pre samples. We used dry lipid mass as the response, and total dry body mass, natal habitat quality, sex, and all two-way interactions as predictors. To test for an effect of natal habitat quality on dry protein mass, we repeated the same analysis using dry protein mass as the response.

We tested whether dry lipid mass, and dry protein mass changed while backswimmers were in the adult tanks using separate ANOVAs with dry lipid mass, or dry protein mass as the response variable, and dry body mass, natal habitat quality, sex, date of preservation, and the two way interactions between date of preservation and the other factors as predictors.

We tested whether the quality of the adult environment influenced dry lipid mass, and dry protein mass by conducting linear mixed models (LMMs) on the "post" samples. We used dry lipid mass, or dry protein mass as the response variable, and dry body mass, natal habitat quality, adult habitat quality, sex, and all two-way interactions as fixed predictors. Mesocosm tank was used as a random effect. LMMs were built using the R package lme4 (Bates et al. 2015).

Results

Survival during the juvenile diet manipulation

Survival was much lower in the low quality natal habitat than in the high quality natal habitat 2 (χ 1 =165.2, p = < 0.0001; Figure D1). Survival did not depend on whether backswimmers were 2 collected as third or fourth instars (χ 1 =0.64, p = 0.4254).

173

Figure D1. Proportion of backswimmers surviving +/- 95% CI through time, as a function of the natal habitat quality treatment.

Development rate

Backswimmers from the high quality natal habitat developed faster than those from the low 2 quality natal habitat (χ 1 =162.24, p < 0.0001; Figure D2). Backswimmers collected as fourth 2 instars matured before those collected as third instars (χ 1 =59.29, p < 0.0001).

174

Figure D2. Proportion of backswimmers that are still juveniles +/- 95% CI through time, as a function of the natal habitat quality treatment.

Dry body mass

Backswimmers from the high quality natal treatment were heavier than those from the low quality natal treatment (Table D1). Backswimmers from the low quality natal treatment gained weight during the dispersal experiment (Table D2) while backswimmers from the high quality natal treatment did not (Table D3). Body mass was a positive function of adult habitat quality (Table D4).

Table D1. Analysis of variance table for the effects of natal habitat treatment and sex on dry body mass in the "pre" samples. Effect df(num) df(den) F p natal quality 1 37 47.155 <0.0001 sex 1 37 0.210 0.650 natal quality × sex 1 36 0.0263 0.872

175

Table D2. Analysis of variance table for the effects of date of preservation (pre, mid, or post) and sex on dry body mass of backswimmers from the low quality natal treatment. Effect df(num) df(den) F p sex 1 52 2.184 0.146 preservation date 2 52 11.427 0.0001

Table D3. Analysis of variance table for the effects of date of preservation (pre, mid, or post) and sex on dry body mass of backswimmers from the high quality natal treatment. Effect df(num) df(den) F p sex 1 60 2.323 0.133 preservation date 2 60 0.392 0.678

Table D4. Analysis of variance table for the effect of natal habitat quality, adult habitat quality, and sex on dry body mass of the "post" samples. Effect df χ2 p natal quality 1 2.641 0.104 sex 1 7.890 0.0050 adult quality 2 9.235 0.0099 natal quality × sex 1 0.0591 0.808 natal quality × adult quality 2 3.160 0.206 Sex × adult quality 2 0.244 0.885

176

Body size

Backswimmers from the high quality natal treatment were larger than those from the low quality natal treatment, and females were larger than males (Table D5).

Table D5. Analysis of variance table for the effect of natal habitat quality, and sex on body width (a proxy for structural body size) of the "pre" samples. Effect df(num) df(den) F p natal quality 1 37 9.964 0.0032 sex 1 37 35.946 <0.0001 natal quality × sex 1 36 0.0022 0.963

Wing area

Wing area was strongly related to fresh body mass (Table D6). After controlling for body mass, there was no effect of natal habitat quality on wing area (Table D6).

Table D6. Analysis of variance table for effect of natal habitat quality and fresh mass on wing area in "pre" samples. Effect df(num) df(den) F p natal quality 1 18 1.297 0.270 fresh body mass 1 18 44.004 <0.0001 natal quality × fresh body mass 1 17 3.440 0.0811

Lipid mass

Immediately after the juvenile diet manipulation, backswimmers had very low dry lipid mass, and dry lipid mass did not differ between the high and low quality natal habitat treatments (Table D7, Figure D3). Dry lipid mass increased through time in both natal habitat treatments (Table D8, Figure D3). Dry lipid mass was independent of dry mass at pre and mid, but a positive function of dry mass at post (Table D8, Figure D3). Combining all pre, mid, and post samples, 177

we see an effect of natal treatment and sex: the low quality natal treatment had higher dry lipid mass, and females had higher dry lipid mass (Table D8).

