Phenotypic Plasticity and the Post-Modern Synthesis: Integrating Evo-Devo and Quantitative Genetics in Theoretical and Empirical Studies

Home , Fitness landscape, Phenotypic plasticity

Utah State University DigitalCommons@USU

All Graduate Theses and Dissertations Graduate Studies

12-2008

Alison G. Scoville Utah State University

Follow this and additional works at: https://digitalcommons.usu.edu/etd

Part of the Biology Commons, and the Genetics Commons

Recommended Citation Scoville, Alison G., "Phenotypic Plasticity and the Post-Modern Synthesis: Integrating Evo-Devo and Quantitative Genetics in Theoretical and Empirical Studies" (2008). All Graduate Theses and Dissertations. 212. https://digitalcommons.usu.edu/etd/212

This Dissertation is brought to you for free and open access by the Graduate Studies at DigitalCommons@USU. It has been accepted for inclusion in All Graduate Theses and Dissertations by an authorized administrator of DigitalCommons@USU. For more information, please contact [email protected]. PHENOTYPIC PLASTICITY AND THE POST-MODERN SYNTHESIS:

INTEGRATING EVO-DEVO AND QUANTITATIVE GENETICS

IN THEORETICAL AND EMPIRICAL STUDIES

Alison G. Scoville

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

DOCTOR OF PHILOSOPHY

Biology

Approved:

______Dr. Michael E. Pfrender Dr. Edmund D. Brodie, Jr. Major Professor Committee Member

______Dr. Todd A. Crowl Dr. Carol D. von Dohlen Committee Member Committee Member

______Dr. Paul G. Wolf Dr. Byron Burnham Committee Member Dean of Graduate Studies

UTAH STATE UNIVERSITY Logan, Utah

2008 2

Phenotypic Plasticity and the Post-Modern Synthesis:

Integrating Evo-devo and Quantitative Genetics

in Theoretical and Empirical Studies

Alison G. Scoville, Doctor of Philosophy

Utah State University, 2008

Major Professor: Dr. Michael E. Pfrender Department: Biology

Mainstream evolutionary biology lacks a mature theory of phenotype. Following from the

Modern Synthesis, researchers tend to assume an unrealistically simple mapping of genotype to phenotype, or else trust that the complexities of developmental architecture can be adequately captured by measuring trait variances and covariances. In contrast, the growing field of evolutionary developmental biology (evo- devo) explicitly examines the relationship between developmental architecture and evolutionary change, but lacks a rigorous quantitative and predictive framework. In my dissertation, I strive to integrate quantitative genetics and evo-devo, using both theoretical and empirical studies of plasticity. My first paper explores the effect of realistic development on the evolution of phenotypic plasticity when there is migration between two discrete environments. The model I use reveals that nonadditive developmental interactions can constrain the evolution of phenotypic plasticity in the presence of stabilizing selection. In my second paper, I examine the manner in which the genetically controlled responsiveness of traits to each other is shaped by selection and can in turn shape the phenotypic response to selection. Here, results indicate that developmental entanglement through plasticity can facilitate rapid multivariate adaptation in response to a novel selective pressure. In my final paper, I examine patterns of gene expression underlying ancestral plasticity and adaptive loss of melanin in Daphnia melanica. My results indicate that the developmental mechanism underlying ancestral plasticity has been co-opted to facilitate rapid adaptation to an introduced predator. (71 pages) 4

To four generations of my family, for their tireless support:

Gavin and Gwyndolyn Scoville

Gerald W. Scoville

Robert and Elizabeth Gabel

and

William and Harriet Gabel 5 ACKNOWLEDGMENTS

Thanks to my advisor, Michael Pfrender, for his longstanding support and guidance. Thanks also to my committee members, Edmund Brodie, Jr., Todd Crowl, Carol von Dohlen, and Paul Wolf, for their encouragement and constructive comments. Leigh Latta and Megan Kanaga provided magnanimous assistance in the wet lab. Greg White contributed to identifying Daphnia genes of interest, Darl Flake developed the melanin assay, and Michelle White participated in real-time quantitative PCR. Shaun

Bushman provided expertise and access to equipment for real-time quantitative PCR. Ronald Knapp shared his detailed knowledge of the alpine lakes of the Sierra Nevada. Funding was provided by a

National Science Foundation grant to Michael Pfrender, as well as two Vice President for Research

Fellowships, a Zobell Scholarship, and a Dissertation Writing Grant.

My heartfelt thanks go to Gerald Scoville for his munificent patience, assistance, and faith. My heartfelt thanks also go to Robert and Elizabeth Gabel, grandparents extraordinaire, who have each contributed their own strengths. William and Harriet Gabel have provided generous and much-appreciated support throughout. Most of all, thanks to Gavin and Gwyndolyn Scoville for providing exuberance, creativity, and many thoughtful, late-night hours.

Alison G. Scoville 6 CONTENTS

Page

ABSTRACT……………………………………………………………………………………… iii

DEDICATION…………………………………………………………………………………… iv

ACKNOWLEDGMENTS……………………………………………………………………….. v

LIST OF TABLES………………………………………………………………………………. vii

LIST OF FIGURES……………………………………………………………………………… viii

CHAPTER

I. INTRODUCTION…………………………………………………………….. 1

II. EPISTASIS, STABILIZING SELECTION, AND CONSTRAINTS ON THE EVOLUTION OF PHENOTYPIC PLASTICITY: A DEVELOPMENTAL PERSPECTIVE……………………………………….. 3

III. GENETIC ACCOMMODATION FACILITATES RAPID ADAPTATION ALONG CRYPTIC GENETIC LINES OF LEAST RESISTANCE………….. 26

IV. FROM ADAPTIVE PLASTICITY TO RAPID ADAPTATION: EVOLUTION OF DAPHNIA GENE EXPRESSION IN RESPONSE TO AN INTRODUCED PREDATOR………………………………………. 37

IV. CONCLUSION……………………………………………………………….. 53

REFERENCES ………………………………………………………………………………….. 54

CURRICULUM VITAE………………………………………………………………………… 62 7 LIST OF TABLES

Table Page

1 Primers used for real-time quantitative PCR ………………………………………………… 42

2 ANOVA results for phenotypic traits ...... ……………………………………………………. 45

3 Means and standard errors of all traits …………………………………………...... 47

4 ANCOVA results for time to maturity ……………………………………………………... 48

5 MANOVA results for gene expression ……………………………………………………… 48

6 ANOVA results for gene expression …………………………………………………………. 50

7 ANCOVA results for melanin ……………………………………………………………….. 50 8 LIST OF FIGURES

Figure Page

1 Phenotypic landscape with plasticity and epistasis ………………………………………… 8

2 Evolutionary trajectory under three different models ……………………………………... 16

3 Genetic (co)variances estimated at two different points on the phenotypic landscape……………………………………………………..………………... 17

4 Plasticity and maladaptation as a function of migration rate………………………………. 18

5 The effect of directional selection, stabilizing selection, and gene flow…………………... 19

6 Reaction norms at equilibrium …………………………………………………………….. 21

7 Trajectory of mean DVM and melanin ……………………………………………………. 31

8 Evolutionary change in plasticity ………………………..……………...... 32

9 Response of melanin and gene expression to UV radiation …………...... 46

10 Patterns of gene expression in 3-dimensional space…….………...... 49 9 CHAPTER 1

INTRODUCTION

Phenotypic plasticity, or the ability of a single genotype to alter its phenotype in response to the environment, is arguably a ubiquitous phenomenon. Nearly every organism is plastic in at least one trait, and nearly every trait is plastic in at least one organism. Very early in the history of evolutionary biology, researchers recognized that the phenotype of an organism is a flexible and dynamic system that interacts with the environment in a genetically determined, and thus evolvable manner (e.g., Wolterek 1909). In addition, the potential for plasticity to influence the rate and direction of evolution has been championed from the earliest days of evolutionary research (Baldwin 1896; Waddington 1942, 1952, 1953). In contrast to this view, the work of Gregor Mendel in the mid-1800s emphasized the deterministic nature of genes

(Sarkar 1999). The traits that Mendel studied, such as seed texture and flower color, were expressed according to a strict pattern of allelic inheritance, unaffected by environmental variation within his experiments.

Following the publication of The Origin of Species by Means of Natural Selection (Darwin 1859), the apparent conflict between Mendel’s discrete traits and Darwin’s emphasis on continuous traits engendered a fierce division within evolutionary biology. This divide was bridged by The Modern

Synthesis of the 1930’s, following the realization that continuous variation in traits could result from the effect of many different discrete alleles (reviewed by Provine 1971). The Modern Synthesis allowed for immense progress, including the development of modern-day population and quantitative genetics. In the process, however, most evolutionary biologists lost sight of the importance of developmental processes, including phenotypic plasticity. To date, population geneticists have focused on relatively simple models for the translation of genetic information into phenotype, while quantitative geneticists have substituted the black box of genetic variances and covariances (the G-matrix) for any detailed understanding of developmental interactions involved in the formation of phenotypic traits.

Since the 1980s, the subdiscipline of evolutionary developmental biology (evo-devo) has grown exponentially, out of a desire to re-integrate evolution and development. Researchers from this tradition have primarily focused on elucidating the proximate developmental mechanisms (e.g., changes in Hox 10 genes) underlying macroevolutionary shifts in phenotype. In contrast, population and quantitative geneticists tend to focus on ultimate mechanisms (e.g., selection and mutation) shaping microevolutionary change in phenotype. From an evo-devo perspective, most evolutionary biologists lack a comprehensive understanding of the translation of genotype to phenotype, and thus fail to see how this relationship might fundamentally shape evolutionary patterns. From a population and quantitative genetic perspective, evo- devo provides intriguing stories, but lacks a rigorous quantitative and predictive theoretical framework.

Phenotypic plasticity, a developmental phenomenon that can clearly play a role in microevolution, provides a natural meeting point between evo-devo and mainstream evolutionary biology. In chapters two and three, I explore microevolutionary change in plasticity through two theoretical studies. These studies employ an expansion of the mathematical framework of population and quantitative genetics that explicitly incorporates the manner in which developmental factors interact in the formation of a phenotype. In chapter four, I use empirical data to examine whether the developmental mechanism underlying ancestral plasticity in Daphnia melanica has been co-opted to facilitate rapid adaptation to an introduced predator. 11 CHAPTER 2

EPISTASIS, STABILIZING SELECTION, AND CONSTRAINTS ON THE EVOLUTION OF

PHENOTYPIC PLASTICITY: A DEVELOPMENTAL PERSPECTIVE

Evolutionary developmental biology (evo-devo) has recently expanded from a focus on macroevolutionary patterns to include both microevolution (micro-evo-devo; Johnson 2007) and ecological interactions (eco-devo; Callahan et al. 1997; Dusheck 2002). This shift in focus has propelled phenotypic plasticity, the nexus of genetics, development, ecology, and natural selection, into a central and broadly integrative role in evolutionary research (DeWitt and Scheiner 2004; Sultan 2005; Ackerly and Sultan

2006). The last several years have yielded rapid progress in identifying genes and patterns of genetic interactions underlying plastic responses (e.g., Stinchcombe et al. 2004; Weinig and Schmitt 2004), elucidating the developmental architecture of plastic traits (Schlichting and Smith 2002; Suzuki and

Nijhout 2006), and exploring the functional (Schmitt et al. 1999, 2003) and ecological (Stracke et al. 2002;

Kessler et al. 2004; Peacor et al. 2006) significance of plasticity. This wealth of research highlights an old

(Wolterek 1909), subsequently abandoned (see Sarkar 1999, 2004), but newly reinvigorated view of the phenotype as a flexible and dynamic system that interacts with the environment in a genetically determined, and thus evolvable, manner.

Recent mathematical and theoretical studies of plasticity have focused on including increasingly realistic details of ecology and natural history (e.g., Sultan and Spencer 2002; de Jong and Behera 2002; de

Jong 2005), but have yet to integrate current developmental insights. For example, accumulating evidence suggests that reaction norms are a result of complex interactions between multiple genes and developmental pathways (reviewed by Windig et al. 2004), and thus almost certainly involve nonadditive interactions at one or more levels (Frankel and Schork 1996; reviewed by Templeton 2000). In addition, several groundbreaking empirical studies confirm the existence of epistatic interactions in the formation of plastic phenotypes (Caicedo et al. 2004; Remold and Lenski 2004). Despite this, models so far have failed to incorporate nonadditive interactions in the construction of plastic traits: the “epistasis” models of the past

(sensu Scheiner and Lyman 1989; Scheiner 1998) involve G x E interactions but assume additive gene action within environments. The need to integrate epistasis into plasticity research has been acknowledged 12 for some time (Berrigan and Scheiner 2004), but traditional modeling structures have failed to yield a clear way forward. Although gametic models provide an explicit way to incorporate the details of genetic architecture, it is currently difficult to determine how to parameterize the relationship between individual loci and complex, multilocus plastic traits in a way that is either realistic or general enough to lead to widely interesting conclusions. Meanwhile, the quantitative genetic framework developed during the

Modern Synthesis, originally incompatible with plasticity (Sarkar 1999, 2004; Sultan 1992), is assumed to supervene over the molecular and developmental details of plastic trait formation, even though formal testing of this assumption yields conflicting results (Czesak et al. 2006) or, more often, is not attempted at all.

In this study, I explore the effect of nonadditive interactions on the evolution of phenotypic plasticity, making use of a revolutionary mathematical and conceptual framework derived by Rice (2002,

2004a, 2004b; see Wolf 2002). This framework encompasses quantitative genetics as a special case but is far more flexible, allowing us to explore the effect of different types of selection, under essentially any form of genetic architecture. Analysis centers on the concept of a phenotypic landscape, which is constructed by plotting the value of a phenotype against the values of the underlying genetic and environmental factors that interact in its development. Calculation of selection vectors and resultant evolution reveals the interplay between how selective pressure on a phenotype translates into change in the underlying genetic factors, and how developmental architecture influences the trajectory of evolution. In this study, I employ the Rice framework to combine a developmental perspective with a quantitative genetic approach to model the evolution of phenotypic plasticity in the presence of population structure and spatial environmental heterogeneity.

The synthesis of development and quantitative genetics allows for a fresh perspective on several outstanding issues in plasticity research. For example, following the observation that organisms do not generally possess the ability to respond optimally to all possible environments, costs and constraints to the evolution of plasticity have received a great deal of attention (reviewed by Berrigan and Scheiner 2004; van

Kleunen and Fischer 2005). Here I demonstrate that epistasis represents a currently unrecognized constraint on the evolution of plasticity. In addition, comparison of the effect of nonadditive genetic interactions with nonadditive interactions between genotype and environment reveals that a similar 13 constraint applies whenever the precise value of environmental cues varies within a single selective environment, even in the absence of epistatic genetic interaction; this provides a novel way to view and analyze the relationship between cue reliability and the evolution of plasticity. Both results rely on the explicit incorporation of stabilizing selection within environments, and comparison between them makes use of the close conceptual kinship between epistasis and plasticity, as well as the benefit of integrating approaches used in these two fields of research (Brodie 2000; Rice 2004a). Finally, I reveal that the variance-covariance matrix (G), assumed to capture the relevant aspects of genetic architecture in traditional quantitative genetic character-state (Falconer 1952; Via and Lande 1985) and polynomial models (Scheiner and Lyman 1989; Gavrilets and Scheiner 1993a, 1993b), is unstable across the phenotypic landscape, and yields dynamics at odds with my more developmentally explicit model.

