<%BANNER%>

Characterization of the Distribution of Developmental Instability


PAGE 1

CHARACTERIZATION OF THE DISTRIBUTION OF DEVELOPMENTAL INSTABILITY By GREGORY ALAN BABBITT A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Gregory Alan Babbitt

PAGE 3

This document is dedicated to my grandmothe r, Sarah Miller, who taught me to admire the natural world.

PAGE 4

iv ACKNOWLEDGMENTS I would like to thank Rebecca Kimball, Susan Halbert, Bernie Hauser, Jane Brockmann, and Christian Klingenberg for helpful discussions and comments; Susan Halbert and Gary Steck (Division of Plant Industry, State of Florida) for specimen ID; Marta Wayne for use of her microscope and digital camera; Glenn Hall (Bee Lab, University of Florida) for a plentiful suppl y of bees; and the Department of Entomology and Nematology (University of Florida) for use of its environmental chambers. I thank Callie Babbitt for help in manuscript prepar ation and much additional love and support.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................viii LIST OF FIGURES...........................................................................................................ix ABSTRACT......................................................................................................................x ii CHAPTER 1 SEARCHING FOR A CONSISTENT INTERPRETATION OF DEVELOPMENTAL INSTABILITY? A GENERAL INTRODUCTION................1 2 ARE FLUCTUATING ASYMMETRY STUDIES ADEQUATELY SAMPLED? IMPLICATIONS OF A NEW MO DEL FOR SIZE DISTRIBUTION......................10 Introduction.................................................................................................................10 Developmental Stability: Definition, Measurement, and Current Debate..........10 The Distribution of Fluctuating Asymmetry.......................................................12 Methods......................................................................................................................17 Results........................................................................................................................ .22 Discussion...................................................................................................................30 The Distribution of FA........................................................................................30 Sample Size and the Estimation of Mean FA......................................................32 The Basis of Fluctuating Asymmetry..................................................................33 Conclusion...........................................................................................................34 3 INBREEDING REDUCES POWER-LAW SCALING IN THE DISTRIBUTION OF FLUCTUATING ASYMMETRY: AN EXPLANATION OF THE BASIS OF DEVELOPMENTAL INSTABILITY........................................................................36 Introduction.................................................................................................................36 What is the Basis of FA?.....................................................................................36 Exponential Growth and Non-No rmal Distribution of FA..................................38 Testing a Model for the Basis of FA...................................................................40 Methods......................................................................................................................41 Model Development............................................................................................41 Simulation of geometric Brownian motion..................................................41

PAGE 6

vi Simulation of fluctuating asymmetry...........................................................44 Inbreeding Experiment........................................................................................45 Morphometric analyses.......................................................................................48 Model selection and inference.............................................................................49 Results........................................................................................................................ .49 Model Simulation................................................................................................49 Experimental Results...........................................................................................50 Model Selection and Inference............................................................................52 Discussion...................................................................................................................54 Revealing the Genetic Component of FA...........................................................54 Limitations of the Model.....................................................................................56 The Sources of Scaling........................................................................................56 Potential Application to Cancer Screening..........................................................57 Conclusion...........................................................................................................58 4 TEMPERATURE RESPONSE OF FLUCTUATING ASYMMETRY TO IN AN APHID CLONE: A PROPOSAL FOR DETECTING SEXUAL SELECTION ON DEVELOPMENTAL INSTABILITY........................................................................60 Introduction.................................................................................................................60 The Genetic Basis of FA.....................................................................................60 The Environmental Basis of FA..........................................................................61 Temperature and FA in and Aphid Clone...........................................................63 Methods......................................................................................................................65 Results........................................................................................................................ .68 Discussion...................................................................................................................72 5 CONCLUDING REMARKS AND RECOMMENDATIONS..................................75 How Many Samples Are Enough?.............................................................................76 What Measure of FA is best?......................................................................................76 Does rapid growth stabilize or destabilize development?..........................................78 Can fluctuating asymmetry be a sexually selected trait?............................................78 Is fluctuating asymmetry a valuab le environmental bioindicator?.............................79 Scaling Effects in Statistical Di stributions: The Bigger Picture.................................80 APPENDIX A LANDMARK WING VEIN INTERSECTIONS CHOSEN FOR ANALYSISOF FLUCTUATING ASYMMETRY..............................................................................84 B USEFUL MATHEMATICAL FUNCTIONS.............................................................86 Asymmetric Laplace Distribution...............................................................................86 Half-Normal Distribution...........................................................................................86 Lognormal Distribution..............................................................................................86 Double Pareto Lognormal Distribution......................................................................86

PAGE 7

vii LIST OF REFERENCES...................................................................................................88 BIOGRAPHICAL SKETCH.............................................................................................96

PAGE 8

viii LIST OF TABLES Table page 2-1. Maximized log-likelihood (MLL), numbe r of model parameters (P) and Akaike Information Criterion differences ( AIC) for all distribut ional models tested. Winning models have AIC difference of zer o. Models with nearly equivalent goodness-of-fit to winners are underscored ( AIC <3.0). 4.0 < AIC <7.0 indicates some support for specified model. AIC >10.0 indicates no support (Burnham and Anderson 1998). Distances between landmarks (LM) used for first univariate size FA, aphids LM 1-2, bees LM 1-4 and l ong-legged fly LM 36 and for second univariate FA aphids LM 2-3, bees LM 2-6 and long-legged fly LM 4-5......................................................................................................................23 2-2. Best fit parameters for models in Table 2-1. Parameters for univariate size FA are very similar to multivariate size FA and are not shown. Skew indexes for asymmetric Laplace are also not shown...................................................................24 3-1. Distribution parameters and model fit for multivariate FA in two wild populations and four inbred lines of Drosophila simulans and one isogenic line of Drosophila melanogaster Model fits are AIC for unsigned centroid size FA (zero is best fit, lowest number is next best fit).................................................51

PAGE 9

ix LIST OF FIGURES Figure page 2-1. Schematic representation of mathem atical relationships between candidate models for the distribution of fluctu ating asymmetry. Mixtures here are continuous................................................................................................................14 2-2. Distribution of multivariate shape FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting lognormal (dashed line, lower inset), and double Pareto lognormal (solid line, uppe r inset) are indicated......................................................25 2-3. Distribution of multivariate centroid size FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting half-normal (dashed line, lower inset) and double Pareto lognormal distribution (solid line upper inset) are indicated..................................26 2-4. Distribution of univariate uns igned size FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting half-normal (dashe d line, lower inset) and double Pareto lognormal distribution (solid line upper inset) are indicated..................................27 2-5. Distribution of sample sizes (n) from 229 fluctuating asymme try studies reported in three recent meta-analyses (Vollest ad et al. 1999, Thornhill and Mller 1998 and Polak et al. 2003). Only five studies had sample sizes greater than 500 (not shown)......................................................................................................................28 2-6. Relationship between sample size and % error for estimates of mean FA drawn from best fitting size (dashed line) and shape (solid line) distributions using 1000 draws per sample size. All ru ns use typical winning double Pareto lognormal parameters (shape FA = -3.7, = 0.2, = 1000, = 9; for size FA = 1.2, = 0.7, = 4.0, = 4.0).................................................................................29 2-7. The proportion and percen tage (inset) of individuals with visible developmental errors on wings (phenodeviants) are show n for cotton aphids (A) and honeybees (B) in relation to distribution of shape FA (Procrustes distance). Average FA for both normal and phenodeviant aphids (C) and honeybees (D) are also given.........30 3-1. Ordinary Brownian motion (low er panel) in N simulated by summing independent uniform random variables (W) (upper panel)......................................42

PAGE 10

x 3-2. Geometric Brownian motion in N and log N simulated by multiplying independent uniform random variables. This was generated using Equation 1.5 with C = 0.54............................................................................................................43 3-3. Geometric Brownian motion in N a nd log N with upward drift. This was generated using Equation 1.5 with C = 0.60............................................................44 3-4. Model representation of Reed and Jo rgensens (2004) physic al size distribution model. Variable negativ e exponentially distributed stopping Times of random proportional growth (GBM with C = 0.5) create double Pareto lognormal distribution of size....................................................................................................46 3-5. A model representation of developm ental instability. Normally variable stopping times of random proportional gr owth (GBM with C > 0.5) create double Pareto lognormal di stribution of size...........................................................47 3-6. Simulated distributions of cell popul ation size and FA for different amounts of variation in the termination of growth (variance in normally distributed growth stop time). The fit of simulated data to the normal distribution can determined by how closely the plotted points foll ow the horizontal line (a good fit is horizontal)................................................................................................................50 3-7. Distribution of fluctuating asymme try and detrended fit to normal for two samples of wild population collected in Ga inesville, FL in summers of 2004 and 2005 and four inbred lines of Drosophila simulans derived from eight generations of full-sib crossing of the w ild population of 2004. Also included is one isogenic line of Drosophila melanogaster (mel75). All n = 1000. The fit of data to the normal distribution can dete rmined by how closely the plotted points follow the horizontal line (a good fit is horizontal).................................................53 4-1. Predicted proximate and ultimate leve l correlations of te mperature and growth rate to fluctuating asymmetry are different Ultimate level (evolutionary) effects assume energetic limitation of individua ls in the system. Proximate level (growth mechanical) effects do not. No tice that temperature and fluctuating asymmetry are negatively correlated in the proximate model while in the ultimate model they ar e positively correlated..........................................................64 4-2. Cotton aphid mean development time SE in days in relation to temperature (n= 531)....................................................................................................................68 4-3. Distribution of isogenic size, size based and shape based FA in monoclonal cotton aphids grown in controlled en vironment at different temperatures Distributions within each temperature treatment are similar to overall distributions shown here...........................................................................................69 4-4. Mean isogenic FA for (A.) centroid si zebased and (B.) Pr ocustes shape-based) in monoclonal cotton aphids (collected in Gainesville FL) grown on isogenic cotton seedlings at different temperatures................................................................70

PAGE 11

xi 4-5. Kurtosis of size-based FA in monoc lonal cotton aphids gr own on isogenic cotton seedlings at differe nt temperatures...........................................................................71

PAGE 12

xii Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy CHARACTERIZATION OF THE DI STRIBUTION OF DEVELOPMENTAL INSTABILITY By Gregory Alan Babbitt May 2006 Chair: Rebecca Kimball Cochair: Benjamin Bolker Major Department: Zoology Previous work on fluctuating asymmetr y (FA), a measure of developmental instability, has highlighted its controversial relationship w ith environmental stress and genetic architecture. I suggest that conflict may derive from th e fact that the basis of FA is poorly understood and, as a consequen ce, the methodology for FA studies may be flawed. While size-based measures of FA have been assumed to have half-normal distributions within populations, developmen tal modeling studies have suggested other plausible distributions for FA. Support for a non-normal distribution of FA is further supported by empirical studies that often record leptokurtic (i.e., fat or long-tailed) distributions of FA as well. In this disse rtation, I investigate a series of questions regarding the both the basis and distributi on of FA in populations. Is FA normally distributed and therefore likel y to be properly sampled in FA studies? If not normal, what candidate model distribution best fits th e distribution of FA? Is the shape of the distribution of FA similar to a simple and specific growth model (geometric Brownian

PAGE 13

xiii motion)? Does reducing individual variati on in populations through inbreeding affect follow the prediction of this model? Ho w does this shape respond to environmental factors such as temperature when ge netic variation is controlled? In three species of insects (cotton aphid, Aphis gossipyii Glover; honeybee, Apis mellifera ; and long-legged fly, Chrysosoma crinitus (Dolichipodidae)), I find that FA was best described by a double Pareto lognor mal distribution (DPLN), a lognormal distribution with power-law tails. The larg e variance in FA under this distri bution and the scaling in the tails both act to slow convergence to the mean, suggesting that many past FA studies are under-sampled when the dist ribution of FA is assumed to be normal. Because DPLN can be generated by geometric Brownian motion, it is ideal for describing behavior of cell populations in growing tissue. I demonstrated through both a mathematical growth model and an inbreeding experiment in Drosophila simulans that the shape of the distribution of FA is highl y dependent on the level of genetic redundancy or heterogeneity in a populati on. In monoclonal lines of cott on aphids, I also demonstrate that FA decreases with temp erature and that a shift in kurtosis is associated with temperature induced phenotypic plasticity. Th is supports the prediction of a proximate model for the basis of FA and also suggests shape of the distribution of FA responds to environmentally induced changes in gene expression on the same genetic background.