Figure D3. Dry lipid mass (mg) +/- 95% CI as a function of dry body mass (mg) and natal habitat quality for A) "pre" samples (animals that were preserved immediately after the juvenile diet manipulation), B) "mid" samples (animals that were preserved after the acclimation period in the adult tanks, but before the dispersal experiment), and C) "post" samples (animals that were preserved after the dispersal experiment).

178

Table D7. Analysis of variance table for the effects of dry body mass, natal habitat quality, and sex on dry lipid mass for the "pre" samples. Effect df(num) df(den) F p dry body mass 1 36 0.596 0.445 natal quality 1 36 2.573 0.117 sex 1 36 1.551 0.221 dry body mass × natal quality 1 33 0.0226 0.882 dry body mass × sex 1 33 0.377 0.543 natal quality × sex 1 33 0.0383 0.846

Table D8. Analysis of variance table for the effects of dry body mass, natal habitat quality, sex, and date of preservation (pre, mid, or post) on dry lipid mass. Effect df(num) df(den) F p dry body mass 1 114 26.239 <0.0001 natal quality 1 114 7.183 0.0084 sex 1 114 4.820 0.0302 preservation date 2 114 5.130 0.0074 dry body mass × preservation date 2 108 5.537 0.0051 natal quality × preservation date 2 108 0.0909 0.913 Sex × preservation date 2 108 1.819 0.167

The effect of adult habitat quality on dry lipid mass interacted with dry body mass (Table D9, Figure D4). The slope of dry lipid mass on dry body mass was shallower in the low quality adult treatment than in the high and medium adult treatments (Table D9, Figure D4).

179

Figure D4. Dry lipid mass (mg) +/- 95% CI as a function of dry body mass (mg) and adult habitat quality. Data shown are from "post" samples only (individuals preserved after the dispersal experiment).

Table D9. Analysis of variance table for the effects of dry body mass, natal habitat quality, adult habitat quality, and sex on dry lipid mass of the "post" samples. Effect df χ2 p dry body mass 1 17.772 <0.0001 natal quality 1 1.266 0.261 sex 1 3.545 0.0600 adult quality 2 0.0550 0.973 dry body mass × natal quality 1 1.128 0.288 dry body mass × sex 1 11.166 0.0008 dry body mass × adult quality 2 10.163 0.0062 natal quality × sex 1 2.098 0.148 natal quality × adult quality 2 0.412 0.814

Sex × adult quality 2 1.699 0.428

180

Protein mass

Dry protein mass was a function of total dry body mass on each date of preservation (Table D10, Table D11, Table D12, Figure D5). Among "pre" samples, backswimmers from the low quality natal habitat had a steeper slope of dry protein mass on total dry body mass (Table D10, Figure D5). Backswimmers from both natal habitat treatments lost protein mass through time (Table D11, Figure D5). There was no effect of natal or adult habitat quality treatment on dry protein mass among post samples (Table D12, Figure D5, Figure D6).

Figure D5. Dry protein mass (mg) +/- 95% CI as a function of dry body mass (mg) and natal habitat quality for A) "pre" samples (animals that were preserved immediately after the juvenile diet manipulation), B) "mid" samples (animals that were preserved after the acclimation period in the adult tanks, but before the dispersal experiment), and C) "post" samples (animals that were preserved after the dispersal experiment).

181

Table D10. Analysis of variance table for the effects of dry body mass, natal habitat quality, and sex on dry protein mass in the "pre" samples. Effect df(num) df(den) F p dry body mass 1 35 17.862 0.0002 natal quality 1 35 0.242 0.626 sex 1 35 0.299 0.588 dry body mass × natal quality 1 32 6.970 0.0127 dry body mass × sex 1 32 1.836 0.185 natal quality × sex 1 32 0.167 0.686

Table D11. Analysis of variance table for the effects of dry body mass, natal habitat quality, date of preservation (pre, mid, or post), and sex on dry protein mass. Effect df(num) df(den) F p dry body mass 1 113 67.019 <0.0001 natal quality 1 113 0.594 0.443 sex 1 113 0.719 0.398 preservation date 2 113 71.576 <0.0001 dry body mass × preservation date 2 107 5.168 0.0072 natal quality × preservation date 2 107 0.445 0.642 Sex × preservation date 2 107 1.156 0.319

182

Figure D6. Dry protein mass (mg) +/- 95% CI as a function of dry body mass (mg) and adult habitat quality. Data shown are from "post" samples only (individuals preserved after the dispersal experiment).