Theoretical Construct

My conceptual framework is adapted from Via and Lande’s (1985) genetic model of subdivided populations under soft selection. This model considers the evolution of a single plastic trait when there is some migration between two discrete environments with different phenotypic optima. The expression of this trait in the two environments is treated as two different quantitative traits, termed character states, with some level of genetic correlation (Falconer 1952; Via and Lande 1985). Each generation, migration between environments is followed by within-environment mating, reproduction, and selection; iteration of this process yields the evolutionary trajectory over time. Population size is regulated locally, so that each population contributes the same proportion of migrants each generation, regardless of its relative fitness. In keeping with quantitative genetic theory, assumptions of this model include a joint multivariate normal distribution of the two character states (Lande 1979), as well as large population sizes and weak selection, so that genetic variation depleted by selection is replaced by mutation (Lande 1976, 1980).

Here, I introduce the simplest level of developmental realism into Via and Lande’s (1985) construct by incorporating the observation that many plastic traits involve a complex series of developmental interactions. In particular, the formation of such traits often involves the production of at least one internal signal, such as a hormone, that acts on a separate part of the organism to trigger creation of the phenotype (Callahan et al.1997; Schlichting and Smith 2002; reviewed by Windig et al. 2004). A 14 number of studies indicate that the level of signal can evolve independently from the sensitivity to that signal (e.g., Hau 2007 and references therein), so that the phenotype produced by any particular individual depends on its environment, the level of internal signal produced in response to that environment, and the phenotypic response to that level of signal. In one recent example, Suzuki and Nijhout (2006) examined the hormonal regulation of melanin synthesis in a mutant line of tobacco hornworm (Manduca sexta). In this case, heat shock induces an increase in juvenile hormone (JH), which affects an upstream regulatory control of melanin synthesis to produce green instead of black coloration. The authors were able to show that a line selected for greater plasticity showed increased juvenile hormone in response to heat shock, whereas a line selected for less plasticity showed decreased sensitivity to JH. In general, if the level of internal signal and the response to that signal interact in a nonadditive fashion, the phenotype depends upon an epistatic interaction between those developmental factors. In this paper, I explore the very simple case of a linear shape to the reaction norm of phenotype in response to internal signal value.

Analysis of my models is accomplished through use of a mathematical framework derived by Rice

(2002). Each generation, I calculate two selection vectors that determine how both directional and stabilizing selection translate into change in the mean values of the underlying genetic factors controlling the level of signal produced in response to the environment and the sensitivity to that signal. These selection vectors depend on three elements: 1) the shape of the phenotype landscape, which is created by plotting the expected phenotype against both genetic and environmental underlying factors that influence its production; 2) the distribution of variation in the underlying factors; and 3) the shape of the individual fitness landscape. Following Rice (2002), my analysis relies on the use of tensors, which can be thought of as a generalization of vectors and matrices and represented as an array of numbers. The shape of the phenotypic landscape is captured in a series of tensors (termed D tensors) that contain all possible partial derivatives of that landscape. The distribution of variation is described by an additional series of tensors

(termed P tensors) that contain all the moments and mixed moments of the distribution of underlying factors. As in traditional quantitative genetic analysis, the degree to which selection translates into change in the next generation depends on the heritability of each underlying factor. In this case, the response vector is determined by multiplying the sum of the selection vectors by a heritability matrix in which environmental factors are given a heritability of 0. Although this framework would also allow us to 15 calculate the effect of selection on higher moments of the distribution of underlying factors, as well as the effect of additional selection vectors, I leave these aspects to future studies.

Mathematical Framework

I examine the evolution of a single trait of interest (φ) when there is some level of migration between two populations (i and j) that inhabit different selective environments (A and B, respectively). In each environment, individual fitness (w) is described by a Gaussian surface with an optimum (θ) that differs between environment A and environment B. For each individual, a discrete cue associated with its environment elicits the production of an internal signal, which then acts to trigger development of the phenotype according to a linear reaction norm. The level of internal signal produced in each environment is treated as a separate quantitative trait, consistent with the character state approach (e.g., Via and Lande

1985), whereas the slope and intercept of the reaction norm to signal value are treated as two separate quantitative traits, consistent with the polynomial approach (e.g., Scheiner 1998). In this case, each individual produces a phenotype according to the following functions:

! A = s Au1 + u2 , (1) !B = sBu1 + u2

where ! A is the phenotype that would be produced in environment A and !B is the phenotype that would be produced in environment B. The factors sA and sB represent the internal signal produced in response to the discrete environmental cues associated with the two environments, whereas the factors u1 and u2 represent the slope and intercept of the reaction norm of the phenotype produced in response to signal value. Epistatic interaction between the signal value and slope is evident in the multiplication of these factors (sA or sB, and u1), so that the effect of a change in one depends upon the value of the other. The nonadditive effect of this interaction is illustrated by curvature in the resultant phenotypic landscape (Fig.

1), as additive interactions give rise to only planar phenotypic landscapes. This curvature also illustrates a 16

Figure 1. Phenotypic landscape with plasticity and epistasis. The surface represents a three-dimensional slice, taken in a single environment where the intercept of the reaction norm is equal to 0. Regions of greater phenotypic plasticity correspond to areas of increased slope, where population-level variance in signal value and slope of the reaction norm translate into greater population-level phenotypic variance.

change in phenotypic variance with movement across the phenotypic landscape, indicating a role for stabilizing selection in affecting the means of the underlying factors. The signal value in each environment is scaled so that the midpoint between them is equal to 0; u2 thus defines the intercept of the reaction norm at that point, and can be interpreted as a measure of constitutive phenotypic expression, independent of environmental context. All four underlying factors (sA, sB, u1, and u2) are given a within-population variance of 1, for simplicity, and the covariance between them is set to 0, due to within-population panmixis and assumed genetic independence. The mean values of these factors in population i and j comprise the 8 dynamical variables of this system, and are contained in the vectors ui and u j .

Each generation, migration takes place between the populations such that mi represents the proportion of population i made up of breeding immigrants from population! j and! vice versa. For the m vector of underlying factors, the mean value in population i after migration and reproduction ( u i ) is

! 17 m ui = miu j + [1" mi ]u i ; (2)

the subscripts! are reversed for population j (adapted from Hendry et al. 2001). The effect of migration on higher moments of the distribution of underlying factors is ignored in this study in order to simplify my

construct and make my results more comparable with earlier studies, including Via and Lande (1985). On

this simple phenotypic landscape, however, if the two populations differ in the values of underlying factors,

and if each population is closer to its respective optimum, taking the increased variance and covariance due

to migration into account is expected to increase the effect of both directional and stabilizing selection (see

use of P tensors below); I thus present a conservative estimate of the ability of selection to counter the

effect of migration and constrain the evolution of phenotypic plasticity.

After migration and reproduction, selection acts separately on each population. Following Rice

(2002, 2004a), I calculate the vector Q1, which captures the effect of directional selection, and the vector

Q1,2 , which captures the effect of stabilizing selection, according to the following functions:

1 "w Q = P 2,D1 1 w "# . (3) 1 " 2w Q = P 4 ,D1 $ D2 1,2 2w "# 2

! "w " 2w Here, w is the mean fitness of the population, and and are the first and second derivatives of "# "# 2

the individual fitness surface, evaluated at the mean phenotype. D1 contains the partial first derivatives of φ ! with respect to each underlying factor, ! and is thus! a rank one tensor (i.e., a vector) that points in the direction of maximum change in the phenotype. Multiplying D1 by the first derivative of the individual

fitness surface gives a vector that points in the direction of maximum change in fitness (Rice 2004a),

analogous to the selection gradient β used in multivariate quantitative genetics (Lande 1979; Lande and

Arnold 1983). The tensor P2, which contains the second central moments of the distribution of underlying

factors, is simply the covariance matrix of these factors; taking the inner product of D1 and P2 provides a 18 way to measure the vector D1 in terms of the variation present in the underlying factors, similar to the effect of multiplying β and the G matrix to determine a response vector in multivariate quantitative genetics

(see Lande and Arnold 1983; Rice 2004a). In this case, all of the diagonal elements of P2 (i.e., variances) are equal to 1, whereas all of the off-diagonal elements are equal to zero. The vector Q1 thus gives the selection differential for each of the underlying factors due to directional selection on the phenotype. This vector is analogous to the single selection differential examined by Via and Lande (1985) as well as the majority of quantitative genetic models.

As discussed by Rice (2004a), the length of the vector D1 captures the slope of the phenotypic landscape, and is thus a measure of the phenotypic variance produced by interactions among the underlying factors. If the underlying factors act in a nonadditive manner to produce the phenotype, as in my models, the phenotypic landscape will be curved. Movement to an area of lesser slope on such a landscape corresponds to the evolution of canalization, where the phenotype is buffered from variation in the underlying factors, and such movement will be favored by stabilizing selection. The derivative of the length of D1 with respect to each underlying factor describes the direction in which the slope of the phenotypic landscape changes most quickly, and is captured by the outer product of D1 and the tensor D2, which contains all of the second derivatives of the phenotypic landscape with respect to the underlying factors (Rice 2004a). Taking the inner product of D1 " D2 and the tensor P4, which contains all of the fourth central moments of the distribution of underlying factors, allows for measurement of the change in slope of the phenotypic landscape in terms! of the variation present in the underlying factors. Multiplication of this result by the second derivative of the individual fitness surface yields a description of the direction of maximum change in fitness due to change in phenotypic variation. Ignoring change in higher moments of the distribution of underlying factors due to migration, all of the P4(k,k,k,k) elements are equal to 3, whereas all of the P4(k,k,l,l) elements are equal to 1, keeping in mind that P tensors are symmetrical (e.g.,

P4(k,k,l,l) = P4(k,l,k,l) = 1). Other elements of P4 are zero.

Under the assumptions outlined above, the vectors Q1, and Q1,2 are obtained for population i using

$ u1 i' &2s Ai # & ) 2 $ ! 1 "w 0 1 ' w 0 i & ) i $ ! , (4) Q1 = Q1,2 = 2 w i "#i &s Ai ) 2wi '(i $2u1i ! & ) $ ! % 1 ( % 0 "

where! the mean fitness of population i is calculated as the product of wi and the joint distribution of sA, u1, and u2, integrated analytically over u2 and sA, and numerically over u1. The vectors Q1, and Q1,2 are obtained for population j using

$ 0 ' & 0 # & ) 2 $ ! 1 "w u1 j 1 ' w j 2s j & ) $ Bj ! , (5) Q1 = Q1,2 = 2 w j "# j & sBj ) 2w j '( j $2u1 j ! & ) $ ! % 1 ( % 0 "

where! the mean fitness of population j is obtained in a manner similar to population i, but using wj and the joint distribution of sB, u1, and u2.

In each population, the vector of mean values of the underlying factors after migration, reproduction, and selection ( u ms ) is then calculated as

ms m !u = u + H (Q1 ) (6)

when only directional selection is considered, and

ms m u = u + H (Q1 + Q1,2 ) (7)

20 when both directional and stabilizing selection are taken into account. In these equations, H is a

heritability matrix where each underlying factor is given a heritability of 1, for simplicity (lower and more

realistic values change only the rate of evolution, not the trajectory). The off-diagonal elements of H,

which correspond to genetic correlation between underlying factors, are set to 0 to eliminate genetic

constraints unrelated to the form of developmental interaction under investigation. The vector u ms

becomes the new vector of mean values of the underlying factors u in the next generation, and the entire

cycle is iterated in order to determine the evolutionary trajectory.

! Analysis

I compared the evolutionary trajectory predicted by my developmental model to that predicted by

the classical character-state and polynomial models in order to assess differences in dynamics as well as

end-point equilibria. For this purpose, I arbitrarily chose the following parameter values: fitness optima θA

= 2 and θB = -2; variance of each fitness surface = 64. To maximize insight into the effect of migration, I

used unequal migration rates arbitrarily set at mi = 0.05 and mj = 0.01. Initially, I assumed that populations

i and j were sitting at their local optima, with no mean plasticity in internal signal value ( sA = sB = 0 and

u2 = θA or θB for population i and j, respectively), and essentially no mean plasticity in the response to

signal value ( u1 = 0.01; this value was chosen to avoid an unstable equilibrium! at u1 =0). The variance of

! 1 for all underlying factors means that a population with no mean plasticity in internal signal value or

response! to signal value still contains variance in both elements of overall! reaction norm. The methodology for determining the evolutionary trajectory for the classical character-state and polynomial models is

described below. To test the sensitivity of my model to migration rate, I also obtained equilibrium

phenotypes for each population for a range of migration rates between 0 and panmixis (0.5), maintaining

the same starting point and, in this case only, keeping mi = mj. Given that the system of equations used in

my model was nonlinear and an analytical solution was not tractable, I explored the parameter space for

additional equilibria by running 100 Monte Carlo simulations, both with and without stabilizing selection,

using 100 sets of eight values randomly chosen from the interval -5 to 5 as starting values for the four 21 underlying genetic factors in the two populations. Equilibrium mean values of both the underlying factors and the resultant phenotypes were recorded.

Following the observation that response in any plastic trait is triggered by an environmental cue, which may vary within environments as well as differ in mean value between environments, I made use of the near equivalence of genetic and environmental factors under the framework derived by Rice (2002) to compare my epistatic model with a scenario that involves nonadditive interactions only between genotype and environment. In this case, I transformed the internal signal into a nonheritable environmental cue by

m m ignoring the effect of migration (so that sA = sA and sB = sB ), changing elements of H that correspond to heritability of sA and sB from 1 to 0, and giving mean values such that sA = 2 and sB = "2 . This construct corresponds to a situation! in which! the environmental cue is normally distributed within each environment,! ! rather than discrete, and any individual’s phenotype! depends !upon both the particular cue value it experiences and the underlying genetic factors that determine a continuous reaction norm of phenotype in response to cue value.

To assess the trajectory predicted by classical character state and polynomial models, I used simulated populations to obtain estimates of variance and covariance of both the character states and the slope and intercept of the phenotypic reaction norm, ignoring the underlying genetic architecture. These estimates were obtained at the start point and used throughout the simulation. However, to assess the stability of G1 and G2 across the phenotypic landscape, I compared these estimates with those obtained by assuming that both populations were sitting at the joint plastic optimum (sA = 2; sB = -2; u1 = 1; u2 = 0).

Estimates were calculated in the following manner: for 10 different Monte Carlo replicates, I created a simulated population of 5000 individuals with the defined distribution of underlying factors, and randomly chose 1000 pairs of individuals with replacement. Each pair was allowed to produce a single offspring with values of the underlying factors precisely intermediate to those of its parents, in keeping with my heritability of 1 for underlying factors. Midparent and offspring values were then calculated for phenotype within each environment, as well as the slope and intercept of the phenotypic reaction norm across environments. Averaging across Monte Carlo replicates, the mean midparent-offspring phenotypic covariance within and between environments was used to estimate the variance-covariance matrix G1 22 emplyed in character-state models (Falconer 1952; Via and Lande 1985), whereas the mean midparent-

offspring covariance of and between phenotypic reaction norm slope and intercept was used to estimate the

variance-covariance matrix G2 used in polynomial models (Scheiner and Lyman 1989; Gavrilets and

Scheiner 1993b). The mean variance of phenotype (V1), phenotypic reaction norm slope (V2), and

phenotypic reaction norm intercept (V3), were obtained for each parental population as well.