PAGE 14

1 CHAPTER 1 SEARCHING FOR A CONSISTENT INTERPRETATION OF DEVELOPMENTAL INSTABILITY? A GENERAL INTRODUCTION Everywhere, nature works true to sc ale and everything ha s its proper size. -DArcy Thompson There have been many incarnations of th e idea that stability and symmetry are somehow related. Hippocrates (460-377 B.C. ) was the first to postulate internal corrective propertie s that work in the presence of disease. Waddington (1942) suggested existence of similar homeostatic buffering agai nst random and presumably additive errors occurring during development. The term f luctuating asymmetry, first introduced by Ludwig (1932), was later adopted by Mather (1953), Reeve (1960) and Van Valen (1962) to describe a measurable form of morphological noise repres enting a hypothetical lack of buffering that is always present during develo pment of organisms. Recently, biologists have become very interested both in fluctu ating asymmetrys poten tial usefulness as a universal bioindicator of environmental health (Parsons 1992) and in its potential as an indicator or even an overt signal of an indi viduals overall genetic quality (Mller 1990). However, over a decade of work has left the field with no clear relationship between increased fluctuating asymmetry and either e nvironmental or genetic stress (Bjorksten et al. 2000, Lens et al. 2002). Despite this fact fluctuating asymmetry is still often assumed to indicate of developmental instability. Recently, Debat and David (2001) page 560 define developmental stability as a set of mechanisms historically selected to keep the phenotype constant in spite of small random developmental i rregularities potentially

PAGE 15

2 inducing slight differences among homologous parts wi thin individuals. ( I have italicized the aspects of this definition that I think are left undefined. ) Fluctuating asymmetry is defined and meas ured as the average right minus left difference in size or shape of morphological characters in a population and has been generally accepted as an indicator of devel opmental instability because both sides of a bilaterally symmetric organism have been de veloped by the same genetic program in the same environment (Mller and Swaddle 1997). Fluctuating asymmetry is measured as left-right side differences in the size or shape of paired b ilaterally symmetric biological structures of organisms. In a character trait that demonstrates fluctu ating asymmetry, it is assumed that the distribution of signed leftright differences is near ze ro and that there is no selection for asymmetry (Palmer and Str obek 1986, 2003). Other types of asymmetry do exist and are thought to indi cate selection against symmet ry. Directional asymmetry denotes a bias towards left or right side dness that causes the population mean to move away from zero. In antisymmetr y, left or right side biases occur equally at the individual level creating a population that is bimodal or platykurtic. While much attention has been directed toward the possibl e genetic basis of fluctuating asymmetry (reviewed by Leamy and Klingenberg 2005, Woolf and Markow 2003), response of fluctuating asymmetry to stress (reviewed by Hoffman and Woods 2003) and correlation of fluctuating asymmetr y with mate choice (Mller and Swaddle 1997); little scientific effort ha s been directed toward investig ating its basis or origin at levels of organization lower than the individu al. A few theoretical explanations for the basis of fluctuating asymmetry have been developed (reviewed by Klingenberg 2003). Only two of these offer a causal explanati on for the increased levels of fluctuating

PAGE 16

3 asymmetry that are sometimes observed under periods of environmental stress. While not mutually exclusive, these two explanations differ at the organizational level at which they are thought to act. Mller and Pomiankowski (1993) fi rst suggested that strong natural or sexual selection can remove regulatory steps controlling the symmetric development of certain traits (e.g., morphology us ed in sexual display). They suggest that with respect to sexually selected traits (a nd assuming that they are somehow costly to produce), individuals may vary in their ability to buffer ag ainst environmental stress in proportion to the size of their own energetic rese rves. These reserves are often indicative of the genetic quality of indi viduals. Therefore, high geneti c quality is expected to be associated with low fluctuating asymmetry. Emlen et al. (1993) present another and more proximate explanation of the basis of fluctuating asymmetry. They do not invoke sexual selection but hypothesize that fluctu ating asymmetry is due largely to the nonlinear dynamics of signaling a nd supply that may occur during growth. Here fluctuating asymmetry is thought to result from th e scaling up of compounding temporal asymmetries in signaling between cells duri ng growth. In their model, hypothetical levels of signaling compounds (morphogens) and or growth precursors used in the construction of cells vary randomly over time. When growth suffers less interruption, in other words, when it occurs faster and under le ss stress, there is al so less complexity (and fractal dimension) in the dynamics of si gnaling and supply. This is should reduce fluctuating asymmetry. Only a few other models have since b een proposed. A model by Graham et al. (1993) suggests that fluctuati ng asymmetry in the individual can also be the net result of compounding time lags and chaotic behavior be tween hormonally controlled growth rates

PAGE 17

4 on both sides of an axis of bilateral symmet ry. More recently Klingenberg and Nijhout (1999) present a model of morphogen diffusi on and threshold response that includes genetic control of each component. They de monstrate that fluctuating asymmetry can result from genetically modulated expression of variation that is en tirely non-genetic in origin. In other words, even without specific genes for fluctuating asymmetry, interaction between genetic and non-genetic sources of variation (G x E) can cause fluctuating asymmetry (Klingenberg 2003). While all these theories for the basis of fluctuating asymmetry have proven useful in making some predictions about fl uctuating asymmetry in relation to sexual selection and growth rate/trait size, none are grounded in any known molecular mechanisms. The search for any single molecular mechanism that stabilizes the developmental process has proven elusive. A recent candidate was the heat shock protein, Hsp90, which normally target conformati onally plastic proteins that act as signal tranducers (i.e., molecular sw itches) in many developmental pathways (Rutherford and Lundquist 1998). Because Hsp90 recognizes protei n folding, it can also be diverted to misfolded proteins that are denatured during environmental stress. Therefore, Hsp90 can potentially link the developmental process to the environment and these authors suggest it may also capacitate the evolution of novel morphology during times of stress by revealing genetic variation previously hidden to selection in non-stre ssful environments. However, additional research by Milton et al. (2003) shows that while Hsp90 does buffer against a wide range of morphologic changes and does mask the effect of much hidden genetic variation in Drosophila it does not appear to affect average levels of fluctuating asymmetry through any single Hsp90 depe ndent pathway or process.

PAGE 18

5 It is said that wisdom begins with the naming of things. The case of fluctuating asymmetry reminds us that the act of giving na mes to things in science can, in fact, lend a false impression that we have achieved a tr ue understanding of that which has been named. While fluctuating asymmetry has had several specific descriptive definitions, we do not really know how to define it at a f undamental level because we do not understand exactly how development is destabilized under certain conditions of both gene and environment. All we can say for now is that fluctuating asymmetry is a mysterious form of morphological variation. Mary Jane West-Eberhard ( 2003) has dubbed it the dark side of variation because it may represent th at noisy fraction of the physical-biological interface that is still free of selection, and not under direct control of the gene. Any study of natural variation would do well to begin by simply observing its shape or its distribution in full. In nature, statistical distributions come mostly in two flavors: those generated by large systems of independent additive components and those generated by large systems of intera cting multiplicative components (Vicsek 2001, Sornette 2003). When large systems are composed of independent subunits, random processes result in the normal distribution, th e cornerstone of classical statistics. The normal distribution is both unique and extrem e in its rapidly decaying tails, its very strong central tendency and its sufficient de scription by just two parameters, the mean and variance, making it, in this sense, th e most parsimonious description of random variation. However, the normal distribution does not describe all kinds of stochastic or random behavior commonly observed in the na tural sciences. When subunits comprising large systems interact, random processes are best described by models that underlie statistical physics (sometimes called Levy st atistics) (Bardou et al 2003. Sornette 2003).

PAGE 19

6 Fat-tailed distributions are often the result of propagation of error in the presence of strong interaction. These type s of distributions include th e power function, Pareto, Zipf and the double Pareto or log-Laplace distri butions, all characterized by a power law and an independence of scale. Collective effect s in complex interacting systems are also often characterized by these power laws (Wilson 1979, Stanley 1995). Examples include higher order phase transitions, self-organized criticality and percolation. During second order phase transition at the critical temperature between physical phases, external perturbation of the network of microscopic interactions between molecules results in system reorganization at a macroscopic level far above that of interacting molecules (e.g., the change from water to ice). This results in collective imitation that propagates among neighboring molecules over long distances. Exactly at these critical temperatures, imitation between neighbors can be observed at all scales creating regions of similar behavior that occur at all sizes (Wilson 1979). Thus, a self-similar power-law manifests itself in the interacting systems susceptibility to perturbation and re sults, in this case, from the multiplicity of intera ction paths in the interacti on network (Stanley 1995). As the distance between two objects in a netw ork increases, the nu mber of potential interaction pathways increas es exponentially and the corr elation between such paths decreases exponentially. The constant cont inuous degree of change represented by the power law is the result of a combined eff ect between both an exponentially increasing and decreasing rate of change. This highlig hts the fact that power laws can be easily manifested from combinations of exponen tial functions which are very common to patterns of change in many natural popul ations (both living and non-living).

PAGE 20

7 Therefore, it is important to note that power laws in statistica l distributions do not have to always indicate strong interacti on in a system and that many other simple mechanisms can create them (Sornette 2003), especially where the behavior of natural populations are concerned. For example, an apparently common method by which power laws are generated in nature is when st ochastic proportional (geometric) growth is observed randomly in time (Reed 2001). Power law size distributions in particle size, human population size, and economic factors are all potentially explained by this process (Reed and Jorgensen 2004). Here we also have exponential increa se in size opposing an exponential decrease in the pr obability of observation or te rmination that results in a gently decaying power law. This process can also explain why power law scaling can occur through the mixing of certain distribu tions where locally exponentially increasing and decreasing distributions overlap. For example, superimposing lognormal distributions results in a lognormal distri bution with power law tails (Montroll and Shlesinger 1982,1983). Given that exponential relati onships are so common in th e natural world, we should assume that in observing any large natural population outside of an experiment, there is probably some potential for a power-law scaling effect to occur. Therefore some degree of non-normal behavior may be likely to be observed, often in the underlying distributions tail. If we assume an u nderlying normal distribution, and sample it accordingly, we are likely to under-sample th is tail. And so we rarely ever present ourselves with enough data to challenge our assumption of normality, and we risk missing the chance to observe a potentially important aspect of natural variation.

