Evolution of Rattlesnake Venom involves Geographically Structured Coevolution and Local Adaptation to Prey

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

By

Matthew Landon Holding

Graduate Program in Evolution, Ecology and Organismal Biology

The Ohio State University

2017

Dissertation Committee:

H. Lisle Gibbs, Advisor

Bryan Carstens

Marymegan Daly

Stuart Ludsin

Copyrighted by

Matthew Landon Holding

2017

Abstract

Predators and prey coevolve to produce some of the most fascinating phenotypic characteristics of animals. However, coevolution is not a simple race toward the most extreme traits. The occurrence, strength, and outcomes of coevolution are hypothesized to be determined by multiple factors; some are environmental and others are intrinsic to the species involved. Although this complexity has been recognized and studied in fast- evolving hosts-parasite systems, testing the key predictions of coevolutionary theory in natural populations of predators and prey has remained a difficult task. I evaluated the effects of two key factors that impact coevolving rattlesnake venom and ground squirrel venom resistance–mechanisms of interaction and population demography–and I provide evidence that the broader composition of the small mammal prey community exerts selection on the venom phenotype as well. Toward this end, I collected Northern Pacific rattlesnake (Crotalus oreganus) venom and California ground squirrel (Otospermophilus beecheyi) blood serum (which contains venom inhibitors) from multiple populations in

California where California ground squirrels have evolved resistance to the venom. I developed an experiment to test for population-level adaptation of venom metalloproteinases in their interaction with ground squirrel venom inhibitors. I demonstrated local adaptation in a snake-prey relationship for the first time, where venom metalloproteinase enzymes have evolved to overcome ground squirrel resistance.

ii

Furthermore, the existence of local adaptation in these biochemical traits suggests that the mechanism of coevolution involving venom is not an escalating arms race as previously thought, but rather a phenotype matching-based interaction involving a molecular lock- and-key mechanism between multiple snake venom proteins and prey inhibitor molecules. The high levels of medically-significant intraspecific venom variation seen in snakes must now be also viewed in terms of a geographic mosaic of molecular coevolution with resistant prey species, and not merely in terms of broad scale variation in the prey species consumed. Yet, the extent that each population of snakes is adapted to local squirrels varied, and two demographic factors, population size and the relative amount of gene flow in each species, are predicted to explain this variation. I generated thousands of DNA-based SNP loci for both rattlesnakes and squirrels using RAD-seq to test the prediction that rattlesnakes, as the locally adapted interacting species, will have higher effective population sizes and less gene flow than ground squirrels, facilitating the adaption of the snakes over squirrels. Using coalescent-based population genetic analysis,

I supported the prediction that the difference in effective population sizes between rattlesnakes and ground squirrels is positively associated with the population-specific signals of local adaptation. This work represents the first analysis of theoretically- predicted impacts of demography on coevolution outside of a host-parasite system, and the first quantitative demonstration of a relationship between effective population size and local adaptation in any coevolving system in nature. Finally, ground squirrels are present at every site I sampled, while the broader mammalian prey community differs substantially. To understand the potential importance of prey community variation in

iii venom divergence among populations, I measured quantitative differences in the venom protein expression profiles of each population. Both population genetic differentiation based on RADseq loci and differences in the prey community combine to predict over

70% of the between-population variation in venom, thus supporting a role for the prey community in driving divergence and suggests that isolation by environment in the heterogeneous landscape of California may drive correlated levels of population genetic differentiation and adaptive divergence in venom composition. Overall, my work will help in understanding how diverse evolutionary and ecological factors influence the coevolutionary process to produce the planet’s diversity of species and their traits.

iv

Dedication

To my parents, Linda and Harold Holding, for supporting my dreams and never stifling

my curiosity. And to my wife, Sloane Henningsen, for unwavering love despite the

distances.

v

Acknowledgments

This work would not have been possible without the assistance of many people to whom I owe a serious debt of gratitude. First, I would like to thank my advisor, Lisle

Gibbs. Lisle supported by efforts to find success in a new study system at every step, taught me to think critically while still thinking big, and has been a model for both professionalism and enthusiasm for discovery in science. I also thank my dissertation committee, Bryan Carstens, Meg Daly, and Stu Ludsin, who have offered valuable advice while challenging me to improve my science and myself. I thank the members of my lab,

Rob Denton, David Salazar, Tony Fries, Sarah Smiley-Walters, and Mike Sovic for being the first filter for my bad ideas and for helping build the good ones. Several other Ohio

State colleagues deserve recognition for their invaluable support and assistance, including

Isaac Ligocki, Eric McCluskey, Jason Macrander, Erin Lindstedt, Jordan Satler, Paul

Blischak, Megan Smith, and Destiny Palik, Andreas Chavez, and Ian Hamilton, and Amy

Kulesza, and Judy Ridgway. My field collection efforts were greatly assisted by Margaret

Earthman, Kayla Hammer, Josh Zajdel, Tony Frazier, Amber Branske, Emily Taylor,

Steve Van Middlesworth, Bree Putman, and Rulon Clark. I am grateful to Jose Diaz, Jim

Biardi, Mark Margres, Margaret Seavy, and Darin Rokyta, who provided assistance for several of the laboratory analyses included herein. Discussions with Scott Nuismer provided key insights about how to test predictions from coevolutionary theory with empirical data. I would like to thank the University of California (UC) Sedgwick vi

Reserve, UC McLaughlin Reserve, Hopland Research Extension Center, San Joaquin

Experimental Range, Wind Wolves Preserve, Big Chico Creek Ecological Reserve, and

Vandenberg Air Force Base for both allowing me to conduct research at their facilities and to do so on a tight budget. Key persons made access to my field sites easy and enjoyable, including Mike Westphal, Howard Hamman, Kate McCurdy, Cathy Koehler,

Kathryn Purcell, Rhys Evans, Ryan Bourque, and Bob Stafford. This work was made possible by a National Science Foundation Graduate Research Fellowship, Ohio State

University Presidential Fellowship, a Theodore Roosevelt Memorial Grant from the

American Museum of Natural History, a Student Research Award from The American

Society of Naturalists, a Grant-in-Aid from the American Society of Mammalogists, an

American Society of Ichthylogists and Herpetologists’ Gaige Award, a Grant-in-aid of research from Sigma Xi, the Herpetologists’ League Jones-Lovich Grant, an Alumni

Grant from Ohio State University, a Graduate Student Research Grant from the Chicago

Herpetological Society, and funding from the California Bureau of Land Management and the Ohio State University.

vii

Vita

2005...... Yorktown High School, Yorktown, IN

2005-2009 ...... B.S. Biology, Ball State University

2009-2011 ...... M.S. Biological Sciences

...... California Polytechnic State University

2010-2014 ...... National Science Foundation Graduate

Research Fellow, California Polytechnic

State University and Ohio State University

2011, 2014-2015 ...... Graduate Teaching Associate, Department

of Evolution, Ecology, and Organismal

Biology, The Ohio State University

2016-2017 ...... Presidential Fellow, Department of

Evolution, Ecology, and Organismal

Biology, The Ohio State University

Publications

Hudson, P., Denton, R.D., Holding, M.L., and Gibbs, H.L. 2016. Repeatability of locomotor endurance in the Smallmouth Salamander (Ambystoma texanum). Herpetological Review 47:583-586

viii

Holding, M.L., Drabeck, D.H., Jansa, S.A., and Gibbs, H.L. 2016. Venom resistance as a model for understanding the molecular basis of coevolutionary adaptations. Integrative and Comparative Biology 56: 1032-1043.

Holding, M.L., Biardi, J.E., Gibbs, H.L. 2016. Coevolution of venom function and prey resistance in a rattlesnake predator and its squirrel prey. Proceedings of the Royal Society B: Biological Sciences 283:28-41.

Saccucci, M., Denton, R.D., Holding, M.L., Gibbs, H.L. 2016. Polyploid unisexual salamanders have higher tissue regeneration rates than diploid sexual relatives. Journal of Zoology 300: 77-81.

Pomento, A.M., Perry, B.W., Denton, R.D., Gibbs, H.L., Holding, M.L. 2016. No safety in the trees: Local and species-level adaptation of an arboreal squirrel to the venom of sympatric rattlesnakes. Toxicon 118:149-155

Holding, M.L., Kern, E.H., Denton, R.D., and Gibbs, H.L. 2016. Fixed prey cue preferences among Dusky Pigmy Rattlesnakes (Sistrurus miliarius barbouri) raised on different long-term diets. Evolutionary Ecology 30:1-7.

Holding, M.L., Denton, R.D., Kulesza, A., and Ridgway, J.S. 2014. Confronting scientific misconceptions by fostering a classroom of scientists in the introductory biology lab. The American Biology Teacher 76: 218-523.

Holding, M.L., Owen, D.A.S., and Taylor, E.N. 2014. Evaluating the thermal effects of translocation in a large-bodied pitviper. Journal of Experimental Zoology A 321: 442-449.

Owen, D.A.S., Carter, E., Holding, M.L., Islam, K., Moore, I.T. 2014. Roads are associated with a blunted stress response in a North American pit viper. General and Comparative Endocrinology 202:87-92.

Holding, M.L., Frazier, J.A., Pollock, N., et al. 2014. Wet- and dry-season steroid hormone profiles and stress reactivity of an insular dwarf snake, the Hog Island Boa (Boa constrictor imperator). Physiological and Biochemical Zoology 87:363- 373

Holding, M.L., Frazier, J.A., Dorr, S.W., Henningsen, S.N., Moore, I.T., and Taylor, E.N. 2014. The physiological and behavioral effects of repeated handling and short- distance translocation on free-ranging Northern Pacific rattlesnakes (Crotalus o. oreganus). Journal of Herpetology 48: 233-239.

Holding, M.L., Frazier, J.A., Taylor, E.N., and Strand, C.R. 2012. Experimentally-altered navigational demands induce changes in the cortical forebrain of free-ranging

ix

Northern Pacific Rattlesnakes (Crotalus o. oreganus). Brain, Behavior, and Evolution 79:144-154.

Fields of Study

Major Field: Evolution, Ecology and Organismal Biology

Minor Field: College and University Teaching

x

Table of Contents

Abstract ...... ii

Acknowledgments...... vi

Vita ...... viii

Publications ...... viii

Fields of Study ...... x

Table of Contents ...... xi

List of Tables ...... xvi

List of Figures ...... xviii

Chapter 1: General Introduction ...... 1

Studying Snake-Prey Coevolution by Measuring Local Adaptation ...... 2

Chapter 2: Coevolution of venom function and venom resistance in a rattlesnake predator and its squirrel prey ...... 8

Abstract ...... 8

Introduction ...... 9

Methods ...... 13

Study Sites and Sample Collection ...... 13 xi

Detecting Geographic Variation in Predator and Prey Performance ...... 15

Testing for Local Adaptation ...... 16

Results ...... 19

Geographic Variation in SVMP Activity and Squirrel Resistance ...... 19

Detecting Local Adaptation ...... 22

Discussion ...... 28

Coevolution of Molecular Traits ...... 28

A Role for Phenotype Matching ...... 29

The Predator is Ahead ...... 32

Environmental Effects on Coevolution ...... 34

Conclusions ...... 35

Chapter 3: Demographic differences predict patterns of local adaptation in a coevolving vertebrate predator and its prey...... 36

Abstract ...... 36

Introduction ...... 37

Methods ...... 44

Populations and Sample Processing ...... 44

Bioinformatic Processing of Sequence Reads...... 45

Genetic Structure of Snake and Squirrel Populations ...... 46

xii

Joint Estimation of m and Ne by Coalescent Simulation ...... 48

Quantifying Local Adaptation ...... 52

Data analysis ...... 55

Results ...... 55

SNP identification in Rattlesnakes and Ground Squirrels ...... 55

Genetic Structure of Rattlesnakes and Ground Squirrels ...... 56

The geographic scale of local adaptation ...... 58

Comparing Demographic Parameters between Species ...... 61

Local Adaptation and Demography ...... 64

Discussion ...... 65

Lineage effects on local adaptation ...... 66

Genetic demography and local adaptation in coevolving systems ...... 68

Conclusion ...... 74

Chapter 4: Assessing biotic and abiotic drivers of divergence in venom composition among populations of the Northern Pacific rattlesnake...... 76

Introduction ...... 77

Methods ...... 83

Sampling ...... 83

Assessing venom composition using HPLC and SDS-PAGE ...... 83

xiii

Differentiation in venom composition between populations ...... 86

Biotic - Prey community variation ...... 88

Abiotic - Environmental variation ...... 89

Genetic- Population differentiation ...... 89

Assessing drivers of population differentiation in venom ...... 90

Results ...... 91

Differentiation in venom composition ...... 91

Drivers of population differentiation ...... 94

Discussion ...... 100

Extent of Geographic Variation in Northern Pacific Rattlesnake Venom ...... 101

Drivers of Variation in Venom Composition ...... 103

Chapter 5: General Conclusions ...... 111

Appendix A: Supplemental Tables ...... 117

A.1: ...... 117

A.2: ...... 120

A.3...... 122

Appendix B: Supplemental Figures ...... 123

B.1: ...... 123

B.2: ...... 125

xiv

B.3...... 126

References ...... 127

xv

List of Tables

Table 1. Geographic variation in the venom and resistance phenotypes of rattlesnakes and ground squirrels according to population...... 20

Table 2. Comparison of twelve models of the change from baseline SVMP activity when in the presence of California ground squirrel blood serum ...... 23

Table 3. Analysis of variance of change in snake venom metalloproteinase activity in the presence of ground squirrel blood serum ...... 24

Table 4. Information theoretic comparison of the three models of population connectivity tested in both the Northern and Southern genetic groups of both Northern Pacific rattlesnakes and California ground squirrels...... 50

Table 5. General linear models results for analysis of local adaptation in the Northern and

Southern genetic groups of rattlesnakes and ground squirrels...... 60

Table 6. Results of PERMANOVA analysis explaining variation in the venom phenotype of Northern Pacific Rattlesnakes from 13 populations in California...... 93

Table 7. Loadings and R2 values for venom peaks with significant correlations with

Principal Components axes 1, 2, or 3 and which have more than 20% of their variance explained by that peak...... 94

xvi

Table 8. Results of Redundancy Analysis of population mean venom variation across 13 rattlesnake populations...... 97

xvii

List of Figures

Figure 1. A predator-prey interaction between the Northern Pacific rattlesnake (Crotalus oreganus) and the California ground squirrel (Otospermophilus beecheyi) ...... 4

Figure 2. Schematic representation of the fully-crossed experiment where ground squirrel serum was tested for its ability to inhibit the venom of sympatric and allopatric rattlesnakes ...... 17

Figure 3. Map of study location and relationships between mean rattlesnake and ground squirrel phenotypes and site elevation...... 22

Figure 4. Local adaptation of rattlesnakes to overcoming ground squirrel resistance...... 26

Figure 5. The difference in venom deviation scores (mean ± 1 S.E.) when sympatric versus allopatric rattlesnakes are paired with ground squirrels from a given population . 27

Figure 6. Decision tree for using demographic information on two putatively coevolving species to make predictions about the effects of each species’ effective population size

(Ne) and migration rate (m) on the direction and magnitude of local adaptation ...... 41

Figure 7. Each of the models of asymmetric gene flow between populations tested in both rattlesnakes and ground squirrels separately in the Northern and Southern genetic groups

...... 49

xviii

Figure 8. Map of twelve sampled locations where samples of northern Pacific rattlesnakes and California ground squirrels were obtained ...... 52

Figure 9. SNAPP trees with posterior probabilities > 0.7 and STRUCTURE plots at K = 2 for rattlesnakes and ground squirrels showing Northern (orange) and Southern (blue) genetic groups ...... 58

Figure 10. Venom adaptation score (performance of sympatric versus allopatric snakes with serum of a given population’s squirrels) of each population ...... 61

Figure 11. Values of Nm estimated for eleven rattlesnake and ground squirrel populations in the Northern and Southern groups ...... 62

Figure 12. Estimates of a) Ne and b) m derived from parameter estimation under an island model in the program FastSimCoal2 for rattlesnakes versus squirrels ...... 63

Figure 13. Scatterplots showing the difference in A) Ne and B) m point estimates at each site for rattlesnake snakes and ground squirrels calculated from FastSimCoal2 versus venom adaptation scores...... 65

Figure 14. Representative HPLC chromatogram showing the 34 venom protein peaks quantified ...... 85

Figure 15. Biplots of first two PCs from robust principal components analysis of venom variation ...... 93

Figure 16. Biplot of first two axes from constrained ordination of ilr-transformed mean venom phenotypes of Northern Pacific rattlesnakes from 13 sites ...... 98

xix

Figure 17. Relationship between pairwise Fst of rattlesnake populations and pairwise distance between the same populations in Euclidean distance in clr-transformed venom phenotype space ...... 99

Figure 18. Relationships between site scores on the first to principal components (PC) of

19 Bioclim variables plus elevation and site scores on the first two non-metric multidimensional scaling axes of prey community variation ...... 100

xx

Chapter 1: General Introduction

Coevolution between natural enemies has been considered a crucial evolutionary force for maintaining the diversity of life (Darwin 1871; Ehrlich & Raven 1964). During antagonistic coevolution, one species evolves to eat, exploit, or parasitize another, and the second species responds in kind with evolved defenses, leading to further evolutionary change through reciprocal selection (Janzen 1980). Yet, coevolution is more complex than dyadic interactions between two species whose phenotypes escalate or oscillate over time. Several characteristics of coevolving species and their environments combine to produce phenotypic diversity across geographic space and maintain range-wide diversity of species over time (Thompson 2005b). For example, variation among locations in the form and intensity of antagonistic coevolutionary selection has produced higher levels of tetrodotoxin resistance in some populations of predatory gartersnakes (Hanifin et al.

2008), variable furanocoumarin resistance in herbivorous webworms (Zangerl et al.

2008), and contributes to allopatric speciation in crossbills that feed on hard-to-crack pine cones (Parchman et al. 2016).

Coevolutionary biology has recently been transformed through the realization that these geographically variable coevolutionary outcomes result because all aspects of the coevolutionary process occur semi-independently among many geographically distinct populations of each species (Thompson 2005b). The same two species may be intensely 1 coevolving in one location, coevolving towards a different outcome elsewhere, or not coevolving at all in still other places due to stronger interactions with the other species present in their local community (e.g. Siepielski & Benkman 2004). The theme of this perspective is that geographical context matters for the outcomes of coevolution, and understanding how various biological factors shape coevolution is now the forefront of this field of evolutionary biology.

The phenotypic variation produced by coevolution often leads to local adaptation

(Lively et al. 2004; Nuismer et al. 2005; Nuismer & Gandon 2008; Thrall et al. 2002;

Vos et al. 2009), where performance during interactions with sympatric populations of enemy species improves when compared to performance in an interaction with an enemy from an allopatric population (Blanquart et al. 2013; Kawecki & Ebert 2004). Local adaptation occurs as species track peaks in the fitness landscape that shift as a function of the other species’ evolution (Blanquart et al. 2012; Forde et al. 2004; Hoeksema & Forde

2008), and thus the direction and magnitude of local adaptation result from the evolutionary potential of local populations to respond to variation in coevolutionary selection.

Studying Snake-Prey Coevolution by Measuring Local Adaptation

Several theoretical studies of coevolving systems have measured the patterns of local adaptation that result from different inputs into the coevolutionary interaction, such as the type of traits involved and the demographic history of the participating species. In doing so, they have provided a set of predictions about local adaptation, that if tested, can

2 elucidate key aspects of the coevolutionary process in natural systems. Two crucial inputs into the coevolutionary process are 1) the mechanism deciding the winner when two species meet (Yoder and Nuismer 2010) and 2) the demographic history of each species, in terms of population size and degree of isolation (Blanquart et al. 2012). If local adaptation is present, arms race dynamics will fail to account for all the functional variation in a system, and instead a phenotype-matching form of interaction must be involved (Nuismer et al. 2007; Ridenhour & Nuismer 2007; Yoder & Nuismer 2010).

Beyond its presence, the magnitude of local adaptation among populations can be used to test the role that population demography plays in local adaptive potential (Blanquart et al.

2013; Gandon 2002; Gandon & Nuismer 2009). However, these factors are difficult to quantify in natural systems, especially in the long-lived vertebrate animals that we hold as hallmark examples of coevolving predators and prey.

My dissertation addresses this challenge by studying the coevolution of venom in

Northern Pacific rattlesnakes (Crotalus oreganus) and venom-resistance in California ground squirrels (Otospermophilus beecheyi; Fig. 1) across the region where the snakes hunt the squirrels in California. Rattlesnake venom is an extreme example of how natural selection on coevolving species can maintain adaptive variation within a species, as venom can change from highly neurotoxic to mostly hemotoxic (disrupting blood chemistry) among snake populations separated by short distances in the order of tens of kilometers (Rokyta et al. 2015). Venom is a complex trait that is nonetheless tractable to study, as an individual snake’s venom can consist of more than 40 types of toxic proteins, yet whose net function can be measured as enzymatic activity.

3

Figure 1. A predator-prey interaction between the Northern Pacific rattlesnake (Crotalus oreganus) and the California ground squirrel (Otospermophilus beecheyi). Photograph by M. Holding.

Venomous snakes bite prey and then let go, allowing prey to die elsewhere.

Therefore, adequate venom resistance can allow prey to survive some encounters.

Ground squirrels evolved resistance in the form of inhibitor proteins in their blood, and often harass the snakes to drive them away (Barbour & Clark 2012). Previous work has yet to jointly study venom and venom resistance among rattlesnake and squirrel populations from a population perspective to take advantage of the system’s tractability for studying the factors impacting predator-prey coevolution. To this end, I have collected venom from rattlesnakes, blood samples from ground squirrels, and DNA 4 samples from both species across multiple populations in California to investigate how geographic and evolutionary context matters for coevolution between this predator and prey.

In my second chapter, I investigated whether the extensive local variation in venom and resistance traits is adaptive, largely to elucidate the mechanisms of molecular interaction between rattlesnake venom and ground squirrel venom inhibitors. Ground squirrels have multiple forms of a protein called inter-α trypsin inhibitor that inactivate toxic metalloprotease venom enzymes. The effectiveness of these inhibitors can be measured in the lab by adding samples of venom from rattlesnake individuals to samples of squirrel blood serum (Biardi et al. 2011b). I tested venom performance of 120 snakes with local and foreign squirrel blood serum inhibitors, generating 1,440 unique combinations of venom and serum to measure the expected outcomes of snakes preying on squirrels in vitro.

Local adaptation should be only possible when the fitness consequences of predator and prey interacting are decided by a broad class of mechanisms classified as

‘phenotype matching’, which can be likened to a lock and key interaction. If the squirrel has the right inhibitor proteins, they bind to the snake’s venom and inhibit it. This is the mechanism hypothesized for the molecular interaction between venom and venom- inhibitors: inhibitors must match and bind to the venom proteins to provide resistance, whereas venom benefits from a mismatch that allows it to avoid the inhibitors and kill the prey. Phenotype matching differs from the escalating ‘arms races’ (also known as phenotype differences mechanisms) that have been widely suggested to play a role in

5 venom evolution (Vonk et al. 2013). In fact, phenotype matching is more likely to maintain trait diversity among populations through oscillating evolutionary cycles than arms race dynamics (Yoder & Nuismer 2010). By testing the matching hypothesis for venom proteins, I gained insights into why venoms are so complex and variable in a single species.

The third chapter of my dissertation uses population-level variation in the magnitude of local adaptation to test the role population genetic demography plays in the local adaptive potential of predator and prey. Every population has a unique demographic history that can influence its ability to adapt to local conditions and keep pace with coevolving enemies. Adaptation can be impeded by random genetic drift, which is stronger in small populations. Populations are also connected by migrants that may introduce new adaptive variation or impede local adaptation by swamping and homogenization of populations (Gandon & Nuismer 2009; Roth et al. 2012).

I used a next-generation genomic sequencing method called restriction site associated DNA (RAD) sequencing to collect data on >1000 genetic loci in 138 rattlesnakes and 127 ground squirrels across the same populations where I assayed venom resistance. I use these high-resolution genetic data to estimate the genetic effective population size of each population and the number of migrants exchanged per generation between populations, and test the prediction that the species with the larger effective population size at a given location should be the locally adapted species.

Venom is a complex trait, and therefore may hold the capacity to respond to multiple forces of natural selection from prey, such as variation in prey community

6 composition, as well as selection from coevolving ground squirrels. My fourth chapter explores whether there is evidence for an association between variation in the composition of the local prey community and between-population differentiation in the venom phenotype.

I measured protein variation in the venom of multiple snakes per population by using high-pressure liquid chromatography to quantify the abundance of venom protein peaks and statistically comparing populations. I showed extensive differentiation in venom across a major phylogeographic boundary in California, as well as population- level differentiation, and then used multivariate statistical techniques to assess the ability of prey community variation, abiotic environmental variation, and genetic data to explain these differences. My results provide evidence for a role of the broader community of local prey in shaping patterns of venom variation, which would therefore operate alongside local adaptation to ground squirrel resistance.

Overall, my dissertation provides a detailed study of how parallel variation among populations of a venomous predator and its resistant prey evolves, and tests of key predictions from coevolutionary theory that have previously been applied only to lab- based model systems or microorganisms. My research highlights the complex nature of the coevolutionary process by demonstrating the importance of the molecular basis of the key traits involved, the recent evolutionary history of the species involved, and the broader community context in which interactions are imbedded to the evolution of biological diversity.

7

Chapter 2: Coevolution of venom function and venom resistance in a rattlesnake predator and its squirrel prey

Note: This chapter has been published as below, and benefitted from contributions of the co-authors:

Holding M.L., Biardi J.E., Gibbs H.L. (2016). Coevolution of venom function and venom resistance in a rattlesnake predator and its squirrel prey. Proceedings of the Royal Society

B: Biological Sciences 283: 28-41.

Abstract

Measuring local adaptation can provide insights into how coevolution occurs between predators and prey. Specifically, theory predicts that local adaptation in functionally-matched traits of predators and prey will not be detected when coevolution is governed by escalating arms races, whereas it will be present when coevolution occurs through an alternate mechanism of phenotype matching. Here, we analyze local adaptation in venom activity and prey resistance across 12 populations of Northern

Pacific rattlesnakes and California ground squirrels, an interaction that has often been described as an arms race. Assays of venom function and squirrel resistance show substantial geographic variation (influenced by site elevation) in both venom

8 metalloproteinase activity and resistance factor effectiveness. We demonstrate local adaptation in the effectiveness of rattlesnake venom to overcoming present squirrel resistance, suggesting that phenotype matching plays a role in the coevolution of these molecular traits. Further, the predator was the locally adapted antagonist in this interaction, arguing that rattlesnakes are evolutionarily ahead of their squirrel prey.