In the character-state formulation, change in the vector of mean phenotypes expressed in each

environment ( z ) was calculated each generation according to the functions

m ! z i = miz j + [1" mi ]z i , (8) ms m m z i = z i + G1i#(z i )

with the! subscripts reversed for population j (adapted from Hendry et al. 2001 and Via and Lande 1985). In m m this equation, "(z i ) represents the selection gradient evaluated at z i and was calculated as the derivative

of the mean fitness surface, in which mean fitness was obtained by convolution of a normal distribution

with! a variance V1 with the Gaussian individual fitness surface! (Lande and Arnold 1983). In the polynomial formulation, change in the vector containing the mean overall reaction norm

slope and intercept ( g ) was calculated each generation according to the functions

m ! gi = mig j + [1" mi ]gi , (9) ms m # 1& m gi = g i + G2i% () (gi ) $ a'

where! a is the value of the environment, set at 2 in environment A and -2 in environment B, and the subscripts are reversed for population j (adapted from Gavrilets and Scheiner 1993b, equation A17). This

time, " was calculated using a phenotypic variance V estimated each generation using the equation

2 ! V = V2 + 2Ca + V3a , (10)

! 23

where C is the estimated covariance between slope and intercept (adapted from Gavrilets and Scheiner

1993b, equation 10).

Results

The character-state, polynomial, and developmental models yield significantly different evolutionary trajectories (Fig. 2). Plasticity is predicted to evolve most rapidly under the polynomial model, due to high heritability of the slope and intercept and little genetic correlation between the two (Fig.

3). The rate of evolution is slightly slower under the character-state model, due to positive correlation between the two character states, as estimated at the start point. In contrast, my developmental model predicts a very slow initial increase in plasticity, indirectly reflecting the tendency for selection to favor two opposing combinations of signal value (s) and slope (u1) (e.g., a higher value of the signal in environment A and a positive slope, or a lower value of the signal in environment A and a negative slope) that are broken up each generation by recombination. Both traditional models eventually yield equilibrium at the joint plastic optimum, in concert with the results of Via and Lande (1985), wheras the equilibrium level of plasticity under the developmental model remains less than half of that at the joint plastic optimum.

Plotting the equilibrium level of plasticity versus migration rate reveals consistency in this trend: the equilibrium level of plasticity increases sharply from zero to almost half the optimum level of plasticity as the proportion of immigrants increases from zero to approximately ten percent, but does not increase appreciably beyond that, even at panmixis (Fig. 4).

Examination of the selection vectors reveals an explanation for the failure of populations to reach the joint plastic optimum under my developmentally explicit model. Each generation, migration pulls the means of the underlying factors in the two populations towards each other. Due to the simple shape of this particular phenotypic landscape, and the fact that pairs of interacting populations converge on similar values of the underlying factors at equilibrium, this change in underlying values also tends to pull the phenotypic means of the populations toward each other (Fig. 5). At the same time, directional selection acts to push each population toward its respective phenotypic optimum. Taken together, these two vectors move both populations toward the joint plastic optimum. As the internal reaction norm slope (u1) diverges 24 further from zero, the within-population variance in signal values translates into greater within-

population phenotypic variance; stabilizing selection within each environment thus acts on the underlying

response variables to reduce the slope of the phenotypic reaction norm.

To assess the effect of selection on the values of the underlying developmental factors and the

resultant phenotypes, I plotted the equilibrium internal reaction norm for each Monte-Carlo simulation,

both with and without stabilizing selection (Fig. 6a, b). When stabilizing selection is ignored, migration

and directional selection together cause both populations to converge on a single stable phenotypic

equilibrium at the joint plastic optimum. However, an infinite number of signal and response value

Figure 2. Evolutionary trajectory under three different models. The predicted trajectory for population i is

shown under the character-state model, polynomial model, and my developmental model, using fitness

optima θA = 2, θB = -2; variance of each fitness surface = 64; and migration rates mi = 0.05, mj = 0.01.

Values for the G matrices used in the character-state and polynomial models were obtained at the start point

of each population at its local optima, with no mean plasticity in internal signal value ( sA = sB = 0 and

u2 = θA or θB for population i and j, respectively), and essentially no mean plasticity in the response to

signal value ( u1 = 0.01; this value was chosen to avoid an unstable equilibrium! at u1 =0).

! ! 25

Figure 3. Genetic (co)variances estimated at two different points on the phenotypic landscape. Estimates were obtained for population i at zero plasticity ( sA = sB = 0, u1 = 0.01, u2 = θA or θB for population i and j, respectively) and at the joint plastic optimum ( sA = 2, sB = "2 , u1 = 1, u2 = 0) for 1) genetic variance of phenotype in environment! A; 2) genetic! covariance ! between phenotypes in environment A and

B; 3) genetic variance of slope of overall !reaction ! norm; 4) !genetic variance! of intercept of overall reaction norm; and 5) genetic covariance between slope and intercept of overall reaction norm. These parameters change continuously over the phenotypic landscape, but I show only the values of the endpoints to reduce computational time.

combinations yield this optimum, explaining the wide variety of equilibrium reaction norms that result from different starting parameters. When stabilizing as well as directional selection is taken into account, a single stable phenotypic equilibrium exists for each population where the forces of migration, directional selection, and stabilizing selection balance each other. By limiting the slope of the reaction norm, 26 stabilizing selection prevents the populations from attaining the joint plastic optimum, and migration between them then holds each population away from its respective optimum. In this case, even though an infinite number of combinations of underlying variable values could produce the equilibrium phenotype, the tendency for stabilizing selection to reduce both the signal value (s) and the response to that value (u1) causes the underlying reaction norms of the two populations to converge on one of two stable equilibria.

Finally, when the internal signal is transformed into a nonheritable environmental cue, there is only a single stable equilibrium for both the phenotype and the underlying factors (Fig. 6c, d). Stabilizing selection again prevents the populations from attaining the joint plastic optimum, although the effect is less pronounced in this case because the values of the environmental cues do not change.

Figure 4. Plasticity and maladaptation as a function of migration rate. The level of plasticity and distance from the local optimum are shown for population i at equilibrium. Values were obtained using fitness optima θA = 2, θB = -2; variance of each fitness surface = 64, and equal migration rates (mi = mj).

Figure 5. The effect of directional selection, stabilizing selection, and gene flow. The effect of selection and migration are shown when populations i and j are at equilibrium. These vectors are superimposed on a fitness landscape obtained from calculating mean fitness across the two environments, in order to illustrate distance from the joint plastic optimum. Directional selection pushes each population toward its local optimum, stabilizing selection reduces plasticity, and gene flow pulls the populations towards each other.

The effect of unequal migration rates is clear in that population j, which receives fewer migrants, persists closer to its phenotypic optimum, where the strength of stabilizing selection is greater relative to directional selection and thus results in a reduced slope of the internal reaction norm. The larger proportion of migrants received by population i pulls it further from its phenotypic optimum, resulting in stronger directional selection. The partial derivative of the phenotype is greater with respect to the slope than the intercept of the internal reaction norm, due to a signal value that is greater than 1. Because of this, directional selection acts to increase the plasticity of population i more than its constitutive phenotype, which is represented by the intercept of the reaction norm to signal value (u2). The result is that population 28 i is more plastic and, counterintuitively, better adapted to environment B than its own environment. In the case of nonheritable environmental cues, population i actually overshoots the optimum in environment

Estimating the measures of genetic variance and covariance utilized in traditional quantitative genetic models illustrates change in these parameters with movement across the phenotypic landscape, despite treatment of the underlying genetic factors as traditional quantitative genetic traits with a fixed variance and zero covariance (Fig. 3). Although these are dynamic variables that change continuously across the phenotypic landscape, I report values only at the end points of zero and optimal plasticity to reduce computation time. Genetic variance in each character state increases dramatically with movement from zero mean plasticity to the joint plastic optimum, as the increase in slope of the internal reaction norm translates the same amount of variance in signal value into much higher phenotypic variance. This increase is accompanied by a shift in covariance between character states from a positive value, reflecting the effect of the intercept of the internal reaction norm on both character states, to a negative value, reflecting the increased importance of the slope of the internal reaction norm when the signal values are different in each environment. In the polynomial framework, the genetic variance of the slope of the phenotypic reaction norm exhibits the most change, starting near zero and increasing with movement to the joint plastic optimum as the effect of environment on signal value becomes larger and consistent in direction.

Discussion

Traditional quantitative genetic studies have indicated that spatial environmental heterogeneity tends to drive evolution toward a joint plastic optimum, given any variation in phenotypic plasticity (Via and Lande 1985), under a wide variety of ecologically realistic parameters (Sultan and Spencer 2002).

However, numerous studies show that migration between populations tends to constrain local adaptation

(e.g., Hendry et al. 2002; Moore et al. 2007), indicating a failure for populations to evolve adaptive plasticity. Understanding factors that limit the evolution of phenotypic plasticity is becoming a major focus of research. Here I show that developmental architecture can also result in constraints on the evolution of 29 a b

c d

Figure 6. Reaction norms at equilibrium. Data are shown for 10 Monte Carlo replicate population pairs started with randomly chosen values of underlying genetic factors in the case of a) epistasis and directional selection, b) epistasis, directional and stabilizing selection, c) environmental variance and directional selection, and d) environmental variance, directional and stabilizing selection. Solid lines represent population i, which receives more migrants, and dotted lines represent population j. Crosses represent the phenotype expressed in environment A and circles represent the phenotype expressed in environment B.

Dashed lines indicate local optima in environments A (2) and B (-2).

phenotypic plasticity, even when underlying genetic factors are treated as quantitative genetic traits with ample additive genetic variance and no genetic correlations. Adding the simplest layer of developmental realism, in the form of a nonadditive interaction between developmental factors or between genotype and environment, changes the degree of plasticity and concomitant constraint on local adaptation expected at equilibrium. No significant plasticity is expected to evolve under very low levels of migration, whereas only partial plasticity is expected to evolve under high levels of migration, including panmixis. The phenotypic equilibrium under directional and stabilizing selection is thus intermediate to that of Via and

Lande (1985), in which populations with any genetic variation in phenotypic plasticity always reach the 30 joint plastic optimum, and Hendry et al. (2001), in which there is no plasticity and gene flow between populations constrains local adaptation: both populations evolve some plasticity, but not enough to produce the optimal phenotype in either environment.

Although generally ignored in plasticity research, the effect of stabilizing selection has long been a central focus of studies on epistasis. The novel results of my model manifest only when the effect of stabilizing selection is taken into account, and reflect the benefit of integrating these two areas of research.

Once plasticity is viewed through a developmental perspective, it becomes clear that, by definition, plasticity involves a phenotypic response to an organism’s context, and therefore shapes the population- level translation of contextual variance into phenotypic variance. Because of this, we can expect reaction norms to be influenced by stabilizing (or disruptive) selection within environments whenever there is variance in either internal or external context and the second derivative of the fitness landscape is nonzero.

In my particular epistatic model, stabilizing selection acts to reduce both the internal signal produced in response to the environment and the sensitivity to that signal, thus preventing evolution to the joint plastic optimum. In addition, stabilizing selection pushes the underlying developmental factors towards one of two stable equilibria, rather than allowing for an infinite number of combinations that produce the equilibrium phenotype. The constraint on plasticity is reduced when the internal signal is replaced by a nonheritable environmental cue, due to the inability of selection to reduce the cue strength, but remains significant, indicating that a developmental perspective can change expectations whenever there is a nonadditive interaction between genotype and environment, even in the absence of epistasis. In this case, within-environment variance in environmental cues is not connected to selective regime, and the constraint on the evolution of plasticity can thus be interpreted as resulting from cue unreliability.

The persistence of population i closer to the optimum of environment B than its own environment is counterintuitive and relevant to the entrenched debate over whether plasticity emerges as a byproduct of selection, or whether selection acts directly on plasticity itself (Scheiner and DeWitt 2004). Because the internal signal values in my epistatic model are only expressed in their respective, discrete environments, change in these parameters emerges as a byproduct of selection. In essence, alleles enter each population through migration; selection within each population acts against alleles that cause constitutive expression of the phenotype appropriate to the opposite environment, but not against alleles that remain silent in the 31 novel environment. The combination of migration and selection causes alleles expressed in only one environment to rise in frequency in both populations, despite a lack of selection on plasticity itself. At the same time, however, selection acts directly on plasticity in two different ways. Stabilizing selection within environments acts to reduce overall plasticity, as discussed above. In addition, the trait value expressed in a single environment is the product of several developmental factors, some of which are plastic with respect to environmental or genetic context. Because altering the reaction norm of plastic developmental factors can alter the trait value expressed in that environment, directional selection at the level of individuals within a single environment can act directly on plasticity itself. This type of selection increases the plasticity of population i, causing it to express a phenotype that is close to, or even overshoots, the optimum in environment B.

Quantitative genetics was founded on the supposition that, at the population level, only the additive (i.e., average) portion of genetic effects determines the parent-offspring covariance, and thus governs the response to selection (reviewed by Hansen 2006). Previous quantitative genetic studies of the evolution of plasticity have thus assumed that, even in the presence of complex genetic and developmental interactions, the relevant aspects of genetic architecture can be captured in the genetic variance-covariance matrix (G) of either the character states or the polynomials that define the shape of a reaction norm

(reviewed by Via et al. 1995). Here I show that, in the presence of a simple non-additive developmental interaction, the G-matrices of both the character-state and the polynomial approaches change fundamentally depending on where the population sits on the phenotypic landscape. In addition, neither approach can take into account the possibility that multiple combinations of developmental factors can create the same phenotype, and thus be equally favored by selection, slowing the overall phenotypic response. For both of these reasons, it appears that traditional G-based approaches cannot be taken as an accurate predictor of evolutionary trajectories in the presence of epistasis, even over the short term.

Current research on the genetic and developmental architecture of plastic traits suggests that they are constructed through highly complex and idiosyncratic interactions. Given this, it is unlikely that my simplified model can adequately reflect any particular empirical system, and the value of my findings lies in the general lessons that they offer. Given the central role of stabilizing selection in producing the novel results presented here, I propose that selection acting directly on the variance of traits is likely to be a 32 common determinant of the evolutionary trajectory of plastic traits and worthy of further investigation.

In addition, following a growing concern over the relevance of G-based approaches (Houle 1991; Gromko

1995; Wolf et al. 2001; Pigliucci 2006), I demonstrate the inability of traditional quantitative genetic approaches to capture the evolutionary dynamics of a plastic trait involving developmental epistasis.