PAGE 21

8 Until now, the basis of fluctuating asymmetry has been addressed only with very abstract models of hypothetical cell signaling, or at the leve l of selection working on the organism with potential mech anism remaining hypothetical. In this dissertation, the underlying common theme is that fluctuating asymmetry must first and foremost be envisioned as a stochastic process occurri ng during tissue growth, or in other words, occurring within an exponen tially expanding populat ion of cells. This expansion process can be modeled by stochastic proportional (geo metric) growth that is terminated or observed randomly over time. As will be explained in subsequent chapters, this generative process can naturally lead to variation that distributes according to a lognormal distribution with power laws in both tails (Reed 2001). In chapter 2, I examine the distribution of fluc tuating asymmetry in the wings of three species of insects (cotton aphid, Aphis gossipyi Glover honeybee, Apis mellifera and long-legged flies, Chrysosoma crinitus ) and test various candida te models that might de scribe the statistical distribution of fluctuating asym metry. I then address whether, given the best candidate model, fluctuating asymmetry studies have b een appropriately sampled. I suggest that much of the current controversy over fluctuat ing asymmetry may be due to the fact that past studies have been under-sample d. In chapter 3, I extend and test a phenomenological model for fluctuating asymme try that is introduced in chapter 2. I present the model and then examine some of its unique predictions concerning the effects of inbreeding on the shape of the dist ribution of fluctuating asymmetry in Drosophila I present evidence that the gene tic structure of a population can have a profound effect on the scaling and shape of the observed distribut ion of fluctuating asym metry. In chapter 4, I characterize the pattern of developmental noise (fluctuatin g asymmetry in the absence

PAGE 22

9 of genetic variation) in two monoclona l populations of cotton aphid cultured under graded environmental temperatures. I inves tigate how developmental noise is altered by this simple change in the environment. I present evidence that the environmental response of size-based FA is di rectly related to developmental time. Lastly, I conclude with a review of my major findings in the context of the introduc tion I have presented here.

PAGE 23

10 CHAPTER 2 ARE FLUCTUATING ASYMMETRY STUDIES ADEQUATELY SAMPLED? IMPLICATIONS OF A NEW MO DEL FOR SIZE DISTRIBUTION Introduction Developmental Stability: Definition, Measurement, and Current Debate Developmental stability is maintained by an unknown set of mechanisms that buffer the phenotype against small random pert urbations during development (Debat and David 2001). Fluctuating asymmetry (FA) the most commonly used assay of developmental instability, is defined either as the average deviation of multiple traits within a single individual (Van Valen 1962) or the deviation of a single trait within a population (Palmer and Strobek 1986, 2003; Parsons 1992) from perfect bilateral symmetry. Ultimately, an individuals developm ental stability is the collective result of random noise, environmental influences, and th e exact genetic archite cture underlying the developmental processes in that indivi dual (Klingenberg 2003; Palmer and Strobek 1986). Extending this to a populat ion, developmental stability is the result of individual variation within each of these three components. Currently, there is conflict in the literature regarding th e effect of both environment and genes on the developmenta l stability of populations. Th e development of bilateral symmetry appears to be destabilized to vari ous degrees by both e nvironmental stressors (review in Mller and Swaddle 1997) and certain genetic arch itectures (usually created by inbreeding: Graham 1992; Lerner 1954; Mather 1953; Messier and Mitton 1996; review by Mitton and Grant 1984). While th e influence of inbreeding on FA is not

PAGE 24

11 consistent (Radwan 2003; Carchi ni et al. 2001; Fowler and Whitlock 1994; Leary et al. 1983,1984; Lens et al. 2002; Perfectti and Camach i 1999; Rao et al. 2002; Vollestad et al. 1999), it has led biologists to us e terminology such as genetic stress or developmental stress when describing inbred populations (Clarke et al. 198 6 and 1993; Koehn and Bayne 1989; Palmer and Strobek 1986). While genetic and environmental stressors have been shown to contribute to developmental instability and FA, the full pict ure is still unclear (B jorksten et al. 2000; Lens et al. 2002). While FA has been proposed as a universal indicator of stress within individual organisms (Parsons 1992), its util ity as a general indicator of environmental stress has been contentious (Bjorksten et al. 2000; Ditchkoff et al. 2001; McCoy and Harris 2003; Merila and Bjorklund 1995; M ller 1990; Rasmuson 2002; Thornhill and Mller 1998; Watson and Thornhill 1994; Wh itlock 1998). Despite many studies, no clear general relationship betw een environmental stress and FA has been demonstrated or replicated through experimentation across different taxa (Bjorksten et al. 2000; Lens et al. 2002). Furthermore, the effects of stress on FA appear to be not only species-specific but also trait-specific and stress-sp ecific (Bjorksten et al. 2000). Several meta-analyses have attempted to unify individual studies on the relation of sexual sele ction, heterozygosity, and trait specificity to FA (Polak et al. 2003; Thornhill and Mller 1998; Vollestad et al. 1999); while some weak general effects have been found, their biolog ical significance is still unresolved. Taken together, the ambiguity of the resu lts from FA studies suggests unresolved problems regarding the definition and/or m easurement of FA. The distribution and overall variability of FA are sometimes disc ussed with regards to repeatability in FA

PAGE 25

12 studies (Whitlock 1998; Palmer and Strobeck 2003), but is seldom a primary target of investigation. Until we can quantify FA mo re reliably and understand its statistical properties, the potential for misinterpret ation of FA is likel y to persist. The Distribution of Fluctuating Asymmetry Although it is always risky to infer unde rlying processes from observed patterns, careful examination of the distribution of FA in large samples may help distinguish between possible scenarios driving FA. For instance, a good fit to a single statistical distribution may imply that the same process ope rates to create FA in all individuals in a population. In contrast, a good f it to a discrete mixture of several different density functions might suggest that highly asymme tric individuals suffe r from fundamentally different developmental processes than thei r more symmetric counterparts. Thin-tailed distributions (e.g., normal or exponential) ma y indicate relative independence in the accumulation of small random developmental e rrors, whereas heavy-tailed distributions may implicate non-independent cascades in the propagation of such error. Despite much interest in the relationship between envir onmental stress and levels of FA, the basic patterns of its distributio n in populations remain largely unexplored. One common distributional attr ibute of FA, leptokurtosis, has been discussed in the literature. Leptokurtosis denotes a distri bution that has many small and many extreme values, relative to the normal distribution. Two primary causes of this kind of departure from the normal distribution are the mixing of di stributions and/or sca ling effects in data. For example, the Laplace or double exponential distribution is leptokur tic (but not heavytailed) and can be represented as a continuous mixture of normal distributions (Kotz et al. 2001; Kozubowski and Podgorski 2001). Just as log scaling in the normal distribution results in the lognormal distribution, log scali ng in the Laplace leads to log-Laplace (also

PAGE 26

13 called double Pareto) distributions (Kozubow ski and Podgorski 2002; Reed 2001), which are both leptokurtic and heavytailed (see Figure 2-1). Several explanations for leptokurtosis in the distribution of FA have been proposed. Both i ndividual differences in developmental stability within a popul ation (Gangestad an d Thornhill 1999) and differences in FA between subpopulations (Houl e 2000) have been suggested to lead to continuous or discrete mixtur es of normal distributions with different developmental variances, which in turn would cause leptoku rtosis (e.g., a Laplace distribution) in the overall distribution of FA. Mixtures of nonnormal distributions may also cause either leptokurtosis or platykurtosis (more intermed iate values than the normal distribution: Palmer and Strobek 2003). A potential exampl e of this is illustrated by Hardersen and Framptons (2003) demonstration that a pos itive relationship between mortality and asymmetry can cause leptokurtosi s. Alternatively, Graham et al. (2003) have argued that developmental error should behave multiplicatively in actively growing tissues, creating a lognormal size distribution in most traits ra ther than the normal distribution that is usually assumed. They argue that this ultimately results in leptokurtosis (but not fat tails) and size dependent expression of FA. Because s imple growth models are often geometric, we should not be surprised if distributions of size-based FA followed the lognormal distribution (see Limpert et al 2001 for a review of lognormal distributions in sciences). Not well recognized within biology is the fact that close interaction of many components can result in powe r-law scaling (distributi onal tails that decrease proportional to x-a rather than to some exponential function of x such as exp(-ax) [exponential] or exp(-ax2) [normal] and hence to heavytailed distributions (Sornette 2003). Powerlaw scaling is often associated with the tail of the lognormal distribution,

PAGE 27

14 especially when log standard deviation is large ( Mitzenmacher unpublished ms.; Montroll and Shlesinger 1982, 1983; Roma n and Porto 2001; Ro meo et al. 2003). NORMAL LOGNORMAL LAPLACELOG-LAPLACE M I X T U R E M I X T U R ENormalLaplace Double Pareto lognormal (DPLN) 1 0 1 1 0 1 012345 y 202 5 additive/linear geometric/logarithmic Figure 2-1. Schematic representation of ma thematical relationships between candidate models for the distribution of fluctu ating asymmetry. Mixtures here are continuous. KEY: Solid line scale of variable (x ln(x)) Dashed line random walk observed at consta nt stopping rate (i.e., negative exponentially distributed stop times) Note: ra ndom walk on log scale exhibits geometric Brownian motion Dotted line convolution of two distributions (one of each type) Block arrow a continuous mixture of distributi ons with stochastic (exponentially distributed) variance (Note Log-Laplace is also called doubl e Pareto by Reed 2001, Reed and Jorgensen 2004)

PAGE 28

15 Because FA in a population may reflect fundamental developmental differences between different classes or groups of individuals, for exam ple stressed and non-stressed, or between different subpopul ations as suggested by Houl e (2000), we might expect discrete mixtures of different distributions to best describe FA. For instance, extreme individuals falling in a heavy upper tail may be those who have exceeded some developmental threshold. Major disruption of development, resulting in high FA, may also reveal the scaling that exists in the underlying gene regulatory netw ork (Albert and Barabasi 2002; Clipsham et al. 2002; Olvai an d Barabasi 2002). Alternatively, if FA is produced by a single process, but to various degrees in diffe rent individuals, then one might expect a continuous mixture model to best describe the di stribution of FA. The possibility of non-normal distribution of FA opens the door to several potential sampling problems. For instance, if the lognormal shape, or multiplicative variance parameter, is large, then broad distribution e ffects may slow the convergence of the sample mean to the population mean as sample sizes are increased (R omeo et al. 2003). An additional, thornier, problem is caused by power-law scaling in the tails of distributions. Many lognormally distributed datasets exhibit pow er-law scaling (or amplification) in th e tail region (Montroll and Shlesi nger 1982, 1983; Romeo et al. 2003; Sornette 2003), sometimes called ParetoLevy ta ils or just Levy tails. As sample sizes grow infinitely large, power-law and Pareto distributions may approach infinite mean (if the scaling exponent is less than three) and infinite variance (if the scaling exponent is less than two), and therefore will not obey the Law of Large Numbers (that sample means approach the population mean as sample sizes increase). Increased sampling actually increases the likelihood of samp ling a larger value in the tail of a Pareto distribution

PAGE 29

16 (Bardou et al. 2003; Quandt 1966), creating more uncertainty in estimates of the mean as sample size increases. The presence of pow er-law tails can slow overall convergence considerably even in distribut ions that are otherwise lognor mal with low variance (which may not look very different from well-beha ved lognormal distributions unless a large amount of data is accumulated). This discussion points to two effects that need to be assessed broad distribution effects, controlled by the shape (lognormal variance) parameter of the body of the FA distribution, and power-law tails, controlled by the scaling exponents of the tails of the FA distribution. To assess these effects, I apply a new statistical model, the double Pareto lognormal (DPLN) distribution (Reed and Jorg ensen 2004). The DPLN distribution is a lognormal dist ribution with power-law behavior in both tails (for values near zero and large positive values). Simila r to the log-Laplace di stributions, the DPLN distribution can be represente d as a continuous mixture of lognormal distributions with different variances. It can also be derived from a ge ometric Brownian motion (a multiplicative random walk) that is stopped or killed at a constant rate (i.e., the distribution of stop times is exponentially distributed: Reed and Jorgensen 2004; Sornette 2003). The parameters of the DPLN di stribution include a lognormal mean ( ) and variance ( 2) parameter which control the loca tion and spread of the body of the distribution, and power-law sca ling exponents for the left ( ) and right ( ) tails. Special cases of the DPLN include the right Pareto lognormal (RPLN) dist ribution, with a powerlaw tail on the right but not the left side ( ); the left Pareto lognormal (LPLN) distribution, with a power-law tail only near zero ( ); and the lognormal distribution, with no power-law tails ( ). For comparison, I also fit normal and half-normal