Phenotype matching needs to be considered as an important mechanism influencing coevolution between venomous animals and resistant prey.

Introduction

The concept of an escalating “arms race” has been repeatedly used to account for the evolution of functionally-paired sets of morphological, physiological, and molecular traits between antagonistic species (Benkman et al. 2003; Brodie et al. 2002; Hanifin et al. 2008; Toju 2008). Yet, theoretical work indicates the arms race concept is one of a number of possible mechanisms that can govern coevolutionary dynamics between predators and their prey (Nuismer et al. 2005). Therefore, a key step in understanding how coevolution has shaped phenotypic variation in offensive traits in predators and defensive traits in prey is assessing the importance of each of these different mechanisms of coevolutionary interaction (Geffeney et al. 2002; Gomulkiewicz et al. 2007; Zangerl et al. 2008).

In an escalating arms race, the winner is determined by “phenotype differences” between antagonists (i.e. which player is bigger, faster, stronger, more toxic; Nuismer et al. 2007). Arms race models can explain trait variation in a number of predator-prey

9 systems in nature: garter snakes have evolved higher resistance to deadly tetrodotoxin when they encounter more toxic newt prey (Brodie et al. 2002; Hanifin et al. 2008), and arms races explain coevolution of morphological adaptations in camellia flowers and weevils that eat them (Toju & Sota 2006). Yet direct assessments of the arms race are absent in other systems where an arms race explanation has been widely suggested (for example venomous animals and their prey; Casewell et al. 2012b) and theory suggests that alternate explanations are possible (Nuismer et al. 2005; Nuismer et al. 2007).

The chief alternative to antagonistic arms races is the “phenotype matching” model, where one coevolutionary participant benefits from a match to a second participant’s phenotype, while this second participant benefits from a mismatch (Nuismer et al. 2007). Therefore, the degree of matching, and not the difference between trait values, decides the fitness outcomes when two individuals meet. This definition of phenotype matching focuses on the degree of ‘fit’ between specific offensive and defensive traits (e.g. Nuismer et al. 2007), and should be distinguished from “trait matching” at the population level, where the population mean trait values of two interacting species are positively correlated (e.g. Anderson & Johnson 2008; Hanifin et al. 2008; Toju 2008). Systems where phenotype matching mediates coevolution of traits include egg mimicry and host recognition in avian brood parasites and their hosts (Davies

2000) and molecular recognition systems in pathogens and their hosts (Dybdahl et al.

2014).

Coevolutionary theory offers a clear prediction to test for a role of phenotype matching in predator-prey coevolution: local adaptation will be generated when

10 phenotypic matching influences the winner of a predator-prey interaction whereas it will not be observed when a phenotype differences mechanism drives trait variation (Nuismer et al. 2007; Ridenhour & Nuismer 2007). The basis of this prediction can be illustrated by considering two hypothetical antagonists with positively correlated values of a key trait among populations that coevolve under an arms race (phenotype differences) mechanism. The populations of predator and prey with the highly exaggerated trait values

(e.g. locations with the most toxic venoms and most resistant prey) will always perform better away from home when performance is compared in sympatric and allopatric combinations of predator and prey, whereas predator-prey populations with low trait values will perform poorly away from home. As a result, the average level of local adaptation across all populations will be zero. In contrast, phenotype matching generates populations of predators and prey that occupy different fitness optima, and so local adaptation will be common when comparing sympatric and allopatric pairs of antagonists

(Nuismer et al. 2007; Ridenhour & Nuismer 2007).

In venomous animals that feed on resistant prey, the arms race dynamic has been repeatedly invoked to explain the high levels of variation in offensive traits in predators

(e.g. venom composition) and defensive traits in prey (e.g. biochemical resistance; Biardi

2008; Casewell et al. 2012b), while the role phenotype matching might play in these systems remains to be investigated. Phenotype matching can be identified through an analysis of local adaptation in predator venom and prey defenses across populations where each coexists. Such analyses are tractable for venomous snakes and their prey through the use of well-developed in-vitro tests that assess venom functional activity and

11 prey resistance (Biardi et al. 2011a; Biardi et al. 2011b). These functional data can then be analyzed using the statistical approaches developed for assessing local adaptation in host-parasite systems (Blanquart et al. 2013; Thrall et al. 2002). Because local adaptation measures the net fit of genotype to environment at a metapopulation level, it can also be used as a measure of the net outcome of coevolutionary interactions, thus estimating the relative success of each partner and revealing which species is evolutionarily ahead in the interaction (Blanquart et al. 2013; Roth et al. 2012).

Here, we focus on the molecular traits that mediate interactions between venomous Northern Pacific rattlesnakes (Crotalus o. oreganus) and their main prey

California ground squirrels (Otospermophilus beecheyi). Multiple biological features of this system argue that coevolutionary interactions play a significant role in the evolution of offensive traits in rattlesnakes and defensive traits in squirrels. Where they coexist, O. beecheyi can comprise the majority of the diet of this rattlesnake, and rattlesnakes are the main squirrel predator (Fitch 1949). Diverse anti-snake behaviors of the ground squirrels

(Coss et al. 1993; Putman et al. 2015; Rundus et al. 2007) and behavioral responses by rattlesnakes (Barbour & Clark 2012) suggest reciprocity in matched behavioral adaptations of prey and predator. At the molecular level, ground squirrels show counter- adaptations against envenomation through serum-based resistance to the activity of snake venom metalloproteinases (SVMP), major constituents of pitviper venom that break down connective tissue and facilitate the penetration of other venom components to target tissues (Biardi 2008). Although both resistance factors in squirrels (Biardi 2008) and venom composition in rattlesnakes (Mackessy 2010) vary geographically, the extent to

12 which this variation is adaptive and the type of coevolutionary interaction generating it remain to be determined.

The proteins in ground squirrel serum inhibit SVMPs by targeted binding to venom proteins and interactions between these two types of proteins offers an in vitro system to study the evolutionary mechanism driving interactions between offensive and defensive traits at the molecular level (Biardi et al. 2011a). Inhibition can be measured by exposing venom to the serum of ground squirrels, which contains the inhibitors. We measured local adaptation in this predator-prey system by functional comparisons of venom performance among sympatric and allopatric combinations of rattlesnake venom and ground squirrel serum. Our approach is to assume the coevolution has influenced the evolutionary dynamics of venom and resistance based on the biological features of the system and then to use measures of local adaptation to infer the potential role of phenotype matching as a driver of coevolution of venom and resistance.

Methods

Study Sites and Sample Collection

We collected rattlesnake venom and ground squirrel serum from at least ten adult rattlesnakes and ten adult squirrels at each of 12 California locations (Fig 3A), under

California Department of Fish and Wildlife permit SC-12027 and using approved procedures described in Ohio State University IACUC protocol 2012A00000015.

Rattlesnakes were located by visual search and venom was extracted and frozen. Adult

13 ground squirrels were captured with live traps and a blood sample collected and serum frozen.

Ten venom and ten serum samples from each location were used in our local adaptation tests. Each rattlesnake was randomly paired with twelve ground squirrels; one sympatric squirrel and one squirrel from each of the eleven allopatric sites. Since there were 10 squirrels collected from every population, each of the 10 individual snakes was tested with a unique squirrel from each population, yielding 1,440 unique combinations of snake and squirrel in our dataset. We used the method of Biardi et al. (2011b) to measure SVMP activity and inhibition by O. beecheyi serum. Briefly, metalloproteinase enzymatic activity was measured using the EnzChek Gelatinase Kit (Life Technologies,

Carlsbad, CA, USA). We followed product protocols, except that the gelatin substrate was diluted to a concentration of 1:100 and 0.3125 ng of venom was used in each test.

We measured fluorescence intensity in relative fluorescence units (RFU) every 1.5 minutes using a FLUOstar Omega microplate reader (BMG Labtech, Ortenberg,

Germany). We calculated the slope (RFU/min) from the linear part of the reaction, which we used as our measure of venom SVMP activity. Larger slopes indicate that the venom is degrading the substrate more quickly, and hence that it is more effective.

Serum-to-venom binding scores calculated during previous work in this system were strongly correlated with venom lethal dose measurements in live ground squirrels

(Poran et al. 1987), suggesting a direct link between levels of serum-based inhibition of venom activity and the fitness outcomes of envenomation for both species. Therefore, determining the reduction in SVMP activity when squirrel venom inhibitors are present

14 provides a functional measure of a snake-squirrel interaction at the molecular level with a probable link to the fitness outcomes of a snake-squirrel interaction in nature.

To do this we first calculated baseline SVMP activity for each snake by testing the venom by itself. Next, we calculated the difference in SVMP activity between the baseline measure and the SVMP activity when venom was incubated with O. beecheyi serum for 30 minutes, yielding the amount of SVMP activity lost to an individual squirrel’s serum inhibitors for each snake individual. Serum samples were first diluted to

5 mg/mL and then incubated with individual venom samples, and their activity measured.

More negative values of inhibition indicate a larger loss of rattlesnake venom activity when squirrel serum is present. In a biological sense, SVMPs injected into a squirrel are acting on target tissues and avoiding inhibitors, so more positive values (less reduction from baseline activity) would be advantageous to the snake.

Detecting Geographic Variation in Predator and Prey Performance

We assessed the presence and pattern of geographic variation in the venom and resistance phenotypes using one-way analysis of variance (ANOVA) with population of origin as the factor. Resistance was defined as above and separately as percent reduction from baseline venom activity for interpretability. Because there are only 10 snakes and squirrels per population, and 66 possible pairwise comparisons between population-mean values for the above variables, we present the post-hoc comparisons of population average phenotypes as protected Fisher’s least significant differences.

15

The SVMP activity of C. o. oreganus is hypothesized to be affected by a trade-off between lethal toxicity and general proteolytic activity of venom possibly mediated by temperature or seasonality at a site (Mackessy 2010). Further, O. beecheyi are thought to be less resistant at higher elevations (Towers & Coss 1990). As such, we explored the possibility that elevation influences both SVMP activity and squirrel resistance. We estimated the elevation of a particular site in ArcMap v.10.2 (ESRI) as the centroid of a minimum convex polygon encompassing all points from which individual snakes were captured at that site. We used these elevation values in linear regressions on the average baseline venom SVMP activity and average squirrel resistance value at each site; fully- factorial multiple regression exploring relationships among SVMP activity, resistance, and elevation resulted in all relationships being non-significant due to combined lack of power and collinearity between predictors, so we present these analyses as separate linear models.

Testing for Local Adaptation

Recent theoretical work on the detection of local adaptation has highlighted the importance of assessing organismal performance in allopatry in the context of relevant environmental variation (Blanquart et al. 2013). Due to the significant relationships we found between elevation and both SVMP activity and resistance levels (see Results below), we partitioned our allopatric combinations of venom and serum based on elevation (Blanquart et al. 2013). We used the median site elevation (422 m a.s.l.) to designate six high and six low elevation sites. Specific snake-squirrel pairings were then

16 designated as sympatric, allopatric-same elevation (AS), or allopatric-different elevation

(AD; Figure 2). The AS designation meant the snake-squirrel comparison was between two high elevation sites or two low elevation sites, whereas the AD designation meant the comparison was between a snake and squirrel collected at different elevations (snake low vs squirrel high or snake high vs squirrel low). This “comparison type” term partitions any contribution of local adaptation in SVMP activity into a portion related to local adaptation at high or low elevation, and a portion attributable to other unknown factors

(Adiba et al. 2010; Blanquart et al. 2013).

Figure 2. Schematic representation of the fully-crossed experiment where ground squirrel serum was tested for its ability to inhibit the venom of sympatric and allopatric rattlesnakes. Curved lines represent peaks and a valley forming an elevation gradient. Allopatric tests were categorized as same type (low to low or high to high elevation; n = 660) or different type (high to low or low to high elevation; n = 660).

17

We fit a linear mixed-effects model (LMM) with resistance as the response variable to test for local adaptation (Blanquart et al. 2013; Thrall et al. 2002). Fixed main-effects in the full model included snake elevation type (SnElev), squirrel elevation type (SqElev), snake population of origin nested within SnElev (SnPop), squirrel population of origin nested within SqElev (SqPop), and our elevation-partitioned local adaptation term, comparison type (S–AS–AD). In addition, snake individual nested within SnPop, and squirrel individual nested within SqPop, were included as random factors to account for the fact that each snake and squirrel individual was used in twelve comparisons. Dunn-Sidak adjusted p-values were used to compare the model-estimated marginal mean resistance values between S, AS, and AD combinations. Larger (more positive) values in sympatric comparisons would show local adaptation of rattlesnakes, whereas smaller (more negative) values in sympatry would indicate local adaptation of ground squirrels.

Finally, we visualized the population-level pattern of performance in sympatric and allopatric populations of snakes or squirrels to determine which sites showed patterns consistent with local adaptation by comparing performance in sympatry versus allopatry

(AS and AD) for each population. We took advantage of the fact that the same individual snake or squirrel was used in multiple functional tests, making our study analogous to a paired or repeated measures design and allowing us to derive a “venom deviation score” for each snake-squirrel combination. The venom deviation score was calculated as the amount that a focal squirrel individual caused a particular snake individual to deviate from its average performance against all squirrels. This approach controls for the large

18 main effect differences among individual snakes, and therefore the populations they are nested within, by measuring performance as the deviation from an individual snake’s average performance. Further, since one squirrel population could be analyzed at a time, main effects of squirrel population are also factored out, resulting in a value for an individual snake’s venom deviation score in sympatry (one value), in all AD comparisons

(6 values), and in all AS comparisons (5 values). For each snake individual, we then subtracted the mean venom deviation scores of all AD and AS comparisons from its sympatric value, generating values for the difference in performance in sympatry versus allopatric sites.

Results

Geographic Variation in SVMP Activity and Squirrel Resistance

We detected extensive among-population variation in both the venom activity of snakes and the capacity to inhibit venom in squirrels. The baseline SVMP activity of venoms varied among rattlesnake populations, with three-fold differences between the least and most enzymatically active populations (one-way ANOVA, F11,119 = 5.49, P <

0.001, R2 = 0.36). Venom activity was reduced by 27%, on average, by the ratio of venom to serum used in our experiments. We found variation in resistance in ground squirrels in both an absolute sense, the average RFU/min of SVMP activity a venom lost

2 in the presence of a given serum (F11,119 = 3.08, P < 0.001, R = 0.24), and as a percentage

2 reduction in SVMP activity (F11,119 = 2.7, P = 0.004, R = 0.22). There was three-fold variation in the average amount of SVMP activity lost to particular squirrel populations.

19

On average, squirrels from the mountains of Humboldt County were least resistant, only causing a 10.8 % reduction in venom activity, while squirrels from Vandenberg A.F.B on the coast and the Sutter Buttes in the northern Sacramento River Valley inhibited venoms by nearly 36% (Table 1).

Table 1. Geographic variation in the venom and resistance phenotypes of rattlesnakes and ground squirrels according to population. To calculate population means for retained activity and squirrel resistance, we first obtained an average for each snake or squirrel individual across all trials in which it was used. From these data, the mean retained activity or squirrel resistance of the ten individual snakes or squirrels was obtained for each site, and is presented below. Superscripts are based on Fisher’s comparison method at α = 0.05. Baseline SVMP Retained SVMP Squirrel Resistance Squirrel Resistance Activity Activity (% of (∆ from Baseline (% Activity Population (RFU/min) Baseline) RFU/min) Reduction) Blue Oaks 849.9 bcd 89.4% b -130.4 a 13.7% cd Cantua Creek 968.0 abc 79.5% cd -207.4 abc 22.3% abcd Chico Creek 737.0 cd 61.1% bc -238.1bcde 24.1% abcd Chimineas 1174.9 a 67.7% cd -155.4 ab 15.4% bcd Hopland 358.9 e 133.2% a -267.8 cde 30.0% a Humboldt 829.7 bcd 68.8% c -124.6 a 10.7% d McLaughlin 1023.4 ab 71.9% bc -240.8 bcde 27.4% abc SJER 671.4 d 62.1% cd -223.1abcd 22.4% abcd Sedgwick 780.2 bcd 76.2% bc -260.4 abcd 27.0% abc Sutter Buttes 657.9 d 48.3% d -315.3 de 35.8% a Vandenberg 681.7 d 80.6% bc -329.3 e 36.0% a Wind Wolves 967.5 abc 68.5% c -264.4 cde 27.8% ab

A large amount of the site-to-site variation in venom SVMP activity was predicted by linear regression on the elevation of a site, where population average SVMP

2 activity was higher at higher site elevation (T = 4.83, df = 10, P = 0.001, R = 70.0%, β =

0.81, Figure 3B). The population average resistance values of ground squirrels were also related to elevation, with squirrels from low elevations having serum better at inhibiting

2 venom (T = 7.21, df = 10, P = 0.038, R = 36.2%, β = -0.00022, Figure 3C). Direct linear

20 regression of average squirrel resistance on average venom SVMP activity among population yields a non-significant relationship (P = 0.12) with a negative slope coefficient. While elevation is clearly an important driver of both overall SVMP activity in snakes and average ability to inhibit SVMPs in squirrels, our test for local adaptation

(below) models the change in SVMP activity to test for interactions between snake and squirrel genotypes on venom function (G x G interactions), which applies regardless of baseline venom activity and the apparent environmental gradient. Thus, the population average functional responses we have measured can be correlated with elevation while local adaptation of predator or prey can still exist, especially if different venom or inhibitor isoforms or expression differences lead to the same global average in function.

21

Figure 3. Map of study location and relationships between mean rattlesnake and ground squirrel phenotypes and site elevation.A) Map of California marking the 12 sites where rattlesnake venom and ground squirrel blood serum were collected. Circles indicate a site greater than 400 m in elevation, while triangles indicate a site that is less than 400 m in elevation. Scatterplots show site elevation versus the B) mean baseline snake venom metalloproteinase (SVMP) activity (RFU/min) of ten rattlesnakes collected from each site and C) mean resistance (% inhibition of SVMP activity compared to a snake’s baseline activity) of ten ground squirrels collected from each site. One standard error of the mean is shown for each population.

Detecting Local Adaptation

We used our linear mixed model approach that factored out population and environmental main effects to test for a difference in performance in sympatric vs. allopatric combinations of venom and serum (Thrall et al. 2002). This model was the best fit to the data based on sample sized-corrected Akaike’s Information Criterion (AICc), when compared to a series of other models that treat local adaptation in alternative ways

(e.g. comparisons of sympatry vs. all allopatric comparisons or consideration of geographic distance as a covariate, Table 2). We detected a significant main effect of squirrel elevation (F1,107 = 16.39, P < 0.001) where squirrels originating in low elevation 22 sites (mean change = -199.7 RFU/min) were ~32% more resistant than those from high elevation sites (mean change = -293.3 RFU/min). The elevation type from which snakes originated had an impact on the amount of SVMP activity they lost (low elevation mean

= -228.6, high elevation mean = -264.3, F1,107 = 10.33, P = 0.002). Both snake population and squirrel population of origin had significant main effects on the amount of SVMP activity lost, showing the existence of variation in the outcome of a snake-squirrel interaction even when controlling for the large general effects of site elevation (Table 3).

Table 2. Comparison of twelve models of the change from baseline SVMP activity when in the presence of California ground squirrel blood serum. All models included main effects of snake elevation, squirrel elevation, snake population, squirrel population, and snake and squirrel individual, except for the “only random effects model”. Models are named based on the terms added in addition to main effects and are listed in ascending order by their AICc values.

a b c d e Model K Adaptation P-value AICc ΔAICc w i S-AS-AD 29 < 0.001 17301.6 0 0.88 SnElev x SqElev 28 <0.001 17306.3 4.7 0.08 Same or Diff. Elevation 28 <0.001 17307.7 6.1 0.04 S-AS-AD + Geo Dist. 30 S-AS-AD: <0.001 17322.2 20.6 < 0.001 SnElev*SqElev +Geo Dist. 29 ExE: < 0.001 17327.2 25.6 < 0.001 Sympatric vs. Allopatric 28 0.1 17330.6 29 < 0.001 Main Effects Only 27 n/a 17339.2 37.6 < 0.001 S-AS-AD*Geo Dist 31 S-AS-AD: 0.122 17339.3 37.7 < 0.001 Sym/Allo + Geo Dist 29 Sym/Allo: 0.061 17350.4 48.8 < 0.001 Geo Dist 28 0.947 17360.2 58.6 < 0.001 SnElev*SqElev*Geo Dist. 32 SnElev*SqElev: 0.064 17380.1 78.5 < 0.001 Only Random Effects 3 n/a 19790.4 2488.8 < 0.001 a All models included intercepts b Number of parameters in model including the intercept and deviance estimate. c P-value for the fixed effect in each model that would measure the contribution of local adaptation in explaining between-subjects variance. This term was used to name each model. d Akaike’s Information Criterion corrected for finite sample size. 23 e Model weights, indicating relative likelihood that the model is best given the data.

Table 3. Analysis of variance of change in snake venom metalloproteinase activity in the presence of ground squirrel blood serum. Error mean squares and degrees of freedom for elevation and population variables were obtained via a linear combination of several terms using the Satterthwaite approximation. Snake and squirrel population terms were nested within elevation terms. Snake and squirrel ID terms were nested within population terms. The S-AS-AD term is the local adaptation test: S, Sympatric; AS, Allopatric-Same Elevation; AD, Allopatric-Different Elevation. Significance values are indicated with asterisks (* P < 0.05, P < 0.001, ***P < 0.0001) Source df F-value

Snake Elevation 1 10.3*

Sn. Elev. Error 107

Squirrel Elevation 1 16.4***

Sq. Elev. Error 107

Snake Population (Sn. Elev.) 10 7.4***

Sn. Pop. Error 107

Squirrel Population (Sq. Elev.) 10 2.0*

Sq. Pop. Error 107

Snake Id (Sn. Pop.) 108 35.7***

Squirrel Id (Sq. Pop.) 108 22.5***

S – AS – AD 2 13.5***

Error 1199

24

After variation in these geographic factors was controlled for, the local adaptation term (S – AS – AD) was a significant predictor of the amount of SVMP activity lost

(Table 3). Post-hoc testing to determine whether snakes or squirrels were locally adapted showed that the S and AS comparisons, where selection pressures may be similar, were comparable in the magnitude of change in baseline venom activity, while snakes lost ~

10% more venom activity in the AD comparisons than in S or AS trials (Figure 4a). This indicates that C. o. oreganus is locally adapted to overcoming the resistance of O. beecheyi in a way that is structured by elevation, where snakes are locally adapted to overcoming squirrel inhibitor proteins when tested against an allopatric squirrel from a site that differs in elevation. Venom deviation scores, the residual difference from an individual snake’s average performance across all squirrels, shows this pattern of local adaptation best: high elevation snakes perform best with high elevation squirrels, and low elevation snakes perform best with low elevation squirrels (Figure 4b).

25

Figure 4. Local adaptation of rattlesnakes to overcoming ground squirrel resistance shown as a) model-estimated marginal mean changes (± 1 S.E.) in SVMP activity in the presence of California ground squirrel serum for sympatric comparisons, allopatric comparisons that involved snakes and squirrels from similar elevations, and allopatric comparisons involving snakes and squirrels from different elevations and b) venom deviation score (mean ± 1 S.E.) of high (circles) and low (triangles) elevation rattlesnakes based on whether they were paired with ground squirrels from a high or low elevation. Different letters indicate significant differences.

A direct illustration of variation in the outcome of interactions between snakes and squirrels comes from comparisons of the venom deviation scores across populations.

Consistent with the result of the overall analysis, in ten of twelve populations, the mean value of the venom deviation score was more positive when calculated as S minus AD

26 than when calculated as S minus AS, demonstrating environmentally-structured local adaptation in rattlesnakes (Figure 5). However, three sites (San Joaquin Experimental

Range, McLaughlin Reserve, and Chico Creek) showed patterns more consistent with local adaptation of ground squirrels to inhibiting snake venoms (Figure 5).

Figure 5. The difference in venom deviation scores (mean ± 1 S.E.) when sympatric versus allopatric rattlesnakes are paired with ground squirrels from a given population. For individual squirrels, the amount they caused a randomly chosen snake to deviate from its average venom activity was calculated, and mean of this score for all allopatric comparisons was subtracted from the sympatric comparison. Comparisons are broken up by whether the allopatric snakes was from the same type of habitat (left bars) or the alternative type of habitat (right bars).

27

Discussion

Coevolution of Molecular Traits

In addition to the biological features of this system, the three-fold variation in rattlesnake venom activity and squirrel resistance across populations and local adaptation of snakes to overcoming squirrel resistance provide support for a coevolutionary interaction between rattlesnakes and squirrels that varies across space due to a selection mosaic (Thompson 2005b). Nonetheless, pattern alone does not provide direct evidence for coevolution since other evolutionary forces can produce similar geographic patterns

(Gomulkiewicz et al. 2007). For example, local adaptation of rattlesnakes to independently-generated geographic variation in ground squirrel resistance is a possible alternative to reciprocal coevolutionary selection. Under this scenario, geographic variation in squirrel resistance would be an evolutionary response to selection independent of rattlesnakes, while rattlesnakes still evolve local adaptation in response to selection from squirrels.

In other systems coevolution is supported when coevolutionary cold spots with no reciprocal selection (e.g. areas where one species is absent) are detected and the phenotype in the present species differs from that in similar locations where both species occur (Brodie et al. 2002; Nuismer et al. 2007; Toju 2008). Cold spots appear to present in this system: Poran et al. (1987) assessed squirrel resistance in the Great Central Valley of California, where snakes are rare. These sites were all low in elevation (< 100 m a.s.l.), and so based on the negative correlation between resistance and elevation we have shown, these populations should be highly resistant, if resistance was primarily

28 determined by site elevation. Instead, both venom resistance and the behavioral response to a live rattlesnake (Poran et al. 1987) are reduced in these Valley floor populations, indicating that the effect of elevation on the squirrel resistance phenotype depends on selection from rattlesnakes. The resistance phenotype in low resistance Central Valley cold-spots supports the presence of a geographic mosaic of coevolutionary selection in this system. Taken together the geographic patterns in venom, resistance, and behavioral traits all provide strong support for our assumption that coevolution plays an important role in the evolution of offensive and defensive traits in this predator and its prey, while more definitive proof would come from further work which uses methods to partition the strength of selection on venom and resistance traits (Ridenhour 2005; Toju & Sota 2006).