Finally, I suggest that the framework derived by Rice (2002) holds great promise for elucidating the effects of developmental architecture on traits that involve nonadditive interactions at any level between genotype and environment, although determining the level of detail necessary to obtain realistic results, and the degree to which general conclusions can be extracted from this approach, are a major challenge for future research. 33 CHAPTER 3

GENETIC ACCOMMODATION FACILITATES RAPID MULTIVARIATE ADAPTATION ALONG

CRYPTIC GENETIC LINES OF LEAST RESISTANCE

More than a century after its initial assertion, the concept that phenotypic plasticity can facilitate adaptive evolution remains highly contentious (reviewed by Pigliucci and Murren 2003; de Jong 2005;

Ghalambor et al. 2007). On one side, an extensive body of work centers on plasticity as an evolvable trait that responds to selection and can therefore produce adaptive solutions, much like any other trait (reviewed by Berrigan and Scheiner 2004; de Jong 2005). This view is grounded in the mathematical framework of quantitative and population genetics, and emphasizes phenotypic plasticity as an evolved adaptation to environmental variation. On the other side, plasticity is ascribed a major role in facilitating evolution itself, including aiding adaptation to novel environments, promoting speciation, and shaping macroevolutionary patterns (West-Eberhard 2003, 2005; Pigliucci and Murren 2003; Schlicting 2004). Although this view has a long and extensive history (e.g., Baldwin 1896; Waddington 1942, 1952, 1953; Simpson 1953; Robinson and Dukas 1999; Pigliucci and Murren 2003; Price et al. 2003), its staunchest current supporters are rooted in developmental biology, far removed from the theoretical framework established during the Modern

Synthesis (e.g., West-Eberhard 2003, 2005; Gerhart and Kirschner 2007).

The quantitative and population genetic perspective on plasticity derives potency from the ability to create and rigorously test hypotheses at a microevolutionary (and thus directly observable) scale, based on a framework that has remained robust from the 1940’s throughout the genetic and even genomic revolution (Mayr 2004; Lynch 2007). Researchers from this tradition note that little to no empirical evidence exists to support a major role of plasticity in facilitating adaptive evolution, and that claims derived from the developmental perspective rely largely on plausibility arguments and a limited number of illustrative examples, open to criticism as mere “evolutionary storytelling” (de Jong 2005; Lynch 2007;

Hoekstra and Coyne 2007). On the other hand, empirical evidence (e.g., Czesak et al. 2006) and models that incorporate developmental realism (e.g., Chapter 2) reveal that essential assumptions of the quantitative genetic approach, including constancy of the genetic variance-covariance matrix, are likely to be violated. Furthermore, the population and quantitative genetic approach has been used almost 34 exclusively to investigate single traits (but see Carroll and Klaussen 1997; Parsons & Robinson 2006), whereas the developmental perspective is fundamentally tied to a multivariate, and thus more realistic, view of adaptation. In particular, this perspective emphasizes that organisms are flexible, developmentally organized systems, so that a novel environmental or genetic input may spark a cascade of plastic adjustments across multiple traits (Schlichting and Pigliucci 1998; West-Eberhard 2005; Breuker et al.

2006; Gerhart and Kirschner 2007).

Here I integrate quantitative genetic and developmental approaches in order to explore the manner in which developmental interactions between two traits can influence adaptation to a novel selective pressure. To provide a concrete example, I incorporate simple aspects of natural history into evolution of an in silico population of Daphnia melanica in response to an introduced fish predator, using knowledge gained from a system of high alpine lakes in the Sierra Nevada Mountains (Fisk et al. 2007; Latta et al.

2007). I examine the evolutionary trajectory of two traits subject to selection due to fish predation: diel vertical migration (DVM, the tendency to migrate downwards in the water column during the daytime) and melanic pigmentation, which are developmentally entangled via their respective effect on and reaction to exposure to ultraviolet radiation (UV). Analysis of my construct is accomplished using a revolutionary but currently underutilized mathematical and theoretical framework derived by Rice (2002, 2004a; see also

Wolf 2002), which extends the framework of population and quantitative genetics to explicitly incorporate the effect of developmental architecture in the formation of phenotypes. This framework allows us to examine interplay between the role of selection in shaping developmental interactions and the effect of developmental interactions in shaping evolutionary trajectories. This interplay is arguably the central focus of the evo-devo synthesis (Hendrikse et al. 2007).

Although I use D. melanica as an example, the construct I examine is relevant to any set of traits that are developmentally entangled through plastic responsiveness to each other. I suggest that behavioral traits, which often mediate exposure to environmental conditions, may be particularly likely to relate to a suite of other traits in this manner. In contrast to researchers who emphasize the likelihood of developmental entanglement giving rise to adaptive “phenotypic accommodation” (West-Eberhard 2003,

2005; Breuker et al. 2006; Gerhart and Kirschner 2007), I take a conservative approach in that I do not assume any particular direction of initial plasticity with respect to fitness. Instead, I explore the manner in 35 which genetic accommodation, particularly change in the genetically controlled responsiveness of traits to each other, is shaped by selection and can in turn shape the rate and direction of phenotypic response to selection. My results, firmly rooted in the framework of the Modern Synthesis, indicate that developmental entanglement through plasticity can facilitate rapid multivariate adaptation in response to a novel selective environment.

Theoretical Construct

D. melanica inhabiting alpine lakes of the Sierra Nevada are normally large-bodied, highly pigmented, and tend to inhabit the upper portion of the water column during the daytime, making them highly visible (Fisk et al. 2007; Latta et al. 2007). A novel visual predator, trout (Oncorhynchus spp.), was introduced into a high percentage of these lakes over the last 100 years, driving numerous populations to extinction (Bradford et al. 1998; Knapp et al. 2001, 2005). Populations that persist show evidence of rapid evolution in a number of traits that influence visibility, including an increase in DVM (Pfrender, unpubl. data) and reduction in melanin (see Chapter 4), consistent with previous studies on selection pressure due to fish predation (Pijanowska et al. 1993; De Meester 1996; Cousyn et al. 2001). Increased DVM reduces exposure to UV radiation, which is filtered out by the water column. Because melanin deposition tends to be plastic with respect to UV exposure (see Chapter 4), a change in DVM can result in secondary change in pigmentation, even in the absence of genetic change directly affecting melanin pathways.

My simulation represents a population of D. melanica that experiences a shift in fitness optimum toward reduced melanin and increased DVM, following the introduction of trout. The population begins with a high level of melanin (m = 10) and little tendency for DVM (d = 1). In the presence of trout, I assume that the fitness effects of melanin and DVM are given by separate Gaussian functions with a variance of 100 and respective optima at m = 1 and d = 10. The fitness effects of melanin and DVM are averaged to give a measure of individual fitness. This choice of individual fitness function is deliberately simple, reflecting absence of detailed knowledge of the joint fitness effects of DVM and melanin, and avoids the unlikely assumption that the fitness effect of a change in one trait would be less if the second trait is far from its fitness optimum, as is the case for a fitness surface based on a bivariate Gaussian distribution. 36 I compare the evolutionary trajectory of this population when melanin and DVM are assumed to

be independent traits to the trajectory when developmental entanglement between them is taken into

account. First, I treat melanin (m) and DVM (d) as separate quantitative genetic traits with a variance of 1

and covariance of 0. In this case, change from generation t to generation t+1 due to directional selection is

given by the equation

" d % " d % " 1 )w *1 - 1 )w *0 -% $ ' = $ ' + H ($ , / + , /' (1) m m 0 1 # & t +1 # & t # w )d + . w )m + .&

where! H is a heritability matrix, w is individual fitness, and w is mean population fitness. H is assumed to be the identity matrix, given that lower and more realistic heritabilities change the rate of evolution in

! both trajectories, leaving! the relationship between! them unaffected. As with any! case that does not involve nonadditive interactions, this equation, given by the Rice framework, is equivalent to the multivariate

breeder’s equation (Lande and Arnold 1983).

Second, I calculate the expected evolutionary trajectory taking into account the effect of increased

DVM on melanin, as mediated by a reduction in UV exposure. For simplicity, I assume a linear shape to

the reaction norm of melanin in response to DVM. In this case, melanin is calculated as

m = z " p ! d (2)

where z is the amount of melanin produced in response to the amount of UV light available at the top of

the water column (d = 0) and p, the slope of the linear reaction norm, is a measure of plasticity, or tendency

! for melanin production to drop as UV exposure decreases. The underlying developmental factors z, p, and d are all treated as quantitative genetic traits with a variance of 1 and covariance of 0, consistent with the

traditional polynomial approach to modeling the evolution of phenotypic plasticity (Scheiner and Lyman

1989; Gavrilets and Scheiner 1993b). Following Rice (2002, 2004), change from generation t to generation

t+1 due to directional selection is given by 37

" z % " z % " *0 - * 1 -% $ ' $ ' $ 1 )w , / 1 )w , /' p = p + H ( 0 + 0d (3) $ ' $ ' $ w )d , / w )m , /' d d $ ' # & t +1 # & t # +,1 ./ +,0 p./&

when developmental! entanglement is taken into account. Throughout the evolutionary trajectory, the covariance between DVM and melanin is equal to p. At the beginning of the trajectory, p is given an average value of 0, yielding a variance of less than 1 for melanin and a covariance of 0 between DVM and melanin. This represents a case where there is no average effect of plasticity on the mean of melanin, even though there is heritable variation in plasticity.

Results

With a traditional quantitative genetic approach, the population is predicted to evolve directly to the new joint fitness optimum, at a rate that depends solely upon mean fitness and slope of the individual fitness landscape (Fig. 7). Under the developmental model, however, individuals with greater DVM and low pigmentation are favored. Of the individuals with greater DVM, those that exhibit a greater reduction in melanin as UV exposure decreases (i.e., are more plastic) tend to be lighter and have even higher expected fitness. Selection for reduced melanin thus increases plasticity with respect to DVM (Fig. 8), creating a covariance between the two traits that is aligned with the direction of selection. In addition, once the population evolves a mean reduction in melanin in response to increased DVM, selection purely on melanin favors individuals with increased DVM, adding to the DVM selection differential. Both of these factors increase the rate of response to selection. As the slope of the reaction norm (and thus covariance) dips below -1, evolution proceeds faster in melanin than DVM, allowing for a particularly rapid approach to the optimal level of pigmentation. After melanin levels are optimal, directional selection on DVM creates a correlated change in melanin that overshoots the melanic optimum. Directional selection to increase melanin then translates into decreased plasticity with respect to DVM, decreasing the correlation between the two traits and allowing for unimpeded evolution to the DVM optimum.

38 Conclusion

Previous hypotheses for the role of plasticity in facilitating evolution fall into two broad categories

(Ghalambor et al. 2007). In one proposed mechanism, plasticity moves the mean phenotype closer to an adaptive peak, allowing a population to persist long enough to respond to directional selection through a combination of standing variation, novel mutation, and recombination (e.g., Baldwin 1896; Baker 1974;

Robinson and Dukas 1999; Schlichting 2004; reviewed in Pigliucci 2001). The developmental view primarily involves this type of shift in trait means, but emphasizes mutual developmental adjustments among many traits that together produce a viable, adaptive phenotype in response to a novel environmental

Figure 7. Trajectory of mean DVM and melanin. DVM and melanin are treated as simple quantitative traits with no correlation (open arrows), as well as traits that are developmentally entangled through their respective effect on and responsiveness to UV radiation (filled arrows). Trajectories are superimposed on the individual fitness landscape.

39 input (West-Eberhard 2003, 2005; Breuker et al. 2006; Gerhart and Kirschner 2007). In a second type of mechanism, plastic responses to a stressful environment result in the expression of cryptic variance that selection can act upon (reviewed by Badyaev 2005). For example, the heat-shock protein Hsp90 has been shown to conceal a substantial amount of genetic variation that is revealed only in response to an extreme environmental stimulus (Rutherford and Lindquist 1998; Rutherford 2000, 2003; Queitsch et al. 2002).

Some of these variants can be stabilized by selection to promote constitutive expression, demonstrating their potential to provide for adaptive change (Rutherford and Lindquist 1998). Under both mechanisms, plasticity is thought to facilitate adaptation to novel environmental conditions (Baldwin 1896; Baker 1974;

Robinson and Dukas 1999; Pigliucci 2001; Schlichting 2004) or to aid shifts between adaptive peaks on a rugged fitness surface (Robinson and Dukas 1999; Pigliucci and Murren 2003; Price et al. 2003;

Schlichting 2004; Amarillo-Suarez and Fox 2006).

Figure 8. Evolutionary change in plasticity. The trajectory of plasticity (i.e., the slope of the reaction norm of melanin with respect to DVM) is plotted over time. This measure of plasticity is also equivalent to the covariance between melanin and DVM. 40 The simulation I examine here provides a novel perspective on the role of phenotypic plasticity in facilitating adaptation. I show that, even in the absence of an adaptive plastic shift in mean phenotype, directional selection on multiple traits can drive genetic accommodation (i.e., genetic change in plasticity) that alters the responsiveness of one trait to another. The result is the creation of genetic covariance that is largely aligned with the direction of selection, increasing the rate and influencing the direction of adaptive evolutionary change. In this case, as the responsive trait reaches its phenotypic optimum, the genetic covariance created by increased plasticity is no longer aligned with the direction of selection, and selection acts instead to reduce responsiveness. This mimics the pattern of evolutionary change in mean phenotype followed by reduced plasticity that is posited by proponents of genetic assimilation (sensu Pigliucci and

Murren 2003), and is consistent with empirical data on divergent populations in disparate environments that maintain some plasticity in the direction of divergence, but not enough to produce the optimal phenotype in the alternate environment (e.g., Day et al. 1994; Losos et al. 2000; Gomez-Mestre and Buchholz 2006). In the present example, the overall effect of developmental entanglement via plasticity is to create a cryptic genetic line of least resistance, expressed only during a transient increase in plasticity, that facilitates rapid adaptation in multi-trait space.

The approach I employ is likely to be widely applicable, given that developmental entanglement of traits via plasticity is a well-established and arguably ubiquitous phenomenon of organisms. Across a wide variety of taxa, changes in behavior and diet have been shown to result in correlated plastic changes in suites of morphological and physiological traits (e.g., Smits et al. 1996; Thompson 1999; Losos et al. 2000;

Stauffer and Gray 2004; reviewed by West-Eberhard 2005). In the D. melanica population I simulate, a more complete picture might include phenomena such alterations in life-history traits due to reduced food availability in deeper water, or compensation for reduced melanin through alternate pathways of UV repair.

Even when traits are not entangled through an environmental mediary such as UV radiation, they may be related through internal developmental pathways, as the complex sequence of steps in development itself has been posited as a form of serial reaction norms to internal conditions (Wolpert 1994; Sarà 1996;

Schlichting and Pigliucci 1998). In addition, the initial form of plasticity does not need to be adaptive, or even result in any mean change in trait value, as long as some heritable variation for plasticity exists. 41 My results are significant for several reasons. First, a number of studies suggest that the rate of adaptive change in response to directional selective is related to the probability of extinction (e.g., Haldane

1957; Gomulkiewicz and Holt 1995; Lande 1998), and thus has potentially profound ecological consequences. By increasing the rate of adaptation, developmental entanglement through plasticity may increase the probability of persistence for populations experiencing a dramatic shift in environment, whether from novel invasive species, change in the physical environment, or invasion of a new environment. Although related to the first mechanism proposed for the facilitation of evolution itself— plasticity allowing for persistence in a novel selective environment—my result involves a change in the rate of evolution, rather than a jump-start in an adaptive direction, and does not require an initially adaptive reaction norm. Second, the degree to which a specific form of developmental entanglement is common across populations can affect the tendency for multiple populations to adopt a parallel pattern of adaptive change, thus affecting macroevolutionary patterns (see Brakefield 2006). This form of genetic channeling differs from that proposed in previous studies (e.g., Cheverud 1984; Maynard Smith et al. 1985; Schluter

1996; Steppan et al. 2002) in that the genetic lines of least resistance do not necessarily exist beforehand, but rather can be created by directional selection itself. Finally, I use the framework derived by Rice

(2002) in order to integrate quantitative genetic and developmental perspectives on the evolution of plasticity, revealing that plasticity can both respond to directional selection and, in turn, facilitate the response to that selection. This approach illustrates one way in which the divide between these viewpoints can be productively bridged, providing novel insights as well as hypotheses that are rooted in a rigorous theoretical framework and can be tested on an ecological timescale.