PAGE 30

17 distributions as well as the asymmetric La place distribution to the data on FA. See Figure 2-1 for schematic representation of re lationships between these candidate models for FA. In the following study, I directly test the f it of different distri butions to large FA datasets from three species of insects. I include a lab cultured monoclonal line of cotton aphid, Aphis gossipyii Glover, in an attempt to isolate the distribution of developmental noise for the first time. I also analyze da ta from a semi-wild population of domestic honeybee, Apis mellifera taken from a single inseminated single queen colony, and from a large sample of unrelated w ild-trapped Long-legged flies ( Chrysosoma crinitus : Dolichopididae). I address two primary groups of goals in this study. First, I investigate what distributions fit FA data best and how the parameters of th ese distributions vary across species, rearing conditions, and levels of genetic relatedness I also address whether outliers (individuals with visi ble developmental errors) appear to result from discrete or continuous processes. Secondly, I determin e how accurate the estimates of population mean FA are at various sample sizes to de termine whether past studies of FA been adequately sampled to accurately estimate mean or average FA in populations. In addition to these two primary goals, I also compare the best-fitting distributions and level of sampling error for three of the most co mmon methods of measuring FA: a univariate and a multivariate size-based metric of asymmetry, and a multivariate shape-based method. Methods Wings were collected and removed from three populations of insects and dry mounted on microscope slides. These populat ions included a m onoclonal population of

PAGE 31

18 1022 Cotton Aphid ( Aphis gossipyii Glover) started from a singl e individual collected from citrus in Lake Alfred, Florida, 1001 honeybees Apis mellifera maintained at University of Florida and 889 long-legged flies ( Chrysosoma crinitus :Dolichopodidae). All species identifications were made through the State of Florida Department of Plant Industry in Gainesville and voucher specimens remain available in their collections. Aphid cultures were maintained on potted plants in reach-in environmental chambers at 15C with constant 14/1 0 hour LD cycle generated by 4 20 watt fluorescent Grolux brand bulbs. Aphid cultures were cult ured on approximately 10 day old cotton seedlings ( Gossipium ) and allowed to propagate until crowded. Crowding stimulated alate formation (winged forms) in later generations which were collected every twenty days with a fine camel hair br ush wetted in ethanol. New plants were added every ten days and alates were allowed to move freely from plant to plant starting new clones until they were collected. The temperature at which the colony was maintained created a low temperature dark morph Co tton Aphid which still propagated on host plants quickly but was larger than high temperature light morphs that form at temperatures greater than 17C. Dark mo rphs colonize stems on cotton whereas light morphs colonized the undersides of leaves. The bees were collected in June 2004 by Dr. Glenn Hall at the University of Floridas Bee Lab from a single inseminated single queen colony. They were presumably all foragers and haplodiploid sist ers collected as they exited th e hive into a collection bag. The bag was frozen for three hours and then the bees were placed in 85% ethanol. Long-legged flies were trapped from a wild population using 14 yellow plastic water pan traps in southwest Gainesville, Florida, during May 2003 and May-June 2004.

PAGE 32

19 A very small amount of dishwashing deterg ent was added to the water to eliminate surface tension and enhance trapping. Traps were checked every three hours during daylight and set up fresh ever y day of trapping. All insect specimens were dried in 85% ethanol, and then pairs of wings were dissected (in ethanol) and air-dried to the gl ass slides while the ethanol evaporated. Permount was used to attach cover slips. This technique prevente d wings from floating up during mounting, which might slightly di stort the landmark configuration. Dry mounts were digitally photographed. All landmarks were identified as wing vein intersections on the digital images (six landmarks on each wing for aphids, eight for honeybees and Dolichopodid flies). See Appendi x A for landmark locations on wings for each species. Wing vein intersections were digitized three times each on all specimens using TPSDIG version 1.31 (Rohlf, 1999). All measures of FA were taken as the average FA value of the three replicate measurements for each specimen. Specimens damaged at or near any landmarks were disc arded. Fluctuating asymmetry was measured in three ways on all specimens. First, a common univariat e metric of absolute unsigned asymmetry was taken for two different landmarks: FA = abs(R L) where R and L are the Euclidean distances between the same two landmarks on either wing. In aphids, landmarks 1-2 and 2-3 were used; in bees, landmarks 1-4 and 2-6 were used; and in long-legged flies, landmarks 3-6 and 4-5 were used. Two multivariate geometric morphometrics using landmark-based methods were performed using all landmarks shown in Appendix A. A multivariate size-based FA (FA 1 in Palmer a nd Strobek 2003) was calculated as absolute value of (R L) where R and L are the centr oid sizes of each wing (i.e., the average of

PAGE 33

20 the distances of each landmark to their combin ed center of mass or cen troid location). In addition, a multivariate shape-based measure of FA known as the Procrustes distance was calculated as the square root of the sum of all squared Eucl idean distances between each left and right landmark after two-dimensional Procrustes fitting of the data (Bookstein 1991; Klingenberg and McIntyre 1998; FA 18 in Palmer and Strobeck 2003; Smith et al. 1997). Procrustes fitting is a three step process including a normalization for centroid size followed by superimposition of two sets of landmarks (right and left) and rotation until all distances between each landmark set is minimized. Centroid size calculation, Euclidean distance calculation and Procru stes fitting were performed using yvind Hammers Paleontological Statistics program PAST version 0.98 (Hammer 2002). For assessing measurement error (ME) of FA (or more specifically, the digitizing error), we conducted a Procrustes ANOVA (in Microsoft Excel) on all pairs of wing images resampled three times each for every species (Klingenberg and McIn tyre 1998). Percent measurement error was computed as (ME/average FA) x 100 where 3 / ) 3 1 3 2 2 1 ( FA FA FA FA FA FA ME All subsequent statistical analyses were performed using SPSS Base 8.0 st atistical software (SPSS Inc.). The fits of all measures of FA to eight distributional models (normal, half-normal, lognormal, asymmetric Laplace, double Pareto lognormal (DPLN), two limiting forms of DPLN, the right Pareto lognormal (RPLN) a nd the left Pareto lognormal (LPLN) and a discrete mixture of lognormal and Pareto) were compared by calculating negative log likelihoods and Likelihood Ratio Test (LRT ) if models were nested and Akaike Information Criteria (AIC) if not nested (Burnham and Anderson 1998; Hilborn and Mangel 1998). Both of these approaches penalize more complex models (those with

PAGE 34

21 more parameters) when selecting the best-fit distributional model for a given dataset. (The Likelihood Ratio Test doe s not technically apply when the nesting parameter is at the boundary of its allowed region, e.g., when for the DPLN, but Pinheiro and Bates (2000) suggest that the LRT is conser vative, favoring simpler models, under these conditions.) Best fitting parameters were obtained by maximizing the log-likelihood function for each model (Appendix B). Th e maximization was performed using the conjugate gradient method within unconstrai ned solve blocks in the program MathCad by MathSoft Engineering and Education Inc ( 2001), and was also confirmed using NelderMead simplex algorithm or quasi-Newt on methods in R version 2.0.1 (2003), a programming environment for data analysis and graphics. Phenodeviants were defined as individuals demonstrating missing wing veins, extra wing veins or partial wing veins on either one or both wings. All phenodeviants in honeybees involved absence of the vein at land mark 6 (LM 6). Phenodeviants in aphids were more variable but mostly involved abse nce of wing vein intersections at LM 2 or LM 3. Procrustes distances were estima ted for phenodeviants by omitting the missing landmarks (caused by the phenodeviance) and cont rolling for the effect of this removal on the sums of squares. I added an average of the remaining sums of squares in place of the missing sums of squares so that the calcu lated Procrustes distance is comparable to normal specimens (i.e., six landmarks). In almost all phenodeviants, this involved omission of only one set of landmark valu es. The frequency of phenodeviants was examined across the range of the FA distribut ion (i.e., Procrustes distance), and mean values of the FA for phenodeviants were co mpared to normal individuals in order to

PAGE 35

22 assess whether phenodeviants tended to show higher than normal levels of FA in characters that were not affected by the missing, partial, or extra wing veins. The best fitting parameters of the best fitting models were used to build a distributional model under which repeated sa mpling was simulated at various sample sizes. Average error in estimation of the m ean FA was calculated as a coefficient of variation ( 100 ) / ( x s) for 1,000 randomly generated datasets. Lastly, comparison were made of the estimation errors given the best fi tting distributions of FA to a dist ribution of sample sizes from 229 FA studies published in three recent meta-analyses (Polak et al. 2003; Thornhill and Mller 1998; Vollestad et al. 1999). Results In the distributions of shape-based FA in monoclonal cotton aphids (n = 1022), domesticated honeybees (n = 1001), and wild trapped long-legged flies (n = 889), AIC and LRTs always favored DPLN or RPLN m odels by a large margin. All size-based FA distributions favored DPLN or LPLN by a larg e margin (Table 2-1). All variants of discrete mixture models we tried had very poor results (data not shown). Figures 2-2 through 2-4 demonstrate best fitting models for multivar iate shape FA (DPLN and lognormal), multivariate centroid size FA (D PLN and half-normal), and univariate size FA (DPLN and half-normal) for all three spec ies. FA was often visually noticeable in aphids, where the mean shape FA (Procrus tes distance) was thre e times higher (0.062 0.00050) than in bees (0.023 0.00026) or flies (0.019 0.00028). I note that distribution of size FA in aphids and bees fit half-normal distribut ion in the upper tails fairly well but fit relatively poorly among i ndividuals with low FA. Long-legged flies exhibit poor fit to half-normal in both tails.

PAGE 36

23 Table 2-1. Maximized log-likelihood (MLL), number of model parameters (P) and Akaike Information Criterion differences ( AIC) for distributional models tested. Winning models have AIC zero. Models with goodness-of-fit nearly equal to winners are underscored ( AIC <3.0). 4.0 < AIC <7.0 indicates some support for specified model. AIC >10.0 indicates no support (Burnham and Anderson 1998). Distances between la ndmarks for first univariate size FA, aphids LM 1-2, bees LM 1-4 and long-legged fly LM 3-6 and for second univariate FA aphids LM 2-3, bees LM 2-6 and long-le gged fly LM 4-5. Species/model Multivariate shape FA Multivariate size FA First univariate size FA Second univariate size FA Cotton Aphid P MLL AIC MLL AIC MLL AIC MLL AIC DPLN 4 255.117 0.000 542.906 1.236 545.725 2.815 614.123 2.333 RPLN 3 264.72 17.207 547.44 8.303 549.789 8.944 617.498 7.083 LPLN 3 256.723 1.213 543.288 0.000 545.317 0.000 613.957 0.000 LNORM 2 734.726 955.218 1261 1433 1265 1438 1413 1596 NORM 2 2241 3968 2847 4606 3352 5612 3534 5838 HNORM 1 3523 6530 2370 3650 2893 4691 3047 4862 LAPLACE 3 637.044 761.855 1638 2189 1883 2676 1950 2671 Honey Bee DPLN 4 132.908 7.369 536.278 2.744 454.337 0.000 515.93 6.329 RPLN 3 130.224 0.000 549.641 27.47 460.51 10.346 525.675 23.82 LPLN 3 136.361 12.275 535.906 0.000 455.483 0.292 513.765 0.000 LNORM 2 523.667 784.888 1262 1451 1081 1249 1211 1393 NORM 2 3371 6480 2180 3286 1709 2505 2302 3574 HNORM 1 5319 10370 1775 2475 2687 1771 1924 2816 LAPLACE 3 1148 2036 1387 1703 1162 1413 1445 1863 Long-legged Fly DPLN 4 146.603 37.971 452.728 0.000 453.958 0.000 412.007 0.000 RPLN 3 128.617 0.000 454.056 0.656 460.261 10.606 422.993 20.182 LPLN 3 149.585 41.935 455.818 4.18 456.237 2.559 412.902 0.212 LNORM 2 506.553 753.871 1055 1201 1064 1216 985.894 1144 NORM 2 2985 5710 1760 2610 1806 2701 1668 2508 HNORM 1 5147 10030 1084 1256 1252 1590 1099 1368 LAPLACE 3 1041 1825 940.241 973.02 1016 1121 997.28 1169 For multivariate shape analysis, right tail power-law exponents ( ) were very high (thousands), left-t ail exponents ( ) from 3.9-9.9, while the dispersion parameter ( ) was narrowly distributed from 0.310 to 0.356. Thus, shap e FA exhibited little scaling in tails (i.e., nearly lognormal). For size-based FA, dispersion wa s much larger (0.57-0.74), and power-law exponents more variable for univari ate and multivariate size-based FA (left