A Role for Phenotype Matching

Our results identify phenotype matching as an important force governing the coevolutionary interactions of rattlesnakes and ground squirrels, because phenotype matching is the only coevolutionary mechanism that can generate a signature of local adaptation that is detectable in a reciprocal crossing study (Nuismer et al. 2005; Nuismer et al. 2007; Ridenhour & Nuismer 2007). In a statistical sense, among-population comparisons are dominated by population main effects during arms race coevolution, leading to the least escalated populations always appearing locally adapted and the most escalated populations always appearing locally maladapted. On the other hand, phenotype matching through a “lock-and-key” mechanism of target recognition can generate populations of snakes that show local adaptation when comparisons are made between

29 sympatric and allopatric pairs of the interacting predator and prey, because the unique resistance phenotype of each squirrel population represents a distinct adaptive peak for the local snakes (Ridenhour & Nuismer 2007).

The role of arms race dynamics as a general mechanism for generating spatial variation in snake venom and prey resistance remains unclear. On one hand, some of the best-studied cases of antagonistic arms races have shown that these interactions generate positive correlations between the trait values of the interacting species (Brodie et al.

2002; Hanifin et al. 2008; Toju & Sota 2006). We did not find such a correlation between population-level measures of SVMP activity in snakes and SVMP resistance in squirrels.

However, theoretical work suggests that regression analysis of phenotype values is not a strong test of the arms race hypothesis, since there is only a weak expectation for arms races to generate positively correlated trait means under strong coevolutionary selection

(Nuismer et al. 2010). Furthermore, our data are based on a functional interaction

(enzyme activities and their inhibition) that closely-matches the mechanism envisioned for phenotype matching (see below; Biardi 2008; Biardi et al. 2011a). In contrast, the concentrations of SVMPs and their inhibitors may only be loosely related to our enzymatic measures if various isoforms of these proteins have different activities, a result that has been confirmed for the SVMPs of other pitvipers (Bernardoni et al. 2014).

Concentrations of venom proteins and their inhibitors may better fit the arms race model and would be a logical next place to search for signs of an arms race mediated by a phenotypic differences mechanism involving venom and resistance.

30

The implication that phenotype matching influences the fitness outcomes of an interaction between venomous and resistant species is a key step in explaining patterns of diversification in this coevolving system. We suggest that interactions between venom and resistance molecules studied here are analogous to the targeted recognition or gene- for-gene coevolution seen with innate immunity to pathogens, which is a form of coevolution governed by phenotype matching (Dybdahl et al. 2014). This explanation is consistent with our understanding of the mechanism by which these proteins interact: resistance molecules in prey bind and inactivate venom protein targets, whereas the venom must bypass these molecules to be effective, which is similar to pathogens evading the immune system (Biardi 2008). Targeted binding of inhibitors and the evolution of molecular mechanisms by which venom overcomes resistance requires complexity in the molecular underpinnings of both the venom and resistance phenotypes

(e.g. allelic variation). Consistent with this idea, multiple SVMP isoforms exist within individual C. o. oreganus (Mackessy 2010). In contrast, the level of variation at O. beecheyi resistance loci is unknown although the small number of resistance proteins identified to date suggests if genetically-based variation present it is likely limited (Biardi et al. 2011a). Phenotype matching, compared to arms races, is more likely to promote phenotypic diversification among populations (Yoder & Nuismer 2010), so may better explain the high levels of intraspecific variation observed in snake venoms (Mackessy

2010; Sunagar et al. 2014).

31

The Predator is Ahead

Rattlesnakes appear to be ahead in the coevolving interactions involving SVMPs and their inhibitors, because venom, and not resistance, showed a pattern of local adaptation in our reciprocal crossing study (Blanquart et al. 2013). Predator local adaptation is unexpected given the potential asymmetry in selection intensity and evolvability between predator and prey, as summarized in the Life-Dinner Principle

(1979). Prey are expected to be under stronger selection because being eaten entails a higher fitness cost than missing a meal, and because they are in general expected to evolve more rapidly due to higher reproductive rates. The life history characteristics of rattlesnakes and squirrels match these expectations: O. beecheyi reproduce annually beginning in their second year and live to around 5 years of age (Fitch 1948), while rattlesnakes reproduce semi-annually after their third year, and have a maximum lifespan of 20 years or older (Fitch 1949). Yet, despite these demographic differences snakes are evolutionarily ahead of their prey at least for the molecular interaction we characterized.

This pattern has also been found in other antagonistic interactions: predatory garter snakes are ahead in the arms race dynamic occurring with their toxic newt prey (Hanifin et al. 2008), while pipefish hosts are ahead of Vibrio bacterial parasites despite generation times that differ by many orders of magnitude (Roth et al. 2012). Our results support Abrams’ (1986) suggestion that the outcome of antagonistic coevolution in predator-prey systems will be a product of ecological and evolutionary factors specific to each system and cannot be predicted from general concepts such as the Life-Dinner

Principle.

32

Two such factors that theoretical and empirical studies emphasize are relative levels of gene flow of the interacting species and the genetic architecture of the key traits mediating the interaction (Gandon 2002; Macpherson et al. 2015). In host-parasite systems, the “winner” in a coevolutionary interaction is often the species with higher gene flow, as higher migration can facilitate more rapid adaption due to the increased spread of evolutionarily advantageous alleles (Gandon 2002). Population genetic studies of congeneric species suggest that ground squirrels show higher levels of genetic structure than Crotalus rattlesnakes (Bushar et al. 2015; Gavin et al. 1999), which is consistent with the prediction that the winning partner will experience greater levels of gene flow, but this remains to be documented for the species studied here.

The architecture of key traits at the phenotypic interface of the interaction may also influence whether predators or prey are ahead (Brodie & Brodie 1999). In particular, the participant with traits that are less evolutionarily constrained or have a more evolvable genetic architecture may be at an advantage (Brodie & Brodie 1999). Both evolvability and constraint were invoked to explain why garter snakes outpaced newts in their tetrodotoxin-based arms race: single-nucleotide substitutions can have large effects on snake resistance, while bio-availability of toxin precursors may constrain newt toxicity levels (Hanifin et al. 2008). Venom is a modular phenotype contained in a secretory venom gland (Margres et al. 2015a), so it may be more evolvable than the blood-based inhibitor proteins of ground squirrels, which could interact pleiotropically with other physiological processes, constraining their ability to respond over evolutionary timescales to changes in venom. We emphasize that our results to date only involve snake SVMP

33 proteins and their inhibitors in squirrels. Whether similar patterns occur for the coevolutionary dynamics of other venom proteins, such as the serine protease and phospholipase A2 proteins to which ground squirrels may be resistant, is an important direction for future work.

Environmental Effects on Coevolution

Local adaptation can be missed if the geographic location of study sites is not congruent with the scale at which local adaptation occurs, or if variable selection from other sources (e.g. the abiotic environment) among allopatric sites leads to large variance in the outcomes of allopatric tests (Blanquart et al. 2013). In our system, controlling for elevation through comparisons of snake-squirrel populations at the same (AS) versus different (AD) elevations was essential to detecting local adaptation of rattlesnake venom to squirrel resistance. Not doing so leads to much larger variance in allopatric comparisons and a difference between sympatric and allopatric performance of venom that is marginally non-significant (Table 2). This result adds to a growing body of work that identifies environmental differences associated with elevation as key influences on the selection mosaic in coevolutionary systems, such as newts and garter snakes (Stokes et al. 2015) and camellias and weevils (Toju 2008). More broadly, our study highlights the need to understand the environmental context of variation in traits involved in species interactions when interpreting their evolutionary history.

Hypoxic conditions at higher elevations are one candidate for the environmental cause of a selection mosaic acting on rattlesnakes and ground squirrels. Mammalian

34 serum albumins are non-specific inhibitors of rattlesnake venom activity (Clark & Voris

1969) that are reduced in abundance in some mammals at high elevations to compensate for increased hematocrit (Crait et al. 2012). Thus, reduced expression in non-specific protein binding agents like albumin may account for the substantial reduction in the venom-inhibitory capacity of high elevation squirrel serum (the squirrel elevation main effect), with isoforms of alpha-1-antitrypsin involved in the remaining variation to which snakes have become locally adapted. Identifying the physiological underpinnings of elevational differences in resistance, and how these may lead to variable coevolutionary selection between snakes and squirrels, are promising avenues for future work.

Conclusions

Our study demonstrates the value of using local adaptation studies to assess the coevolutionary mechanism that is responsible for trait variation at the molecular level in venomous animals and their resistant prey. We used this approach to draw the novel conclusion that phenotype matching is a driver of coevolution between venom and resistance traits involved in a rattlesnake-prey interaction. We also show that despite biological differences between snakes and squirrels that favor squirrels winning the interaction, snakes appear to be evolutionarily ahead of their prey. The evolutionarily labile venom phenotype, combined with phenotype matching and its propensity to promote diversification (Yoder & Nuismer 2010), may provide a mechanism by which coevolution generates biologically and medically important diversity in animal venoms.

35

Chapter 3: Demographic differences predict patterns of local adaptation in a coevolving vertebrate predator and its prey

Abstract

Coevolutionary theory predicts that differences in the genetic demography of interacting species can influence patterns of local adaptation by effecting the potential of local populations to respond to selection. Explicit tests of this hypothesis are rare in natural systems. Here, we assess the power of differences in two key demographic factors – effective population size (Ne) and migration (m) – to predict patterns of local adaptation in a coevolving vertebrate predator-prey system where the predator appears to be the more locally adapted species. Initially, we used RADseq data to assess phylogeographic structure and generate population-specific estimates of Ne and m for multiple populations of venomous rattlesnakes and their resistant ground squirrel prey across California. We then showed that the sign and magnitude of local adaptation of venom enzymatic activity and ground squirrel resistance differs among largely coincident phylogenetic lineages of these interacting species: we only find evidence for local adaptation in populations south of San Francisco Bay. In these populations, rattlesnakes had larger Ne values than squirrels at most locations and a significant positive relationship existed between the difference in rattlesnake and ground squirrel Ne and the magnitude of rattlesnake local adaptation. In contrast, local differences in m values between species at a given site did

36 not predict the level of local adaptation. These results explain why rattlesnakes are the locally adapted player in this coevolutionary interaction–the larger population sizes of snakes allow them to adapt more rapidly than the ground squirrels with which they interact.

Introduction

Coevolution leads to local adaptation when focal species track the adaptive evolution of other species in geographically distinct populations (Blanquart et al. 2012;

Forde et al. 2004; Hoeksema & Forde 2008). The direction and magnitude of local adaptation is impacted by the evolutionary potential of local populations to respond to variation in reciprocal selection pressures across space and time. Variation in local selection has been shown to account for differences in the magnitude of local adaptation in a wide range of antagonistic coevolutionary systems such as experimental cultures of bacteria and viruses (Buckling & Rainey 2002; Friman & Buckling 2013; Koskella et al.

2011; Lopez Pascua et al. 2014; Vogwill et al. 2009), natural populations of hosts and parasites (Bernays & Graham 1988; Forde et al. 2004; Laine 2006; Lively et al. 2004;

Sicard et al. 2007) and vertebrate predators and their prey (Brodie & Ridenhour 2003;

Talluto & Benkman 2014). As such, identifying the evolutionary and ecological characteristics of interacting species that influence each species’ ability to locally adapt can provide a more complete understanding of the context-dependence of coevolutionary outcomes (Hoeksema & Forde 2008; Thompson 2005b).

An important prediction of coevolutionary theory is that the genetic demography of interacting populations affects the ability of a species to adapt to coevolving enemies

37 in local populations (Gandon & Michalakis 2002; Gandon & Nuismer 2009; Morgan et al. 2005). In particular, genetically effective population size (Ne) and migration rates among local populations (m) may have a significant impact on the potential of populations to mount an evolutionary response to coevolving enemies. Effective population size is important because larger populations are less susceptible to the effects of genetic drift, which can act to oppose selection. Selection can therefore act more efficiently in generating adaptation in large versus small populations (Allendorf 1986;

Fisher 1930; Nei et al. 1975).

The role of Ne in coevolutionary interactions is supported by theoretical simulations (Gandon & Michalakis 2002; Gandon & Nuismer 2009; Thompson &

Burdon 1992), where the interacting species with larger Ne is predicted to have higher levels of local adaptation. Most empirical investigations of the susceptibility of small, less genetically diverse populations to attack by enemies are from studies of disease ecology. Higher genetic diversity of both individuals and populations often coincides with higher overall resistance to parasites in taxa such as birds and frogs (Becker et al.

2014; Savage & Zamudio 2011; Whiteman et al. 2006). While these studies are indicative of the role Ne plays in local population health, comparative estimates of Ne between natural populations of sympatric, coevolving taxa are necessary to empirically link differences in species’ Ne to patterns of local adaptation, but such studies are rare.

Social parasitic bees have smaller Ne than their larger host bees suggesting that drift constrains their response to host selection and yet parasites continue to persist in host nests (Shokri Bousjein et al. 2016), but these authors did not measure adaptation of one

38 species to the other directly. Investigating the role of comparative Ne among coevolving species would be valuable to test predictions from theory on its importance to local adaptation.

While coevolutionary theory suggests larger Ne generally facilitates local adaption, the expected impact of m on local adaptation is dependent on specific features of the genetic demography of the interacting species (Gandon & Michalakis 2002;

Gandon & Nuismer 2009; Nuismer et al. 2010; Nuismer et al. 2007). Migration is traditionally viewed as a process that homogenizes genetic divergence among populations and therefore decreases levels of local adaptation (Slatkin 1985). Population genetic models of coevolving systems that consider large local population sizes or relatively high levels of migration between populations (Nm > 1) predict that migration reduces levels of local adaptation (Nuismer & Thompson 2006; Ridenhour & Nuismer 2007; Yoder &

Nuismer 2010). In contrast, population genetic models that assume finite population sizes and reduced migration rates (Nm < 1) demonstrate that drift can significantly reduce levels of adaptive variation in local populations. In such cases, migration can enhance local adaption by importing adaptive variation into a population from which it has been lost (Forde et al. 2007a; Gandon & Michalakis 2002; Gandon & Nuismer 2009; Morgan et al. 2005). Under these “small population” models of population structure, differences in m between two coevolving antagonistic species can determine the direction and magnitude of local adaptation between them, where the species with higher m are the locally adapted species (Blanquart et al. 2012; Gandon & Michalakis 2002; Gandon &

Nuismer 2009).

39

In natural populations, higher values of m have been shown to coincide with local adaptation of Melampsora fungi to its host plant (Thrall et al. 2002), local adaptation of the plant Silene latifolia to a fungal pathogen (Kaltz et al. 1999), local adaptation of the island lizard Gallotia galloti to a blood parasite (Oppliger et al. 1999) and local adaptation of the pipefish Syngnathus typhle to a bacterial pathogen (Roth et al. 2012).

Yet, other studies have found a maladaptive impact of migrants, in that higher gene flow coincided with local maladaptation (Farkas et al. 2016; Ganz & Washburn 2006; Start &

Gilbert 2016). Two separate meta-analyses of host-parasite systems have attempted to link differences in m between natural populations of coevolving antagonists to the magnitude of local adaptation (Greischar & Koskella 2007; Hoeksema & Forde 2008).

Each found that higher gene flow in parasites was often associated with parasite local adaptation to their hosts, but the sign and magnitude of this effect varied considerably between studies. The inconsistent impacts of gene flow on local adaptation could reflect differences in the genetic demography of the interacting species studied, in terms of differences in Nm, yet this was not assessed.

These considerations mean that linking theory with predictions about the impact of genetic demography on adaptation in natural coevolving systems requires prior knowledge about population structure of the interacting species. Specifically, when Nm >

1 and selection is a more significant evolutionary force than drift, the prediction is that larger Ne should have a positive effect on levels of adaptive whereas higher m will have a negative impact. In contrast, in populations with small Ne and limited migration (Nm < 1) genetic drift plays a dominant role in determining levels of adaptive variation. Under

40 such scenarios, differences in migration will play a more significant role in driving levels of local adaptation than differences in Ne (Fig. 6). Thus, in natural populations, the a priori determination of population structure in terms of whether Nm values for interacting species are > or < 1 is essential to link predictions from theory about the impacts genetic demography to levels of local adaptation of coevolving species.

Figure 6. Decision tree for using demographic information on two putatively coevolving species to make predictions about the effects of each species’ effective population size (Ne) and migration rate (m) on the direction and magnitude of local adaptation (LA). Predictions are based on the collective theoretical results of Gandon et al. (1996), Gandon (2002), Gandon and Michalakis (2002), Gandon and Nuismer (2009), and Nuismer et al. (2010).

To assess the predicted impacts of genetic demography on local adaptation in natural coevolutionary systems, empirical estimates of Ne and m need to be coupled with 41 quantitative measures of the degree to which populations of interacting species are locally adapted (Holding et al. 2016a; Roth et al. 2012; Thrall et al. 2002). The recent availability of methods for generating genome-wide SNP datasets combined with demographic modeling approaches now allow us to jointly and accurately estimate both m and Ne among populations in non-model taxa (Bryson et al. 2016; Kraus et al. 2013;

Sovic et al. 2016; Sullivan et al. 2014; Wang et al. 2016; Waples 2016). We can use these approaches to a priori assess the magnitude of Nm values to identify the appropriate predictive framework for how genetic demography should impact levels of local adaptation (See Figure 6). The value of this approach is that it provides a more comprehensive and context-appropriate assessment of predictions from coevolutionary theory about how the genetic demography of interacting species impacts levels of local adaptation.

Here, we adopt this approach by comparing estimates of genetic demography and local adaption in a putatively coevolving vertebrate predator-prey system involving northern Pacific rattlesnakes (Crotalus oreganus oreganus) and California ground squirrels (Otospermophilus beecheyi). A recent analysis of local adaptation among 12

California populations of these species showed that the rattlesnake’s venom is locally adapted to overcoming ground squirrel venom resistance (Chapter 2; Holding et al.

2016a). This analysis led to the inference that geographic variation in rattlesnake venom is partly a response to population-level variation in the ground squirrel resistance phenotype, and supports the long-standing hypothesis that rattlesnakes and ground squirrels are coevolving in response to reciprocal selection pressures (Biardi 2008, 2000;

42

Coss et al. 1993; Poran et al. 1987). Although the rattlesnakes were found to be the more locally adapted species on average, the direction and magnitude of sympatric versus allopatric contrasts of venom performance (Blanquart et al. 2013) varied among populations, with 4 of 12 sites displaying a pattern consistent with local advantage in the ground squirrels (Holding et al. 2016a). Here, we assess the hypothesis that this variation could be explained by local differences in gene flow or effective population size that impact the evolutionary potential of rattlesnake and ground squirrel populations, creating local differences in each species’ ability to keep pace with its coevolving enemy.

Specifically, we combine the established measures of local adaptation in several populations (Holding et al. 2016a) with estimates of m and Ne inferred from RADseq data to compare values of these demographic factors between rattlesnakes and squirrels as previous studies of coevolving taxa have done (Couchoux et al. 2016; Shokri Bousjein et al. 2016; Sutton et al. 2016). First, we use the Nm values obtained for each species to predict how effective population size and migration should impact local adaptation in this system. Next, we test two specific hypotheses: 1) Larger Ne facilitates local adaptation of predator and prey, so the difference in Ne values will be positively correlated with the extent of local adaptation in rattlesnakes; and 2) higher m is maladaptive, so the differences in m values between rattlesnakes and ground squirrels for each sampled population will be negatively correlated with the extent of local adaptation in rattlesnakes. These quantitative tests for associations between demographic differences and local adaptation extend traditional comparisons of genetic demographic factors

43 between species by providing a more concrete link between the demography of populations and their ability to locally adapt to sympatric enemies.

Methods

Populations and Sample Processing

As described in Holding et al. (2016a), we collected 10 rattlesnakes and 10 ground squirrels at each of 12 locations spread over the northern two-thirds of the state of

California. Samples were either the same animals used by Holding et al. (2016a) or different animals collected at the same time. Snakes were located by visual search of suitable habitat, while ground squirrels were live-trapped in the same locations. We spread the traps among multiple squirrel colonies to minimize the potential for trapping closely related animals. Venom was manually extracted from each snake using previously published protocols (Holding et al. 2016a), and a blood sample was taken from the caudal vein for genetic analysis. Ground squirrels were anesthetized via isoflurane inhalation, during which time a blood sample was obtained via cardiac puncture and an ear clip was taken for DNA. All animals were released at the site of original capture following processing. All procedures were conducted under protocols approved by the Ohio State

Institutional Animal Care and Use Committee (protocol #2012A00000015).

We extracted genomic DNA from snake blood and squirrel ear tissue using DNA

Blood and Tissue Kits (Qiagen, Valencia, CA, USA) or standard phenol-chloroform protocols. We generated double-digest RADseq libraries for each individual following the methods of Sovic et al. (2016). Briefly, we digested genomic DNA with the EcoRI

44 and SbfI restriction enzymes (New England Biolabs, Ipswich, MA, USA) and electrophoresed each individual’s genomic DNA on a single lane of a 2% (w/v) low melt point agarose gel to select fragments of between 300 and 450 base-pairs. The size- selected fragments were extracted from the gel with MinElute gel extraction kits

(Qiagen), amplified via PCR, and purified with AmPure beads. KAPA library quantification kits (KAPA Biosystems, Wilmington, MA) were then used to quantify the number of molecules present in each sample for equimolar pooling into a final library.

Each pooled library was then subjected to 50-bp, single end sequencing on an Illumina

HiSeq 2500.

Bioinformatic Processing of Sequence Reads.

We assembled reads, identified SNPs, and assigned genotypes to individual snakes and squirrels using AftrRAD v.5.1 (Sovic et al. 2015). For each species, we performed three separate AftrRAD analyses. First, we conducted a global analysis using all samples to generate a set of loci common to all individuals that was used for genetic clustering analyses across all samples. Second, we isolated loci polymorphic within the

Northern and Southern genetic groups (see below) in both species that were identified based on the global clustering analysis and used these data for demographic analyses within each group. Following Sovic et al. (2016), our analyses only included loci with no missing data across all samples to minimize the impact of unscored genotypes on estimates of population genetic parameters (Arnold et al. 2013).

45

Genetic Structure of Snake and Squirrel Populations

We conducted genetic analyses of both snakes and squirrels to assess the genetic distinctiveness within each species at different evolutionary scales. This approach is important when studying independent yet interacting species because local adaptation is a metapopulation process that occurs among populations connected by gene flow

(Blanquart et al. 2013; Nuismer & Gandon 2008; Penczykowski et al. 2016). Thus, the scale at which local adaptation is measured should correspond to the extent to which populations are evolutionarily connected across a landscape (Blanquart et al. 2013; Kaltz et al. 1999). Phylogenetic structure associated with the San Francisco Bay/Sacramento-

San Joaquin River Delta has been suggested for both rattlesnakes (Goldenberg 2013) and ground squirrels (Phuong et al. 2014) yet both studies used limited genetic datasets for their analyses. Therefore, one goal of our genetic analyses of both species was to confirm this coincident pattern of divergence using genomic scale data and assess whether the patterns of local adaptation based on a range wide analysis in Holding et al. (2016a) was similar or different between sets of interacting populations in different phylogeographic lineages.

To assess levels of genetic structure between samples collected at different sites we calculated pairwise FST values among all pairs of populations and among genetic groups within each species. We used the pairwise.WCfst function in the package hierfstat

(Goudet 2005) in R (R Development Core Team 2015) and used the boot.ppfst function from the same package to test the significance of the observed values by calculating 99% confidence intervals from 10,000 bootstrap replicates.

46

Next, to evaluate genetic relationships among populations we used both clustering and lineage-based analyses. To assess whether populations formed sets of genetically similar clusters we ran Bayesian clustering analysis in the program STRUCTURE

(Pritchard et al. 2000). Three separate iterations were run for each K value ranging from

1 to 13, with population (1 of 12 sites) as a prior and a burn-in of 100,000 iterations out of 1,000,000 total Markov chain Monte Carlo iterations following Gilbert et al. (2012).

The best number of K clusters from the STRUCTURE analyses of each species was determined using the Delta-K method (Evanno et al. 2005) as calculated in Structure

Harvester (Earl & Vonholdt 2012).

To assess whether sets of populations form phylogenetically distinct lineages as suggested by previous work (see above), we conducted species tree analyses using the program SNAPP (Bryant et al. 2012). For this analysis, we used the first biallelic SNP from each polymorphic locus to construct data sets for the two species. The computational demands of the SNAPP analyses did not permit including all samples from all populations, so we randomly chose two samples to represent each population (Sovic et al. 2016). Mutation rates were set as u = v = 1 and the coalescence rate was sampled from a starting value of 10 (A. Leaché, pers. comm.). To derive realistic values for α and β, we calculated the average pairwise substitutions per site (rattlesnakes = 0.006; ground squirrels = 0.013), and used these values to set α to 1 and β to a number that yielded an appropriate expectation for θ. β was set to 166 for rattlesnake and 80 for ground squirrels.

We set kappa to 1 and sampled λ from a gamma distribution with a minimum of 2, maximum of 400, start value of 5, α = 2, and β = 80. This λ distribution includes a

47 reasonable estimate of tree height for both species based on the number of substitutions per site between the northernmost and southernmost populations of rattlesnakes or ground squirrels. The SNAPP runs for both species had a burn-in of 100,000, a chain length of 10,000,000, and sampled trees every 1,000 iterations. We evaluated the convergence of the run for each species using Tracer (Rambaut et al. 2014). ESS values for all parameters were > 100 with the exception of theta estimates for the two smallest snake populations. Finally, we used TreeAnnotator

(http://beast.bio.ed.ac.uk/treeannotator) to generate maximum clade credibility consensus trees, and visualized these trees and associated uncertainty in DensiTree (Bouckaert

2010).