Explanation of Calculations

Equations 1 and 3 are consistent with the framework derived by Rice (2002), which provides a way to calculate the manner in which selection on a trait translates into change in the distribution of developmental factors that interact in the production of that trait. Calculation of the vector of selection differentials of the underlying factors requires knowledge of the shape of the phenotypic landscape, which is a plot of the trait value versus the values of all underlying developmental factors under consideration.

Particular combinations of developmental factors are represented by combinations of derivatives of the 42 phenotypic landscape, and each nonzero set of derivatives has a corresponding vector that describes the contribution of that combination of developmental factors to the total selection differential. The general equation for the vector of selection differentials corresponding to the set of derivatives [a1...an ] is

# b % " w 1 a a a Q P +# i , D 1 D n , (4) a1 .....an = # b !L! w "$

where " = 1 !ai!!bj!, a denotes the degree of differentiation, !b is the total number of derivatives

1+! ai under consideration, P is a tensor of rank 1+ !ai that contains all the 1+ !ai -th moments and

an mixed moments of the distribution of underlying factors, and D is a tensor of rank an that contains all

possible an th derivatives of the phenotypic landscape (Rice 2002 equation 7). The total change in the vector of mean values of the underlying factors is then calculated as the sum of all possible Q vectors multiplied by a heritability matrix H, as long as no change is expected in the mean values due to processes other than differential reproduction (modified from Rice 2002 equation 8).

In both cases examined here, I am concerned solely with the effect of directional selection on melanin and DVM. Direction selection involves the first derivative of the individual fitness surface, and thus only Q vectors containing a single derivative of the phenotypic landscape. Equation 1 involves additive landscapes for both melanin and DVM, so that only D1, the vector containing the first derivative of the phenotypic landscape with respect to the corresponding trait, is nonzero for each trait. Equation 3 involves a nonadditive interaction between plasticity and DVM such that D2 is also nonzero for melanin.

3 However, the P tensor in the equation for the vector Q2 is zero, due to the assumption of a symmetric distribution of the underlying factors, yielding a Q2 vector of 0. In each case and for each trait, I thus limit my analysis to the selection differential described by the vector Q1, where

1 "w Q = P 2,D1 (5) 1 w "#

! 43

(Rice 2002 equation 13).

2 In calculating the Q1 vectors, I use the same vector of underlying factors (and thus the same P

tensor) for each trait. In each case, the assumption of a phenotypic variance of 1 and covariance of 0

between underlying factors yields a P2 tensor that is the identity matrix. In equation 1, melanin and DVM

map directly to their respective phenotypes, and are thus treated as “underlying factors” for themselves, in

order to preserve a structure that is parallel to that of equation 3. In this case, the phenotypic landscapes

" d% 1 "1 % can be described by the equations m = m and d = d , the vector of underlying factors is $ ' , Dd = $ ' , #m & #0 &

"0 % and 1 . When plasticity is taken into account, the phenotypic landscapes can be described by Dm = $ ' ! ! #1 & ! ! " z% "0 % # 1 & $ ' 1 $ ' 1 % ( equation 2 and d = d , the vector of underlying factors is p , D = 0 , and D = "d . To obtain ! $ ' d $ ' m % ( #$ d&' #$1 &' $%" p'(

the !selection differential due to selection acting independently on both traits, the Q1 vectors for the two traits are summed. To determine the change in! mean ! values of the underlying! factors each generation, the result is multiplied by the heritability matrix H. 44 CHAPTER 4

FROM ADAPTIVE PLASTICITY TO RAPID ADAPTATION: EVOLUTION OF DAPHNIA GENE

EXPRESSION IN RESPONSE TO AN INVASIVE PREDATOR

The proposed role of phenotypic plasticity in promoting adaptation through genetic assimilation or, more inclusively, genetic accommodation, has generated intense interest but little direct evidence

(reviewed by Pigliucci and Murren 2003; de Jong 2005; Ghalambor et al. 2007). Genetic assimilation, defined as the evolutionary conversion of an inducible plastic phenotype into an environmentally invariant phenotype, was first suggested as a mechanism of adaptation over a century ago (Baldwin 1896;

Waddington 1942, 1953). More recently, the broader phenomenon of genetic accommodation, or genetic change in the regulation or form of a trait induced by a novel environmental or genetic input, has been championed as a major factor in adaptive change (West-Eberhard 2003, 2005). Under the current model of evolution through genetic accommodation, ancestral plasticity allows for the production of an induced phenotype, and selection on this phenotype leads to subsequent genetic change in the shape of the reaction norm (Gomez-Mestre and Buchholz 2006). The proposed outcome is congruence in the direction of within-group plasticity and between-group adaptive change, generated by a shared developmental mechanism.

A number of artificial selection experiments demonstrate rapid genetic accommodation within lineages (Waddington 1953; Gibson and Hogness 1996; Rutherford and Lindquist 1998; Suzuki and

Nijhout 2006), providing support for the plausibility of this phenomenon in nature. In addition, multiple studies reveal patterns of plasticity and adaptive change at the population (Gurevitch 1988, 1992; Gurevitch and Schuepp 1990; Sword 2002) and species level (Day et al 1994; Losos et al. 2000; Gomez-Mestre and

Buchholz 2006) that are consistent with evolution through genetic accommodation (reviewed by Pigliucci and Murren 2003; Gomez-Mestre and Buchholz 2006). In particular, these studies reveal plasticity within populations or closely related species that echoes the direction of adaptive divergence between them.

Although taken as indirect evidence for the role of genetic accommodation in adaptation, these studies fall short of providing direct evidence in several ways. First, lack of knowledge of the ancestral reaction norm precludes evaluation of the hypothesis that plasticity provides an initially adaptive phenotype for selection 45 to act upon (but see Gomez-Mestre and Buchholz 2006). Second, the developmental mechanisms underlying plasticity and divergence remain largely unknown, so that it is impossible to assess the role of plasticity in providing the machinery for adaptive divergence. Finally, genetic accommodation is thought to allow for rapid adaptation, making direct observation of the process largely unattainable (Pigliucci and

Murren 2003).

Here I examine the role of genetic accommodation in the rapid adaptation of independent populations of Daphnia melanica to the introduction of a novel fish predator. Normally darkly pigmented and thus highly visible, D. melanica inhabit naturally fishless alpine lakes of the Sierra Nevada Mountains and exhibit plasticity in melanic pigmentation with respect to ultraviolet radiation (UV). Salmonids

(Onchorhynchus, Salmo, and Salvelinus spp.), which are visual predators of Daphnia, have been introduced into many of these lakes during the last century, resulting in either local extinction of D. melanica

(Bradford et al. 1998; Knapp et al. 2001, 2005) or rapid, parallel adaptation in a suite of morphological and life-history traits (Fisk et al. 2007; Latta et al. 2007). Consistent with known patterns of selection pressure from fish predation (Saegrov et al. 1996), populations of D. melanica that coexist with introduced fish exhibit reduced pigmentation. I characterize the reaction norm of pigmentation with respect to UV in populations that retain the ancestral fishless state as well as populations that have persisted over 53-91 years of exposure to fish predation. In order to test whether adaptation has proceeded through cooption of the developmental mechanisms involved in the production of phenotypic plasticity, I also characterize the reaction norms of gene expression for genes known to be involved in melanin synthesis (Ddc and ebony;

Wittkopp et al. 2002; reviewed by True 2003) or cuticular sclerotization (bursicon; Taghert and Truman

1982; Gade and Hoffmann 2005). My results show a pattern of genetic accommodation at the phenotypic level that is associated with a combination of genetic assimilation and constitutive upregulation at the level of gene expression.

Methods

Study Populations

Individual genotypes were collected from four permanent lakes in the Central Sierra Nevada

Mountains during the summer of 2004 (see Latta et al. 2007; Fisk et al. 2007). All four lakes, located at 46 elevations ranging from 3150-3632 meters, are naturally fishless. Frog Lake (ID# 52103; UTM Zone 11:

351079 E, 4124432 N) and Source Lake (UTM Zone 11: 349988 E 4125708 N), retain their naturally fishless condition, and I refer to these as fishless lakes. Puppet Lake (UTM Zone 11: 346277 E, 4127817

N) has been stocked with golden trout (Oncorhynchus mykiss aguabonita) every other year since 1951

(California Dept. of Fish and Game, unpublished stocking records), resulting in 53 years of predation on the resident D. melanica population at the time of collection. Evelyn Lake (UTM Zone 11: 295393 E,

4186659 N) was first stocked with brown trout (Salmo trutta) in 1913, followed by repeated introductions of Brook trout (Salvelinus fontinalis) and rainbow trout (Oncorynchus mykiss) (Elliot and Loughlin 2005; detailed in Latta et al. 2007), resulting in 91 years of fish predation at the time of collection. Together, I refer to Puppet and Evelyn Lakes as fish lakes.

Clone Establishment and Maintenance

Daphnia were collected from each study lake using a 30 CM conical tow net and maintained at

4°C for a maximum of 2 weeks during transportation to the laboratory. The low temperature prevented individuals born after collection from maturing, ensuring that all mature females originated from the lakes and therefore comprised a degree of genetic variation representative of the natural populations. Individual mature females were isolated in 250mL of filtered well-water and allowed to establish clonal lines that were maintained by parthenogenic reproduction under constant conditions of temperature (15°C), photoperiod (16L:8D), and food (a vitamin-supplemented, pure culture of the green alga Scenedesmus obliquus). My experiment began after approximately 20 generations of parthenogenic reproduction.

Life-table Assay

I assayed all traits using modification of a standard experimental design (Lynch 1985; Pfrender and Lynch 2000). Mature females from independent clonal lines were isolated in 250mL beakers. All progeny from these females were isolated and maintained individually through maturity and reproduction to standardize maternal effects. Clutches of progeny from these females were placed in separate beakers on the day after birth and randomly assigned to either control or UV conditions. I recorded data only from this third, experimental generation. Throughout the experiment, I maintained all Daphnia in the same controlled temperature room at 18°C with a 16L:8D photoperiod, in 200mL of filtered well-water that was 47 supplemented with a constant concentration (135,000 cells/mL) of S. obliquus and replaced every other day. Control beakers were placed under standard fluorescent lights as well as plastic Llumar screens that block all UV radiation. UV treatment beakers were placed 10.5 cm below Reptisun 5.0 UVB fluorescent lightbulbs (Zoo Med), and fans were used to ensure that UV treatment and control beakers retained the same 18°C temperature. Opaque screens separated control and UV beakers. To ensure that food was not a limiting resource, I allowed a maximum of five experimental-generation clutch mates to coexist within a single beaker; clutches of more than five individuals were split between beakers. Placement of the beakers was randomized within the growth chamber and changed every other day to minimize micro-environmental variation.

I recorded the time from birth to maturity for the fastest-maturing member of each clutch, defining maturity as the first instar with deposition of eggs into the brood pouch. When the first clutch of eggs from experimental generation individuals developed enough to produce two separate eyes (i.e., just before adult molting), I randomly assigned these individuals to preservation for either melanin or RNA extraction.

Preservation only at this stage reduced variance, given that Daphnia deposit melanin throughout each instar and become darkest just before molting. Individuals assigned to melanin extraction were measured for length and placed into 4% formalin. Individuals assigned to RNA extraction were placed into RNAlater

(Qiagen). In both cases, Daphnia samples were stored at -20°C until extraction. I also recorded the number of clutch failures (i.e., clutches in which all individuals died before maturity).

Melanin Assay

I incubated each Daphnia preserved for melanin extraction in 100uL of 5M NaOH for 4 days at

40°C, measured absorbance of the supernatant at 350nm, and divided by the length of the Daphnia to yield absorbance/mm. Subtraction of an average measure of absorbance/mm obtained from three nonmelanic

Daphnia pulex was used to account for absorbance due to compounds other than melanin. Comparison with a standard curve derived from commercial melanin dissolved in 5M NaOH was used to convert measures of absorbance to ug melanin/mm Daphnia.

48 Gene Expression Assay

Assay of gene expression involved four reference genes, including the elongation factor 1-alpha gene (EF-1α), glyceraldehyde-3-phosphate dehydrogenase gene (G3PD), α-tubulin gene, and RNA polymerase II gene (RNAP II). Genes of interest included two genes known to participate in the insect melanin pathway, Ddc and ebony (Wittkopp et al. 2002; True 2003), and one gene known to participate in cuticular sclerotization, bursicon (Taghert and Truman 1982; Gade and Hoffmann 2005). All genes were identified by BLASTing homologous Drosophila or Crustacean sequence against the Daphnia pulex genome. The resultant Daphnia sequence was then BLASTed back into the NCBI Database to check for homology. Only sequence that recovered homologous sequence from the desired gene was accepted.

Introns were identified by alignment with D. pulex cDNA sequence and, whenever possible, primers were designed to amplify a region spanning at least one intron to provide an additional method to check for amplification of contaminating DNA. Only primer pairs that yielded amplification efficiency greater than or equal to 97% were used for quantitation (Table 1).

For each clone-treatment combination, I isolated total mRNA from an average of 6 (range 1-12) individual Daphnia using an RNeasy Plus Mini kit (QIAGEN) and confirmed the absence of DNA contamination by attempted PCR amplification of the RNA extraction in parallel with a positive control.

Synthesis of cDNA was accomplished from 200ng total RNA using the iScript cDNA Synthesis Kit (Bio

Rad), followed by quantitative real-time PCR performed in duplicate on 3uL of a 1:10 dilution of the cDNA synthesis reaction. I used a PCR mix with the following final concentrations: 0.2X SYBR Green dilution (10,000X SYBR Green I from Invitrogen, diluted to 5X in DMSO), 1X PCR Buffer, 0.2mM each dNTP, 0.4µM forward primer, 0.4µM reverse primer, 2mM MgCl2, and 0.04 U Immolase DNA

Polymerase (Bioline). Thermal cycling conditions were 95°C for 2 min. 30 s, followed by 40 cycles of

95°C for 20 s, annealing for 20 s, 72°C for 30 s, and a plate read, followed by a melting curve from 60-

90°C. Optimal annealing temperatures were determined empirically and the melting curve was examined for each reaction to confirm absence of DNA contamination and amplification of a single template. Using the program geNorm, based on the methodology described by Vandesompele et al. (2002), the three reference genes G3PD, α-tubulin, and RNAP II were determined to be the most stably expressed across all 49 samples, and were geometrically averaged to calculate a gene expression normalization factor for each sample.