PAGE 37

24 tail ( ) and right tail ( ) were generally low, indicating moderate power-law scaling in both tails) as shown in Table 2-2. Table 2-2. Best fit parameters for models in Table 1. Parameters for univariate size FA are very similar to multivariate size FA and are not shown. Skew indexes for asymmetric Laplace are also not shown. Multivariate shape FA Multivariate size FA Cotton Aphid Location Dispersion/ Shape Right tail Left Tail Location Dispersion/ Shape Right tail Left Tail DPLN -2.62 0.356 1160 4.04 1.45 0.735 6.24 3.01 RPLN -3.01 0.415 7.03 1.07 0.800 4.58 LPLN -2.62 0.353 3.90 1.73 0.699 2.23 LNORM -2.87 0.434 1.28 0.824 NORM 0.062 0.027 4.92 3.94 HNORM 0.009 0.059 0.163 6.31 LAPLACE 0.049 0.058 0.657 2.19 Honey Bee DPLN -3.74 0.310 8380 9.80 1.32 0.565 10.4 1.57 RPLN -4.02 0.274 5.52 0.571 0.829 4.82 LPLN -3.71 0.305 7.75 1.52 0.510 1.35 LNORM -3.84 0.327 0.777 0.850 NORM 0.023 0.008 2.95 2.14 HNORM 0.007 0.018 0.078 3.64 LAPLACE 0.021 0.018 0.550 2.24 Long-legged Fly DPLN -3.93 0.346 4940 10.3 2 -0.067 0.683 3.09 5.14 RPLN -4.28 0.231 3.73 -0.302 0.698 2.75 LPLN -3.92 0.341 9.91 2 0.319 0.743 3.85 LNORM -4.02 0.351 0.060 0.783 NORM 0.019 0.008 2.00 2.24 HNORM 0.007 0.015 0.075 2.11 LAPLACE 0.017 0.017 0.346 1.06 Figure 2-5 shows the distributi on of FA sample sizes from 229 studies published in three recent meta-analyses (Polak et al. 2003; Thornhill and Mller 1998; Vollestad et al. 1999).

PAGE 38

25 00.020.040.060.080.10.120.140.160.180.20.220.24 0 5 10 15 20 25 30 35 00.050.10.150.20.25 0 20 40 60 expDPLN obsDATAA. Cotton Aphid 00.050.10.150.20.25 0 0.1 0.2 0.3 expLGN obsDATA 00.010.020.030.040.050.060.070.080.09 0 10 20 30 40 50 60 70 80 90 00.020.040.060.08 0 5 10 15 expDPLN obsDATAB. Domestic Honeybee 00.020.040.060.08 0 0.05 0.1 expLGN obsDATA 00.010.020.030.040.050.060.070.08 0 20 40 60 80 100 00.020.040.060.08 0 5 10 expDPLN obsDATAC. Long-legged Fly 00.020.040.060.08 0 0.02 0.04 0.06 expLGN obsDATADPLN DPLN DPLN Lognormal Lognormal Lognormal Figure 2-2. Distribution of multivaria te shape FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting lognormal (dashed line, lower inset), and double Pareto lognormal (solid line, uppe r inset) are indicated.

PAGE 39

26 02468101214161820222426 0 5 10 15 20 25 30 35 40 45 50 55 60 051015 10 expDPLN obsDATAA. Cotton Aphid 051015 0 10 expHNORM obsDATA 012345678910111213 0 5 10 15 20 25 30 35 40 45 50 0510 20 40 expDPLN obsDATAB. Domestic Honeybee 0510 0 5 10 expHNORM obsDATA 012345678 0 10 20 30 40 50 60 70 80 90 100 110 120 0246810 5 10 expDPLN obsDATAC. Long-legged Fly 0246810 0 5 10 expHNORM obsDATADPLN DPLN DPLN Half-normal Half-normal Half-normal Figure 2-3. Distribution of multivariate centroid size FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting half-normal (d ashed line, lower inset) and double Pareto lognormal distribution (solid line, upper inse t) are indicated.

PAGE 40

27 0510152025303540 0 5 10 15 20 25 30 35 40 45 50 55 60 010203040 20 40 expDPLN obsDATAA. Cotton Aphid (LM 1-2) 010203040 0 20 40 expHNORM obsDATA 051015202530354045 0 5 10 15 20 25 30 35 40 45 50 55 60 010203040 20 40 expDPLN obsDATAA. Cotton Aphid (LM 2-3) 010203040 0 20 40 expHNORM obsDATA 0123456789 0 5 10 15 20 25 30 35 40 45 50 55 60 02468 5 expDPLN obsDATAB. Domestic Honeybee (LM 1-4) 02468 0 5 expHNORM obsDATA 0 1 2 3 4 5 6 7 8 9 10 11 12 0 5 10 15 20 25 30 35 40 45 02468 5 10 expDPLN obsDATAB. Domestic Honeybee (LM 2-6) 0510 0 5 10 expHNORM obsDATA 012345678910 0 10 20 30 40 50 60 70 80 90 100 110 120 n 0246810 5 10 expDPLN obsDATAC. Long-legged Fly (LM 3-6) 0246810 0 5 10 expHNORM obsDATA 012345678910 0 5 10 15 20 25 30 35 40 45 50 55 60 0246810 5 10 expDPLN obsDATAC. Long-legged Fly (LM 4-5) 0246810 0 5 10 expHNORM obsDATADPLN Half-normal Half-normal Half-normal Half-normal Half-normal Half-normal DPLN DPLN DPLN DPLN DPLN Figure 2-4. Distribution of univariate unsigned size FA of A) cotton aphid ( Aphis gossipyii ) B) domestic honeybee ( Apis mellifera ) and C) long-legged fly ( Chrysosoma crinitus ). Best fitting half-normal (dashed line, lower inset) and double Pareto lognormal distribution (solid line, upper inset) are indicated.

PAGE 41

28 sample or treatment size4 50 4 00 3 50 3 00 2 5 0 200 1 50 1 00 50 5frequency40 30 20 10 0 Figure 2-5. Distribution of sample sizes (n) from 229 fluctuating asymmetry studies reported in three recent meta-analyses (Vollestad et al. 1999, Thornhill and Mller 1998 and Polak et al. 2003). Only five studies had sample sizes greater than 500 (not shown). Nearly 70% of the 229 FA studies have sa mple or treatment sizes less than 100. Figure 2-6 demonstrates the hypot hetical error levels (coeffi cients of variations) in estimated mean FA at various sample sizes. Approximate best fit parameters were used to estimate the coefficient of variation (C V) under the DPLN distribution (for shape FA, = -3.7, = 0.35, = 1000, = 9; for size FA, = 1.2, = 0.70, = 4.0, = 4.0). For the same set of landmarks, multivariate shape FA measures lead to the least amount of error in estimating mean FA under DPLN at any sample size. Both univariate and multivariate size FA perform more poorly in terms of both convergence and overall percentage error. I found that while phenodeviants occurred in almost all regions of the distribution range of FA, the percentage of phenodeviant individuals increased dramatically with

PAGE 42

29 increasing FA (Figure 2-7; aphids, r = 0.625 p = 0.013 ; bees, r = 0.843 p = 0.001). I also found that individuals with phenodeviant wings (both aphids and bees) showed significantly higher levels of FA across those wing landmarks unaffected by the phenodeviant traits (p < 0.002 in both aphids and bees). Only a single case of phenodeviance was sampled in long-legged flies. 0 50 100 150 200 250 300 350 400 450 500 0 5 10 15 20 25 30 35 40 45 Legend shape FA size FA SAMPLE SIZE % ERROR IN ESTIMATION OF THE MEAN (CV) Figure 2-6. Relationship between sample si ze and % error for estimates of mean FA drawn from best fitting size (dashed line ) and shape (solid line) distributions using 1000 draws per sample size. All runs use typical winning double Pareto lognormal parameters (shape FA = -3.7, = 0.2, = 1000, = 9; for size FA = 1.2, = 0.7, = 4.0, = 4.0). Percent measurement error for shape FA was 1.41% in aphids, 1.63% in bees, and 2.42 % in flies while for size FA, it was ~4.5% in aphids, ~5% in bees, and ~6.5 % in flies. In a Procrustes ANOVA (Klingenberg and McIntyre 1998) the mean squares for the interaction term of the ANOVA (MSInteraction) was highly significant p<0.001 in all three species indicating that FA variation was significantly larger than variation in measurement error (ME). The distribution of signed ME was normal and exhibited

PAGE 43

30 moderate platykurtosis for all types of FA in all species examined. Measurement error was very weakly correlated to FA in all samples (0.01 < r2 <0.07). 125 1070 N =APHID WINGSPHENODEVIANT NORMAL WINGMean +1 SE Procrustes Distance.064 .062 .060 .058 .056 .054 .052 65 1035 N =BEE WINGSPHENODEVIANT NORMAL WINGMean +1 SE Procrustes Distance.030 .029 .028 .027 .026 .025 .024 .023 C D 0 t o 0 0 1 0 0 1 t o 0 0 2 0 0 2 t o 0 0 3 0 0 3 t o 0 0 4 0 0 4 t o 0 0 5 0 0 5 t o 0 0 6 0 0 6 t o 0 0 7 0 0 7 t o 0 0 8 0 0 8 t o 0 0 9 0 0 9 t o 0 1 0 1 t o 0 1 1 0 1 1 t o 0 1 2 0 1 2 t o 0 1 3 0 1 3 t o 0 1 4 0 1 4 t o 0 1 5 0 1 5 t o 0 1 6 0 1 6 t o 0 1 7 0 1 7 t o 0 1 8 0 1 8 t o 0 1 9 0 1 9 to 0 2 0 50 100 150 200 250 300 Legend phenodeviants normal Procrustes Distance FrequencyA 0 t o 0 0 0 4 5 0 0 0 4 5 t o 0 0 0 9 0 0 0 9 t o 0 0 1 3 5 0 0 1 3 5 t o 0 0 1 8 0 0 1 8 to 0 0 2 2 5 0 0 2 2 5 to 0 0 2 7 0 0 2 7 to 0 0 3 1 5 0 0 3 1 5 t o 0 0 3 6 0 0 3 6 to 0 0 4 0 5 0 0 4 0 5 t o 0 0 4 5 0 0 4 5 t o 0 0 4 9 5 0 0 4 9 5 t o 0 0 5 4 0 0 5 4 t o 0 0 5 8 5 0 0 5 8 5 t o 0 0 6 3 0 0 6 3 t o 0 0 6 7 5 0 0 6 7 5 t o 0 0 7 2 0 0 7 2 t o 0 0 7 6 5 0 0 7 6 5 t o 0 0 8 1 0 0 8 1 t o 0 0 8 5 5 0 0 8 5 5 t o 0 0 9 0 50 100 150 200 250 300 Legend phenodeviants normal Procrustes DistanceB Figure 2-7. The proportion and percentage (inset) of individuals with visible developmental errors on wings (phenode viants) are shown for cotton aphids (A) and honeybees (B) in relation to distribution of shape FA (Procrustes distance). Average FA for both nor mal and phenodeviant aphids (C) and honeybees (D) are also given. Discussion The Distribution of FA The data demonstrate a comm on pattern of distribution in the FA in wing size and shape of three different species whose populations existed under very different environmental conditions (lab culture, free-li ving domesticated, and wild) and genetic structure (monoclonal, haplodipl oid sisters, and unrelated). The similarities across the very different species and r earing conditions used in this study suggest that the