Joint Estimation of m and Ne by Coalescent Simulation

To obtain joint, long-term estimates of m and Ne for both snake and squirrel populations, we used a model-based approach incorporated in the coalescent-based program FastSimCoal2 (Excoffier et al. 2013; Sovic et al. 2016). We modeled migration between populations using separate stepping-stone migration models for the Northern and

Southern genetic groups separately where gene flow was estimated between geographically proximate sites (Fig. 7). We evaluated three separate models for each species in both the North and South, a full model, a model without gene flow across the

Central Valley of California (Hedin et al. 2013), and a model without gene flow across the Cascades Range in the North, or Transverse Ranges and Sierra Nevada in the South

(Fig. 7). Models were compared with AIC (Burnham & Anderson, 2002), and the full

48 model was more than 2 ∆AIC units above the other models for all comparisons in rattlesnakes and ground squirrels, so parameters were estimated under these full models

(Table 4).

Figure 7. Models of asymmetric gene flow between populations tested in both rattlesnakes and ground squirrels separately in the Northern and Southern genetic groups.

49

Table 4. Information theoretic comparison of the three models of population connectivity tested in both the Northern and Southern genetic groups of both Northern Pacific rattlesnakes and California ground squirrels. Squirrels North Model K ln Likelihood AIC ∆AIC Akaike Weight Full_North 14 -5939.6 27381.1 0.0 0.99 No Trans-Valley North 12 -5946.3 27407.7 26.7 < 0.01 No Cascades Migrants 10 -5944.8 27396.9 15.8 < 0.01

Squirrels South Model K ln Likelihood AIC ∆AIC Akaike Weight Full_South 12 -7350.0 33872.1 0.0 0.99 No Trans-Valley South 10 -7360.1 33914.7 42.5 < 0.01 No Sierra Migrants 10 -7370.8 33963.6 91.5 < 0.01

Snakes North Model K ln Likelihood AIC ∆AIC Akaike Weight Full_North 14 -4507.8 20787.0 0.0 0.99 No Trans-Valley North 12 -4513.5 20809.4 22.4 < 0.01 No Cascades Migrants 10 -4521.9 20843.9 56.9 < 0.01

Snakes South Model K ln Likelihood AIC ∆AIC Akaike Weight Full_South 12 -4724.3 21780.2 0.0 0.99 No Trans-Valley South 10 -4733.4 21818.2 38.0 < 0.01 No Sierra Migrants 10 -4742.2 21858.6 78.4 < 0.01

Parameters were estimated based on observed pairwise joint site frequency spectra that were generated using AftrRAD, and that contained a maximum of one SNP from each locus. When evaluating models, values for Ne were selected from a uniform distribution from 10 to 100,000, while Nm values were selected from a log-uniform

50 distribution ranging from 0.01 to 15 and converted to m by dividing Nm by estimates of

Ne. For both Northern and Southern rattlesnakes and ground squirrels, we performed 50 replicate runs of the full dataset with 40 ECM cycles and 100,000 simulations per run, and a mutation rate of 2.5 x 10-8 mutations per site per generation. From these runs, we chose the one with the highest likelihood value for our point estimates of all parameters.

To calculate a single value of Nm or m for a given population, we summed the values for each incoming migration path in our model. For example, in the inset example of three populations in Figure 8, the sum of m31 and m21 would give the overall m value for population 1.

To generate estimates of variation for each parameter estimate, we used a non- parametric bootstrapping approach. We resampled the matrix of SNP loci for each species with replacement to create 25 replicate multi-dimensional site frequency spectra with the same number of loci as the original. Each replicate dataset was subjected to 25 independent runs of FastSimCoal2 that were otherwise identical to the original run. We chose the run with the highest likelihood for each resampled dataset, and calculated interquartile ranges for each parameter from the 25 values generated.

51

Figure 8. Map of 12 sampled locations where samples of Northern Pacific rattlesnakes and California ground squirrels were obtained. The locations that belong to the Northern and Southern genetic groups in both species are shown in blue and red, respectively, whereas Sutter Buttes, which belonged to the Northern snake cluster and the Southern squirrel cluster, is colored yellow. The black lines connecting sites represent migration paths for which bidirectional migration rates were estimated using demographic modeling in FastSimCoal2 (see Methods). Inset: an example of the FastSimCoal2 model with bidirectional gene flow estimates for three Southern populations.

Quantifying Local Adaptation

We quantified levels of local adaptation in the ability of venom to carry out its enzymatic activity in the presence of ground squirrel venom inhibitors using data from Holding et al. (2016a). As previously described (Holding et al. 2016a) each snake was paired with a randomly chosen squirrel from its sympatric population and one squirrel from each

52 allopatric site. Dilute whole venom samples from individual snakes (10 per population) were incubated with dilute ground squirrel serum prior to measurement of snake venom metalloproteinase activity with the EnzChek Gelatinase Kit (Life Technologies, Carlsbad,

CA), which measures the activity of snake venom metalloproteinases (Biardi et al.

2011b).

Holding et al. (2016a) used a linear mixed-effects model (LMM) to detect a range-wide signal of rattlesnake local adaptation across the 12 locations in Figure 8. We reanalyzed the same data to determine whether the signal of rattlesnake local adaptation previously detected is present within both the Northern and Southern genetic groups or originates in only one region. Whereas Holding et al. (2016a) compared performance of ground squirrels with sympatric rattlesnakes and rattlesnakes in 11 allopatric sites across

California, we limited the allopatric comparisons to those within the same genetic group

(Northern or Southern) as that of a given site. We then separately applied the same LMM to these functional measures of the rattlesnake-ground squirrel interaction in the Northern and Southern regions. Briefly, the model includes main effects of snake population of origin, ground squirrel population of origin, snake individual, squirrel individual, and a local adaptation factor. Additionally, elevation is included because it is an important abiotic driver of the pattern of snake-squirrel local adaptation (see Holding et al. 2016a).

The local adaptation factor (termed S-AS-AD) parses each snake-squirrel pairing based on whether it was sympatric (S), allopatric but involving populations of similar elevation

(AS), or allopatric but involving major differences in elevation (AD).

53

We derived scores for local adaptation in individual populations of rattlesnakes and squirrels that account for placement of our sampled populations in the Northern or

Southern genetic groups, again following Holding et al. (2016a), but only considered combinations of rattlesnake venom and ground squirrel serum within and not between genetic groups. We first calculated mean venom activity of each rattlesnake in the presence of all ground squirrel sera with which it was tested (mean venom performance).

Then, we determined the extent to which each squirrel caused each rattlesnake to deviate from its mean venom performance (venom deviation score). For each squirrel, we then subtracted the average venom deviation score based on all allopatric rattlesnakes from the venom deviation score of the sympatric rattlesnake (individual squirrel adaptation score).

Finally, we calculated a “venom adaptation score” as the population average of the individual squirrel adaptation scores. The venom adaptation score is analogous to a residual value that controls for variation among snakes in baseline activity and among squirrel population in average amount of venom resistance (Holding et al. 2016a). These adaptation scores summarize the extent to which sympatric rattlesnakes outperform allopatric rattlesnakes in the presence of sympatric squirrels with which they have coevolved (if the value is positive), or vice versa (for negative values indicating squirrel local adaptation). These analyses were performed separately for the Northern and

Southern genetic groups.

54

Data analysis

We made a quantitative comparison of both Ne and m between rattlesnakes and ground squirrels to test the predictions that rattlesnakes should have larger Ne and smaller m, because they are the locally adapted species range-wide (Holding et al. 2016a). To do this, we determined whether the point estimates of each parameter in each population derived from the full empirical dataset in FastSimCoal25 were included in the interquartile ranges of the non-parametric bootstrapping estimates of parameters in

FastSimCoal2 generated for each population of the other species.

Finally, we assessed the prediction that larger Ne or smaller m in rattlesnakes compared to ground squirrels at a given site would lead to stronger local adaptation at a site. Differences in Ne and m for each population were calculated by subtracting the respective values in ground squirrels from that of rattlesnakes, leading to the specific prediction of positive relationships between venom adaptation score and the difference in

Ne and a negative relationship between the venom adaptation score and difference in m.

We conducted simple linear regressions in R to assess these predictions with an accepted

Type I error rate of α = 0.05.

Results

SNP identification in Rattlesnakes and Ground Squirrels

For the rattlesnakes, we generated a total of 64,596,311 reads with a mean read depth of 652,488 per individual across 99 individuals. Our mean and median read depth per locus were 79.5 and 52 reads, respectively. We identified a total of 22,709 non-

55 paralogous loci, of which 5,012 were polymorphic. For clustering and SNAPP analyses, we ran AftrRAD with all individuals included (Northern and Southern) to generate a dataset of 917 polymorphic loci with 1,556 SNPs scored in all individuals. For demographic model choice and parameter estimation, we performed separate AftrRAD runs on the Northern and Southern populations alone. The four Northern populations yielded a dataset of 875 polymorphic loci with 1,212 SNPs scored in all individuals, whereas the 7 Southern populations yielded 869 polymorphic loci with 1,395 SNPs in all individuals.

For ground squirrels, we generated a total of 85,816,430 reads with a mean read depth of 866,832 per individual across 99 individuals. Our mean and median read depth per locus were 37.6 and 27, respectively. We identified a total of 30,182 non-paralogous loci, of which 6,484 were polymorphic. For clustering and SNAPP analyses, we ran

AftrRAD with all individuals included (Northern and Southern) to generate a dataset of

1,380 polymorphic loci with 2,178 SNPs scored in all individuals. For demographic model choice and parameter estimation, we performed separate AftrRAD runs on the north and south populations alone. The four Northern populations yielded a dataset of

1,121 polymorphic loci with 1,379 SNPs scored in all individuals, whereas the 7

Southern populations yielded 1,278 polymorphic loci with 1,819 SNPs in all individuals.

Genetic Structure of Rattlesnakes and Ground Squirrels

Across all scales, ground squirrels displayed greater population genetic structure compared to rattlesnakes. The overall FST values across all squirrel populations (FST =

56

0.40) was 2.2 times greater than the value across all snake populations (FST = 0.18). This pattern also holds for comparisons between and within Northern and Southern groups of both species. Northern and Southern ground squirrels are more differentiated (FST = 0.57) than Northern and Southern rattlesnakes (FST = 0.18) and squirrel populations are more distinct among both Northern (FST = 0.21 vs. 0.09) and Southern populations (FST = 0.12 vs. 0.10).

Both the Bayesian clustering and species tree analyses support the existence of

Northern and Southern genetic groups in both species that are split across the San

Francisco Bay and Sacramento/San Joaquin River Delta. For the Structure analysis, comparisons of K values using the Delta K approach (Evanno et al. 2005) indicates that

K = 2 clusters are optimal for both rattlesnakes and ground squirrels with populations forming clusters North and South of San Francisco Bay (Fig. 9). Likewise, the species tree analysis of phylogenetic relationships between populations using SNAPP shows that the deepest node defines a Northern and Southern clade in both species congruent with the clusters defined by Structure (Fig. 9).

The location of the eastern boundary of the Northern and Southern groups is not congruent between rattlesnakes and ground squirrels in that Sutter Buttes snakes, just north of Sacramento, are grouped with the Northern snake group, while ground squirrels from the same location are grouped with the Southern squirrel group (Figure 9, “SB”).

Since it is unclear how to include this site in our current statistical framework for analyzing local adaptation based on allopatric versus sympatric comparisons, we removed it from subsequent analysis.

57

Figure 9. SNAPP trees with posterior probabilities > 0.7 and STRUCTURE plots at K = 2 for rattlesnakes and ground squirrels showing Northern (orange) and Southern (blue) genetic groups. Note the incongruent group membership of the Sutter Buttes animals (SB).

The geographic scale of local adaptation

Next, we assessed whether patterns of local adaptation differed between populations of snakes and squirrels in the Northern and Southern genetic groups of each species. Our analysis shows differences in the observed level of local adaptation in each lineage suggesting that rattlesnake and ground squirrel populations may be coevolving

58 differently within each lineage, although the effects of sample size could also account for the difference. Specifically, we detected statistically significant rattlesnake local adaptation among the Southern sites (S-AS-AD term: F2,349 = 4.57, P = 0.011, Table 5).

Tukey’s post-hoc tests showed that snake venom from Southern sites performed significantly better in sympatric populations (deviation from baseline = -243 RFU) than in allopatric sites of different elevation class (-270) and a contrast of sympatric versus all allopatric combinations suggests local adaptation regardless of environmental structure (P

= 0.08). In contrast, the four Northern sites did not show a pattern consistent with local adaptation of rattlesnakes or ground squirrels (F2,79 = 1.01, P = 0.368, Table 5). These results are reflected in the venom adaptation scores calculated for the performance of each ground squirrel population with sympatric and allopatric snakes (Fig.10): six of seven Southern sites (86%) showed positive venom adaptation scores consistent with rattlesnake local adaptation, while only two of four Northern sites (50%) had positive venom adaptation scores.

It is unclear whether our result is due to differences in sample size (four versus seven populations) impacting our ability to detect local adaptation or if rattlesnakes and ground squirrels experience different evolutionary dynamics in each region. Regardless, because we are interested in assessing the impact of demographic factors on levels of local adaptation we focus subsequent analyses on populations in the Southern lineage where local adaption is statistically detectable and limit our analysis of the quantitative relationship between demographic factors and the strength of local adaptation to these sites.

59

Table 5. General linear models results for analysis of local adaptation in the Northern and Southern genetic groups of Northern Pacific rattlesnakes and California ground squirrels. Terms included snake elevation class (snake elev), squirrel elevation class(squirrel elev), snake population (snake pop), squirrel population (squirrel pop), snake individual (snake_ID), squirrel individual (squirrel_ID), and a local adaptation defining the type of venom-serum combination as sympatric (S), allopatric same elevation class (AS), or allopatric different elevation class (AD). Northern Sites Source DF SS MS F-Value P-Value Snake Elev 1 1692041 1692041 21.98 0 Squirrel Elev 1 117397 117397 2.2 0.148 Snake Pop(Snake Elev) 2 883266 441633 5.74 0.007 Squirrel Pop(Squirrel Elev) 2 368998 184499 3.45 0.043 Snake_ID(Snake Pop) 36 2136394 59344 9.23 0 Squirrel_ID(Squirrel Pop) 36 1500025 41667 6.48 0 S-AS-AD term 2 12996 6498 1.01 0.368 Error 79 507717 6427 Total 159 7948420 R2 = 87.2

Southern Sites Source DF AdjSS AdjMS F-Value P-Value Snake Elev 1 258398 258398 1.64 0.205 Squirrel Elev 1 705817 705817 8.18 0.006 Snake Pop(Snake Elev) 5 3980062 796012 4.94 0.001 Squirrel Pop(Squirrel Elev) 5 1248617 249723 2.83 0.023 Snake_ID(Snake Pop) 63 8750587 138898 21.66 0 Squirrel_ID(Squirrel Pop) 63 4827174 76622 11.95 0 S-AS-AD term 2 58617 29308 4.57 0.011 Error 349 2238264 6413 Total 489 24660471 R2 = 87.1

60

60 Northern Southern 50

40

30

20

10

0

-10 Venom Adaptation Score Adaptation Venom -20

-30

Figure 10. Venom adaptation score (performance of sympatric versus allopatric snakes with serum of a given population’s squirrels) of each population. A positive venom adaptation score indicates local rattlesnake advantage, while a negative score indicates local ground squirrel advantage.

Comparing Demographic Parameters between Species

Estimates of genetic demographic parameters for rattlesnakes and ground squirrels from both the Northern and Southern groups indicates they consist of large and highly connected populations. Specifically, summed values of Nm, calculated as the sum of the point estimates for all Nm values for each incoming migration path, suggest Nm >

1 for all rattlesnake populations and all but one ground squirrel population, and Nm > 1.8 for all Southern populations of both species (Fig. 11). Therefore, if genetic demographic factors impact local adaptation, then coevolutionary theory predicts that the locally

61 adapted species (in this case rattlesnakes) should have larger Ne, and differences in Ne and not migration between rattlesnakes and ground squirrels should be positively associated with the magnitude of rattlesnake local adaptation (Fig. 6). We tested these predictions below using the Southern groups only, where local adaptation was statistically significant.

13 12 11 10 9 8

7 m

N 6 5 4 3 2 1 0 0 Rattlesnake1 Nm Squirrel2 Nm 3

Figure 11. Values of Nm estimated for eleven rattlesnake and ground squirrel populations in the Northern (gray) and Southern (black) groups. The horizontal dashed line indicates Nm = 1.

In terms of differences in population size, our demographic analyses showed that when comparisons are made between species at given locations, rattlesnakes have similar or larger Ne than ground squirrels. Based on comparisons of interquartile ranges with point estimates for each species, five rattlesnake populations are larger than sympatric ground squirrel populations, five populations of rattlesnakes and ground squirrels are not

62 distinguishably different in size, and only one population of snakes is smaller than its sympatric squirrel population (Fig. 12A). Based on point estimates alone, Ne values were higher for rattlesnakes than ground squirrels in 7 of 11 locations. In contrast, point estimates of m did not show a trend toward higher migration rates in rattlesnakes or ground squirrels. The interquartile ranges for non-parametric bootstrap estimates of each parameter included the point estimate of m from the other species in all but five comparisons of rattlesnake and ground squirrel m, and m was higher in the ground squirrels for three of the five where the species differed (Fig. 12B). For the point estimates of m, 7 of 11 sites had higher point estimates of m for rattlesnakes than for ground squirrels.

Figure 12. Estimates of a) Ne and b) m derived from parameter estimation under an island model in the program FastSimCoal2 for rattlesnakes versus squirrels. Northern sites are shown in gray, while Southern sites are black. Intervals around each point estimate of a parameter are the interquartile range of the same parameters from 25 bootstrapped datasets run under the same model in FastSimCoal2. The diagonal dashed line represents a one to one relationship between the parameters in each species. 63

Local Adaptation and Demography

Finally, we assessed whether local differences in Ne between Southern populations of rattlesnakes and ground squirrels were associated with variation in the degree of local adaptation in terms of venom adaptation scores. We found support for a relationship between local adaptation and Ne, but not m, consistent with predictions derived from theory when Nm is large for coevolving populations (Fig. 6). Specifically, we detected a significant positive correlation between the difference between rattlesnake

Ne and ground squirrel Ne at a given site and the venom adaptation score at the site (F1,5 =

16.2, P = 0.01, R2 = 76.4%, Fig. 13A). Functionally, this correlation indicates that for every increase in Ne of 1,000 snakes relative to a fixed squirrel population, rattlesnakes show a 1.8-unit increase in venom adaptation score. This effect size would translate to a

10.8% increase to the mean venom adaptation score and 3.7% of the highest venom adaptation score displayed by rattlesnakes at Chimineas Ranch Ecological Reserve, a location where rattlesnakes Ne is larger than ground squirrel Ne by nearly 24,000 animals.

While we predicted that higher relative migration rates should have a maladaptive effect on levels of local adaptation, the difference between rattlesnake m and ground squirrel m was not significantly associated with venom adaptation score (F1,5 = 1.58, P = 0.26, Fig.

13B), although the slope of a linear regression on these data was negative. Together, these results support the prediction from theory that, in coevolving species with Nm values >> 1, genetic demographic variation in the form of differences in how Ne impact the ability of local populations to adapt to coevolving enemies.

64

Figure 13. Scatterplots showing the difference in A) Ne and B) m point estimates at each site for rattlesnake snakes and ground squirrels calculated from FastSimCoal2 versus venom adaptation scores.

Discussion

We have combined measures of local adaptation with population genetic inferences from genome-wide SNPs to show that 1) the magnitude of coevolutionary interactions in the form of local adaptation between a rattlesnake predator and its squirrel prey is lineage-dependent and 2) within populations where increased local adaptation is present in the rattlesnake predator relative to its prey, there is a quantitative link between the variation in the genetic demography of snakes compared to squirrels and the degree to which rattlesnakes show local adaptation to squirrel resistance. The latter result is consistent with how demographic factors are predicted to impact coevolutionary interactions in large, high gene flow populations. Specifically, larger relative Ne in rattlesnakes versus squirrels quantitatively predicts the extent of rattlesnake local

65 adaptation to overcoming venom resistance. We discuss the significance of these findings below.

Lineage effects on local adaptation

Our analysis shows significant differences in the degree of local adaptation in populations from largely coincident, phylogeographically distinct lineages of rattlesnakes and squirrels. In particular, we are unable to detect significant local adaptation of either species in the Northern lineage, whereas the Southern groups show higher levels of local adaptation in rattlesnakes consistent with the pattern described in Holding et al. (2016a).

We suggest several possible reasons for this difference. First, it could be a simple consequence of differences in sample sizes of analyzed populations between lineages, such that sampling additional populations in the Northern genetic group would lead to similar patterns of functional variation in both the North and South. Second, if real this difference could be due to ecological or environmental differences that result in reciprocal selection being weak or absent across much of the North. For example, we find evidence of a lower abundance of ground squirrels at three of four of the North populations relative two Southern populations (M. Holding, unpublished data). This could lead to broad-scale coevolutionary cold-spots in the North because of local snakes having reduced interactions with ground squirrels due to targeting alternative prey species (Macartney 1989). Furthermore, previous studies of venom resistance in an

Oregon population of California ground squirrels, where rattlesnakes are rare showed 40- fold reductions in resistance (Poran et al. 1987). Taken together, overall lower abundance

66 of one or both species could lead to lower strength of coevolutionary selection in the north and render it a coevolutionary cold-spot (Gomulkiewicz et al. 2000; Thompson

2005a). Future work involving denser sampling of Northern and Southern populations will be necessary to determine if altered patterns of local adaptation exist at the regional phylogeographic scale. Such a pattern would facilitate a multi-level reciprocal study of the impacts of lineage formation for geographic mosaics of coevolution.

Regardless of the specific causes, our finding of differences in the strength of local adaptation in rattlesnake and squirrel population in different lineages supports previous host-parasite studies that emphasize careful consideration must be given to the appropriate scale at which local adaptation should be measured given both the evolutionary history of the focal taxa and the biological question at hand (Burdon &

Thrall 2000; Cogni & Futuyma 2009; Koskella et al. 2011; Laine 2005; Penczykowski et al. 2016; Tack et al. 2014). For example, patterns of local adaptation vary within and between geographically separated metapopulations of plants and pathogens (Laine 2005;

Tack et al. 2014) and in hyperparasitic bacteriophages and bacteria among but not within the same host tree (Koskella et al. 2011). In terms of our previous result for this system

(Holding et al. 2016a), including a major phylogeographic break within our sampled region likely introduced a historical component of lineage splitting and evolution in allopatry that obscures our understanding population-level local adaptation as it is impacted by levels of gene flow among closely related populations or demes. This problem is exemplified by work on resistance to pathogens that show a non-linear relationship between genetic distance to source hosts and host resistance, produced by

67 contrasting impacts of local adaptive evolution and evolutionary divergence in allopatry

(Antonovics et al. 2013), and recently hypothesized to operate among coevolving rattlesnake and squirrel species (Pomento et al. 2016).

To describe processes contributing to the geographic mosaic of coevolution among populations, it is important to focus on sets of populations that exchange migrants via gene flow (Penczykowski et al. 2016), while controlling for any deeper evolutionary divergences between populations. Planning major sampling of populations using both natural history data and surveys of phylogeographic and population genetic structure in the interacting species can guide researchers to the appropriate geographic scale for addressing specific coevolutionary questions about interacting populations

(Penczykowski et al. 2016; Richardson et al. 2014).

Genetic demography and local adaptation in coevolving systems

The expected role of gene flow in local adaptation is strongly modulated by Nm

(Gandon & Nuismer 2009; Yoder & Nuismer 2010), since adaptive gene flow hinges on the potential to reintroduce variation previously lost to genetic drift. When genetic drift is weak, migration is most likely to be a maladaptive force counteracting local adaptation by homogenizing adaptive variation (Forde et al. 2007b; Slatkin 1985). The Ne estimates of rattlesnakes and ground squirrels indicate large population sizes in excess of 1,000 individuals with Nm values > 1 for both species. Therefore, variation in these large populations is unlikely to be lost to drift and low frequency venom or resistance alleles should remain available for future adaptive evolution if local selective regimes shift.

68

Under these “large population” conditions (Fig. 6), theory predicts that the degree of difference in Ne between rattlesnakes and ground squirrels strongly predicts the direction and magnitude of local adaptation in the venom-driven predator-prey interaction and that migration is largely maladaptive.

We found support for both predictions. For the seven Southern populations, the four sites where rattlesnake Ne was higher than squirrel Ne were the sites with the highest magnitudes of rattlesnake local adaptation, while the San Joaquin Experimental Range and Cantua Creek sites had larger populations of ground squirrels and rattlesnakes that were the least locally adapted (local rattlesnake maladaptation in the case of the San

Joaquin Experimental Range). When combined, these results show a significant positive relationship between differences in rattlesnake and squirrel population size and degree of snake local adaptation. We interpret this result to reflect the fact that larger Ne facilitates local adaptation by increasing the efficiency of selection (Fisher 1930; Petit & Barbadilla

2009; Slotte et al. 2010). In other words, a given population of rattlesnakes is therefore expected to track the evolution of its local squirrel prey in accordance with the difference in population sizes. In contrast, we did not find a significant effect of the difference in m values between local rattlesnakes and ground squirrels on the magnitude of local adaptation. However, the slope of the regression was negative, suggesting migration is either unimportant or potentially maladaptive for local adaptation in this system. These results underscore the importance of a comprehensive knowledge of population demography in studies that propose to test theoretical predictions about coevolutionary adaptation, since the predictions change both quantitatively and qualitatively under

69 different demographic contexts. The direct links between continuous measures of population demography and variation in local adaptation we provide are rare, with previous studies limited to comparisons of the physical distance among populations

(Adiba et al. 2010; Kaltz et al. 1999). Our approach allows both a direct test of the predictions from various coevolutionary hypotheses and allows estimation of the effect size on local adaptation in a particular system.

Life history characteristics support the result that a rattlesnake predator can have a larger Ne than a rodent prey, which is a nonintuitive result because prey are generally expected to have larger census sizes than predators due to differences in life history characteristics and trophic position. However, these differences in genetic demography are consistent with ecological characteristics of these specific taxa. First, ground squirrels are more specialized in their habitat requirements as burrow-dwellers, preferring open oak woods with suitable soil for burrowing while avoiding thick chaparral, open grassland with shallow soil, and steep rocky areas (Evans & Holdenried 1943; Fitch

1948; Owings & Borchert 1975). Rattlesnakes, on the other hand, can be found in areas ranging from thick chaparral to dry rocky slopes. Hence, squirrels can be locally abundant, but rattlesnakes are more wide spread with higher levels of connectivity

(supported by our Nm estimates) due to broader habitat preferences. Second, the local census size of rattlesnakes may approach or exceed that of ground squirrels in some areas. The density of rattlesnakes and ground squirrels was estimated on the San Joaquin

Experimental Range by intensive field surveys in the 1940s (Fitch 1948, 1949) in which there were 2.9 rattlesnakes and 3.7 ground squirrels per hectare in this area. These high

70 rattlesnake densities relative to prey densities are possible for sit-and-wait foraging ectotherms like these snakes that require only two or three meals per year (Diller &

Wallace 1996; Secor & Nagy 1994; Taylor et al. 2005; Wallace & Diller 1990). Third, these life history features may also make ground squirrels more susceptible to frequent population size fluctuations induced by environmental factors, such as California’s frequent droughts (Griffin & Anchukaitis 2014), which can lead to reduced ratios of Ne to census size (Luikart et al. 2010; Vucetich et al. 1997).