Statistical Procedures

I performed multiple analysis of variance (MANOVA) and analysis of variance (ANOVA) on all traits to test for an effect of clone nested within population, population nested within fish history, fish history, UV treatment, and fish history by UV treatment interaction, with clone treated as a random effect

(Minitab® Statistical Software). Performing a MANOVA on all traits allowed me to control the experiment-wide error rate, and I followed a significant result with separate ANOVAs on each trait. I also performed MANOVA on only my three genes of interest in order to elucidate the overall pattern of change in gene expression. For both MANOVAs and ANOVAs, I interpret a significant effect of fish treatment as evidence for parallel evolution in response to fish predation, and a significant UV-by-fish-history interaction as evidence for parallel evolution of the reaction norm to UV in response to fish predation. A

Table 1: Primers used for real-time quantitative PCR.

Locus Primer Name: Primer Sequence Annealing Temp

U35 5’- CAC CTT GGA ACA CTC TGA CT -3’ α-tubulin 59 L358 5’- AGG GTC GCA TTT GAC CAT CT -3’

F8 5’- GCC AAT TAA CTC CCG TTA TTC A -3’ bursicon 64 R108 5’- CTG GAC GTA GCT GGT GCA T -3’

F12 5’- CAT CTG CGA AGA GTA TCG TCA -3’ Ddc 64 R120 5’- CCA CAT GGC AGA ACA GTC AA -3’

F19 5’- GAA TTT CTC GAG GCG ATT GA -3’ ebony 64 R191 5’- CCT TTC TGT ATC GCC AGC TC -3’

U17 5’- CCA AGT TCT ACG TCA CTA TCA TC -3’ EF-1α 59 L157 5’- TCG CGG GTC TGT CCG TTC -3’

33F 5’- TTA TCA CCT CCT CAA CTT C -3’ G3PD 59 222R 5’- CTT CTT CCT TCA CTT CTC C -3’

F756 5’- GGT GGT CGT GAA GGT CTC AT -3’ RNAP II 64 R883 5’- CCA ACC GAG TTA CGG ACT GT -3’

Note: All primers yielded amplification efficiency greater than or equal to 97%, as assessed through

amplification of a standard curve in triplicate.

50 significant effect of UV treatment indicates overall plasticity with respect to UV. Significant results for

UV-by-fish-history interaction were followed by relevant ANOVAs to evaluate plasticity within each type of fish treatment, using clone nested within population, population, and UV as main effects, or to evaluate differences between fish and nonfish populations within each type of UV treatment, using clone nested within population, population nested within fish treatment, and fish treatment as main effects. I used caution in interpretation rather than formal correction of p-values for multiple testing. Visualization of gene expression differences was accomplished by plotting 20% contours of the three-dimensional kernel density using the ks package written in R (Duong 2007 a and b, Team 2007). The densities were computed using kernel discriminate analysis in three dimensions on all four combinations of fish history and UV treatment.

I evaluated the relationship between gene expression and melanin, as well as melanin and time to maturity, using a general linear model (i.e., ANCOVA) that incorporated both discrete factors and continuous variables (Minitab® Statistical Software). I proceeded with this test only after confirming the absence of significant interactions between each factor and continuous variable that were examined together. In the first case, the discrete factors examined in my ANOVAs were hypothesized to affect the expression of genes that control melanin, creating the expectation that these factors and gene expression should covary. I approached this problem by predicting melanin as a function of all significant factors, as evaluated by ANOVA, plus expression of ebony, Ddc, and bursicon, and all three possible interactions between these genes. This ANCOVA provided a conservative test of the association between gene expression and melanin, by evaluating whether gene expression can explain variance in melanin that remains even after accounting for the effect of all experimental factors. To test whether changes in gene expression could explain the effect of experimental factors on melanin, I examined the direction of significant associations between gene expression and melanin for consistency with the direction and significance of their respective association with experimental factors. I also repeated the ANCOVA after eliminating factors that covaried with gene expression and melanin in the expected direction. I proceeded to analyze the relationship between melanin and time to maturity in the same manner.

To directly compare plasticity and adaptive change in expression of ebony and Ddc, I examined the angle between the vectors that describe the plastic response of nonfish populations and the adaptive 51 response of fish populations. Following the approach of McGuigan et al. (2005), I fit a MANOVA model that predicts gene expression of fish populations as a function of clone nested within population, population, and UV treatment, in order to obtain the SSCP matrix for the main effect of UV treatment. I then fit a MANOVA model that predicts gene expression under UV as a function of clone nested within population, population nested within fish treatment, and fish treatment in order to obtain the SSCP matrix for the main effect of fish treatment. After scaling by the appropriate error SSCP matrix (Rencher 1998;

McGuigan et al. 2005), the first principal component of these matrices yielded, respectively, the vector that describes the ancestral plastic response to the absence of UV and the vector that describes evolutionary change due to fish predation, when measured under UV conditions. The angle between these vectors represents the level of congruence in the developmental mechanisms associated with plastic loss of melanin and adaptive loss of melanin. I used 1000 bootstrap replicates in order to obtain 95% confidence intervals around the calculated angles between vectors.

Results

MANOVA on all traits together revealed a highly significant effect of all factors examined, including clone, population, fish treatment, UV treatment, and fish treatment by UV treatment interaction

(all p < 0.002).

Melanin

Levels of melanin showed significant plasticity with respect to UV radiation both overall (Table 2) and in the fish (F1,56 = 16.61, p < 0.001) and fishless (F1,65 = 84.99, p < 0.001) populations separately.

However, populations that coexist with fish exhibited a highly significant reduction in both overall melanin and plasticity of melanin, as evidenced by the effect of fish treatment and the UV x fish treatment interaction (Table 2, Fig. 9). Consistent with the hypothesis that melanin provides protection from UV, a significant fish by UV interaction on clutch failure revealed that fish populations experience a greater increase in mortality when exposed to UV (Table 3).

I investigated the relationship between melanin and time to maturity using ANCOVA, following confirmation of insignificant interaction effects between melanin and all experimental factors. Despite a 52 highly significant effect of UV on time to maturity when analyzed by ANOVA, neither UV nor melanin were significant predictors of time to maturity when both were included in the same model (Table 4).

When UV, which is highly correlated with melanin, is eliminated, however, melanin shows a significant positive effect on time to maturity. This result indicates that the possible effects of UV and melanin production cannot be disentangled using these data. However, both melanin and time to maturity increase more under UV light in the fishless populations than the fish populations (Table 3), even though the fishless populations show less stress under UV, as measured by clutch failure.

Gene Expression

MANOVA on gene expression data revealed a highly significant effect of clone and fish treatment, as well as a highly significant fish-by-UV-treatment interaction (Table 5, Figure 10). There was no significant effect of population. Subsequent ANOVAs revealed a highly significant effect of fish predation on ebony and a highly significant fish by UV interaction for Ddc (Table 6, Figure 9). Expression of bursicon exhibited both a highly significant effect of fish treatment and a significant fish-by-UV-treatment interaction. Both Ddc and bursicon showed a significant effect of clone, but population was not a significant source of variation for any gene. Subsequent comparisons via ANOVA indicated that fishless populations exhibit significant plasticity in Ddc with respect to UV (F1,46 = 16.29, p = 0.002), but fish

Table 2: ANOVA results for phenotypic traits.

Trait

Melanin Clutch Failure Time to Maturity

Effect df F p df F p df F p

Clone 48 1.21 0.199 47 1.91 0.014 51 1.28 0.129 Population 2 15.37 0.000 2 2.80 0.071 2 12.05 0.000 Fish 1 180.46 0.000 1 8.74 0.005 1 0.10 0.758 UV 1 92.31 0.000 1 79.98 0.000 1 9.93 0.002 UV x Fish 1 27.72 0.000 1 5.72 0.021 1 3.09 0.081

Note: Shown are the degrees of freedom (df), F-values (F) and p-values (p). Significant results (p < 0.05) are indicated in bold. 53 populations do not (F1,46 = 1.61, p = 0.23). In addition, expression of Ddc is significantly different between fish and nonfish populations when measured under UV conditions (F1,46 = 5.84, p = 0.025), but not whenUV light is absent (F1,46 = 2.60, p = 0.123). Finally, expression of bursicon is plastic with respect to

UV light in fish populations (F1,46 = 5.61, p = 0.037), but not nonfish populations (F1,46 = 0.87, p = 0.371).

My ANCOVA analysis predicted melanin as a function of all significant main effects, plus expression of and interaction between all three genes of interest. Despite the conservative nature of this test, it revealed a significant effect of ebony, Ddc, and ebony-by-Ddc interaction (Table 7). In addition, the direction of this effect was consistent with the direction in which fish and UV treatments are significantly associated with both gene expression and melanin. Consequently, when fish and UV treatments were dropped from the model, p-values decreased for all three terms. This less conservative test assumed that fish and UV treatments affect melanin entirely through their influence on expression of the three genes I examined here.

Figure 9. Response of melanin and gene expression to UV radiation. Lines show reaction norms of fishless populations (Source and Frog) and fish populations (Puppet and Evelyn). Melanin is reported in ug/mm Daphnia length and gene expression is reported as a percent of the highest expression sample in the experiment. 54 Table 3. Means and standard errors of all traits.

Treatment

No UV UV

Exposure Trait Population N Mean (SE) N Mean (SE) Time Source 0 13 0.51 (0.03) 10 0.96 (0.07) Frog 0 15 0.72 (0.03) 14 1.15 (0.05) Melanin Puppet 53 13 0.42 (0.03) 12 0.50 (0.04) Evelyn 91 10 0.20 (0.02) 11 0.38 (0.03) Source 0 11 0.01 (0.01) 11 0.17 (0.06)

Clutch Frog 0 15 0.06 (0.03) 15 0.29 (0.05) Failure Puppet 53 13 0.05 (0.03) 12 0.38 (0.07) Evelyn 91 12 0.12 (0.04) 12 0.45 (0.07) Source 0 13 8.28 (0.11) 11 8.99 (0.39)

Time to Frog 0 16 8.95 (0.15) 15 10.54 (0.40) Maturity Puppet 53 13 9.56 (0.26) 12 10.07 (0.70) Evelyn 91 13 8.15 (0.13) 11 8.30 (0.48) Source 0 6 0.41 (0.05) 6 0.31 (0.06)

Bursicon Frog 0 6 0.37 (0.03) 6 0.38 (0.05) Expression Puppet 53 6 0.64 (0.07) 6 0.74 (0.12) Evelyn 91 6 0.61 (0.06) 6 0.70 (0.07) Source 0 6 0.41 (0.12) 6 0.19 (0.06)

DDC Frog 0 6 0.55 (0.09) 6 0.34 (0.03) Expression Puppet 53 6 0.32 (0.02) 6 0.46 (0.09) Evelyn 91 6 0.38 (0.05) 6 0.40 (0.06) Source 0 6 0.46 (0.06) 6 0.39 (0.08)

Ebony Frog 0 6 0.33 (0.06) 6 0.34 (0.02) Expression Puppet 53 6 0.61 (0.08) 6 0.50 (0.06) Evelyn 91 6 0.45 (0.05) 6 0.57 (0.05)

Note: Clutch failure is given as a percent of total clutches, time to maturity is given in days, and melanin is given in ug melanin/mm of Daphnia length. Gene expression is a relative measure, given as a percent of the highest expression sample in the data set. 55 Table 4: ANCOVA results for time to maturity.

Effect df F p df F p Population 3 6.39 0.001 3 6.34 0.001 Melanin 1 0.65 0.420 1 6.34 0.014 UV 1 2.57 0.112

Note: Statistics are shown for models with and without UV included as a factor.

Shown are the degrees of freedom (df), F-values (F) and p-values (p). Significant results (p < 0.05) are indicated in bold.

Table 5. MANOVA results for gene expression.

Effect dF F p

Clone 60/60 2.47 0.000 Population 6/40 1.41 0.236 Fish 3/20 34.62 0.000 UV 3/20 1.85 0.171 UV x Fish 3/20 5.51 0.006

Note: Shown are the degrees of freedom (dF), F-values, and p-values based on Wilks’ Lambda. Significant results (p < 0.05) are indicated in bold.

Table 6. ANOVA results for gene expression.

Gene

Ebony DDC Bursicon

Effect dF F p dF F p dF F p

Clone 20 1.44 0.203 20 2.36 0.026 20 4.17 0.001 Population 2 0.97 0.395 2 1.35 0.281 2 0.10 0.904 Fish 1 9.44 0.006 1 0.07 0.798 1 25.48 0.000 UV 1 0.09 0.767 1 3.00 0.097 1 0.90 0.354 UV x Fish 1 0.25 0.621 1 13.19 0.001 1 5.31 0.031

Note: Shown are the degrees of freedom (dF), F-values, and p-values based on Wilks’ Lambda. Significant results (p < 0.05) are indicated in bold. 56

Figure 10. Patterns of gene expression in 3-dimensional space. Differences in expression within each combination of fish treatment and UV treatment are visualized for ebony, Ddc, and bursicon. Colored areas represent 20% contours of the 3-dimensional kernal density estimate. Evolution of gene expression in response to fish is evident in the separation of warm and cool colors. Differences in the plasticity of expression with respect to UV radiation are evident from the lack of congruence in vectors that point from expression in non-UV conditions (light color) to expression under UV (dark color) for fish and nonfish populations, respectively.

Table 7. ANCOVA results for melanin.

Effect coefficient df F p

Pop(FishT) 2 8.64 0.001 FishT 1 39.42 0.000 UV 1 26.11 0.000 Ebn -1.45 1 5.23 0.028 Ddc -1.39 1 6.06 0.019 Burs 0.01 1 0.00 0.983 Ebn*Ddc 3.27 1 6.49 0.015 Ebn*Burs 0.18 1 0.07 0.799 Burs*Ddc -0.68 1 0.47 0.498

Note: Shown are the degrees of freedom (dF), F-values, and p-values based on Wilks’ Lambda.

Significant results (p < 0.05) are indicated in bold. 57 Discussion

Given the existence of replicated salmonid introductions, Sierra Nevada D. melanica provide an unusually penetrating view of rapid adaptation in response to an invasive predator. In this study, the pattern of a significant effect of fish predation on either the mean or plasticity of all traits provided strong evidence that measured differences between fish and fishless populations represent adaptive change (Bull et al. 1997; Cooper et al. 2001, 2003). Particularly when coupled with a lack of significant population-level effects, this pattern indicated that the fishless populations represent the most likely ancestral state of Sierra

D. melanica, and that the independent fish populations have adapted through parallel or convergent evolution. In addition, I present evidence that increased melanin is associated with decreased clutch failure under UV, but a general increase in time to maturity within populations. Melanin thus provides a benefit in the presence of UV but likely a general cost that would make it maladaptive in the absence of UV; together, these results demonstrate an adaptive nature of the ancestral reaction norm of melanin to UV. In concert with the proposed first step of evolution through genetic accommodation, the ancestral reaction norm is expected to expose induced, lighter phenotypes to selection, given seasonal variation in UV (Rautio and

Korhola 2002) as well as variation in DVM that can translate into altered levels of melanin through an indirect effect on UV exposure.