PAGE 44

31 distribution of size and shape FA ma y have universal parameters (e.g., 0.35 for shape FA and 0.7 for size based FA). The data co nfirm that although size FA sometimes exhibits reasonable fit to half-normal in the upper tail, a nd shape FA is reasonably well fit by lognormal distributions, large datasets of FA in both size and shape are always best described by a double Pareto lognormal distribut ion (DPLN) or one of its limiting forms, LPLN and RPLN. Multivariate shape FA demo nstrates narrow distri bution with a large right tail, including the top fe w percent of the most extremely asymmetric individuals, that is best fit by DPLN or RPLN. Both univariate and multivariate size FA exhibit a considerably broader distribution with moderate leptokurtosis that is best fit by DPLN or LPLN. The data suggest that the DPLN di stribution and its limiting forms are generally the most appropriate models for the dist ribution of FA regardless of method of measurement. Evidence that distribution of FA clos ely follows DPLN, a continuous mixture model, and appearance of phenodeviance across n early the entire range of data suggested that developmental errors may be cause d by a similar process across the entire distribution of FA in a populat ion. In other words, variati on in FA may have a single cause in most of the data. Although phenode viance is significantly related to increased levels of FA and is more prevalent in the ri ght tail region of the shape FA distribution, it does not appear associated exclusively with th e right tail, as a thre shold model for high FA might predict. The very poor fits to all variations on the discrete mixture model also suggest a lack of distinct pro cesses creating extreme FA in th e three datasets. However, I caution that use of maximum like lihood methods to fit data to discrete mixtures is often technically challenging. I found no block e ffects (e.g., no differences in FA levels

PAGE 45

32 between long-legged fly samples collected on different weeks or aphid samples collected from different pots or growth chambers), so it appears there is no obvious sample heterogeneity that could result in a discre te mixture. With the usual caveats about inferring process from pattern, I do not find obvious thresholds in the distribution of asymmetries at the population level that would suggest threshold eff ects at a genetic or molecular level. Lastly, based on the appearance of only one phenodeviant among our wild trapped long-legged fly population (as oppos ed to many in the bee and lab reared aphid populations), I speculate that mortality related to phenodeviance (and perhaps high FA) in wing morphology may be relaxed in lab cu lture and domestication. But this could be confounded by other differences between the three datasets including genetic redundancy and species differences. Furthe r comparisons among populations of single species under different conditions woul d be needed to test this idea. Sample Size and the Estimation of Mean FA In random sampling under DPLN, we found th at broad distributi on effects due to the shape parameter were minimal in thei r effect of slowing convergence to the population mean in multivariate shape-based FA ( 0.35 for all three datasets). However, these effects are considerable for univariate and multivariate size-based FA (where 0.70). The effects of scaling in th e tails of the distribution, which cause divergence from the mean, appear to have little effect in th e right tail of the distribution of shape-based FA. However, larger effect s in the tails of the size FA create more individuals with very low a nd very high asymmetry than expected under the assumption of normality. The point estimates of the sc aling exponents for size FA are close to the range where very extreme values may be sampled under the di stribution tails (if < 3 or <3), greatly affecting confidence in the estim ate of mean FA. With a sample size of 50

PAGE 46

33 and the best fitting DPLN parameters typica l for asymmetry in multivariate shape, we found that the coefficient of variation for m ean FA is about 5%, whereas size-based mean FA fluctuates about 13% from sample to sample. At a sample size of 100, these coefficients of variation ar e 3.2% and 7.5% respectively. Un less experimental treatment effects in most FA studies are larger than this, which is unlikely in studies using sizebased measures of FA of more canalized trai ts, statistical power and repeatability will be low. Given the sample size range of most previous studies (n = 30-100) and their tendency to favor size-based measurement me thods, our results suggest that many past FA studies may be under-sampled. Furthermore, it is also likely that given the small sample sizes in many FA studies, particularly involving vertebrate s where n < 50, the tail regions of natural FA distributions are often severely under sampled and sometimes truncated by the removal of outliers. These factors may artificially cause non-normal distributions to appear normal, also potential ly resulting in inaccurate estimation of mean FA. The Basis of Fluctuating Asymmetry The surprisingly good fit of FA distribu tions to the DPLN model in our study suggests that the physical basis of FA ma y be created by the combination of random effects in geometrically expa nding populations of cells on either side of the axis of symmetry (i.e., geometric Browni an motion). Studies in the Drosophila wing indicate that cell lines generally compete to fill a prescribed space during development with more rapidly dividing lines out-competing weaker ones (Day and Lawrence 2000). Because regulation of the growth of such cell popul ations involves either nutrients and/or signaling substances that stop the cell cycl e when exhausted, it is likely that the distribution of numbers of cells present at the completion of growth follows an

PAGE 47

34 exponential distribution. Reed and Jorgense n (2004) demonstrate that when a population of repeated geometric random walks is ki lled at such a constant rate, the DPLN distribution is the natural result. There are many other examples of growth processes in econometrics and physics where random pr oportional change combined with random stopping/observation create size distributions of the kind described here (Reed 2001; Kozubowski and Podgorski 2002). In the future, when applying this model to instability during biological growth, it would be very interesting to inves tigate how genetic and environmental stress might affect the parameters of this model. If scaling effects are found to vary with stress, then leptokurtosis may potentially be a be tter candidate signal of developmental instability than increased mean FA. Conclusion Although size-based FA distributions can sometimes appear to fit normal distributions reasonably well as previous definitions of FA suppose, I demonstrate that three large empirical datasets all support a ne w statistical model for the distribution of FA (the double Pareto lognormal distribution), which potentially exhibits power-law scaling in the tail regions and leadi ng to uncertain estimation of tr ue population mean at sample sizes reported by most FA studies. The assumption of normality fails every time candidate models are compared on large data sets. Failure of this assumption in many datasets may have been a major source of disc ontinuity in results of past FA studies. Future work should attempt to collec t larger sample/treatment sizes (n 500) unless the magnitude of treatment effects on FA (and thus the statistical power of comparisons) is very large. Our results demonstrate that multivariate shape-based methods (Klingenberg and McIntyre 1998) result in more repeatab le estimates of mean FA than either multivariate or univariate size-based me thods. I would also recommend that

PAGE 48

35 methodology be re-examined even in large sample studies of FA. For example, because Drosophila are usually reared in ma ny replicates of small tube s with less than 50 larvae per tube, many large studies may still be co mpromised by individual sizes of replicate samples. I also suggest that authors of pa st meta-analyses and reviews of FA literature reassess their conclusions after excluding studi es in which under-sampling is found to be problematic. Careful attention to distributional and sampling i ssues in FA studies has the potential both to mitigate problems with repeatability and possibly to suggest some of the underlying mechanisms driving variation in FA among individuals, populations, and species.

PAGE 49

36 CHAPTER 3 INBREEDING REDUCES POWER-LAW SC ALING IN THE DISTRIBUTION OF FLUCTUATING ASYMMETRY: AN EX PLANATION OF THE BASIS OF DEVELOPMENTAL INSTABILITY Introduction Fluctuating asymmetry (FA) is the average difference in size or shape of paired or bilaterally symmetric morphol ogical trait sampled across a population. The study of FA, thought to be a measure of developmental instability, has a controversial history. Fluctuating asymmetry is hypothesized by some to universally represent a populations response to environmental and/or genetic stress (Parsons 1992, Clarke 1993, Graham 1992). It is also generally accepted that FA ma y be co-opted as an indicator or even a signal of individual genetic buffering cap acity to environmen tal stress (Moller 1990, Moller and Pomiankowski 1993). R ecent literature reviews reveal that these conclusions are perhaps premature and anal yses of individual studies often demonstrate conflicting results (Lens 2002, Bjorksten 2000). Babbitt et al (2006) demonstrate that this conflict may be caused by under-sampling due to a fals e assumption that FA always exhibits a normal distribution. Also, FA may be responding to experimental treatment in a complex and as yet unpredictable fashion. Until th e basis of FA is better understood, general interpretation of FA studies remains difficult. What is the Basis of FA? Earlier studies of sexual selection and FA conclude that FA is ultimately a result of strong selection against th e regulation of the developmen t of a particular morphology (e.g., morphology used in sexual display). T hus in some instances FA may increase and

PAGE 50

37 therefore become a better signal and/or more honest indicator of good genes. Sexually selected traits tend to have increased FA (Moller and Swaddle 1997), however the exact mechanism by which FA increases remains unexplained (i.e., a black box). More recently, theoretical attempts have been made to explain how FA may be generated but none have been explicitly tested. Models for the phenomenological basis of FA fall into two general categories: reaction-diffusion models and diffusion-threshold models (reviewed in Klingenberg 2003). The former class of models involves the chaotic and nonlinear dynamics in the regulation or negative feedback among neighboring cells (Emlen et al. 1993) or adjacent bilateral morphology (Graham et al. 1993). The latter class of models combines morphogen diffusion and a threshold response, the parameters of which are controlled by hypothetical genes and a small amount of random developmental noise (Klingenberg and Nijhout 1 999). The result of this latter class of model is that different genotypes respond di fferently to the same amount of noise, providing an explanation for gene tic variation in FA response to the same environments. Traditionally, models for the basis of FA assume that variation in FA arises from independent stochastic events that influence the regulati on of growth through negative feedback rather than processes that may fuel or promote growth. None of these models explicitly or mathematically address the effect of stochastic behavior in cell cycling on the exponential growth curve. More recently Graham et al. (2003) makes a compelling argument that fluctuating asymmetry often results from multiplicative errors during growth. This is consistent with one part icular detail about ho w cells behave during growth. For several decades there has been evidence that during development, cells actually compete to fill prescribed space unt il limiting nutrients or growth signals are

PAGE 51

38 depleted (Diaz and Moreno 2005, Day and La wrence 2000). Cell populations effectively double each generation until signaled or forced to stop. It has also been observed that in Drosophila wing disc development, that synchro ny in cell cycling does not occur across large tissue fields but rather extends only to an average cl uster of 4-8 neighboring cells regardless of the size and stage of development of the imaginal disc (Milan et al. 1995). The assumption of previous models, particul arly the reaction-diffusion type, that cell populations are collectively controlling their cell cyc ling rates across a whole developmental compartment is probably unreali stic. Regulatory control of fluctuating asymmetry almost certainly does occur, but probably at a higher level involving multiple developmental compartments where competing cells are prevented from crossing boundaries. However, given that indivi dual cells are behaving more or less autonomously during growth within a single de velopmental compartment, I suggest that variation in fluctuating asy mmetry can be easily generated at this level by a process related to stochastic exponen tial expansion and its termination in addition to regulatory interactions that probably act at higher levels in the organism. In this paper, I explore and test simple model predictions regarding th e generation of fluctuating asymmetry though multiplicative error without regulatory feedback. Exponential Growth and Non-No rmal Distribution of FA In previous work, Babbitt et al. (2006) de monstrate that the distribution of unsigned FA best fits a lognormal distribution with scaled or power-law tails (double Pareto lognormal distribution or DPLN). This distribution can be generated by random proportional (exponential) growth (or geometri c Brownian motion) that is stopped or observed randomly according to a negative expo nential probability (Reed and Jorgensen 2004). I suggest that this source of power-law scaling in the ta ils of the FA distribution is