Since the work of Gandon and his colleagues’ which showed that gene flow can facilitate local coevolutionary adaptation (Gandon 2002; Gandon et al. 1996; Gandon &

Michalakis 2002), a series of natural host-parasite systems have used measures of genetic distance, FST, optimal genetic clustering results, or life history information to infer higher average gene flow in either host or parasite and used this information to predict which participant will be locally adapted (Brandt et al. 2007; Couchoux et al. 2016; Ganz &

Washburn 2006; Oppliger et al. 1999; Prugnolle et al. 2005; Roth et al. 2012; Sullivan &

Faeth 2004; Sutton et al. 2016). These studies provide important comparisons of host and parasite population structure and sometimes pair it with information on the sign and magnitude of local adaptation. However, we see two issues with this past approach. First, neutral differentiation between populations is the joint product of levels of gene flow and the power of genetic drift within each population, so genetic differentiation alone is insufficient evidence for differential rates of gene flow between two species since such differences may also be due to differences in Ne. Coalescent analyses that jointly model migration and effective population size (Beerli & Palczewski 2010; Excoffier et al. 2013)

71 are necessary to accurately infer differences in gene flow between species. Converting

FST to Nm assumes, often incorrectly, large population sizes and symmetrical gene flow

(Whitlock & Mccauley 1999), and comparing the number of genetic clusters between hosts and parasites can be confounded by smaller Ne in one species, different histories of lineage splits and admixtures, and different strengths of selection against migrants for reasons external to the coevolutionary process.

Second, as discussed above, the prediction that the species with higher migration will be locally adapted is context-dependent. For situations with large Ne and highly connected populations, the theory instead predicts that gene flow is maladaptive and the species with the higher Ne should be the locally adapted species, all else being equal

(Gandon & Nuismer 2009; Nuismer et al. 2010). Two meta-analyses of host-parasite interactions support a role for higher relative gene flow in facilitating local adaptation to enemy species (Greischar & Koskella 2007; Hoeksema & Forde 2008), but both use population structure as an indirect measure of gene flow. The high gene flow-increased local adaptation relationship may include cases where Ne is much higher in one species, representing a hidden variable that leads to both local adaptation and the outward appearance of higher connectivity. Carlsson-Granér and Thrall (2015) provided a robust link between gene flow and adaptation to enemy attack by surveying populations of the plant Viscaria alpina that varied in their degree of connectivity from isolated to continuous, and assaying resistance to the anther-smut fungus (Microbotryum violaceum). Resistance was higher in the more connected plant populations suggesting a role for connectivity in adaptation. However, the authors point out that selection for

72 resistance might be stronger in the connected populations because the fungus is more common, and the connected populations also tended to have larger population sizes.

Clearly, hypothesis-driven empirical work on the factors leading to local adaptation in species interactions would be most productive when predictions from coevolutionary theory are assessed in the context of empirical estimates of the genetic demography of interacting species as we describe in Figure 6.

Demographic advantage in Ne and demographic equivalence in m help to explain local adaptation of rattlesnake predators to their ground squirrel prey, but are not an exclusive explanation for this pattern. These demographic advantages likely combine with the high evolvability of rattlesnake venom and potentially lower evolvability of ground squirrel venom inhibitors to produce the patterns observed here and by Holding et al. (2016a). Venom is produced within an isolated gland that minimizes any pleiotropic effects of adaptive mutations in venom genes and hence constraints on evolvability

(Casewell et al. 2012a; Casewell et al. 2012b; Margres et al. 2015b). Meanwhile, serum- based resistance factors occur in the blood, and may be subject to physiological trade-offs with other key functions of blood in transport or disease resistance (Eastman et al. 2012;

Westergaard et al. 1970). Venom is a complex secretion, where an individual rattlesnake can express up to 40 proteins in its venom (Mackessy 2010). In contrast, to date only two proteins have been identified as responsible for venom resistance in squirrel serum

(Biardi et al. 2011a). This difference is significant because trait complexity (Macpherson et al. 2015) and a lack of functional constraints (Schluter 1996) may facilitate the rate at which adaptations can evolve, contributing to the evolutionary advantage of rattlesnakes

73 over ground squirrels. Clearly, simplifications such as the Life-Dinner Principle

(Dawkins & Krebs 1979) or comparisons of generation time will often fail to predict the more locally adapted player in a set of coevolving species since this requires detailed knowledge of the phenotypes involved in an interaction and the demography and evolutionary history of the participant species (Abrams 1986; Brodie & Brodie 1999;

Thompson 2005b).

Conclusion

Studies in several natural systems have quantified the degree of local adaptation in several individual populations (e.g. Kaltz et al. 1999; Roth et al. 2012; Thrall et al.

2002), but only qualitatively link the pattern of local adaptation to demographic or life history characteristics of the species involved. We have demonstrated how combining high resolution genetic data with estimates of the magnitude of local adaptation can provide a quantitative link between key factors theorized to impact local adaptation and actual measures of local adaptation in natural systems. Without this type of analysis, we must await meta-analyses to evaluate quantitative relationships between population sizes or gene flow estimates and local adaptation (Greischar & Koskella 2007; Hoeksema &

Forde 2008; Yates & Fraser 2014). This study supports a growing body of theoretical studies that suggests differences in Ne between antagonists are crucial to coevolutionary adaptation across various classes of interaction, while larger m values are only adaptive in a subset of cases (small populations and discrete traits that are often lost to genetic drift;

Nuismer et al. 2007; Ridenhour & Nuismer 2007; Yoder & Nuismer 2010). Many

74 predators and prey interact with complex chemical secretions or morphological adaptations such as tooth and claw, which may tend to involve more quantitative traits, and the frequent genetic bottlenecks associated with parasite ecology are less common in other organisms. Predator-prey and host-parasite interactions are antagonistic, exploitative, and often compared, but the two classes of interaction may fundamentally differ with respect to the importance of population size dynamics versus gene flow in deciding the outcomes of local coevolution.

75

Chapter 4: Assessing biotic and abiotic drivers of divergence in venom composition among populations of the Northern Pacific rattlesnake.

Abstract

Assessing the biotic and abiotic correlates of population divergence in functional traits can provide insights into the evolutionary mechanisms that generate local adaptation among populations. An example of such a trait is the venom used by predators to capture prey. Venom shows significant divergence among populations of the same species across many taxa and yet the underlying causes of this general pattern remain unclear. Here, we assess patterns of population differentiation at the protein specific level for individual venoms from Northern Pacific rattlesnakes (Crotalus oreganus) from 13 locations across California and then assess the relative importance of major biotic (prey species community composition), abiotic (temperature, precipitation, and elevation) and genetic (genetic distance) factors as drivers of population divergence in venom phenotypes. We found that over half of the variation in venom is due to divergence among populations and that this differentiation occurred along axes that define previously observed functional trade-offs between classes of venom proteins that have neurotoxic and myotoxic (phospholipase A2 – PLA2) and hemorrhagic (metalloproteinase – MP) functions. A multivariate analysis shows that biotic, abiotic and genetic measures all explain a significant amount of population divergence in venom but vary in their relative

76 importance. Surprisingly, genetic differentiation among populations was the best predictor of venom divergence accounting for 46% of overall variation, whereas differences in prey community composition and abiotic factors explained smaller amounts of variation (23% and 19%, respectively). We suggest the strong impact of genetic differentiation on venom composition may reflect an isolation by environment effect where selection against recent migrants is strong when they come from populations locally adapted to different environments. This results in a correlation between neutral genetic differentiation and venom differentiation due to selection and not genetic drift.

Our analyses suggest that using coarse estimates of prey community composition can be useful in understanding the potential selection pressures acting on patterns of venom protein expression and that in these rattlesnakes, which have previously been shown to be locally adapted to a single prey species can simultaneously adapt to broader range of prey. Overall, our results provide a clear example of how multiple factors drive differences between populations in venom composition through distinct mechanisms and suggest that factors other than adaptation to spatial variation in prey abundance need to be considered as explanations for population divergence in venom.

Introduction

Functional traits that show high levels of genetically-based divergence between populations represent adaptations that have evolved through evolutionary processes such as local adaptation (Hoekstra et al. 2006; Rundell & Price 2009). Understanding the biological and environmental factors that are responsible for population differentiation in

77 functional traits can provide insights into the selection pressures that generate adaptive differences between populations that in turn influence fundamental evolutionary processes such as local adaptation and ecological speciation (Nachman et al. 2003;

Schluter 2009; Thompson 2005b). Animal venoms represent one such type of functional trait that shows high levels of divergence at the population level, the causes of which are poorly-understood (Casewell et al. 2012b; Gibbs & Chiucchi 2011; Holding et al.

2016b). Venoms are complex biochemical secretions that mediate predator-prey interactions in animals as diverse as cnidarians, arthropods, mollusks, and vertebrates

(Casewell et al. 2012b). In venomous predators, the venom is injected directly into potential prey and interacts with aspects of prey physiology to produce the fitness outcome of the interaction for both participants (i.e. prey death or prey escape). Animal venom therefore provides an explicit link in an adaptive phenotype between a predator and its prey, and provides the opportunity to elucidate the contribution of various aspects of prey, the abiotic environment, and other evolutionary forces to the generation and maintenance of divergence in complex phenotypes at the intraspecific level.

Past work has emphasized three main factors as determinants of intraspecific venom divergence: biotic factors in the form of selection pressures related to population differences in diet (biotic), specific environmental factors such as temperature and elevation (abiotic) and genetic factors reflecting population history (genetic). Because venom is a predatory adaptation, selection related to biotic factors in the form of diet variation is thought to be the major driver of venom evolution. For example, population- level differentiation in venom composition of Malayan pitvipers (Calloselasma

78 rhodostoma) was best predicted by differences in the relative abundances of birds, mammals, reptiles, and amphibians in the diets of these snakes (Daltry et al. 1996).

Among-population variation in venom composition has since been linked to geographic differences in diet in other snakes (Creer et al. 2003) and cone snails (Conus sp.; Chang et al. 2015; Duda et al. 2009). Given evidence for prey-specific effects of venom differences (Barlow et al. 2009; Bernardoni et al. 2014; Pawlak 2006) the population differences in venom seem to reflect functional differences related to the utilization of different prey (Casewell et al. 2012b). The inferences drawn in previous studies may have been effected by inherent biases that can occur in diet studies, such as collection efforts focused on habitats that are easy to access or effects that diet may have on their detectability. Both are problematic for accuracy in quantifying diet, as collection efforts often involved collecting snakes as they crossed roads and particularly large prey items will encumber snake movement and induce them to remain hidden. Studies that consider variation in the local prey community as a whole can circumvent these biases and determine whether and how venom responds to differences in the available prey species.

Other studies have suggested two other types of factors as potential drivers of venom divergence: abiotic environmental variation that selects for predigestive and other non-killing functions of venom (Mackessy 2008, 2010) and neutral evolutionary processes related to the amount of time that populations have been isolated from each other (Williams et al. 1988). Abiotic environmental variation has been hypothesized to play a direct role in venom evolution because ectotherms inhabiting areas with suboptimal ambient temperature regimes may be selected to maintain pre-digestive

79 functions of venom (Mackessy 2008, 2010). Indeed, Holding et al. (2016a) found that baseline venom metalloproteinase activity among Northern Pacific rattlesnake populations was associated with site elevation, and thus potentially temperature regimes.

Furthermore, another study of Northern Pacific rattlesnake venom found that temperature and precipitation data outperformed genetic and geographic distance measures as predictors of venom variation (Gren et al. 2017). Finally, Williams et al. (1988) demonstrated that for island populations of tiger snakes (Notechis sp.) the best predictor of venom differentiation was the depth of the seas between islands, suggesting that isolation time, and therefore neutral evolutionary processes is the main force driving venom differentiation in these snakes. While Williams et al. (1988) used a geologic measure of isolation time to approximate neutral divergence, most studies that reject neutral divergence as a causal hypothesis use mitochondrial cytochrome b sequence data

(Creer et al. 2003; Daltry et al. 1996; Gren et al. 2017). Given the smaller effective population sizes, maternal inheritance, and single-gene nature of previous mitochondrial studies (Palumbi & Baker 1994), a multi-locus estimate of neutral genetic differentiation would be valuable to explore the degree to which genetic differentiation explains venom differentiation.

Although the bulk of current evidence focuses on explaining venom differences between populations due to adaptation to different prey, the previously mentioned biases in diet studies and potential problems associated with single gene mitochondrial estimates of genetic differentiation leave room for additional studies of population-level divergence in venom. Here, we fill this gap through work that quantifies between-population

80 differentiation in venom among 13 populations of a venomous snake, and then assesses biotic, abiotic, and neutral genetic correlates of the patterns we observe. We first report the extent of population differentiation in venom composition and identify the proteins involved in differentiation. We then use constrained ordination for an integrated test of three hypotheses regarding venom compositional differentiation: that differentiation is driven by 1) geographic differences in prey community composition, 2) geographic differences in the abiotic environment, and 3) the degree of genetic divergence between populations.

Specifically, we studied the Northern Pacific rattlesnake (Crotalus oreganus), a large rattlesnake species distributed along the West Coast of the United States. The venom of this snake is known to be hemotoxic and myotoxic, and is largely comprised of type I and type III snake venom metalloproteinases (MP-I and MP-III), serine proteinases

(SP), disintegrins, and small basic polypeptide myotoxins (Mackessy 2008, 2010). A recent analysis of venom variation in snakes from across Central California, spanning the distributions of C. oreganus and closely related C. helleri, showed that abiotic environmental variation was superior to mitochondrial genetic distance and geographic distance as a predictor of venom compositional variability (Gren et al. 2017). While this work supports adaptive population-level venom divergence, the authors concede that abiotic environmental variation could exert selection on venom directly or act indirectly by impacting which prey species are present in an area. Gren et al. (2017) did not attempt to estimate local prey community variation, leaving this question for future study. Also, genetic distance at a single mitochondrial gene may poorly estimate neutral

81 differentiation, so the strength of genetic differentiation as a predictor of venom variation in these rattlesnake species merits further consideration.

To assess diet as an explanation for venom differentiation, we focus on assessing the small mammal community across multiple populations of C. oreganus because this snake is a small mammal specialist (90% of known diet items). A meta-analysis of several local diet studies revealed at least 21 small mammal species are consumed, with each study reflecting diet differences associated with regional variability in the species present (Sparks et al. 2015), suggesting that these snakes feed opportunistically on all small mammal taxa in a given location. Habitat heterogeneity is high along the West

Coast, leading to rapid changes in community composition over short distance (e.g. transitions from dry San Joaquin valley grassland to cooler conditions in the high elevations of the Coast Ranges and Sierra Nevada) (Kerr & Packer 1997). The presence of well-known phylogeographic breaks at the Transverse Ranges, San Francisco Bay, and

Cascades Ranges enhance the effects of habitat heterogeneity to produce rapid changes in mammal communities across the region (Calsbeek et al. 2003; Davis et al. 2008), and these may predict variation in the composition of C. oreganus venom. The dynamic landscape of California also presents significant amounts of abiotic environmental variation across elevational and latitudinal gradients, which could directly impact venom through selection for pre-digestive function where temperature varies or act indirectly by influencing the prey community. However, if environmental variation is simply a proxy measure for prey community variation, then we would expect direct information on prey community composition to outperform abiotic environmental information as a predictor

82 of venom divergence. Finally, we have previously shown (Chapter 3) that there is a relatively high level of genetic differentiation among our study locations with pairwise

FST values ranging from 0.04 to 0.27 (also see Goldenberg 2013), raising the possibility that genetic drift impacts venom divergence across these variably isolated rattlesnakes.

Our approach yields an integrated assessment of whole venom divergence and its potential causes in C. oreganus.

Methods

Sampling

We collected venom samples in the field at 13 locations in California. At each site, we conducted visual searches of suitable habitat to locate rattlesnakes and captured them using metal tongs and cloth bags. Within 24 hours of capture, the snakes were induced to bite a plastic-covered beaker to obtain a venom sample, which was immediately frozen in liquid nitrogen. The snakes were then released at their site of capture. We limited our collection to snakes greater than 60 cm snout-to-vent length

(SVL) to control for the significant ontogenetic shift in venom composition that occurs in this species as adult-size snakes shift toward a mostly mammalian diet (Mackessy 1988).

We obtained at least 10 adult venoms samples per site at all locations except San Joaquin

Experimental Range (n = 7). We released all snakes at their site of original capture after samples were obtained. The Ohio State Institutional Animal Care and Use Committee

(protocol #2012A00000015) approved our capture and sampling procedures.

Assessing venom composition using HPLC and SDS-PAGE

83

We analyzed the protein composition of each venom sample with reversed-phase high-performance liquid chromatography (RP-HPLC) on a Beckman System Gold HPLC

(Beckman Coulter, Fullerton, CA) and used Beckman 32 Karat Software v.8.0 for peak abundance quantification. Sample preparation and chromatographic separation methods were identical to those of Margres et al. (2014). We identified 34 separate RP-HPLC peaks in the focal venom samples from among our entire set of 127 C. oreganus venom samples (Fig. 14). We quantified the area under each peak relative to the total area of all identified peaks, which is a useful proxy for the relative amount of protein in a specific peak by weight (Margres et al. 2015b).

84

Figure 14. Representative HPLC chromatogram showing the 34 venom protein peaks quantified. Peaks that were defined as regions as opposed to individual peaks, or those that were missing in this venom sample are marked with a number and bracket showing the elution region over which that particular peak was measured.

To identify the proteins present in our focal samples, we loaded a NuPAGE 10%

Bis-Tris precast gel (Invitrogen) with each of the HPLC fractions collected from a pooled sample of six individual C. oreganus from across the species range which held all of the

34 peaks quantified (Appendix B.1). We rehydrated these lyophilized fractions in 10 ul of 85

HPLC-grade water, then combined 6.5 ul of this venom dilution with 2.5 ul of LDS buffer and 1ul of NuPAGE reducing agent (Invitrogen). The mixture was heated to 70°C for 10 min. We electrophoresed the samples in MES SDS Running Buffer at 100 V for

120 mins alongside a Mark12 Unstained Standard protein ladder (Life Technologies). We then stained gels with Simply Blue SafeStain (Invitrogen) using the manufacturer’s protocol and obtained photographs of each gel. Using these images, we estimated the molecular weight of each band using the GelAnalyzer 2010a software

(www.gelanalyzer.com) and used the protein size ranges of Mackessy (2008) to assign bands in each peak to a particular protein class (Appendix A.1). Many collected peaks yielded more than two gel bands. In these cases, we report the putative identity of the two most intense bands.

Differentiation in venom composition between populations

Our first goal was to characterize the geographic differentiation in venom composition present, and to quantify the variation explained by clade (North or South) versus specific population of origin. Venom differentiation among phylogenetically distinct clades (Gibbs et al. 2013; Lomonte et al. 2014) and geographically distinct populations (Daltry et al. 1996; Gibbs & Chiucchi 2011; Margres et al. 2015b) of snakes, as well as individual snake size (Mackessy 1988; Wray et al. 2015) and sex (Furtado et al. 2006) have all been reported to covary with the composition of snake venoms. We used permutational multivariate analysis of variance (PERMANOVA) as implemented in the adonis function of the vegan package (Dixon 2003) in R v. 3.2 (R Development Core

86

Team 2015) to test the significance and magnitude of each of these effects on venom composition in a linear modelling framework. Our PERMANOVA analysis used

Euclidean distances on the ilr-transformed venom peak abundances and calculated statistical significance of each factor using 10,000 permutation of the raw data. The permutations were stratified across clades, because the individual populations are nested within the Northern and Southern clades (Wray et al. 2015). After confirming overall population differentiation in the venom phenotype existed, we determined which peaks differed among populations. To achieve this we carried out separate two-way ANOVAs on the clr-transformed mean abundances of each HPLC peak, with clade (North vs.

South) and population of origin (nested within clade) as factors.

We assessed the differentiation in venom associated with geographic factors

(clade and population) and then identified venom peaks responsible for this differentiation via ordination. Ordination was accomplished through robust principal components analysis of the compositional data (Filzmoser et al. 2009) in R, using the pcaCoDa function in the package robCompositions (Templ et al. 2011). We used linear regression of the clr-transformed abundance of each peak versus the PC1, PC2, and PC3 scores as an assessment of significant contribution to each PC, and we report the peaks with p > 0.05 and R2 > 20% instead of choosing an arbitrary loading value cutoff for reporting peaks as associated with a given PC (Gibbs & Chiucchi 2011).

87

Biotic - Prey community variation

To quantify variation in local prey community composition, we consulted published accounts of C. oreganus diet to establish which species to consider prey for rattlesnakes at a given site (Fitch & Twining 1946; Sparks et al. 2015; Wallace & Diller

1990). Based on these studies, we included the following small mammal prey species in our analysis: two rabbits (genus Sylvilagus), five shrews (genus Sorex), and 37 small to medium sized rodents (Appendix A.2).

We classified each of these species as present or absent at each of our study sites based on range map overlap. Range map overlap is a coarse measure of species presence, and therefore can systematically overestimate biodiversity at small scales due to habitat heterogeneity, but performs acceptably at coarser scales and produces estimates that are strongly positively correlated with true diversity across scales (D'amen et al. 2015;

Pineda & Lobo 2012). The California Wildlife Habitat Relationships project (California

Department of Fish and Wildlife) provides detailed, georeferenced range maps for each terrestrial vertebrate in the state, and we viewed these maps in ArcGIS v. 10 (ESRI,

Redlands, CA, USA) along with points at the geographic center of each of our study sites.

A species was marked as present if its range included a given site.

Next, we obtained scores in multivariate prey space for each of our study sites by subjecting the matrix of binary presence/absence data for 44 small mammals to non- metric multidimensional scaling (NMDS) using the metaMDS function in the vegan package (Dixon 2003), and we recorded the NMDS axes 1 and 2 scores for each location.

88

Abiotic - Environmental variation

To quantify variation in the abiotic environment at each location, we followed the approach used by Gren et al. (2017) on the same rattlesnake species by extracting principal component scores from a matrix of temperature, precipitation, and elevation variables for each site. We downloaded 30 arc-second raster grids of all 19 Bioclim temperature and precipitation variables from the WorldClim database

(www.worldclim.org) and extracted the values at each of our sampling locations. The site elevation at the centroid of each of our sampling locations was added to the Bioclim data, to generate a 13 site x 20 abiotic environmental variable matrix. We subjected this environmental matrix to standard principal components analysis using the prcomp function in the R base package (R Development Core Team 2015), and extracted the PC1

(46% of variation), PC2 (38%), and PC3 (7%) scores for each location.

Genetic- Population differentiation

Finally, we used RADseq SNP data to estimate genetic divergence between snake populations. To use genetic data as an independent explanatory variable in redundancy analysis, the SNP dataset had to be reduced to a limited measure for each population where differences would reflect genetic differentiation between the populations. Principal components analysis of the SNP matrix is increasingly being used to extract key axes of differentiation between populations, summarizing these data into a few principal components (Bryc et al. 2010; François et al. 2010; Ma & Amos 2012; Price et al. 2006).

We used PCA to obtain each population’s average position in putatively neutral genotype

89 space using the ddRADseq loci for each location generated as described in Chapter 3, with the addition of 8 snakes from Montana de Oro State Park that were not included in the previous work. We used the dapc function in the R package adegenet (Jombart 2008) to generate principal components of our ddRADseq data, and recorded the average scores of each rattlesnake population on genetic PC1 and PC2, which collectively explained

33% of the total genetic variation present.

Assessing drivers of population differentiation in venom

Using the ordination scores from the first two axes of prey community, environmental, and genetic variation, we tested the relative ability of each of these factors in explain variation in average venom composition among populations. We calculated the population average abundance of each peak in ilr-transformed venom space, and used conditioned Redundancy Analysis (RDA) implemented through the rda function in the vegan package for the analysis. The conditioned RDA controls for the effects of one set of variables before performing traditional RDA on the residual matrix. In this case, we conditioned on genetic variation before testing for impacts of the biotic prey community or abiotic environmental data. Past studies have used Mantel tests to examine relationships between venom differentiation and attributes of populations (Gibbs &

Chiucchi 2011; Gren et al. 2017) but this approach has been criticized as inappropriate when distance measures (e.g. genetic distance) are analyzed instead of dissimilarity measures (Legendre & Fortin 2010; Legendre et al. 2015), and conditional RDA has been suggested as more appropriate analog of the partial Mantel test (Legendre et al. 2011).

90

The RDA approach estimated the explanatory power of each independent variable and tested for significance using 1000 permutations.

Based on high variance inflation factors and bivariate correlations, we found the first two NMDS scores of the prey community to strongly covary with the first two PC scores of environmental variation, which we interpret to mean they summarize similar underlying information. To account for this, we conducted two separate RDAs, one including the prey scores and the second including the abiotic environmental scores, while conditioning the ordinations on the neutral genetic PCs. Comparing the results in terms of the amount of variance explained allowed us to determine whether prey community or abiotic environmental data is the better predictor of venom variation.

Results

Differentiation in venom composition

The venom composition of Northern Pacific rattlesnakes varied with both the geographic location of sampling and, to a small extent, the size of the snake, with location explaining over half of the variation in our dataset. Our PERMANOVA analysis quantified the amount of multivariate venom variation explained by three factors: clade

(North vs. South), population, and snake size (Table 6). Over 50% of the variation among individual HPLC venom profiles is explained by the two geographic factors: 18.5% of the variation associated with the Northern and Southern clades present on opposite sides of

San Francisco Bay and an additional 31.9% is explained by the specific population from which the venom sample was collected. Despite including venom from only adult

91 rattlesnakes, snake size explained a small (2.9%) but significant amount of venom general compositional variation, suggesting that the ontogenetic shift in this species’ venom is not fully complete at 60 cm. Rattlesnake sex was not a statistically significant predictor of venom variation (p = 0.19) as it has been in some other viper species

(Menezes et al. 2006). Our post-hoc analyses of each individual venom peak show that each peak varied significantly between clades, between populations within each clade, or both (Appendix B.2 and B.3).