183

Table D12. Analysis of variance table for the effects of dry body mass, natal habitat quality, adult habitat quality, and sex on dry protein mass in the "post" samples. Effect df χ2 p dry body mass 1 19.599 <0.0001 natal quality 1 1.234 0.267 sex 1 0.0047 0.945 adult quality 2 3.985 0.136 dry body mass × natal quality 1 0.815 0.367 dry body mass × sex 1 0.396 0.529 dry body mass × adult quality 2 4.190 0.123 natal quality × sex 1 0.0019 0.965 natal quality × adult quality 2 0.346 0.841 Sex × adult quality 2 4.275 0.118

184

Dispersal

Figure D7. Dispersal status at the end of the experiment for each individual backswimmer, as a function of fresh body mass and sex. Note that individuals either dispersed (1) or did not disperse (0), but points are jittered with respect to the y axis to improve visibility. Lines and bands represent the probability of dispersal ± 95% confidence intervals as a function of fresh body mass and sex. Regression lines were generated using a generalized linear model with a binomial error structure.

References

Bates, D., M. Maechler, B. Bolker, and S. Walker. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 67: 1-48. Therneau, T. M. 2018. Coxme: Mixed effects Cox models. R package version 2.2-7. https://CRAN.R-project.org/package=coxme

185

Appendix E: Supplementary material for chapter 6 Section 1: Phenology of backswimmer population density and mite infection

Figure E1. Backswimmer density (dashed black lines) and the proportion of backswimmers infected (solid gray lines) in each of the eight ponds from which backswimmers were marked in the mark-release-recapture study, from May through August in 2017. Four ponds were classified as having high mite incidence (mean incidence = 22%, maximum incidence > 45%) and four were classified as having low mite incidence (mean incidence = 3%, maximum incidence < 30%).

186

Section 2: Effect of body size and sex on the probability of backswimmers being infected with Hydrachnidia mites

Methods

I used data from the mark-release-recapture (MRR) dataset to test whether size and sex influence the probability of backswimmers being infected with Hydrachnidia mites. The methods of the MRR study are described in sections 4.3 and 6.2.3. I used data from 2016, the year in which we measured backswimmer body size. I did not use body mass data from 2017 because attached mites may bias the body mass measurement.

Statistical analysis

I restricted the analyses to backswimmers in infected ponds in order to estimate the effect of size and sex on the probability of being infected, when there is a risk of parasitism. I used a general linear mixed model (R package ‘lme4’; Bates et al. 2015) with a binomial error structure. Mite status (infected or not infected) was the response variable. I used width, sex, and their interaction as fixed predictors. I included the identity of the original capture site, and the date of original capture as random effects. The nonsignificant interaction term (p > 0.05) was removed before evaluating the significance of the main effects. This analysis was conducted in R v 3.4.3 (R Core Team 2017).

Results

Neither body size nor sex influenced the probability that backswimmers were infected with mites (Table E1, Figure E2).

187

Table E1. Results of the GLMM of the effects of the body size, sex, and their interaction on the probability of backswimmers being infected with Hydrachnidia mites. Effect Estimate s.e. χ2 df p body width -1.242 1.124 1.205 1 0.272 sex -0.183 0.385 0.222 1 0.637 width × sex -3.990 2.295 3.093 1 0.079

Figure E2. Probability of being infected with mites ± 95% confidence intervals as a function of body width and sex. Data are from 2016 only. Sample sizes for uninfected females, uninfected males, infected females, and infected males were 103, 122, 24, and 29, respectively.

References

Bates D., M. Maechler, B. Bolker, and S. Walker (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 67:1-48. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

188

Copyright Acknowledgements

My publication in Ecosphere is open access and does not require permission to reproduce. I have received permission from Journal of Animal Ecology to reproduce my publication in this thesis.

189