PAGE 52

39 also the cause of leptokurtosis th at is often observed empirically in the distribution of FA. Kurtosis is the value of the standardized four th central moment. Like the other moments, (location, scale, and skewness), kurtosis is best viewed as a concept that can be formalized in multiple ways (Mosteller and Tukey 1977). Leptokurtosis is best visualized as the location and scale-free movement of probability mass from the shoulders of a symmetric distribution towa rds both its center and tail (Balanda and MacGillivray 1988). Both the Pareto and power-function distributions have shapes characterized by the power-law and a larg e tail and therefore exhibit a lack of characteristic scale. Both because kurtosis is strongly affected by tail behavior, and because leptokurtosis involves a diminishing of characteristic scal e in the shape of a distribution, the concepts of scaling and kurtosis in real data can be, but are not necessarily always, inte r-related. Both in the past and very recently, leptokur tosis in the distribution of FA has been attributed to a mixture of normal FA distri butions caused by a mixi ng of individuals, all with different genetically-based developmen tal buffering capacity, or in other words, different propensity for expressing FA (Gangestad and Thornhill 1999, Palmer and Strobek 2003, Van Dongen et al. 2005). Alt hough not noted by these authors, continuous mixtures of normal distributions generate the Laplace distribution (Kotz et al. 2001, Kozubowski and Podgorski 2001) and can be distinguished from other potential candidate distributions by using appropriate model selection techniques, such as the Akaike Information Criterion technique (Burnham and Anderson 1998). Graham et al. (2003) rejects the typical e xplanation of leptokurtosis th rough the mixing of normal distributions by noting that differences in random lognormal variable can generate

PAGE 53

40 leptokurtosis. Babbitt et al. (2006) also reject the explanation that leptokurtosis in the distribution of FA is caused by a mixture of normal distributions because the double Pareto-lognormal distribution, not the Laplace di stribution, always appears the better fit to large samples of FA. Therefore, leptokur tosis, often observed in the distribution of FA, may not be due to a mixing process, but in stead may be an artifact of scaling in the distribution tail, which pr ovides evidence of geometric Brownian motion during exponential expansion of populations of cells. Testing a Model for the Basis of FA I propose that the proximate basis for variat ion of FA in a population of organisms is due to the random termination of stochas tic geometric growth. The combination of opposing stochastic exponential functions re sults in the slow power-law decay that describes the shape of the distributions tail. In this paper, I present a model for FA and through simulation, test the prediction that ge netic variation in the ability to precisely terminate growth will lead to increased kur tosis and decreased s caling exponent in the upper tail of the distribution. Then, I asse ss the validity of this model by direct comparison to the distribution of FA within large samples of wild and inbred populations of Drosophila Under the assumption that a less heterozygous population will have less variance in the termination of growth, I w ould predict that inbr eeding should act to reduce power-law scaling effects in the distribution of FA in a population. Inbreeding should also reduce the tail we ight (kurtosis) and mean FA assuming inbred individuals have lower variance in the times at which th ey terminate growth. I also assume that inbreeding within specific lines does not act to amplify FA due to inbreeding depression. It has been demonstrated that Drosophila melanogaster do not increase mean FA in response to inbreeding (Fowler and Whitlock 1994) and it is suggested that large

PAGE 54

41 panmictic populations typical of Drosophila melanogaster may not harbor as many hidden deleterious recessive mutations as other species (Houle 1989), making them resistant to much of the typi cal genetic stress of inbreeding. Therefore, the absence of specific gene effects duri ng inbreeding suggests that Drosophila may be a good model for investigating the validity of our model as an explana tion of the natural variation occurring in population level FA. Methods Model Development Simulation of geometric Brownian motion Ordinary Brownian motion is most easily simulated by summing independent Gaussian distributed random numbers or white noise (iX ). See Figure 3-1. (3.1) n i iX X W1) ( which simulated in discrete steps is (3.2) 1 1 t t tW N N where N = cell population size, t = time st ep and W = a random Gaussian variable. Exponential or geometric Brownian motion, a ra ndom walk on a natural log scale, can be similarly simulated. Geometric Brownian motion is described by the stochastic differential equation (3.3) ) ( ) ( ) ( ) ( t dW t Y dt t Y t dY or also as (3.4) ) ( ) ( ) ( t dW dt t Y t dY

PAGE 55

42 ORDINARY BROWNIAN MOTION-600 -400 -200 0 200 400 600 800 1000 124477093116139162185208231254277300323346369392415438461484 WHITE NOISE-150 -100 -50 0 50 100 150 121416181101121141161181201221241261281301321341361381401421441461481 timeW N Figure 3-1. Ordinary Brownian motion (l ower panel) in N simulated by summing independent uniform random variables (W) (upper panel). where W(t) is a Brownian motion (or Weiner process) and and are constants that represent drift and volatility respectively. Equation 3.3 has a lognormal analytic solution (3.5) ) ( ) 2 (2) 0 ( ) (t W te Y t Y A simulation of geometric Brownian mo tion in discrete form follows as (3.6) 1 1 1 t t t tW N N N where N = cell population size, t = time step and W = a random Gaussian variable. See Figure 3-2. Equation 3.6 is iden tical to the equation for multip licative error in Graham et al. (2003). I modify equa tion 3.6 slightly by letting W range uniformly from 0.0 to 1.0 with W= 0.5 and adding the drift constant C that allows for stocha stic upward drift (at C > 0.5) or downward drift (at C < 0.5).

PAGE 56

43 (3.7) 1 1 1 1 1) ( t t t t t tN W C W N CN N At C = 0.5, eqns. 3.6 and 3.7 behave identical ly. Geometric Brownian motion with upward drift is shown in Figure 3-3. 0 0.00005 0.0001 0.00015 0.0002 0.00025 0.0003 0.00035 0.0004 0.00045 13569103137171205239273307341375409443477 time step 0.0000001 0.000001 0.00001 0.0001 0.001 0.01 0.1 1 13671106141176211246281316351386421456 time step Log N Ntime Figure 3-2. Geometric Brownian motion in N and log N simulated by multiplying independent uniform random variables. This was generated using Equation 1.5 with C = 0.54.

PAGE 57

44 0 500000000 1000000000 1500000000 2000000000 2500000000 3000000000 3500000000 4000000000 4500000000 13773109145181217253289325361397433469 time step 0.00001 0.001 0.1 10 1000 100000 1000000 0 1E+09 1E+11 13467100133166199232265298331364397430463 time step Log N Ntime Figure 3-3. Geometric Brownian motion in N and log N with upward drift. This was generated using Equation 1.5 with C = 0.60. Simulation of fluctuating asymmetry Using the MathCad 13 (Mathsoft En gineering and Education 2005), two independent geometric random walks were performed and stopped randomly at mean time t= 200 steps with some variable normal probabilit y. The random walks result in cell population size equal to tN(orL tN and R tN on left and right sides respectively). Fluctuating asymmetry (FA) was defined as the difference in size resulting from this random proportional growth on both sides of th e bodies axis of symmetry plus a small degree of random uniformly distributed noise or

PAGE 58

45 (3.8) rv N N FAR t L t where (rv) was uniformly distribute d with a range of .1(R t L tN N ). Using a MathCad-based simulation in VisSim LE, the generation of individual FA values was repeated until a sample size of 5000 was reached. The random noise (rv) has no effect on the shape of the FA distribution but fills empty bins (gaps) in distri bution tails. Because rv is small in comparison toR t L tN N its effect is similar to that of measurement error (which would be normally distributed rather th an uniform). Schematic representation of the simulation process for Reed and Jo rgensen (2004) model and simulation of fluctuating asymmetry are shown in Figure 4 an d 5. The simulation of the distribution of fluctuating asymmetry was compared at norma l standard deviation of termination of growth (t) ranging from = 0.5, 0.8, 1.2,3 and 7 with a drift constant of C = 0.7. Inbreeding Experiment In May 2004 in Gainesville, Florida, 320 free-living Drosophila simulans were collected in banana baited traps and put into a large glass jar and cultured on instant rice meal and brewers yeast. After two genera tional cycles 1000 indivi duals were collected in alcohol. This was repeated again in June 2005 with 200 wild trapped flies. Lines of inbred flies were created from the May 2004 wild population through eight generations of full sib crosses removing an estimated 75% of the preexisting heterozygosity (after Crow and Kimura 1970) Initially, ten in dividual pairs were isolated from the stock culture and mated in pint mason jars with media and capped with coffee filters. In each generation, and in each line, and to ensure that inbred lines were not accidentally lost t hough an inviable pairing, four pairs of F1 sibs from each cross were then mated in pint jars. Offs pring from one of these four crosses were

PAGE 59

46 randomly selected to set up the next genera tion. Of the original ten lines, only four remained viable after eight generations of full sib crossing. These remaining lines were allowed to increase to 1000+ individuals in 1 quart mason jars and then were collected for analysis in 85% ethanol. This generall y took about 4 generations (8 weeks) of open breeding. One completely isogenic (balancer) line of Drosophila melanogaster was obtained from Dr. Marta Wayne, University of Florida and also propagated and collected. GROWTH STOP TIME NEGATIVE EXPONENTIAL VARIATION IN GROWTH STOP TIME 0100200300400500 0 50 100 150 200 0 1 e lDOUBLE PARETO LOGNORMAL DISTRIBUTION REPEAT SAMPLING 0100 0 2 0 10 20 30 40 50 60 70 80 90 100 1E-006 1E-005 0.0001 0.001 0.01 RANDOM WALKS WITH GEOMETRIC BROWNIAN MOTION Figure 3-4. Model representation of Reed and Jorgensens (2004) physical size distribution model. Variable nega tive exponentially distributed stopping Times of random proportional growth (GBM with C = 0.5) create double Pareto lognormal distribution of size.