We used principal components analysis to visualize the geographic differences in venom composition present and to reveal specific venom peaks associated with this differentiation (Fig. 15). The first principal component (PC1) of venom variation accounted for 26% of the variation present, and delineates rattlesnakes from the North and South clades on either side of the San Francisco Bay, although there is also significant overlap with rattlesnakes from the eastern side of major river drainages in

California’s Central Valley (Sutter Buttes and San Joaquin Experimental Range). The second principal component (PC2; 17% of venom variation) separates individual populations in both the North and South regions. Peaks containing each of the major rattlesnake venom components (Disintegrins, SPs, MPs, and PLA2) loaded heavily and significantly onto these first two PC axes (Table 7), indicating that mixed sets of proteins contribute to these major axes of variation consistent with results from another rattlesnake species, Sistrurus catenatus (Gibbs & Chiucchi 2011). Both PC1 and PC2 were characterized by the previously-documented tradeoff between MP and PLA2 venom proteins that is well-documented in rattlesnake venoms (Mackessy 2010). Higher scores

92 on PC1 were associated with higher amounts of four MP peaks and lower amounts of four PLA2 peaks while higher scores on PC2 were associated with lower amounts of three

MP-rich peaks and higher amounts of two PLA2 peaks.

Table 6. Results of PERMANOVA analysis explaining variation in the venom phenotype of Northern Pacific Rattlesnakes from 13 populations in California. Factor d.f. Mean F P-value R2 Squares Clade 1 25629.5 43.6 < 0.0001 18.5 Population 11 4018 6.83 < 0.0001 31.9 SVL 1 4097 6.97 < 0.0001 2.9 Sex 1 765.3 1.30 0.192 N/A Residual 108 588 46.0

Figure 15. Biplots of first two PCs from robust principal components analysis of venom variation. Left: Samples coded a Northern (blue) or Southern (red) clade of origin. Right: Samples coded by population of origin. Confidence ellipses show one standard deviation from the mean of each group.

93

Table 7. Loadings and R2 values for venom peaks with significant correlations with Principal Components axes 1, 2, or 3 and which have more than 20% of their variance explained by that peak. Axis (% Peak I.D. R2 Axis Loading Variance) PC 1 (26%) 7 – PLA2 59.7 -0.392 8 – PLA2 38.8 -0.239 9 – PLA2 48.9 -0.196 11 – Disintegrin 30.5 -0.230 12 – PLA2 37.7 -0.355 13 – SVMP, SP 42.4 0.124 18 – SP 29.9 0.222 19 – SP 24.2 0.198 20 – PLA2 38.9 0.277 24 – SVMP 20.3 0.099 25 – SVMP 25.1 0.193 29 – SVMP 22.5 0.250 31 – SVMP 35.2 0.269

PC 2 (17%) 8 – PLA2 29.7 0.243 11 – Disintegrin 23.5 0.243 12 – PLA2 28.1 0.406 23 – SVMP 39.2 -0.360 28 – SVMP, 22.6 -0.287 LAAO 33 – SVMP 52.0 -0.632

PC 3 (13%) 10 – PLA2, 34.2 -0.120 Lectin 12 – PLA2 21.3 0.428 15 – SP 28.3 -0.246 18 – SP 26.5 0.378 27 – SVMP 56.6 -0.550

Drivers of population differentiation

Both genetic differentiation among populations and local prey community composition are powerful predictors of geographic variation in Northern Pacific rattlesnake venom composition (Table 8; Fig. 16). Genetic differentiation between 94 populations was the best predictor of ilr-transformed population venom variation in the conditioned RDA analysis. Specifically, the first two PCs that summarize population genetic variation explained nearly half of the among-population variation in rattlesnake venom (Table 8). A plot of pairwise FST against pairwise venom dissimilarity for all pairs of sites (Fig. 17) shows a non-linear relationship between genetic and venom differentiation. Specifically, there is a positive association between pairwise FST and venom differentiation when comparisons are made within the North and South clades, but the relationship reaches an asymptote at FST ~ 0.15 such that in between-clade comparisons, more genetically differentiated populations are not expected to be more or less different in their venom composition.

When we conditioned the RDA on genetic data, prey community variation explained a moderately large and significant portion of the remaining variation (R2 =

23%, p = 0.02). Of the 23% of overall venom variation explained by the prey community, the first axis of prey community variation accounted for the majority (68%, p = 0.002).

The second axis of prey community variation accounted for the remainder (32%), but it was a marginally non-significant predictor (p = 0.06). Lower scores on the first prey community axis were associated with the presence of three kangaroo rats (Dipodomys nitratoides D. hermanni, and D. ingens) five mice (Peromyscus eremicus, P. fraterculus,

P. crinitus, Perognathus inornatus, Onychomus torridus) and Nelson’s antelope squirrel

(Ammospermophilus nelson), whereas higher scores on this axis were associated with the presence of the chipmunk Tamias sonomae, the squirrel Tamiasciurus douglasii, two voles (Myodes californicus and Arborimus pomo) and several shrews, sciurids, and voles

95 that were found only at the Humboldt County (HC) site. Low scores on the second prey community axis were associated with the presence of D. nitratoides, a fourth kangaroo rat species, D. californicus, and P. inornatus, while high scores were associated with the presence of a chipmunk, Tamias merriami, P. californicus, the woodrat Neotoma macrotis, and a fifth kangaroo rat D. agilis (Appendix A.2).

Low scores on environmental PC1 were associated with warmer winters and stable climates, while higher scores reflected more variable temperature regimes

(Appendix A.3). Low scores on environmental PC2 were associated with hotter, drier climates, higher scores reflected cooler, wetter areas, while scores of PC3 were positively associated with site elevation. We found that site scores on the first PCs of abiotic environmental data were significantly associated with our prey community NMDS axes

(Fig. 18, p < 0.05), suggesting that the environmental (abiotic) and prey (biotic) matrices reflect much of the same information about each site and, not surprisingly, that prey community composition is related to abiotic environmental variation. Comparing the power of each dataset to predict venom variation is therefore an important test to distinguish between the confounding effect of abiotic factors on venom variation compared to direct effects of variation in prey communities. We thus ran a separate RDA analysis conditioned on the genetic PCs, but replacing the prey community NMDS scores with the environmental PC scores. The three environmental PCs were marginally non- significant as predictors of venom variation (p = 0.08), and account for slightly less of the variation in the dataset (22%) compared with the prey community axes. To ensure the weaker performance of environmental variation was not due to over-parameterization due

96 to comparing results from three environmental axes versus two prey community axes, we ran a second environmental RDA with only the first two environmental PCs. This RDA was significant (p = 0.03), which is expected given the association of these two PCs with the prey community scores (Fig. 18). However, the amount of venom variation predicted by the environmental PC scores is even lower (19%) than in the first analysis using the prey community scores directly (23%), suggesting that differences in available prey are important, direct drivers of population-level venom variation.

In summary, we found that genetic differentiation was the best predictor of venom differentiation among populations. When we controlled for this genetic differentiation, direct biotic information about the local prey community outperformed abiotic environmental data in predicting venom variation. Taken together, these results suggest roles for both the isolation of populations and variation in available prey in driving venom divergence over geographic space.

Table 8. Results of Redundancy Analysis of population mean venom variation across 13 rattlesnake populations. The analysis assessed variance explained by the first two NMDS axes of prey community variation while conditioning the analysis on the first five principal components of RAD-seq genetic variation. Factor Inerti Variance P- a Explained value Conditional (Genetic 14.9 47.6 Variation) Constrained (Prey 7.2 23.0 0.02 Community) Unconstrained (Standard 9.2 29.4 PCA) Total 31.3

97

Figure 16. Biplot of first two axes from constrained ordination of ilr-transformed mean venom phenotypes of Northern Pacific rattlesnakes from 13 sites (abbreviated site names on plot indicating site scores. Arrows are eigenvectors of the first two axes of prey community space.

98

Figure 17. Relationship between pairwise FST of rattlesnake populations and pairwise distance between the same populations in Euclidean distance in clr-transformed venom phenotype space. Comparisons are within the southern clade (red), northern clade (green), or between clades (blue).

99

Figure 18. Relationships between site scores on the first to principal components (PC) of 19 Bioclim variables plus elevation and site scores on the first two non-metric multidimensional scaling axes of prey community variation. Associations between environmental and prey community site scores indicate that the two sets of variables carry much of the same information.

Discussion

Numerous studies of snake animal venoms have assessed whether diet variation, abiotic environmental variation, or genetic distance best explain venom variation in their focal taxon with mixed results regarding the relative importance of each factor (Chang et al. 2015; Creer et al. 2003; Daltry et al. 1996; Duda et al. 2009; Gibbs & Chiucchi 2011;

Gren et al. 2017; Margres et al. 2015b; Remigio & Duda Jr 2008; Williams et al. 1988).

We employed a novel method (constrained ordination) which allowed us to partition the variation associated with various factors in a way that allows us to evaluate the relative importance of each. Our major findings are that variation in the local prey community was associated with venom differentiation while genetic variation explains more variation 100 than either prey or abiotic environmental data. Below, we discuss the implications of these results for understanding the underlying evolutionary and ecological drivers of population-level differentiation in venom.

Extent of Geographic Variation in Northern Pacific Rattlesnake Venom

Population-level variation in snake venom has been repeatedly documented

(Creer et al. 2003; Daltry et al. 1996; Gibbs & Chiucchi 2011; Margres et al. 2016;

Massey et al. 2012). Our study complements another study on the same rattlesnake species with fewer individuals and over a smaller latitudinal range that showed venom composition also varies extensively among populations (Gren et al. 2017). Gren et al.

(2017) suggested that the venom of Northern Pacific rattlesnakes is more variable than any snake species previously studied although the quantitative basis for this claim is unclear. Our study provides quantitative support for this claim: comparison of our results with those of a study of congeneric Eastern diamondback rattlesnakes (Crotalus adamanteus) using similar methods showed that population of origin only explained

11.6% of the observed venom variation compared to over 50% of the variation explained by location in C. oreganus. Additional comparisons of venom differentiation in other related species are needed to judge if C. o, oreganus is exceptional, and the comparisons would require both population and deeper phylogeographic splits in the dataset as existed here.

The clearest differentiation in the venom phenotype was between the Northern and Southern rattlesnake clades that lay on either side of the best known phylogeographic

101 break in the California Floristic Province: the San Francisco Bay and San

Joaquin/Sacramento River Deltas. This break has produced phylogeographic divergence in diverse several terrestrial vertebrates (Calsbeek et al. 2003) including these rattlesnakes (Chapter 2, Goldenberg 2013) and appears to also coincide with the major axis in the functional trait of venom. High levels of geographic differentiation of venom in this species may reflect unusually high levels of habitat variation in this region compared with other areas where rattlesnakes are found in the US. In particular, the

California Floristic Province has been called a hotbed of population and phylogeographic diversification (Calsbeek et al. 2003; Hickerson et al. 2010), where steep environmental gradients of elevation and aridity are superimposed on longer term drivers of diversification including mountain uplift, ocean embayment, and glaciation (Matocq et al. 2012). This is reflected in multiple studies of adaptive phenotypic variation which show high levels of variation in adaptive traits such as the mimicry colors of Ensatina eschscholtizii salamanders (Kuchta et al. 2009), the toxicity of Taricha newts and resistance of garters snakes that feed on them (Hanifin et al. 2008) and the corolla shape of Lithophragma politella plants that are both parasitized and pollinated by Greya moths

(Thompson et al. 2013). Mammals which are prey to snakes in this region also show considerable population and phylogeographic structure (Conroy & Neuwald 2008;

Motocq 2002; Phuong et al. 2014) and physiological adaptations to environmental gradients (Eastman et al. 2012) that may result in selection for differences in venom efficiency in different rattlesnake populations.

102

Population differentiation also occurs in two other rattlesnakes that occur west of the Rocky Mountains, namely the Southern Pacific Rattlesnake (C. helleri; Sunagar et al.

2014) and the Mojave Rattlesnake (C. scutulatus; Massey et al. 2012). Both species show trade-offs in the relative abundance of PLA2 and MP proteins in their venom, which characterize a functional trade-off between neurotoxicity and hemotoxic/hemorrhagic mechanisms of killing prey. The Northern Pacific rattlesnake fits this pattern, as the first three principal components of venom composition are characterized by a negative association between the signs of the loadings of PLA2 versus MP protein peaks. The abundance of several MP peaks increases with venom PC 1 scores, which means the venom of Southern clade snakes is more MP rich. Part of the functional significance of

MP variation among these Southern clade populations is tied to local adaptation to variable resistance in California ground squirrels (Holding et al. 2016a) but the fact that

MP variation also covaries with broad scale variation in the prey community composition suggests that MP proteins are under selection to adapt to additional mammal prey.

Whether these other mammal prey also show evidence for a coevolutionary response in terms of resistance is an interesting question for future work.

Drivers of Variation in Venom Composition

Our finding that population genetic divergence between populations was the best predictor of venom divergence is in direct contrast with other studies of intraspecific variation in snake venom that have made similar comparisons (Creer et al. 2003; Daltry et al. 1996; Gibbs & Chiucchi 2011; Margres et al. 2015a; Margres et al. 2015b),

103 including a recent study in the Northern Pacific rattlesnake (Gren et al. 2017). A major difference between our study and others is that we used genome-scale data in the form of

RADseq loci to estimate genetic differentiation between rattlesnake populations, which raises the possibility that previous estimates of genetic divergence were less accurate measures of differentiation. For example, all previous studies used either mitochondrial sequence data or small numbers of microsatellite loci to estimate population genetic differences. Divergence estimates based on a single mitochondrial gene will suffer from a series of issues that can potentially lead to biased estimation of genetic differentiation in nuclear genes among populations. For example, the fact that many viperid snakes including C. oreganus show male-biased dispersal (Putman et al. 2013) and the smaller genetically effective sizes of mitochondrial genes means that mitochondrial data could lead to divergence estimates that do not reflect those of nuclear genes (e.g. Gibbs et al.

2000; Palumbi & Baker 1994). Finally, any single gene study represents a single sample from which to infer levels of differentiation, whereas our dataset of over nine hundred loci provides a genome-wide average to better approximate divergence among populations. Alternatively, there could be real biological differences in the strength of mechanisms driving the relationship between venom and genetic differentiation (see below).

One interpretation of the strong relationship between genetic and venom divergence among populations is that neutral evolutionary processes such as genetic drift or founder effects have had a significant impact on functional venom variation. Changes in climate since the Last Glacial Maximum likely mean that rattlesnakes are more widely

104 distributed today than they were 15,000 years ago, particularly for our Northern clade populations, which may have been founded since the LGM and have smaller effective population sizes (Chapter 3). Hence, founder effects could play a limited role in present day patterns of venom differentiation. However, other analyses show that gene flow between populations is high and effective populations sizes large (> 1,000 individuals;

Chapter 3) suggesting that genetic drift is not a viable mechanism explaining divergence in functional and hence presumably adaptive variation in venom.

A possible alternative explanation for the close association between levels of divergence in venom and neutral genetic variation is that it is due to an Isolation by

Environment mechanism (IBE) (Wang & Bradburd 2014). The process of IBE operates when local adaptation drives correlated divergence in adaptive traits and neutral markers through selection against maladaptive phenotypes in immigrants, with correlated impacts on neutral loci carried by such immigrants. This possibility is supported by the lack of the positive relationship between pairwise FST and distance in venom space when comparisons are on opposite sides of the major genetic break at San Francisco Bay, where venom divergence has likely occurred independent of the exchange of migrants.

To test this hypothesis, one could take advantage of California’s Coast and Sierra Nevada ranges, which are oriented in a North to South direction, and sample populations at similar elevations along the mountain ranges as well as down into the Central Valley, where sampling could cover similar geographic distance but encounter large changes in species composition toward the Valley floor (Wang & Bradburd 2014; Wang et al. 2013).

If IBE is the mechanism responsible for generating varying levels of neutral divergence

105 between these rattlesnake populations then environmental distance should outperform geographic distance as a predictor of FST or similar measures.

If the type of molecular marker used does not explain why neutral differentiation was an important predictor of venom divergence in our study but only a secondary (Creer et al. 2003; Daltry et al. 1996) or poor (Gibbs & Chiucchi 2011) predictor in other studies, then a biological difference must exist between the Northern Pacific rattlesnake and these other snakes. One possible explanation is variation in the strength of an IBE effect among species or geographic locations. The contribution of IBE to genetic divergence has been shown to vary among species. For example, among 17 Caribbean species of anole lizards (Anolis sp.), the amount of population genetic divergence explained by IBE ranged from 0.1 to 48 percent (Wang et al. 2013). Lizards from the island of Hispaniola, where habitat heterogeneity was highest, showed the highest levels of IBE. Among rattlesnake venoms assessed for correlations between population genetic and venom divergence, all occur in areas with lower levels of habitat heterogeneity than

C. oreganus; the Eastern massasauga (Sistrurus catenatus) occurs in areas of the upper

Midwest, the Eastern diamondback rattlesnake in the Coastal Plain of the Southeast, and the Southern Pacific rattlesnake (C. helleri) occurs in Southern California (Conant &

Collins 1998), and all of these areas are less heterogeneous than Central and Northern

California where our study took place (Kerr & Packer 1997). Conditions across the range of C. oreganus may therefore produce local adaptation that is stronger and occurs at a finer scale on average in C. oreganus, and lead to stronger associations between population genetic and adaptive venom differentiation. This hypothesis could again be

106 tested by planned studies of IBE in C. oreganus, particularly if IBE in C. oreganus is compared to that of its close relatives, C. helleri and C. lutosus (Davis et al. 2016;

Goldenberg 2013), which are broadly distributed across more uniform habitats.

The second most important predictor of venom divergence was differences in prey community between sites. This result aligns with conventional ideas that variation in diet is a significant selective force that molds venom differences between populations

(Casewell et al. 2012b; Daltry et al. 1996; Wuster 1999.). The first axis of prey community variation delineated locations with prey communities consisting of rodents that are arid habitat specialists (kangaroo rats, desert mice, and antelope squirrels) from sites with small mammals representative of wetter habitats at high elevation and/or latitude (chipmunks, Douglas squirrels, and voles). While Sciurids and mice in the genus

Peromyscus occur among all locations, the dichotomy between kangaroo rats and voles as major diet items represents a major axis of taxonomic differentiation in rattlesnake prey, and hence may impose distinct selection pressures on venom composition. These species are from divergent families within Rodentia, but are both medium sized and locally abundant, and thus may be key components of prey that differ between xeric and mesic habitats. The three published diet studies support the existence of a kangaroo rat- vole dichotomy among locations. At sites in British Columbia and Washington, voles made up 50 and 57 percent, respectively, of the diet items found in the stomachs of

Northern Pacific rattlesnakes (Macartney 1989; Wallace & Diller 1990), while on the San

Joaquin Experimental Range in the central foothills of the Sierra Nevada, voles made up only 2 percent of the diet whereas kangaroo rats were the second most abundant diet item

107

(16% of prey taken). Reciprocal toxicity tests on voles versus kangaroo rats using the venom of snakes from mesic versus dry environments in California would provide a functional test of the role that these groups of rodents play in shaping the venom phenotype (Pomento et al. 2016), which could be followed by investigating the physiological underpinnings of venom susceptibility in each rodent species.

That abiotic environmental variables were correlated with the prey community variables (Fig. 18) but were outperformed by the prey community as direct predictors of venom differentiation suggests that the prey community directly selects for venom divergence. Previous studies have suggested that the abiotic environment could impact venom directly through an increased requirement for pre-digestive functions of venom in more climatically variable environments (Mackessy 2010), and abiotic environmental variation has predicted both functional and compositional differences in the venom of C. oreganus (Gren et al. 2017; Holding et al. 2016a). However, comparative tests of whether these are direct effects or if environment acts indirectly by structuring prey community composition have been unclear. Our results help resolve this question by showing higher predictive power of a coarse measure of prey community variation to predict venom divergence, compared to a fine grained (1 kM resolution) measure of abiotic environmental variation by suggesting an overriding effect of the available prey species. As we learn more about the hemorrhagic and coagulopathic metalloproteinases in relation to the killing capacity of venoms, we may see that these proteins previously considered as pre-digestive may actually simply be involved in the incapacitation of particular prey types (Bernardoni et al. 2014)

108

Assessing the availability of potential prey using presence-absence data to characterize local mammal community composition, instead of analysis of stomach or fecal contents represents a new approach to evaluating prey diversity as a cause of population-level divergence in venom. Our finding that prey community differences predicted 23% of the variance in venom composition, and that prey community data outperformed abiotic environmental data suggest that small mammal distribution data, a more accessible and unbiased source information that stomach or fecal contents, can be useful for studies of the adaptive significance of venom variation. While stomach content data provide direct information about snake diet, they can be biased in terms of representing an accurate assessment of what snakes eat. For example, smaller prey items are more likely to be detected because snakes with large prey items are encumbered and therefore avoid movement and remain hidden. The intensive field study of Fitch (1949) reported 116 California ground squirrels in a total of 285 prey items (40%), with squirrels making up 70% of rattlesnake prey by weight. Yet, an assessment of 88 prey items in the guts of museum specimens from Central California, which are often collected on roads, only included 3 ground squirrels and likely reflects bias against detecting these larger prey taxa (Sparks et al. 2014). Community variation, using an assessment of which animals represent potential prey based on previous work, may circumvent the biases of stomach content studies. This approach could be refined by incorporating niche modelling to better predict localized absences and weight the presences based on habitat suitability indices (e.g. Aranda & Lobo 2011; D'amen et al. 2015; Pineda & Lobo 2012).

109

In some regions, C. oreganus feeds most heavily one species, the California ground squirrel (Fitch 1949; Fitch & Twining 1946), which comprised the majority of videotaped encounters with potential prey at two sites (Putman et al. 2016). Our previous work suggests that Northern Pacific rattlesnake venom is locally adapted to overcoming

California ground squirrel MP resistance, suggesting adaptive divergence in venom specifically in response to variable resistance in a single species of prey (Holding et al.

2016a). Because here we show that broader prey community variation impacts venom divergence among these same populations, we conclude that population divergence in rattlesnake venom can be a simultaneous response to both broad taxonomic differences in prey communities and variable resistance in coevolving prey species in a geographic mosaic of coevolution (Thompson 2005b). In fact, responses to prey community differences and coevolving resistant species need not be mutually exclusive. Each distinct selection pressure could act on venom such that an optimum in venom phenotype space is reached by a population that is a balance between the relative abundance of available prey and the strength of coevolutionary selection between rattlesnakes and ground squirrels which represents a continuum from opportunism to specialization on ground squirrels. Alternatively, functional modules that may evolve semi-independently have been proposed for snake venoms (Gibbs & Chiucchi 2011; Gibbs et al. 2009) which would allow separate sets of venom proteins to evolve in response to these various sources of selection and maximize a snake’s chance of capturing the different prey in its local environment.

110

Chapter 5: General Conclusions

In this dissertation, I have explored evolutionary and ecological causes of phenotypic variation in the venom of Northern Pacific rattlesnakes (Crotalus oreganus) alongside variation in venom resistance of the California ground squirrel

(Otospermophilus beecheyi). My analyses identified functional local adaptation of a major class of venom proteins (metalloproteinases) to overcoming squirrel resistance, used genetic data to estimate key demographic factors that explain variation in the strength of local adaptation among populations, and considered the potential of the broader community prey community to impact a coevolving venom phenotype. My work represents an example of how studies of geographically variable species interactions can provide insight into the coevolutionary process by evaluating the evolutionary and ecological factors predicted to influence coevolutionary selection and lead to local adaptation (Abrams 2000; Thompson 2005b). In this concluding chapter, I review the key insights of each of my studies and suggest future directions.

In Chapter 2, I documented local adaptation between Northern Pacific rattlesnake venom metalloproteinases and the metalloproteinase inhibitors in California ground squirrel blood serum based on a range-wide samples from coexisting populations of snakes and squirrels. Documenting local adaptation was a first clear suggestion of coevolution between this predator and this prey, as opposed to one-sided local adaptation 111 of squirrel resistance to snake density as previously suggested (Poran et al. 1987). It also linked functional venom variation to the defensive traits of a specific prey species, rather than broader differences in which prey are taken in a given location (e.g. Daltry et al.

1996). Local adaptation provided support for phenotype matching interactions between snake venom proteins and serum-based venom inhibitors, suggesting coevolution in this system occurs through a different mechanism that suggested by the arms race analogy

(Ridenhour & Nuismer 2007). The observed pattern of local adaptation was surprising for two reasons. First, local adaptation was structured across an environmental gradient in this system, such that performance of venom with sympatric squirrels was only significantly better than performance with allopatric squirrels from sites at different elevations. This result supports elevation or an environmental correlate of elevation as key in the production of geographic mosaics of coevolutionary selection in this system.

Second, the rattlesnakes were the locally adapted species, showing that venomous snakes can evolutionarily outpace key prey despite simplistic demographic predictions to the contrary (e.g. the Dinner-Life Principle of Dawkins & Krebs 1979).

Several key questions about the genetic architecture and complexity of the ground squirrel resistance phenotype are raised by Chapter 2, which we can now address due to advances in genomic and proteomic technologies for assessing the molecular basis of adaptive trait variation. Local adaptation suggests phenotype matching, and snake local adaptation suggests that snakes evolutionarily outpace their squirrel prey. These factors lead to the prediction that squirrel serum proteins are geographically variable but that there is more variation and evolutionary potential in snake venom. Testing these

112 predictions will require the identification of the full set of squirrel serum proteins that confer resistance to snake venom metalloproteinases. Once these are known, then sequence capture (Grover et al. 2012) or other population genomic methods can be used to assess the population-level variation in genetic basis of venom and resistance genes, and determine if there are relationships between the frequency or expression of certain metalloproteinase genes and the frequency or expression of matched resistance genes in the squirrels. Affinity chromatography using venom proteins as molecular “baits”

(Calvete 2010; Calvete et al. 2009) is currently being employed to characterize the full squirrel resistance phenotype for this purpose, and the knowledge gained about the degree of protein specificity in venom inhibition could also lead to advances in the treatment of snake bite.