Results from the melanin assay indicate that adaptation to fish predation is characterized by reduction in both the mean and the plasticity of melanin. This produces the classic pattern of plasticity that echoes adaptive change between populations, and is in agreement with the prediction that selection on ancestral plasticity will be followed by genetic change in the shape of the reaction norm. To assess the role of plasticity in providing the developmental mechanism for adaptive change, I measured patterns of expression in the melanin genes Ddc and ebony and the tanning gene bursicon. Ddc exhibits a pattern of plastic regulation associated with ancestral plasticity in melanin as well as a pattern of genetic assimilation associated with reduction of both the mean and plasticity of melanin in fish populations. This provides evidence that the effect of Ddc on producing ancestral plasticity is directly aligned with an effect on producing adaptive divergence between fish and fishless populations, and is congruent with the hypothesis that ancestral plasticity can provide the developmental mechanisms to effect adaptive change. In contrast, 58 expression of ebony is invariant with respect to the environment in both fish and fishless populations, but shows a clear pattern of upregulation in fishless populations. This is associated with an overall reduction in melanin across both non-UV and UV environments. Expression of bursicon is substantially different in fish populations, exhibiting plasticity with respect to UV as well as a higher overall mean. However, it does not appear to be associated with the production of melanin.

The patterns of gene regulation uncovered by my study closely match known functions of the melanin genes Ddc and ebony (reviewed by True 2003). In the insect melanin pathway, Ddc converts

DOPA to dopamine, diverting it from direct conversion to brown or black DOPA melanin by a product from the gene yellow. The gene ebony comes into play further downstream, where its product changes dopamine, which would otherwise be converted into brown or black dopamine melanin, into NBAD, which eventually forms yellowish-tan NBAD sclerotin. Upregulation of ebony is well established as an inhibitor of melanin in Drosophila (Wittkopp et al. 2002). Although the melanin pathways of crustaceans are largely unstudied, information from insects suggests that upregulation of Ddc is likely to decrease melanin by diverting DOPA down the dopamine pathway, and that upregulation of ebony further decreases melanin by diverting dopamine into a pathway that produces light-colored sclerotin. The gene bursicon is known to produce the tanning hormone Bursicon, which plays a critical role in cuticular sclerotization (Taghert and

Truman 1982; Gade and Hoffmann 2005) but no currently known role in melanin synthesis. Given the pattern of strong correlation between ebony and bursicon expression, I propose that upregulation of bursicon may be a correlated response to selection on ebony, although evaluation of this hypothesis will require further study.

Overall, my results provide strong evidence for rapid adaptation through genetic accommodation.

The presence of independent fishless and fish populations allows me to make well-founded inferences regarding the ancestral and derived patterns of phenotypic plasticity and underlying patterns of gene regulation. In particular, I was able to provide evidence that a developmental factor underlying ancestral plasticity, Ddc expression, has been coopted through genetic assimilation to produce adaptive change between populations experiencing disparate selection pressure. Measurement of the magnitude of adaptive divergence experienced under UV suggests that 72% of the difference between fishless and fish populations can be attributed to loss of ancestral plasticity, associated with change in the regulation of Ddc. 59 Additional adaptive change is associated with constitutive upregulation of gene expression, and most likely results from upregulation in the gene ebony. These results are congruent with theoretical models that treat means and slopes of reaction norms as independent quantitative genetic traits (e.g., Gavrilets and

Scheiner 1993a). Because adaptation has proceeded over just 53-91 years of exposure to fish predation, my results highlight the role of genetic accommodation in facilitating rapid evolutionary responses that have potentially profound ecological implications. 60 CHAPTER 5

CONCLUSION

In all three papers presented here, the use of a developmental perspective provides novel answers to fundamental evolutionary questions. Chapters 2 and 3 explore the effect of adding developmental realism to models of the evolution of phenotypic plasticity. Chapter 2 reveals that realistic developmental architecture can result in previously unrecognized constraints on the evolution of phenotypic plasticity. In addition, previous studies of plasticity have rarely considered the effect of stabilizing selection. The structure of the model in Chapter 2 makes it clear that plasticity, by definition, shapes the population-level translation of contextual variance into phenotypic variance. Because of this, reaction norms are expected to be influenced by stabilizing (or disruptive) selection whenever there is variance in either external or internal context. Furthermore, both chapter two and three reveal that, in the presence of a simple nonadditive interaction, the G-matrices currently used to predict evolution in phenotypic plasticity change fundamentally, even over the short term. This change has a profound effect on predicted evolutionary trajectories.

Chapters 3 and 4 both address the rate of adaptation to a novel predator, and thus provide insight into the ability of species to persist following a sudden change in selective regime. In Chapter 3, the presence of plasticity allows directional selection on two different traits to create genetic covariance that is largely aligned with the direction of selection, increasing the rate and influencing the direction of adaptive evolutionary change. In Chapter 4, I was able to provide evidence that a developmental factor underlying ancestral plasticity, Ddc expression, has been co-opted through genetic assimilation to produce rapid adaptive change in populations experiencing predation from introduced fish. Because this adaptation proceeded over just 53-91 years of exposure to fish predation, my results highlight the role of genetic accommodation in facilitating rapid evolutionary responses that have potentially profound ecological implications. 61 REFERENCES

Ackerly, D., and S. E. Sultan. 2006. Mind the gap: the emerging synthesis of plant 'eco-devo'. New Phytologist 170:648-653.

Amarillo-Suarez, A. R., and C. W. Fox. 2006. Population differences in host use by a seed beetle: local adaptation, phenotypic plasticity and maternal effects. Oecologia 150:247-258.

Badyaev, A. V. 2005. Role of stress in evolution: from individual adaptability to evolutionary adaptation. Pages 277-302 in B. Hallgrimsson and B. K. Hall, eds. Variation: A central concept in biology. Elsevier Academic Press, London.

Baker, H. G. 1974. The evolution of weeds. Annual Review of Ecology and Systematics 5:1-24.

Baldwin, J. M. 1896. A new factor in evolution. American Naturalist 30:441-451; 536-553.

Berrigan, D., and S. M. Scheiner. 2004. Modeling the evolution of phenotypic plasticity. Pages 82-97 in T. J. DeWitt and S. M. Scheiner, eds. Phenotypic plasticity: Functional and conceptual approaches. Oxford University Press, New York.

Bradford, D. F., S. D. Cooper, T. M. J. Jenkins, K. Kratz, O. Sarnelle, and A. D. Brown. 1998. Influences of natural acidity and introduced fish on faunal assemblages in California alpine lakes. Canadian Journal of Fisheries and Aquatic Sciences 55:2478-2491.

Brakefield, P. M. 2006. Evo-devo and constraints on selection. Trends in Ecology and Evolution 21:362- 368.

Breuker, C. J., V. Debat, and C. P. Klingenberg. 2006. Functional evo-devo. Trends in Ecology and Evolution 21:488-492.

Brodie, E. D. I. 2000. Why evolutionary genetics does not always add up. Pages 3-19 in J. B. Wolf and E. D. I. Brodie, eds. Epistasis and the evolutionary process. Oxford University Press, New York.

Bull, J. J., M. R. Badgett, H. A. Wichman, J. P. Huelsenbeck, D. M. Hillis, A. Gulati, C. Ho, and I. J. Molineux. 1997. Exceptional convergent evolution in a virus. Genetics 147:1497-1507.

Caicedo, A. L., J. R. Stinchcombe, K. M. Olsen, J. Schmitt, and M. D. Purugganan. 2004. Epistatic interaction between Arabidopsis FRI and FLC flowering time genes generates a latitudinal cline in a life history trait. Proceedings of the National Academy of Sciences, USA 101:15670-15675.

Callahan, H. S., M. Pigliucci, and C. D. Schlichting. 1997. Developmental phenotypic plasticity: where ecology and evolution meet molecular biology. Bioessays 19:519-525.

Carroll, S. B., and S. P. Klassen. 1997. Genetic differentiation of fitness-associated traits among rapidly evolving populations of the soapberry bug. Evolution 51:1182-1188.

Cheverud, J. M. 1984. Quantitative genetics and developmental constraints on evolution by selection. Journal of Theoretical Biology 110:155-171.

Cooper, T. F., D. E. Rozen, and R. E. Lenski. 2003. Parallel changes in gene expression after 20,000 generations for evolution in Escherichia coli. Proceedings of the National Academy of Sciences 100:1072-1077.

Cooper, V. S., D. Schneider, M. Blot, and R. E. Lenski. 2001. Mechanisms causing rapid and parallel losses of ribose catabolism in evolving populations of Escherichia coli B. Journal of Bacteriology 183:2834-2841. 62 Cousyn, C., L. De Meester, J. K. Colbourne, L. Brendonck, D. Verschuren, and F. Volckaert. 2001. Rapid, local adaptation of zooplankton behavior to changes in predation pressure in the absence of neutral genetic changes. Proceedings of the National Academy of Sciences 98:6256-6260.

Czesak, M. E., C. W. Fox, and J. B. Wolf. 2006. Experimental evolution of phenotypic plasticity: how predictive are cross-environment genetic correlations? American Naturalist 168:323-335.

Darwin, C. 1859. The origin of species by means of natural selection. John Murray, London, England.

Day, T., J. Pritchard, and D. Schluter. 1994. Ecology and genetics of phenotypic plasticity: a comparison of two sticklebacks. Evolution 48. de Jong, G. 2005. Evolution of phenotypic plasticity: patterns of plasticity and the emergence of ecotypes. New Phytologist 166:101-118. de Jong, G., and N. Behera. 2002. The influence of life-history differences on the evolution of reaction norms of phenotypic plasticity. Evolutionary Ecology Research 4:1-25.

De Meester, L. 1996. Evolutionary potential and local genetic differentiation in a phenotypically plastic trait of a cyclical parthenogen, Daphnia magna. Evolution 50:1293-1298.

DeWitt, T. J., and S. M. Scheiner. 2004. Phenotypic plasticity: Functional and conceptual approaches. Oxford University Press, New York.

Duong, T. 2007a. ks: Kernel density estimation and kernal discriminant analysis for multivariate data in R. Journal of Statistical Software 21

Duong, T. 2007b. ks: Kernal smoothing. R package version 1.4.10.

Dusheck, J. 2002. It's the ecology, stupid! Nature 418:578-579.

Elliot, M. J., and M. H. Loughlin. 2005. Historical overview of fishery management in Yosemite National Park 1877-1992. Natural Resources Report Draft, Yosemite National Park, California.

Falconer, D. S. 1952. The problem of environment and selection. American Naturalist 86:293-298.

Fisk, D. L., L. C. I. Latta, R. A. Knapp, and M. E. Pfrender. 2007. Rapid evolution in response to introduced predators I: rates and patterns of morphological and life-history trait divergence. BMC Evolutionary Biology 7:1-11.

Frankel, W. N., and N. J. Schork. 1996. Who's afraid of epistasis? Nature Genetics 14:371-373.

Gade, G., and K. Hoffmann. 2005. Neuropeptides regulating development and reproduction in insects. Physiological Entomology 30:103-121.

Gavrilets, S., and S. M. Scheiner. 1993a. The genetics of phenotypic plasticity. V. Evolution of reaction norm shape. Journal of Evolutionary Biology 6:31-48.

Gavrilets, S., and S. M. Scheiner. 1993b. The genetics of phenotypic plasticity. VI. Theoretical predictions for directional selection. Journal of Evolutionary Biology 6:49-68.

Gerhart, J., and M. Kirschner. 2007. The theory of facilitated variation. Proceedings of the National Academy of Sciences 104:8582-8589.

Ghalambor, C. K., J. K. McKay, S. P. Carroll, and D. N. Reznick. 2007. Adaptive versus non-adaptive phenotypic plasticity and the potential for contemporary adaptation in new environments. Functional Ecology 21:394-407. 63 Gibson, G., and D. S. Hogness. 1996. Effect of polymorphism in the Drosophila regulatory gene ultrabithorax on homeotic stability. Science 271:200-203.

Gomez-Mestre, I., and D. R. Buchholz. 2006. Developmental plasticity mirrors differences among taxa in spadefoot toads linking plasticity and diversity. Proceedings of the National Academy of Sciences 103:19021-19026.

Gomulkiewicz, R., and R. D. Holt. 1995. When does evolution by natural selection prevent extinction. Evolution 49:201-207.

Gromko, M. H. 1995. Unpredictability of correlated response to selection: pleiotropy and sampling interact. Evolution 49:685-693.

Gurevitch, J. 1988. Variation in leaf dissection and leaf energy budgets among populations of Achillea from an altitudinal gradient. American Journal of Botany 75:1298-1306.

Gurevitch, J. 1992. Sources of variation in leaf shape among two populations of Achellea larzulosa. Genetics 130:385-394.

Gurevitch, J., and P. H. Schuepp. 1990. Boundary layer properties of highly dissected leaves: an investigation using an electrochemical fluid tunnel. Plant Cell Environment 13:783-792.

Haldane, J. B. S. 1957. The cost of natural selection. Journal of Genetics 55:511-524.

Hansen, T. F. 2006. The evolution of genetic architecture. Annual Review of Ecology, Evolution, and Systematics 37:123-157.

Hau, M. 2007. Regulation of male traits by testosterone: implications for the evolution of vertebrate life histories. BioEssays 29:133-144.

Hendrikse, J. L., T. E. Parsons, and B. Hallgrimsson. 2007. Evolvability as the proper focus of evolutionary developmental biology. Evolution and Development 9:393-401.

Hendry, A. P., T. Day, and E. B. Taylor. 2001. Population mixing and the adaptive divergence of quantitative traits in discrete populations: a theoretical framework for empirical tests. Evolution 55:459-466.

Hendry, A. P., E. B. Taylor, and J. D. McPhail. 2002. Adaptive divergence and the balance between selection and gene flow: lake and stream stickleback in the misty system. Evolution:1199-1216.

Hoekstra, H. E., and J. A. Coyne. 2007. The locus of evolution: evo devo and the genetics of adaptation. Evolution 61:995-1016.

Houle, D. 1991. Genetic covariance of fitness correlates: what genetic correlations are made of and why it matters. Evolution 45:630-648.

Johnson, N. A. 2007. The micro-evolution of development. Genetica 129:1-5.

Kessler, A., R. Halitschke, and I. Baldwin. 2004. Silencing the jasmonate cascade: induced plant defenses and insect populations. Science 305:665-668.

Knapp, R. A., C. P. Hawkins, J. Ladau, and J. G. McClory. 2005. Fauna of Yosemite National Park lakes has low resistance but high resilience to fish introductions. Ecological Applications 15:835-847.

Knapp, R. A., K. R. Matthews, and O. Sarnelle. 2001. Resistance and resilience of alpine lake fauna to fish introductions. Ecological Monographs 71:401-421. 64 Lande, R. 1976. The maintenance of genetic variability by mutation in a polygenic character with linked loci. Genetical Research 26:221-235.

Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, applied to brain: body size allometry. Evolution 33:402-416.

Lande, R. 1980. The genetic covariance between characters maintained by pleiotropic mutations. Genetics 94:203-215.

Lande, R. 1998. Anthropogenic, ecological and genetic factors in extinction and conservation. Researches on Population Ecology 40:259-269.

Lande, R., and S. J. Arnold. 1983. The measurement of selection on correlated characters. Evolution 37:1210-1226.

Latta, L. C. I., J. W. Bakelar, R. A. Knapp, and M. E. Pfrender. 2007. Rapid evolution in response to introduced predators II: the contribution of adaptive plasticity. BMC Evolutionary Biology 7:1-9.