In Chapter 3, I provided a direct test of a relationship between the strength of local adaptation and genetic demographic factors predicted by coevolutionary models to influence coevolutionary potential (Gandon & Michalakis 2002). I collected genomic

SNP data based on RAD-seq to estimate effective populations sizes of and migration rates between rattlesnake and squirrel populations, and showed that the difference between snake and squirrel effective sizes among populations explained variation in the pattern of local adaptation reported in Chapter 2. My approach extended on previous studies that linked demography to patterns of local adaptation because by focusing on variation in demography and local adaptation, instead of simply reporting whether the species with higher average effective sizes or rates of gene flow also happens to be the locally adapted species range wide. Host-parasite studies could also use continuous

113 variation in demography and local adaptation in this way, which may help to explain the numerous examples of host-parasite relationships that seem to break from traditional expectations set for by theoretical models of the impacts of gene flow on local adaptive potential (Hoeksema & Forde 2008).

My work in Chapter 3 also highlights how a comparison of the rattlesnake- squirrel relationship North and South of San Francisco Bay represents an opportunity to study how coevolutionary selection differs across distinct evolutionary lineages. The San

Francisco Bay and Sacramento/San Joaquin River Delta form a phylogeographic break in many taxa (Calsbeek et al. 2003), and my genetic data support the presence of this break in Northern Pacific rattlesnakes and California ground squirrels. The pattern of local adaptation is not present in the North for reasons that remain unclear. If future work confirms a lack of local adaptation in the North, then the geographic differences between

North and South populations of snakes and squirrels could be used as a model to understand how coevolutionary adaptation in different environments interact with the divergence of lineages. For example, a question that could be addressed is did historical events associated with glaciation lead to genetic bottlenecks in Northern snakes and impact their ability to stay ahead of coevolving squirrels, or is coevolution simply more diffuse in the North due to the presence of other key prey? Combining phylogeographic studies of divergence patterns in snakes and squirrels with knowledge of the levels of standing variation in venom and resistance genes in the North and South would suggest if

Southern adaptive variation venom and resistance loci is missing in Northern animals,

114 and could detect if adaptive introgression occurs in the Central Sierra Nevada where

Northern and Southern clades may come into contact (Phuong et al. 2014).

In Chapter 4, I quantified patterns of differentiation in the whole venom phenotype between rattlesnake populations and sought to account for the variation in terms of three factors previously suggested as important drivers of inter-population variation in venom. Following the approach of previous studies of the causes of population-level variation in the venom, I examined whether biotic, abiotic, or genetic variation among localities best explained venom variation, but the data I used were different from previous work. Instead of gut content-based diet data, I estimated variation in the makeup of the local prey community using range maps. Also, instead of geologic estimates of isolation or mitochondrial genetic distance, I used a multi-locus RAD-seq data set to estimate genetic differentiation between populations. I showed that genetic and prey community data explains most of the variation in venom, and importantly that prey community data provided better explanatory power than abiotic environmental data. The power of genetic data challenges the assertion that most venom variation is adaptive by suggesting that genetic drift and isolation are significant mechanisms, but could also be explained in an adaptive context though a pattern of isolation by environment (Wang &

Bradburd 2014). Because functional local adaptation to ground squirrel resistance

(Chapter 2) and the composition of broader prey community (Chapter 4) appear to impact venom variation among these rattlesnake populations, my dissertation supports the idea of modular evolution of the venom phenotype, where sets of venom proteins may respond

115 semi-independently to differing selective pressures (Gibbs & Chiucchi 2011; Gibbs &

Mackessy 2009)

Overall, my research in this system highlights several key factors influencing adaptive evolution in venomous snakes and their prey, while testing predictions of coevolutionary theory in a natural predator-prey system. My work lays the foundation for future studies that will combine field ecology, functional tests of venom action, and genomic and proteomic analyses of the underlying molecular bases of venom and resistance traits to tease apart the contributions of evolutionary history, environmental constraints, and the nature of venom and resistance traits themselves in the maintenance of biological diversity through coevolution.

116

Appendix A: Supplemental Tables

A.1:

Identification of proteins composing each RP-HPLC peak based on band sizes from SDS- PAGE and quantification based on Mackessy, 2008. The two most intense bands from each peak are bolded (1 bolded if only 1-2 bands present). Peak Number Protein (s) Mass (kDa) 1 Disintegrin 8.3 Myotoxin 5.2 2 Disintegrin 8.4 Myotoxin 5.2 3 Disintegrin 5.6 4 Disintegrin 6.3 5 SP 35.4, 32.6, 28.0 MP-PI, CRISP 23.6 Disintegrin 7.5 Myotoxin 4.9 6 SP 35.1, 30.5 7 SP, CRISP 29.1 PLA2 13.0 8 PLA2 13.4 9 MP-PI, CRISP 21.4 PLA2 13.0 10 MP-PIII 45.8 MP-PII, CRISP, 28.5 CTL PLA2 13.5 Disintegrin 10.7 11 PLA2 15.7 Disintegrin 10.4 12 PLA2 15.1 Disintegrin 10.7 Continued

117

A.1: Continued Peak Number Protein (s) Mass (kDa) 13 LAAO 85.8 MP-PIV, Nuc 61.8 MP-PIII, MP-PIV 51.6 SP 35.5, 30.8 14 MP-PIII, MP-IV 49.5 MP-PIII 45.3 SP 32.4 PLA2 16.5 15 SP 34.1, 31.0 16 LAAO, PD 101.9 MP-PIV, Nuc 66.9 SP 34.6, 31.2 PLA2 15.0 17 MP-PIV, Nuc 60.6 SP 31.5 Disintegrin 11.1, 9.0 18 SP 31.7 19 SP 31.7 Disintegrin 14.5, 11.7, 5.8 20 SP 35.1 Disintegrin 13.9 21 LAAO, PD 114.9 MP-PIII, MP-PIV 56.2 SP 32.2 22 Unidentified N/A 23 MP-PI, CRISP 24.3 24 LAAO, PD 120.8 MP-PIII, Nuc 58.6, 43.1 MP-PI, CRISP 24.2 25 LAAO, PD 107.4 MP-IV, Nuc 72.0 MP-III, MP-IV 58.0, 50.3 MP-III 47.6 26 LAAO, PD 108.3 MP-IV, Nuc 66.5 MP-III 47.6 27 MP-III, MP-IV 55.8, 53.9, 49.0 28 LAAO, PD 109.2 MP-PIII 51.1, 46.1 MP-PI, CRISP 22.5 Continued 118

Continued Peak Number Protein (s) Mass (kDa) 29 MP-PIII, MP-PIV 54.3 MP-PIII 46.7 Unknown 37.2 MP-PI, CRISP 23.8 30 LAAO, PD 126.4 MP-PIV, Nuc 61.6 31 MP-PIII, MP-PIV 49.0 32 LAAO, PD 116.8 MP-PIII, MP-PIV 57.7, 49.5 33 MP-PIII, MP-PIV 55.4, 56.3 34 MP-PIII, MP-PIV 54.8, 49.7 SP 35.3, 34.0 MP_PI, CRISP 24.2 Abbreviations: CRISP - Cysteine-rich secretory protein; LAAO, L-amino acid oxidase; MP-PI, Metalloproteinase Type I; MP-PII, Metalloproteinase Type II; MP-PIII, Metalloproteinase Type III; MP-PIV, Metalloproteinase Type IV; Nuc, 5’-nucleotidase; PD, Phosphodiesterase; SP, Serine protease

119

A.1:

Species scores from non-metric multidimensional scaling of presence/absence data for each species across 13 study sites in central California. Species Axis 1 Species Axis 2 Dipodomys nitratoides -0.66 Dipodomys californicus -0.55 Ammospermophilus nelsoni -0.60 Dipodomys nitratoides -0.39 Dipodomys ingens -0.60 Perognathus inornatus -0.36 Onychomys torridus -0.54 Ammospermophilus nelsoni -0.26 Peromyscus eremicus -0.50 Dipodomys ingens -0.26 Peromyscus fraterculus -0.50 Tamias sonomae -0.20 Peromyscus crinitus -0.50 Peromyscus truei -0.17 Perognathus inornatus -0.45 Neotoma fuscipes -0.16 Dipodomys heermanni -0.38 Onychomys torridus -0.12 Chaetodipus californicus -0.27 Tamiasciurus douglasii -0.09 Sorex ornatus -0.26 Myodes californicus -0.09 Tamias merriami -0.26 Arborimus pomo -0.09 Neotoma lepida -0.26 Microtus californicus -0.05 Sylvilagus audubonii -0.25 Spermophilus beecheyi -0.04 Peromyscus californicus -0.20 Thomomys bottae -0.04 Neotoma macrotis -0.20 Reithrodontomys megalotis -0.04 Dipodomys agilis -0.18 Peromyscus maniculatus -0.04 Microtus californicus -0.13 Sylvilagus audubonii -0.02 Dipodomys venustus -0.09 Dipodomys venustus 0.00 Spermophilus beecheyi 0.00 Peromyscus eremicus 0.00 Thomomys bottae 0.00 Peromyscus fraterculus 0.00 Reithrodontomys megalotis 0.00 Peromyscus crinitus 0.00 Peromyscus maniculatus 0.00 Sylvilagus bachmani 0.06 Peromyscus boylii 0.00 Sciurus griseus 0.09 Sylvilagus bachmani 0.00 Dipodomys heermanni 0.11 Peromyscus truei 0.11 Sorex vagrans 0.13 Sciurus griseus 0.21 Sorex pacificus 0.13 Neotoma fuscipes 0.23 Sorex bendirii 0.13 Sorex trowbridgii 0.28 Aplodontia rufa 0.13 Dipodomys californicus 0.30 Tamias senex 0.13 Tamias sonomae 0.52 Glaucomys sabrinus 0.13 Tamiasciurus douglasii 1.05 Arborimus albipes 0.13 Myodes californicus 1.05 Microtus longicaudus 0.13 Arborimus pomo 1.05 Microtus oregoni 0.13 Continued 120

Continued Species Axis 1 Species Axis 2 Sorex vagrans* 1.44 Zapus trinotatus 0.13 Sorex pacificus* 1.44 Sorex ornatus 0.16 Sorex bendirii* 1.44 Peromyscus boylii 0.18 Aplodontia rufa* 1.44 Sorex trowbridgii 0.25 Tamias senex* 1.44 Neotoma lepida 0.28 Glaucomys sabrinus* 1.44 Chaetodipus californicus 0.28 Arborimus albipes* 1.44 Tamias merriami 0.47 Microtus longicaudus* 1.44 Peromyscus californicus 0.59 Microtus oregoni* 1.44 Neotoma macrotis 0.59 Zapus trinotatus* 1.44 Dipodomys agilis 0.80

121

A.2.

Axis loadings from PCA for 19 Bioclim variables (Var.) and elevation (Elev.) across the 13 California study sites. PC1, PC2, and PC3 axes are shown, and each column is sorted in increasing order of loadings to facilitate environmental interpretation (Interp.) of each. Var. PC1 Interp. Var. PC2 Interp. Var. PC3 Interp. Warmer Low Elev., winters, Warmer Hot, Dry Stable bio11 -0.31 climate bio1 -0.30 bio1 -0.44 bio3 -0.30 bio10 -0.23 bio6 -0.33 bio6 -0.29 bio9 -0.22 bio10 -0.23 bio8 -0.29 bio5 -0.20 bio8 -0.22 bio15 -0.22 bio2 -0.16 bio9 -0.21 bio18 -0.04 bio7 -0.16 bio11 -0.21 bio1 0.06 bio4 -0.15 bio17 -0.15 Elev. 0.13 bio15 -0.14 bio13 -0.12 bio14 0.15 bio8 -0.11 bio14 -0.12 bio16 0.15 bio11 -0.06 bio12 -0.11 bio19 0.16 Elev. 0.01 bio19 -0.10 bio12 0.16 bio6 0.04 bio16 -0.09 bio13 0.16 bio3 0.09 bio4 -0.08 bio17 0.17 bio13 0.29 bio5 -0.07 bio10 0.23 bio14 0.29 bio18 -0.05 bio2 0.24 bio17 0.30 bio7 0.04 bio9 0.25 bio19 0.30 bio3 0.18 bio5 0.27 bio16 0.30 bio15 0.21 bio7 0.29 bio12 0.31 bio2 0.29 bio4 0.30 Cooler bio18 0.34 Elev. 0.52 High elev., winters, Cool, Cooler Variable Wet Climate

122

Appendix B: Supplemental Figures

B.1:

123

Images of SDS-PAGE gels used to identify venom proteins from a pooled sample of Northern Pacific rattlesnake venom. Numbers beneath each lane indicate which of the 34 HPLC peaks was run on that lane, and “L” indicates a lane with the molecular weight standard. Some HPLC peaks are ran in multiple lanes.

124

B.2:

clr-transformed mean peak abundance -5 -4 -3 -2 -1 0 1 2 3 4 5 CC * HC Peak 1 - Dis/Myo HR ML * SB Peak 2 - Dis/Myo BO CN † * Peak 3 - Dis CR MD SR * Peak 4 - Dis SW VB † * Peak 5 - SP WW

† Peak 6 - SP

† * Peak 7 - PLA2

† * Peak 8 - PLA2

† * Peak 9 - PLA2

† * Peak 10 - PLA2/MP/CTL

† * Peak 11 - Dis/PLA2

† * Peak 12 - PLA2

† * Peak 13 - SP/MP

† * Peak 14 - SP

* Peak 15 - SP

† * Peak 16 - MP/Nuc/PLA2

* Peak 17 - SP/Dis

Population average of clr-transformed areas for peaks 1-17 in Northern Pacific rattlesnake venom HPLC chromatograms. North-clade populations have bars with dark borders. † indicates significant north/south variation, * indicates significant difference between populations, α = 0.05. 125

B.3

clr-transformed mean peak abundance -5 -4 -3 -2 -1 0 1 2 3 4 5

† * Peak 18 - SP

† * Peak 19 - SP

† * Peak 20 - Dis/PLA2 CC HC * Peak 21 - LAAO/PD/MP HR ML † * Peak 22 - ? SB BO CN † * Peak 23 - MP/CRISP CR MD † * SR Peak 24 - MP/Nuc/CRISP SW VB † Peak 25 - MP

* Peak 26 - MP,Nuc

† * Peak 27 - MP

† * Peak 28 - LAAO/PD/MP

† * Peak 29 - MP

* Peak 30 - MP/Nuc

† * Peak 31 - MP

† * Peak 32 - MP

* Peak 33 - MP/Nuc

* Peak 34 - SP

Population average of clr-transformed areas for peaks 18-34 in Northern Pacific rattlesnake venom HPLC chromatograms. North-clade populations have bars with dark borders. † indicates significant north/south variation, * indicates significant differences between populations, α = 0.05. 126

References

Abrams PA (1986) Adaptive responses of predators to prey and prey to predators: The failure of the arms-race analogy. Evolution 40, 1229-1247.

Abrams PA (2000) The evolution of predator-prey interactions: Theory and evidence. Annual Review of Ecology and Systematics 31, 79-105.

Adiba S, Huet M, Kaltz O (2010) Experimental evolution of local parasite maladaptation. Journal of Evolutionary Biology 23, 1195-1205.

Allendorf FW (1986) Genetic drift and the loss of alleles versus heterozygosity. Zoo Biology 5, 181-190.

Anderson B, Johnson SD (2008) The geographical mosaic of coevolution in a plant- pollinator mutualism. Evolution 62, 220-225.

Aranda SC, Lobo JM (2011) How well does presence-only-based species distribution modelling predict assemblage diversity? A case study of the Tenerife flora. Ecography 34, 31-38.

Arnold B, Corbett-Detig RB, Hartl D, Bomblies K (2013) Radseq underestimates diversity and introduces genealogical biases due to nonrandom haplotype sampling. Molecular Ecology 22, 3179-3190.

Barbour MA, Clark RW (2012) Ground squirrel tail-flag displays alter both predatory strike and ambush site selection behaviours of rattlesnakes. Proceedings of the Royal Society B: Biological Sciences 279, 3827-3833.

Barlow A, Pook CE, Harrison RA, Wuster W (2009) Coevolution of diet and prey- specific venom activity supports the role of selection in snake venom evolution. Proceedings of the Royal Society B: Biological Sciences 276, 2443-2449.

127

Becker CG, Rodriguez D, Toledo LF, et al. (2014) Partitioning the net effect of host diversity on an emerging amphibian pathogen. Proceedings of the Royal Society B: Biological Sciences 281. DOI: 10.1098/rspb.2014.1796.

Beerli P, Palczewski M (2010) Unified framework to evaluate panmixia and migration direction among multiple sampling locations. Genetics 185, 313-326.

Benkman CW, Parchman TL, Favis A, Siepielski AM (2003) Reciprocal selection causes a coevolutionary arms race between crossbills and lodgepole pine. The American Naturalist 162, 182-194.

Bernardoni JL, Sousa LF, Wermelinger LS, et al. (2014) Functional variability of snake venom metalloproteinases: Adaptive advantages in targeting different prey and implications for human envenomation. PLoS ONE 9, e109651.

Bernays E, Graham M (1988) On the evolution of host specificity in phytophagous arthropods. Ecology 69, 886-892.

Biardi JE (2008) The ecological and evolutionary context of mammalian resistance to rattlesnake venoms. In: Biology of the Rattlesnakes (eds. Hayes WK, Beaman KR, Cardwell MD, Bush SP), pp. 557-568. Loma Linda University Press, Loma Linda, California.

Biardi JE, Coss, R. G., Smith, D. G. (2000) California ground squirrel (Spermophilus beecheyi) blood sera inhibits crotalid venom proteolytic activity. Toxicon 38, 719- 721.

Biardi JE, Ho CYL, Marcinczyk J, Nambiar KP (2011a) Isolation and identification of a snake venom metalloproteinase inhibitor from california ground squirrel (Spermophilus beecheyi) blood sera. Toxicon 58, 486-493.

Biardi JE, Nguyen KT, Lander S, Whitley M, Nambiar KP (2011b) A rapid and sensitive fluorometric method for the quantitative analysis of snake venom metalloproteases and their inhibitors. Toxicon 57, 342-347.

Blanquart F, Gandon S, Nuismer SL (2012) The effects of migration and drift on local adaptation to a heterogeneous environment. Journal of Evolutionary Biology 25, 1351-1363.

Blanquart F, Kaltz O, Nuismer S, Gandon S (2013) A practical guide to measuring local adaptation. Ecology Letters 16, 1195-1205. 128

Bouckaert RR (2010) Densitree: Making sense of sets of phylogenetic trees. Bioinformatics 26, 1372-1373.

Brandt M, Fischer-Blass B, Heinze J, Foitzik S (2007) Population structure and the co- evolution between social parasites and their hosts. Molecular Ecology 16, 2063- 2078.

Brodie ED, Brodie ED (1999) Predator-prey arms races. Bioscience 49, 557-568.

Brodie ED, Ridenhour BJ (2003) Reciprocal selection at the phenotypic interface of coevolution. Integrative and Comparative Biology 43, 408-418.

Brodie ED, Ridenhour BJ, Brodie ED (2002) The evolutionary response of predators to dangerous prey: Hotspots and coldspots in the geographic mosaic of coevolution between garter snakes and newts. Evolution 56, 2067-2082.

Bryant D, Bouckaert R, Felsenstein J, Rosenberg NA, Roychoudhury A (2012) Inferring species trees directly from biallelic genetic markers: Bypassing gene trees in a full coalescent analysis. Molecular Biology and Evolution 29, 1917-1932.

Bryc K, Auton A, Nelson MR, et al. (2010) Genome-wide patterns of population structure and admixture in West Africans and African Americans. Proceedings of the National Academy of Sciences 107, 786-791.

Bryson RW, Savary WE, Zellmer AJ, Bury BR, Mccormack JE (2016) Genomic data reveal ancient microendemism in forest scorpions across the california floristic province. Molecular Ecology 25, 3731-3751.

Buckling A, Rainey PB (2002) The role of parasites in sympatric and allopatric host diversification. Nature 420, 496-499.

Burnham KP, Anderson DR. (2002) Model selection and multimodel inference: a practical information-theoretic approach. Springer, New York, NY.

Bushar LM, Bhatt N, Dunlop MC, et al. (2015) Population isolation and genetic subdivision of timber rattlesnakes (Crotalus horridus) in the new jersey pine barrens. Herpetologica 71, 203-211.

Calsbeek R, Thompson JN, Richardson JE (2003) Pattern of molecular evolution and diversification in a biodiversity hotspot: The California Floristic Province. Molecular Ecology 12, 1021-1029. 129

Calvete JJ (2010) Snake venomics, antivenomics, and venom phenotyping: The ménage à trois of proteomic tools aimed at understanding the biodiversity of venoms. In: Toxins and hemostasis: From bench to bedside eds. Kini R, Clemetson KJ, Markland Jr FS, McLane MA, Morita T), pp. 45-72. Springer, Netherlands.

Calvete JJ, Sanz L, Angulo Y, Lomonte B, Gutiérrez JM (2009) Venoms, venomics, antivenomics. FEBS Letters 583, 1736-1743.

Carlsson-Granér U, Thrall PH (2015) Host resistance and pathogen infectivity in host populations with varying connectivity. Evolution 69, 926-938.

Casewell NR, Huttley GA, Wüster W (2012a) Dynamic evolution of venom proteins in squamate reptiles. Nature Communications 3, 1066.

Casewell NR, Wüster W, Vonk FJ, Harrison RA, Fry BG (2012b) Complex cocktails: The evolutionary novelty of venoms. Trends in Ecology & Evolution 28, 219-229.

Chang D, Olenzek AM, Duda TF (2015) Effects of geographical heterogeneity in species interactions on the evolution of venom genes. Proceedings of the Royal Society B: Biological Sciences 282, 20141984

Clark WC, Voris HK (1969) Venom neutralization by rattlesnake serum albumin. Science 164, 1402-1404.

Conant R, Collins JT (1998) A field guide to reptiles and amphibians of eastern and central North America, 3rd edn. Houghton Mifflin, New York.

Conroy CJ, Neuwald JL (2008) Phylogeographic study of the California vole, Microtus californicus. Journal of Mammalogy 89, 755-767.

Coss RG, Gusé KL, Poran NS, Smith DG (1993) Development of antisnake defenses in California ground squirrels (Spermophilus beecheyi): II. Microevolutionary effects of relaxed selection from rattlesnakes. Behaviour 124, 137-164.

Couchoux C, Seppä P, Van Nouhuys S (2016) Strong dispersal in a parasitoid wasp overwhelms habitat fragmentation and host population dynamics. Molecular Ecology 25, 3344-3355.

Crait JR, Prange HD, Marshall NA, et al. (2012) High-altitude diving in river otters: Coping with combined hypoxic stresses. The Journal of Experimental Biology 215, 256-263. 130

Creer S, Malhotra A, Thorpe RS, et al. (2003) Genetic and ecological correlates of intraspecific variation in pitviper venom composition detected using matrix- assisted laser desorption time-of-flight mass spectrometry (maldi-tof-ms) and isoelectric focusing. Journal of Molecular Evolution 56, 317-329.

D'amen M, Pradervand J-N, Guisan A (2015) Predicting richness and composition in mountain insect communities at high resolution: A new test of the sesam framework. Global Ecology and Biogeography 24, 1443-1453.

Daltry JC, Wüster W, Thorpe RS (1996) Diet and snake venom evolution. Nature 379, 537-540.

Darwin C (1871) On the origin of species. Appleton and Co., New York.

Davies NB (2000) Cuckoos, cowbirds, and other cheats T. & A. D. Poyser, London, UK.

Davis EB, Koo MS, Conroy C, Patton JL, Moritz C (2008) The California hotspots project: Identifying regions of rapid diversification of mammals. Molecular Ecology 17, 120-138.

Davis MA, Douglas MR, Collyer ML, Douglas ME (2016) Deconstructing a species- complex: Geometric morphometric and molecular analyses define species in the western rattlesnake (Crotalus viridis). PLoS ONE 11, e0146166.

Dawkins R, Krebs JR (1979) Arms races between and within species. Proceedings of the Royal Society B: Biological Sciences 205, 489-511.

Diller LV, Wallace RL (1996) Comparative ecology of two snake species (Crotalus viridis and Pituophis melanoleucus) in southwestern Idaho. Herpetologica 52, 343-360.

Dixon P (2003) Vegan, a package of r functions for community ecology. Journal of Vegetation Science 14, 927-930.

Duda TF, Jr., Chang D, Lewis BD, Lee T (2009) Geographic variation in venom allelic composition and diets of the widespread predatory marine gastropod Conus ebraeus. PLoS ONE 4, e6245.

Dybdahl MF, Jenkins CE, Nuismer SL (2014) Identifying the molecular basis of host- parasite coevolution: Merging models and mechanisms. The American Naturalist 184, 1-13. 131

Earl DA, Vonholdt BM (2012) Structure Harvester: A website and program for visualizing structure output and implementing the Evanno method. Conservation Genetics Resources 4, 359-361.

Eastman LM, Morelli TL, Rowe KC, Conroy CJ, Moritz C (2012) Size increase in high elevation ground squirrels over the last century. Global Change Biology 18, 1499- 1508.

Ehrlich PR, Raven PH (1964) Butterflies and plants: A study in coevolution. Evolution 18, 586-608.

Evanno G, Regnaut S, Goudet J (2005) Detecting the number of clusters of individuals using the software structure: A simulation study. Molecular Ecology 14, 2611- 2620.

Evans FC, Holdenried R (1943) A population study of the beechey ground squirrel in central California. Journal of Mammalogy 24, 231-260.

Excoffier L, Dupanloup I, Huerta-Sanchez E, Sousa VC, Foll M (2013) Robust demographic inference from genomic and snp data. PLoS Genetics 9, e1003905.

Farkas TE, Mononen T, Comeault AA, Nosil P (2016) Observational evidence that maladaptive gene flow reduces patch occupancy in a wild insect metapopulation. Evolution 70, 2879-2888.

Filzmoser P, Hron K, Reimann C, Garrett R (2009) Robust factor analysis for compositional data. Computers & Geosciences 35, 1854-1861.

Fisher RA (1930) The genetical theory of natural selection: A complete variorum edition Oxford University Press.

Fitch HS (1948) Ecology of the California ground squirrels on grazing lands. American Midland Naturalist 39, 513-596.

Fitch HS (1949) Study of snake populations in central California. American Midland Naturalist 41, 513-579.

Fitch HS, Twining H (1946) Feeding habits of the Pacific rattlesnake. Copeia 1946, 64- 71.

132

Forde S, Xa, E, et al. (2007a) Gene flow reverses an adaptive cline in a coevolving host and parasitoid interaction. The American Naturalist 169, 794-801.

Forde SE, Thompson JN, Bohannan BJ (2007b) Gene flow reverses an adaptive cline in a coevolving host‐parasitoid interaction. The American Naturalist 169, 794-801.

Forde SE, Thompson JN, Bohannan BJM (2004) Adaptation varies through space and time in a coevolving host-parasitoid interaction. Nature 431, 841-844.

François O, Currat M, Ray N, et al. (2010) Principal component analysis under population genetic models of range expansion and admixture. Molecular Biology and Evolution 27, 1257-1268.

Friman V-P, Buckling A (2013) Effects of predation on real-time host–parasite coevolutionary dynamics. Ecology Letters 16, 39-46.

Furtado MFD, Travaglia-Cardoso SR, Rocha MMT (2006) Sexual dimorphism in venom of Bothrops jararaca (Serpentes: Viperidae). Toxicon 48, 401-410.

Gandon S (2002) Local adaptation and the geometry of host-parasite coevolution. Ecology Letters 5, 246-256.

Gandon S, Capowiez Y, Dubois Y, Michalakis Y, Olivieri I (1996) Local adaptation and gene-for-gene coevolution in a metapopulation model. Proceedings of the Royal Society of London. Series B: Biological Sciences 263, 1003-1009.

Gandon S, Michalakis Y (2002) Local adaptation, evolutionary potential and host– parasite coevolution: Interactions between migration, mutation, population size and generation time. Journal of Evolutionary Biology 15, 451-462.

Gandon S, Nuismer SL (2009) Interactions between genetic drift, gene flow, and selection mosaics drive parasite local adaptation. The American Naturalist 173, 212-224.

Ganz HH, Washburn JO (2006) Relative migration rates and local adaptation in a mosquito-protozoan interaction. Journal of Evolutionary Biology 19, 816-824.

Gavin TA, Sherman PW, Yensen E, May B (1999) Population genetic structure of the northern Idaho ground squirrel (Spermophilus brunneus brunneus). Journal of Mammalogy 80, 156-168.

133

Geffeney S, Brodie Jr. ED, Ruben PC, Brodie Iii ED (2002) Mechanism of adaptation in a predator-prey arms race: TTX-resistant sodium channels. Science 297, 1336- 1339.

Gibbs HL, Chiucchi JE (2011) Deconstructing a complex molecular phenotype: Population-sevel variation in individual venom proteins in Eastern massasauga rattlesnakes (Sistrurus c. catenatus). Journal of Molecular Evolution 72, 383-397.

Gibbs HL, Dawson RJG, Hobson KA (2000) Limited differentiation in microsatellite DNA variation among northern populations of the yellow warbler: Evidence for male-biased gene flow? Molecular Ecology 9, 2137-2147.

Gibbs HL, Mackessy SP (2009) Functional basis of a molecular adaptation: Prey-specific toxic effects of venom from Sistrurus rattlesnakes. Toxicon 53, 672-679.

Gibbs HL, Sanz L, Calvete JJ (2009) Snake population venomics: Proteomics-based analyses of individual variation reveals significant gene regulation effects on venom protein expression in Sistrurus rattlesnakes. Journal of Molecular Evolution 68, 113-125.

Gibbs HL, Sanz L, Sovic MG, Calvete JJ (2013) Phylogeny-based comparative analysis of venom proteome variation in a clade of rattlesnakes (Sistrurus sp.). PLoS ONE 8, e67220.

Gilbert KJ, Andrew RL, Bock DG, et al. (2012) Recommendations for utilizing and reporting population genetic analyses: The reproducibility of genetic clustering using the program structure. Molecular Ecology 21, 4925-4930.

Goldenberg J (2013) Thesis. "Multilocus species delimitation and species tree inference within the western rattlesnake (Crotalus viridis) species complex. San Diego State University

Gomulkiewicz R, Drown DM, Dybdahl MF, et al. (2007) Dos and don'ts of testing the geographic mosaic theory of coevolution. Heredity 98, 249-258.

Gomulkiewicz R, Thompson JN, Holt RD, Nuismer SL, Hochberg ME (2000) Hot spots, cold spots, and the geographic mosaic theory of coevolution. The American Naturalist 156, 156-174.

Goudet J (2005) Hierfstat, a package for R to compute and test hierarchical F-statistics. Molecular Ecology Notes 5, 184-186. 134

Greischar MA, Koskella B (2007) A synthesis of experimental work on parasite local adaptation. Ecology Letters 10, 418-434.

Gren ECK, Kelln W, Person CE, et al. (2017) Geographic variation of venom composition and neurotoxicity in the rattlesnakes Crotalus oreganus and C. helleri: Assessing the potential roles of selection and neutral evolutionary processes in shaping venom variation. In: The Biology of the Rattlesnakes II (eds. Dreslik MJ, Hayes WK, Beaupre SJ, Mackessy SP). ECO Herpetological Publishing and Distribution, Rodeo, N.M.

Griffin D, Anchukaitis KJ (2014) How unusual is the 2012–2014 California drought? Geophysical Research Letters 41, 9017-9023.

Grover CE, Salmon A, Wendel JF (2012) Targeted sequence capture as a powerful tool for evolutionary analysis. American Journal of Botany 99, 312-319.

Hanifin CT, Brodie Jr. ED, Brodie Iii ED (2008) Phenotypic mismatches reveal escape from arms-race coevolution. PLoS Biology 6, 471-482.

Hedin M, Starrett J, Hayashi C (2013) Crossing the uncrossable: Novel trans-valley biogeographic patterns revealed in the genetic history of low-dispersal mygalomorph spiders (Antrodiaetidae, Antrodiaetus) from California. Molecular Ecology 22, 508-526.

Hickerson MJ, Carstens BC, Cavender-Bares J, et al. (2010) Phylogeography’s past, present, and future: 10 years after. Molecular Phylogenetics and Evolution 54, 291-301.

Hoeksema JD, Forde SE (2008) A meta‐analysis of factors affecting local adaptation between interacting species. The American Naturalist 171, 275-290.

Hoekstra HE, Hirschmann RJ, Bundey RA, Insel PA, Crossland JP (2006) A single amino acid mutation contributes to adaptive beach mouse color pattern. Science 313, 101-104.

Holding ML, Biardi JE, Gibbs HL (2016a) Coevolution of venom function and venom resistance in a rattlesnake predator and its squirrel prey. Proceedings of the Royal Society B: Biological Sciences 283, 20152841.

135

Holding ML, Drabeck DH, Jansa SA, Gibbs HL (2016b) Venom resistance as a model for understanding the molecular basis of complex coevolutionary adaptations. Integrative and Comparative Biology 56, 1032-1043.

Janzen DH (1980) When is it coevolution? Evolution 34, 611-612.

Jombart T (2008) Adegenet: A r package for the multivariate analysis of genetic markers. Bioinformatics 24, 1403-1405.

Kaltz O, Gandon S, Michalakis Y, Shykoff JA (1999) Local maladaptation in the anther- smut fungus Microbotryum violaceum to its host plant Silene latifolia: Evidence from a cross-inoculation experiment. Evolution 53, 395-407.

Kawecki TJ, Ebert D (2004) Conceptual issues in local adaptation. Ecology Letters 7, 1225-1241.

Kerr JT, Packer L (1997) Habitat heterogeneity as a determinant of mammal species richness in high-energy regions. Nature 385, 252-254.

Koskella B, Thompson JN, Preston GM, Buckling A (2011) Local biotic environment shapes the spatial scale of bacteriophage adaptation to bacteria. The American Naturalist 177, 440-451.

Kraus RHS, Van Hooft P, Megens H-J, et al. (2013) Global lack of flyway structure in a cosmopolitan bird revealed by a genome wide survey of single nucleotide polymorphisms. Molecular Ecology 22, 41-55.

Kuchta SR, Parks DS, Wake DB (2009) Pronounced phylogeographic structure on a small spatial scale: Geomorphological evolution and lineage history in the salamander ring species Ensatina eschscholtzii in central coastal California. Molecular Phylogenetics and Evolution 50, 240-255.

Laine A-L (2006) Evolution of host resistance: Looking for coevolutionary hotspots at small spatial scales. Proceedings of the Royal Society B: Biological Sciences 273, 267-273.

Legendre P, Fortin M-J (2010) Comparison of the mantel test and alternative approaches for detecting complex multivariate relationships in the spatial analysis of genetic data. Molecular Ecology Resources 10, 831-844.

136

Legendre P, Fortin M-J, Borcard D (2015) Should the Mantel test be used in spatial analysis? Methods in Ecology and Evolution 6, 1239-1247.

Legendre P, Oksanen J, Ter Braak CJF (2011) Testing the significance of canonical axes in redundancy analysis. Methods in Ecology and Evolution 2, 269-277.

Lively CM, Dybdahl MF, Jokela J, Osnas EE, Delph LF (2004) Host sex and local adaptation by parasites in a snail‐trematode interaction. The American Naturalist 164, S6-S18.

Lomonte B, Tsai W-C, Ureña-Diaz JM, et al. (2014) Venomics of new world pit vipers: Genus-wide comparisons of venom proteomes across Agkistrodon. Journal of Proteomics 96, 103-116.

Lopez Pascua L, Hall AR, Best A, et al. (2014) Higher resources decrease fluctuating selection during host–parasite coevolution. Ecology Letters 17, 1380-1388.

Luikart G, Ryman N, Tallmon DA, Schwartz MK, Allendorf FW (2010) Estimation of census and effective population sizes: The increasing usefulness of DNA-based approaches. Conservation Genetics 11, 355-373.

Ma J, Amos CI (2012) Principal components analysis of population admixture. PLOS ONE 7, e40115.

Macartney JM (1989) Diet of the northern pacific rattlesnake, Crotalus viridis oreganus, in British Columbia. Herpetologica 45, 299-304.

Mackessy SP (1988) Venom ontogeny in the pacific rattlesnakes Crotalus viridis helleri and C. v. oreganus. Copeia 1988, 92-101.

Mackessy SP (2008) Venom composition in rattlesnakes: Trends and biological significance. In: Biology of the Rattlesnakes (eds. Hayes WK, Beaman KR, Cardwell MD, Bush SP), pp. 495-510. Loma Linda University Press, Loma Linda, California.

Mackessy SP (2010) Evolutionary trends in venom composition in the western rattlesnakes (Crotalus viridis sensu lato): Toxicity vs. Tenderizers. Toxicon 55, 1463-1474.

Macpherson A, Hohenlohe PA, Nuismer SL (2015) Trait dimensionality explains widespread variation in local adaptation. The American Naturalist 282, 20141570. 137

Margres MJ, Mcgivern JJ, Seavy M, et al. (2015a) Contrasting modes and tempos of venom expression evolution in two snake species. Genetics 199, 165-176.

Margres MJ, Mcgivern JJ, Wray KP, et al. (2014) Linking the transcriptome and proteome to characterize the venom of the eastern diamondback rattlesnake (Crotalus adamanteus). Journal of Proteomics 96, 145-158.

Margres MJ, Walls R, Suntravat M, et al. (2016) Functional characterizations of venom phenotypes in the eastern diamondback rattlesnake (Crotalus adamanteus) and evidence for expression-driven divergence in toxic activities among populations. Toxicon 119, 28-38.

Margres MJ, Wray KP, Seavy M, et al. (2015b) Phenotypic integration in the feeding system of the eastern diamondback rattlesnake (Crotalus adamanteus). Molecular Ecology 24, 3405-3420.

Massey DJ, Calvete JJ, Sánchez EE, et al. (2012) Venom variability and envenoming severity outcomes of the Crotalus scutulatus scutulatus (mojave rattlesnake) from southern Arizona. Journal of Proteomics 75, 2576-2587.

Matocq MD, Kelly PA, Phillips SE, Maldonado JE (2012) Reconstructing the evolutionary history of an endangered subspecies across the changing landscape of the Great Central Valley of California. Molecular Ecology 21, 5918-5933.

Menezes MC, Furtado MF, Travaglia-Cardoso SR, Camargo ACM, Serrano SMT (2006) Sex-based individual variation of snake venom proteome among eighteen Bothrops jararaca siblings. Toxicon 47, 304-312.

Morgan AD, Gandon S, Buckling A (2005) The effect of migration on local adaptation in a coevolving host-parasite system. Nature 437, 253-256.

Motocq MD (2002) Phylogeographical structure and regional history of the dusky-footed woodrat, Neotoma fuscipes. Molecular Ecology 11, 229-242.

Nachman MW, Hoekstra HE, D'agostino SL (2003) The genetic basis of adaptive melanism in pocket mice. Proceedings of the National Academy of Sciences 100, 5268-5273.

Nei M, Maruyama T, Chakraborty R (1975) The bottleneck effect and genetic variability in populations. Evolution, 1-10.

138

Nuismer SL, Doebeli M, Browning D (2005) The coevolutionary dynamics of antagonistic interactions mediated by quantitative traits with evolving variances. Evolution 59, 2073-2082.

Nuismer Scott l, Gandon S (2008) Moving beyond common‐garden and transplant designs: Insight into the causes of local adaptation in species interactions. The American Naturalist 171, 658-668.

Nuismer SL, Gomulkiewicz R, Ridenhour BJ (2010) When is correlation coevolution? The American Naturalist 175, 525-537.

Nuismer SL, Ridenhour BJ, Oswald BP (2007) Antagonistic coevolution mediated by phenotypic differences between quantitative traits. Evolution 61, 1823-1834.

Nuismer SL, Thompson JN (2006) Coevolutionary alternation in antagonistic interactions. Evolution 60, 2207-2217.

Oppliger A, Vernet R, Baez M (1999) Parasite local maladaptation in the canarian lizard Ggallotia galloti (Reptilia: Lacertidae) parasitized by haemogregarian blood parasite. Journal of Evolutionary Biology 12, 951-955.

Owings DH, Borchert M (1975) Correlates of burrow location in Beechey ground squirrels. The Great Basin Naturalist 35, 402-404.

Palumbi SR, Baker CS (1994) Contrasting population structure from nuclear intron sequences and mtDNA of humpback whales. Molecular Biology and Evolution 11, 426-435.

Parchman TL, Buerkle CA, Soria-Carrasco V, Benkman CW (2016) Genome divergence and diversification within a geographic mosaic of coevolution. Molecular Ecology 25, 5705-5718.

Pawlak J (2006) Denmotoxin, a three-finger toxin from the colubrid snake Boiga dendrophila (mangrove catsnake) with bird-specific activity. Journal of Biological Chemistry 281, 29030-29041.

Penczykowski RM, Laine A-L, Koskella B (2016) Understanding the ecology and evolution of host–parasite interactions across scales. Evolutionary Applications 9, 37-52.

139

Petit N, Barbadilla A (2009) Selection efficiency and effective population size in drosophila species. Journal of Evolutionary Biology 22, 515-526.

Phuong MA, Lim MCW, Wait DR, Rowe KC, Moritz C (2014) Delimiting species in the genus Otospermophilus (Rodentia: Sciuridae), using genetics, ecology, and morphology. Biological Journal of the Linnean Society 113, 1136-1151.

Pineda E, Lobo JM (2012) The performance of range maps and species distribution models representing the geographic variation of species richness at different resolutions. Global Ecology and Biogeography 21, 935-944.

Pomento AM, Perry BW, Denton RD, Gibbs HL, Holding ML (2016) No safety in the trees: Local and species-level adaptation of an arboreal squirrel to the venom of sympatric rattlesnakes. Toxicon 118, 149-155.

Poran NS, Coss RG, Benjamini E (1987) Resistance of California ground squirrels (Spermophilus beecheyi) to the venom of the northern Pacific rattlesnake (Crotalus viridis oreganus): A study of adaptive variation. Toxicon 25, 767-777.

Price AL, Patterson NJ, Plenge RM, et al. (2006) Principal components analysis corrects for stratification in genome-wide association studies. Nature Genetics 38, 904- 909.

Pritchard JK, Stephens M, Donnelly P (2000) Inference of population structure using multilocus genotype data. Genetics 155, 945-959.

Prugnolle F, Théron A, Pointier JP, et al. (2005) Dispersal in a parasitic worm and its two hosts: Consequence for local adaptation. Evolution 59, 296-303.

Putman B, Coss R, Clark R (2015) The ontogeny of antipredator behavior: Age differences in California ground squirrels (Otospermophilus beecheyi) at multiple stages of rattlesnake encounters. Behavioral Ecology and Sociobiology, 1-11.

Putman BJ, Barbour MA, Clark RW (2016) The foraging behavior of free-ranging rattlesnakes (Crotalus oreganus) in California ground squirrel (Otospermophilus beecheyi) colonies. Herpetologica 72, 55-63.

Putman BJ, Lind C, Taylor EN (2013) Does size matter? Factors influencing the spatial ecology of northern Pacific rattlesnakes (Crotalus oreganus oreganus) in central California. Copeia 2013, 485-492.

140

R Development Core Team (2015) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Rambaut A, Suchard M, Xie D, Drummond A (2014) Tracer v1. 6.

Remigio EA, Duda Jr TF (2008) Evolution of ecological specialization and venom of a predatory marine gastropod. Molecular Ecology 17, 1156-1162.

Ridenhour BJ (2005) Identification of selective sources: Partitioning selection based on interactions. The American Naturalist 166, 12-25.

Ridenhour BJ, Nuismer SL (2007) Polygenic traits and parasite local adaptation. Evolution 61, 368-376.

Rokyta DR, Wray KP, Mcgivern JJ, Margres MJ (2015) The transcriptomic and proteomic basis for the evolution of a novel venom phenotype within the timber rattlesnake (Crotalus horridus). Toxicon 98, 34-48.

Roth O, Keller I, Landis SH, Salzburger W, Reusch TBH (2012) Hosts are ahead in a marine host-parasite coevolutionary arms race: Innate immune system adaptation in pipefish Syngnathus typhle against Vibrio phylotypes. Evolution 66, 2528- 2539.

Rundell RJ, Price TD (2009) Adaptive radiation, nonadaptive radiation, ecological speciation and nonecological speciation. Trends in Ecology & Evolution 24, 394- 399.

Rundus AS, Owings DH, Joshi SS, Chinn E, Giannini N (2007) Ground squirrels use an infrared signal to deter rattlesnake predation. Proceedings of the National Academy of Sciences 104, 14372-14376.

Savage AE, Zamudio KR (2011) MHC genotypes associate with resistance to a frog- killing fungus. Proceedings of the National Academy of Sciences 108, 16705- 16710.

Schluter D (1996) Adaptive radiation along genetic lines of least resistance. Evolution 50, 1766-1774.

Schluter D (2009) Evidence for ecological speciation and its alternative. Science 323, 737-741.

141

Secor SM, Nagy KA (1994) Bioenergetic correlates of foraging mode for the snakes Crotalus cerastes and Masticophis flagellum. Ecology 75, 1600-1614.

Shokri Bousjein N, Gardner MG, Schwarz MP (2016) Small effective population sizes of bee social parasites compared to their hosts raise important questions for evolutionary arms race. Journal of Zoology 299, 17-26.

Sicard D, Pennings PS, Grandclément C, et al. (2007) Specialization and local adaptation of a fungal parasite on two host plant species as revealed by two fitness traits. Evolution 61, 27-41.

Siepielski AM, Benkman CW (2004) Interactions among moths, crossbills, squirrels, and lodgepoll pine in a geographic selection mosaic. Evolution 58, 95-101.

Slatkin M (1985) Gene flow in natural populations. Annual Review of Ecology and Systematics 16, 393-430.

Slotte T, Foxe JP, Hazzouri KM, Wright S (2010) Genome-wide evidence for efficient positive and purifying selection in Capsella grandiflora, a plant species with a large effective population size. Molecular Biology and Evolution 27, 1813-1821.

Sovic MG, Fries AC, Gibbs HL (2015) Aftrrad: A pipeline for accurate and efficient de novo assembly of RADseq data. Molecular Ecology Resources 15, 1163-1171.

Sovic MG, Fries AC, Gibbs HL (2016) Origin of a cryptic lineage in a threatened reptile through isolation and historical hybridization. Heredity 117, 358-366.

Sparks AM, Lind C, Taylor EN (2015) Diet of the northern Pacific rattlesnake (Crotalus o. oreganus) in California. Herpetological Review 46, 161-165.

Start D, Gilbert B (2016) Host–parasitoid evolution in a metacommunity. Proceedings of the Royal Society of London B: Biological Sciences 283, 20160477.

Stokes AN, Ray AM, Buktenica MW, et al. (2015) Otter predation on Taricha granulosa and variation in tetrodotoxin levels with elevation. Northwestern Naturalist 96, 13-21.

Sullivan J, Demboski JR, Bell KC, et al. (2014) Divergence with gene flow within the recent chipmunk radiation (Tamias). Heredity 113, 185-194.

142

Sullivan TJ, Faeth SH (2004) Gene flow in the endophyte Neotyphodium and implications for coevolution with Festuca arizonica. Molecular Ecology 13, 649- 656.

Sunagar K, Undheim EaB, Scheib H, et al. (2014) Intraspecific venom variation in the medically significant southern pacific rattlesnake (Crotalus oreganus helleri): Biodiscovery, clinical and evolutionary implications. Journal of Proteomics 99, 68-83.

Sutton TL, Riegler M, Cook JM (2016) One step ahead: A parasitoid disperses farther and forms a wider geographic population than its fig wasp host. Molecular Ecology 25, 882-894.

Talluto MV, Benkman CW (2014) Conflicting selection from fire and seed predation drives fine-scaled phenotypic variation in a widespread North American conifer. Proceedings of the National Academy of Sciences 111, 9543-9548.

Taylor EN, Malawy MA, Browning DM, Lemar SV, Denardo DF (2005) Effects of food supplementation on the physiological ecology of female western diamond-backed rattlesnakes (Crotalus atrox). Oecologia 144, 206-213.

Templ M, Hron K, Filzmoser P (2011) Robcompositions: An R-package for robust statistical analysis of compositional data. In: Compositional data analysis, pp. 341-355. John Wiley & Sons, Ltd.

Thompson JN (2005a) Coevolution: The geographic mosaic of coevolutionary arms races. Current Biology 15, 992-994.

Thompson JN (2005b) The geographic mosaic of coevolution The University of Chicago Press, Chicago, Illinois.

Thompson JN, Burdon JJ (1992) Gene-for-gene coevolution between plants and parasites. Nature 360, 121-125.

Thompson JN, Schwind C, Guimarães PR, Friberg M (2013) Diversification through multitrait evolution in a coevolving interaction. Proceedings of the National Academy of Sciences 110, 11487-11492.

Thrall PH, Burdon JJ, Bever JD (2002) Local adaptation in the Linum marginale- Melampsora lini host-pathogen interaction. Evolution 56, 1340-1351.

143

Toju H (2008) Fine-scale local adaptation of weevil mouthpart length and Camellia pericarp thickness: Altitudinal gradient of a putative arms race. Evolution 62, 1086-1102.

Toju H, Sota T (2006) Imbalance of predator and prey armament: Geographic clines in phenotypic interface and natural selection. American Naturalist 167, 105-117.

Towers SR, Coss RG (1990) Confronting snakes in the burrow: Snake-species discrimination and antisnake tactics of two California ground squirrel populations. Ethology 84, 177-192.

Vogwill T, Fenton A, Buckling A, et al. (2009) Source populations act as coevolutionary pacemakers in experimental selection mosaics containing hotspots and coldspots. The American Naturalist 173, E171-E176.

Vonk FJ, Casewell NR, Henkel CV, et al. (2013) The king cobra genome reveals dynamic gene evolution and adaptation in the snake venom system. Proceedings of the National Academy of Sciences 110, 20651-20626.

Vos M, Birkett PJ, Birch E, Griffiths RI, Buckling A (2009) Local adaptation of bacteriophages to their bacterial hosts in soil. Science 325, 833-833.

Vucetich JA, Waite TA, Nunney L (1997) Fluctuating population size and the ratio of effective to census population size. Evolution 51, 2017-2021.

Wallace RL, Diller LV (1990) Feeding ecology of the rattlesnake, Crotalus viridis oreganus, in northern Idaho. Journal of Herpetology 24, 246-253.

Wang IJ, Bradburd GS (2014) Isolation by environment. Molecular Ecology 23, 5649- 5662.

Wang IJ, Glor RE, Losos JB (2013) Quantifying the roles of ecology and geography in spatial genetic divergence. Ecology Letters 16, 175-182.

Wang J, Santiago E, Caballero A (2016) Prediction and estimation of effective population size. Heredity 117, 193-206.

Waples RS (2016) Life-history traits and effective population size in species with overlapping generations revisited: The importance of adult mortality. Heredity 117, 241-250.

144

Westergaard H, Jarnum S, Preisig R, et al. (1970) Degradation of albumin and igg at high altitude. Journal of Applied Physiology 28, 728-732.

Whiteman NK, Matson KD, Bollmer JL, Parker PG (2006) Disease ecology in the Galápagos hawk (Buteo galapagoensis): Host genetic diversity, parasite load and natural antibodies. Proceedings of the Royal Society B: Biological Sciences 273, 797-804.

Whitlock MC, Mccauley DE (1999) Indirect measures of gene flow and migration: Fst ≠ 1/(4Nm + 1). Heredity 82, 117-125.

Williams V, White J, Schwaner TD, Sparrow A (1988) Variation in venom proteins from isolated populations of tiger snakes (Notechis ater niger, N. scutatus) in South Australia. Toxicon 26, 1067-1075.

Wray KP, Margres MJ, Seavy M, Rokyta DR (2015) Early significant ontogenetic changes in snake venoms. Toxicon 96C, 74-81.

Wuster W, Daltry, J.C., Thorpe, R.S. (1999.) Can diet explain intraspecific venom variation? Toxicon 37, 253-258.

Yates MC, Fraser DJ (2014) Does source population size affect performance in new environments? Evolutionary Applications 7, 871-882.

Yoder Jeremy b, Nuismer Scott l (2010) When does coevolution promote diversification? The American Naturalist 176, 802-817.

Zangerl AR, Stanley MC, Berenbaum MR (2008) Selection for chemical trait remixing in an invasive weed after reassociation with a coevolved specialist. Proceedings of the National Academy of Sciences 105, 4547-4552.

145