Losos, J. B., D. A. Creer, D. Glossip, R. Goellner, A. Hampton, G. Roberts, N. Haskell, P. Taylor, and J. Ettling. 2000. Evolutionary implications of phenotypic plasticity in the hindlimb of the lizard Anolis sagrei. Evolution 54:301-305.

Lynch, M. 1985. Spontaneous mutations for life-history characters in an obligate parthenogen. Evolution 39:804-818.

Lynch, M. 2007. The frailty of adaptive hypotheses for the origins of organismal complexity. Proceedings of the National Academy of Sciences 104:8597-8604.

Maynard Smith, J., R. Burian, S. Kauffman, P. Alberch, J. Campbell, B. Goodwin, R. Lande, D. Raup, and L. Wolpert. 1985. Developmental constraints and evolution. The Quarterly Review of Biology 60:265-287.

Mayr, E. 2004. 80 years of watching the evolutionary scenery. Science 305:46-47.

McGuigan, K., S. F. Chenoweth, and M. W. Blows. 2005. Phenotypic divergence along lines of genetic variance. American Naturalist 165:32-43.

Moore, J.-S., J. L. Gow, E. B. Taylor, and A. P. Hendry. 2007. Quantifying the constraining influence of gene flow on adaptive divergence in the lake-stream threespine stickleback system. Evolution 61:2015-2026.

Parsons, K. J., and B. W. Robinson. 2006. Replicated evolution of integrated plastic responses during early adaptive divergence. Evolution 60

Peacor, S. D., S. Allesina, R. L. Riolo, and M. Pascual. 2006. Phenotypic plasticity opposes species invasions by altering fitness surface. PLOS 4:2112-2120.

Pfrender, M. E., and M. Lynch. 2000. Quantitative genetic variation in Daphnia: temporal changes in genetic architecture. Evolution 54:1502-1509.

Pigliucci, M. 2001. Phenotypic plasticity: Beyond nature and nurture. Johns Hopkins University Press, Baltimore.

Pigliucci, M. 2006. Genetic variance-covariance matrices: a critique of the evolutionary quantitative genetics research program. Biology and Philosophy 21:1-23. 65 Pigliucci, M., and C. J. Murren. 2003. Perspective: Genetic assimilation and a possible evolutionary paradox: Can macroevolution sometimes be so fast as to pass us by? Evolution 57:1455-1464.

Pijanowska, J., L. J. Weider, and W. Lampert. 1993. Predator-mediated genotypic shifts in a prey population: Experimental evidence. Oecologia 96:40-42.

Price, T. D., A. Qvarnstrom, and D. E. Irwin. 2003. The role of phenotypic plasticity in driving genetic evolution. Proceedings of the Royal Society of London, Series B:1433-1440.

Provine, W. B. 1971. The origins of theoretical population genetics. University of Chicago Press, Chicago.

Queitsch, C., T. A. Sangster, and S. Lindquist. 2002. Hsp90 as a capacitor of phenotypic variation. Nature 417:618-624.

Rautio, M., and A. Korhola. 2002. UV-induced pigmentation in subarctic Daphnia. Limnology and Oceanography 47:295-299.

Remold, S. K., and R. E. Lenski. 2004. Pervasive joint influence of epistasis and plasticity on mutational effects in Escherichia coli. Nature Genetics 36:423-426.

Rencher, A. C. 1998. Multivariate statistical inference and applications. John Wiley and Sons, Inc., New York.

Rice, S. H. 2002. A general population genetic theory for the evolution of developmental interactions. Proceedings of the National Academy of Sciences, USA 99:15518-15523.

Rice, S. H. 2004a. Evolutionary theory: Mathematical and conceptual foundations. Sinauer Associates, Inc., Sunderland.

Rice, S. H. 2004b. Developmental associations between traits: covariance and beyond. Genetics 166:513- 526.

Robinson, B. W., and R. Dukas. 1999. The influence of phenotypic modification on evolution: the Baldwin effect and modern perspectives. Oikos 85:582-589.

Rutherford, S. L. 2000. From genotype to phenotype: buffering mechanisms and the storage of genetic information. BioEssays 22:1095-1105.

Rutherford, S. L. 2003. Between genotype and phenotype: protein chaperones and evolvability. Nature Reviews Genetics 4:263-274.

Rutherford, S. L., and S. Lindquist. 1998. Hsp90 as a capacitor for morphological evolution. Nature 396:336-343.

Saegrov, H., A. Hobaek, and J. H. L'Abee-Lund. 1996. Vulnerability of melanic Daphnia to brown trout predation. Journal of Plankton Research 18:2113-2118.

Sarà, M. 1996. A "sensitive" cell system. Its role in a new evolutionary paradigm. Rivista de Biologia 89:139-156.

Sarkar, S. 1999. From the Reaktionsnorm to the adaptive norm: the norm of reaction, 1909-1960. Biology and Philosophy 14:235-252.

Sarkar, S. 2004. From the Reaktionsnorm to the evolution of adaptive plasticity: a historical sketch, 1909- 1999. Pages 10-30 in T. J. DeWitt and S. M. Scheiner, eds. Phenotyic plasticity: Functional and conceptual approaches. Oxford University Press, New York. 66 Scheiner, S. M. 1998. The genetics of phenotypic plasticity. VII. Evolution in a spatially structured environment. Journal of Evolutionary Biology 11:303-320.

Scheiner, S. M., and T. J. DeWitt. 2004. Future research directions. Pages 201-206 in T. J. DeWitt and S. M. Scheiner, eds. Phenotypic plasticity: Functional and conceptual approaches. Oxford University Press, New York.

Scheiner, S. M., and R. F. Lyman. 1989. The genetics of phenotypic plasticity. I. Heritability. Journal of Evolutionary Biology

Schlichting, C. D. 2004. The role of phenotypic plasticity in diversification. Pages 191-200 in T. J. DeWitt and S. M. Scheiner, eds. Phenotypic plasticity: Functional and conceptual approaches. Oxford University Press, Oxford.

Schlichting, C. D., and M. Pigliucci. 1998. Phenotypic evolution: A reaction norm perspective. Sinauer Associates, Inc., Sunderland.

Schlichting, C. D., and H. Smith. 2002. Phenotypic plasticity: linking molecular mechanisms with evolutionary outcomes. Evolutionary Ecology 16:189-211.

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

Schmitt, J., S. A. Dudley, and M. Pigliucci. 1999. Manipulative approaches to testing adaptive plasticity: phytochrome-mediated shade avoidance responses in plants. American Naturalist 154:S43-S54.

Schmitt, J., J. R. Stinchcombe, M. S. Heschel, and H. Huber. 2003. The adaptive evolution of plasticity: phytochrome-mediated shade avoidance responses. Integrative and Comparative Biology 43:459- 469.

Simpson, G. G. 1953. The Baldwin effect. Evolution 7:110-117.

Smits, J. D., F. Witte, and F. G. Van Veen. 1996. Functional changes in the anatomy of the pharyngeal jaw apparatus of Astatoreochromis alluaudi (Pisces, Cichlidae), and their effects on adjacent structures. Biological Journal of the Linnean Society 59:389-409.

Stauffer, J. R. J., and E. V. S. Gray. 2004. Phenotypic plasticity: its role in trophic radiation and explosive speciation in cichlids (Teleostei: Cichlidae). Animal Biology 54:137-158.

Steppan, S. J., P. C. Phillips, and D. Houle. 2002. Comparative quantitative genetics: evolution of the G matrix. Trends in Ecology and Evolution 17:320-327.

Stinchcombe, J., C. Weinig, M. Ungerer, K. Olsen, C. Mays, S. Halldorsdottir, M. Purugganan, and J. Schmitt. 2004. A latitudinal cline in flowering time in Arabidopsis thaliana modulated by the flowering time gene FRIGIDA. Proceedings of the National Academy of Sciences, USA 101:4712- 4717.

Stracke, S., C. Kistner, S. Yoshida, L. Mulder, S. Sato, T. Kaneko, S. Tabata, N. Sandal, J. Stougaard, K. Szczyglowski, and M. Parniske. 2002. A plant receptor-like kinase required for both bacterial and fungal symbiosis. Nature 417:959–962.

Sultan, S. E. 1992. Phenotypic plasticity and the Neo-Darwinian legacy. Evolutionary Trends in Plants 6:61-71.

Sultan, S. E. 2005. An emerging focus on plant ecological development. New Phytologist 166:1-5.

Sultan, S. E., and H. G. Spencer. 2002. Metapopulation structure favors plasticity over local adaptation. American Naturalist 160:271-283. 67 Suzuki, Y., and F. Nijhout. 2006. Evolution of a polyphenism by genetic accomodation. Science 311:650-652.

Sword, G. A. 2002. A role for phenotypic plasticity in the evolution of aposematism. Proceedings of the Royal Society of London B: Biological Sciences 269:1639-1644.

Taghert, P. H., and J. W. Truman. 1982. The distribution and molecular characteristics of the tanning hormone, bursicon, in the tobacco hornworm Manduca sexta. Journal of Experimental Botany 98:373-383.

Team, R. D. C. 2007. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Templeton, A. R. 2000. Epistasis and complex traits. Pp. 41-57 in J. B. Wolf, E. D. I. Brodie and M. J. Wade, eds. Epistasis and the Evolutionary Process. Oxford University Press, New York.

Thompson, D. B. 1999. Genotype-environment interaction and the ontogeny of diet-induced phenotypic plasticity in size and shape of Melanopus femurrubrum (Orthoptera: Acrididae). Journal of Evolutionary Biology 12:38-48.

True, J. R. 2003. Insect melanism: the molecules matter. Trends in Ecology and Evolution 18:640-647. van Kleunen, M., and M. Fischer. 2005. Constraints on the evolution of adaptive phenotypic plasticity in plants. New Phytologist 166:49-60.

Vandesompele, J., K. De Preter, F. Pattyn, B. Poppe, N. Van Roy, A. De Paepe, and F. Speleman. 2002. Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biology 3:1-12.

Via, S., R. Gomulkiewicz, G. de Jong, S. M. Scheiner, C. D. Schlichting, and P. van Tienderen. 1995. Adaptive phenotypic plasticity: consensus and controversy. Trends in Ecology and Evolution 10:212-217.

Via, S., and R. Lande. 1985. Genotype-environment interaction and the evolution of phenotypic plasticity. Evolution 39:505-522.

Waddington, C. H. 1942. Canalization of development and the inheritance of acquired characters. Nature 150:563-565.

Waddington, C. H. 1952. Selection of the genetic basis for an acquired character. Nature 169:278.

Waddington, C. H. 1953. Genetic assimilation of an acquired character. Evolution 7:118-126.

Weinig, C., and J. Schmitt. 2004. Environmental effects on the expression of quantitative trait loci and implications for phenotypic evolution. Bioscience 54:627-635.

West-Eberhard, M. J. 2003. Developmental plasticity and evolution. Oxford University Press, New York.

West-Eberhard, M. J. 2005. Developmental plasticity and the origin of species differences. Proceedings of the National Academy of Sciences 102:6543-6549.

Windig, J. J., C. G. F. de Kovel, and G. De Jong. 2004. Genetics and mechanics of plasticity. Pages 31-49 in T. J. DeWitt and S. M. Scheiner, eds. Phenotypic plasticity: Functional and conceptual approaches. Oxford University Press, New York. 68 Wittkopp, P. J., J. R. True, and S. B. Carroll. 2002. Reciprocal functions of the Drosophila Yellow and Ebony proteins in the development and evolution of pigment patterns. Development 129:1849- 1858.

Wolf, J. B. 2002. The geometry of phenotypic evolution in developmental hyperspace. Proceedings of the National Academy of Sciences, USA 99:15849-15851.

Wolf, J. B., W. A. Frankino, A. F. Agrawal, E. D. I. Brodie, and A. J. Moore. 2001. Developmental interactions and the constituents of quantitative variation. Evolution 55:232-245.

Wolpert, L. 1994. The evolutionary origin of development: cycles, patterning, privilege and continuity. Development Supplement:79-84.

Wolterek, R. 1909. Weitere experimentelle Untersuchungen über Artveränderung, speziell über das Wesen quantitativer Artunterscheide by Daphniden. Verhandlungender Deutschen Zoologischen Gesellschaft 19:110-192.

CURRICULUM VITAE

Alison Scoville (2008)

EDUCATION

Carleton College, Northfield, MN B.A. Magna cum laude Biology 1997

Utah State University, Logan, UT Ph.D. Candidate Biology 2008

PUBLICATIONS

Grigione, M, A. Scoville, G. Scoville, and K. Crooks. 2007. Neotropical Cats in Southeast Arizona and Surrounding Areas: Past and Present Status of Jaguars, Ocelots, and Jaguarundis. Journal of Neotropical Mammalogy 14(2): 189-199.

Crooks, K.R., M. Grigione, A. Scoville, and G. Scoville. In press. Distribution and relative activity of mammalian carnivores within the Peloncillo and Chiricahua Mountains. The Southwestern Naturalist.

PRESENTATIONS

Scoville, A. 2006. Epistasis constrains the evolution of phenotypic plasticity in a subdivided population: results from a quantitative genetic model. Paper presented at the joint annual meeting of the Society for the Study of Evolution, the Society of Systematic Biologists, and the American Society of Naturalists, Stony Brook, New York.

Scoville, A. and M. Pfrender. 2008. From adaptive plasticity to rapid adaptation: evolution of Daphnia gene expression in response to an introduced predator. Poster presented at the annual IRACDA conference, Chapel Hill, North Carolina.

Scoville, A. and M. Pfrender. 2008. Rapid parallel genetic assimilation of Daphnia gene expression in response to an introduced predator. Paper presented at the joint annual meeting of the Society for the Study of Evolution, the Society of Systematic Biologists, and the American Society of Naturalists, Minneapolis, Minnesota.

Scoville, A. and M. Pfrender. 2008. Genetic assimilation of Daphnia gene expression underlying rapid adaptation. Poster presented at the annual Ecological Genomics Symposium, Kansas City, Kansas.

70 EMPLOYMENT

Teaching Assistant—Utah State University Department of Biology: 2000-2007 Evolution, Quantitative Genetics, Introductory Biology, Animal Behavior, Human Physiology

Research Assistant—Pfrender Laboratory, Utah State University: 2002-2003 Development of quantitative PCR for Daphnia

Research Assistant—Ecological Reserve Design, Round River 2001 Conservation Studies (affiliated with Utah State University), Wyoming

Instructor—Conservation Biology, Field Biology, Statistics. 2000 Round River Conservation Studies, Namibia, Africa.

Instructor—Ecological Literacy, Field Biology. Hemas Council, 1998 Heiltsuk Native American Tribe, British Columbia.

Teaching Assistant—Conservation Biology, Field Biology, Western 1997-1999 Literature. Round River Conservation Studies. Montana, Arizona.

Research Assistant—Ecological Reserve Design, Round River 1998 Conservation Studies, British Columbia

Campus Naturalist—Carleton College, Northfield, Minnesota 1993-1997

ACADEMIC AWARDS AND HONORS

Utah State University Dissertation Writing Fellowship 2007

Claude E. Zobell Scholarship 2005

Vice President for Research Fellowship, Utah State University 2004

NSF Graduate Fellowship Honorable Mention 2000

Vice President for Research Fellowship, Utah State University 1999

Election to Sigma Xi, The Scientific Research Society 1997