PAGE 60

47 Figure 3-5. A model representa tion of developmental instab ility. Normally variable stopping times of random proportional gr owth (GBM with C > 0.5) create double Pareto lognormal distribution of size. All flies were cultured on rice meal a nd yeast at 29C with 12:12 LD cycle in environmentally controlled chambers at th e Department of Entomology and Nematology at the University of Florida. Wings were collected and removed from 1000 flies from each of the two samples of wild population a nd four samples of inbred lines and dry mounted on microscope slides. Specimens were dried in 85% ethanol, and then pairs of Stochastic Geometric Growth1 10000 1E+08 1E+12 1E+16 1E+20 1E+24 1E+28 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748log Cell Number 050100150 0 2 4 GROWTH STOP TIME NORMAL VARIATION IN GROWTH STOP TIME RANDOM WALKS WITH GEOMETRIC BROWNIAN MOTIONSIZE DISTRIBUTION REPEAT SAMPLING 01002003004005 00 0 50 100 150 200

PAGE 61

48 wings were dissected (i n ethanol) and air-dried to the gla ss slides. Permount was used to attach cover slips. This technique prev ented wings from floating up during mounting, which might slightly distort the landmark configuration. Dry mounts were digitally photographed. All landmarks were identified as wing vein intersections on the digital images (eight landmarks on each wing). See Appendix A for landmark locations. Morphometric analyses Wing vein intersections were digitized on all specimens using TPSDIG version 1.31 (Rohlf, 1999). Specimens damaged at or near any landmarks were discarded. Fluctuating asymmetry was measured in two ways on all specimens using landmarkbased multivariate geometric morphometrics. A multivariate size-based FA (FA 1 in Palmer and Strobek 2003) was calculated as ab solute value of (R L) or just R-L in signed FA distributions where R and L are th e centroid sizes of each wing (i.e., the sum of the distances of each landmark to their comb ined center of mass or centroid location). In addition, a multivariate shape-based measure of FA known as the Procrustes distance was calculated as the square root of the su m of all squared Euclidean distances between each left and right landmark after two-dime nsional Procrustes fitting of the data (Bookstein 1991; Klingenberg and McIntyre 19 98; FA 18 in Palmer and Strobeck 2003; Smith et al. 1997). Centroid size calcul ation, Euclidean distance calculation and Procrustes fitting were performed using yvind Hammers Paleontological Statistics program PAST version 0.98 (Hammer 2002). A sub-sample of 50 individuals from the fourth inbred line (pp4B3) was digitized five times to estimate meas urement error. In these cases, measures of FA were taken as the average FA value of the five replicate measurements for each specimen. Percent measurement error was also computed as

PAGE 62

49 (ME/average FA) x 100 where) 5 4 3 2 1 ( FA FA FA FA FA sd ME (as per Palmer and Strobek 2003). For assessing whether measur ement error (ME) interfered significantly with FA, a Procrustes ANOVA (in Micr osoft Excel) was performed on the five replications of the 50 specimen sub-samp le (Klingenberg and McIntyre 1998). Any subsequent statistical analys es were performed using SPSS Ba se 8.0 statistical software (SPSS Inc.). Model selection and inference The fits of unsigned size FA to three distributional models (half-normal, lognormal, and double Pareto lognormal (DPL N)) were compared in the Drosophila lines, by calculating negative log like lihoods and Akaike Informa tion Criteria (AIC) (Burnham and Anderson 1998; Hilborn and Mangel 1998) This method penalizes more complex models (those with more parameters) when selecting the best-fit di stributional model for a given dataset. Best fitting parameters were obtained by maximizing the log-likelihood function for each model (Appendix B). Th e maximization was performed using the conjugate gradient method within unconstrai ned solve blocks in the program MathCAD by MathSoft Engineering and Education Inc (2001). Results Model Simulation The amount of variance in termination times related directly to levels of simulated FA (i.e., low variance in termination time (t) gives low FA and vice versa). I found that not only does amount of FA increase with in creased variance in (t), but so do both kurtosis and the scaling effect in the distri bution tails. In Figure 3-6, normal quantile plots of signed FA are shown fo r different standard deviations in the normal variation of the termination of growth of geometrically expanding cell populations. The degree of the

PAGE 63

50 S or sigmoidal shape in the plot indicates level of leptokurtos is. The leptokurtosis in the quantile plot is reduced greatly with a decrea se in the standard deviation of the normal variation in growth termination times. SIGNED DISTRIBUTION OF FLUCTUATING ASYMMETRY DETRENDED FIT TO NORMAL 933 0 8 0 kurtosis634 0 5 0 kurtosis149 1 2 1 kurtosis812 11 0 3 kurtosis266 27 0 7 kurtosis Figure 3-6. Simulated distri butions of cell population size and FA for different amounts of variation in the termination of gr owth (variance in normally distributed growth stop time). The fit of simulate d data to the normal distribution can determined by how closely the plotted points follow the horizontal line (a good fit is horizontal).

PAGE 64

51 Experimental Results Both mean unsigned size FA and shape FA decreased with inbreeding in all lines. I also observed that the kurtosis of signed size FA and the skewness of both unsigned size and shape FA follow an identical trend. The trend was strongest in kurtosis, which decreased rapidly with inbreeding, indicating th at, as predicted, changes in mean FA are influenced strongly by the shape and tail behavior of the distribution of FA (Table 3-1). Table 3-1. Distribution parameters and m odel fit for multivariate FA in two wild populations and four inbred lines of Drosophila simulans and one isogenic line of Drosophila melanogaster Model fits are AIC for unsigned centroid size FA (zero is best fit, lowest number is next best fit). Kurtosis (signed size FA) Mean (unsigned size FA) Mean (shape FA) Scaling exponent in upper tail AIC HNORM AIC LGN AIC DPLN Wild 2004 28.3 .12 5.73 .38 0.0223 .0003 1.49 90.6 324 0.000 Wild 2005 45.9 .15 4.68 .19 0.0237 .0003 2.87 66.2 261 0.000 Inbred line 1 (OR18D) 2.43 .11 3.27 .09 0.0183 .0002 5.09 0.000 104 15.5 Inbred line 2(OR18D3) 8.85 .15 4.24 .13 0.0212 .0002 3.10 11.50 223 0.000 Inbred line 3 (PP4B2) 2.90 .15 3.31 .09 0.0184 .0002 4.63 0.000 130 44.0 Inbred line 4 (PP4B3) 3.30 .15 3.72 .11 0.0213 .0003 5.78 0.000 228 79.7 Isogenic (mel75) 0.550 .11 3.90 .10 0.0201 .0002 9.99 0.000 84.9 101 While wild populations on the whole, de monstrated increased mean, skew and leptokurtosis of FA compared to inbred lines, significant differences between populations were found in all distributional parameters even between each of the inbred lines (ANOVA mean size FA for a ll lines; F = 27.49, p < 0.001; ANOVA mean size FA for inbred lines only; F = 14.38 p < 0.001; note shape FA also show same result). Overall,

PAGE 65

52 inbred lines demonstrated lowe r kurtosis, just slightly above that expected from a normal distribution (Table 1). They also had lower mean FA. No significant differences were found with respect to these results according to sexes of flies. In Figure 3-7, I show the distribution of FA and detrended normal qua ntile plots for the wild population, four inbred lines and one isogen ic line respectively. As in the simulated data, the degree of th e S shape in the plot indicated level of leptokurtosis. The S shape in quantile plot is reduced greatly with inbreeding and nearly disappears in the isogenic line. Model Selection and Inference The comparison of candidate distributional m odels of FA demonstrated normalization associated with inbreeding in three of the four inbred lines (Table 3-1). In the remaining inbred line, the half-normal candidate mode l was a close second to the double Pareto lognormal distribution. In the wild populati on samples, the distribution of unsigned sizebased FA was best described by the double Pareto lognormal distribution (DPLN), a lognormal distribution with scali ng in both tails. In the best fitting parameters of this distribution there was no observable trend in lognormal mean or variance across wild populations and inbred lines. The sc aling exponent of the lower tail ( in Reed and Jorgensen 2004) was close to one in all lines while the scaling exponent in the upper tail ( in Reed and Jorgensen 2004) increased with inbreeding (Table 3-1). Both samples of the wild population demonstrat ed the lowest scaling expone nt in the upper tail of the distribution. The low sca ling exponents here indicate di vergence in variance (2004 and 2005 where < 3) and in the mean (2004 only where < 2). The inbred lines all show higher scaling exponents that are consistent with converging mean and variance.

PAGE 66

53 21 0 1 3 8 6 6 -.6 -7.8 -1 5 0 -2 2 2 2 1.0 1 3 .8 6 6 .6 7 .8 -15 .0 -22 .2 9 4 2 1 .0 1 3 8 6 6 -.6 -7.8 -1 5 0 -2 2 2 4 2 1 0 13 8 6 6 .6 7 8 1 5 0 2 2. 2 25. 6 2 0. 6 15 6 10.6 5 6 6 -4.4 -9. 4 -14.4 1 9. 4 -24 4 4 21.0 13.8 6.6 .6 7 .8 -15.0 -22.2 23. 1 15. 6 8.1 6 -6.9 -14. 4 -21 .9 9 4 25 28 2004 kurtosis wild 85 45 2005 kurtosis wild 43 2 18 kurtosis D OR inbred 30 3 3 4 kurtosis B PP inbred 90 2 2 4 kurtosis B PP inbred 85 8 3 18 kurtosis D OR inbred 55 0 75 kurtosis mel isogenic Figure 3-7. Distribution of fluctuating asymmetry and detr ended fit to normal for two samples of wild population collected in Gainesville, FL in summers of 2004 and 2005 and four inbred lines of Drosophila simulans derived from eight generations of full-sib crossing of the wild population of 2004. Also included is one isogenic line of Drosophila melanogaster (mel75). All n = 1000. The fit of data to the normal distributi on can determined by how closely the plotted points follow the horizontal line (a good fit is horizontal).

PAGE 67

54 The isogenic Drosophila melanogaster line demonstrated the highest scaling exponent and the most normalized dist ribution of unsigned FA. Measurement error was 7.6% for shape-based FA and 13.0% for centroid sizebased FA. In a Procrustes ANOVA (Klingenbe rg and McIntyre 1998) the mean squares for the interaction term of the ANOVA (MSInteraction) was highly significant p<0.001 indicating that FA variation was significantly larger than variation due to measurement error (ME). Discussion Revealing the Genetic Component of FA While we should be cautious about inferring process from pattern, the very similar results of both the modeling and the inbreeding experiment in Drosophila seem to suggest the presence of a scal ing component in the distribu tion of fluctuating asymmetry that is caused by a random multiplicative grow th process as suggested previously by Graham et al. 2003. This parameter appears to change with the genetic redundancy of the population which is presumably increased by genetic drift and and reduction in heterozygosity during inbreeding. Specifica lly, the scaling exponent (s) of the upper tail ( ) of the unsigned FA distribution, or outer tails of the signed FA distribution, are increased with inbreeding, causing more rapid power-law decay in the shape of the tails. This effect also reduces kurtosi s and apparently normalizes th e distribution of FA in more inbred populations. This sugge sts that individual genetic di fferences in the capacity to control variance in the term ination of random proportional growth (i.e., geometric Brownian motion) may be responsible for determining the shape and kurtosis of the distribution of FA In other words, leptokurtosis (kur tosis >3) in signed FA distribution

PAGE 68

55 indicates genetic variability in the populat ion while normality (kurtosis = 3) indicates genetic redundancy. Because leptokurtosis is very often obse rved in the distribution of FA, genetic variability potentially underlie s a large proportion of the vari ability observed in the FA of a given population. Observed differences or changes in FA are therefore not only a response of development to environmental stre ss, but clearly also can reflect inherent differences in the genetic redundancy of populatio ns. The significantly different levels of mean FA among the inbred lines in this st udy, presumably caused by the random fixation of certain alleles, also suggest s that there is a st rong genetic component to the ability to buffer development against random noise. It is assumed that the differences observed in FA between wild trapped and i nbred populations in this study do not indicate an effect of inbreeding depression in the study for two reasons. First, the four inbred lines analyzed in this study were vigorous in culture so the fixation of random alleles was probably not deleterious. Second, and more important, i nbreeding reduced FA rather than increasing it as would have been expect ed under genetic stress. It is also important to no te that kurtosis is potentially a much stronger indicator of FA than the distribution mean. The lo w scaling exponents found in the non-normal distributions of FA in the wild populations of Drosophila simulans are capable of slowing and perhaps even stopping, the conve rgence on mean FA with increased sample size. Because kurtosis is a fourth order mo ment, estimating it accurately also requires larger sample sizes. However, if kurtosis can be demonstrated to re spond as strongly to environmental stress as it does here to inbreedi ng, its potential streng th as a signal of FA may allow new interpretation of past studies of FA without the collect ion of more data.

PAGE 69

56 This may help resolve some of the current deba te regarding FA as a universal indicator of environmental health and as a potential sexua l signal in good genes models of sexual selection. Limitations of the Model There are certain aspects of the model presented here that may be oversimplifications of the real developmental process. First of all, this model assumes developmental instability is generated left-r ight growing tissue fiel ds with no regulatory feedback or control other than when growth is stopped. It is very likely that left-right regulation is able to occur at higher levels of organization (e.g., across multiple developmental compartments) even though ther e is no evidence of regulated cell cycling rates beyond the distance of 6-8 cells on average within any given developmental compartment (Milan et al. 1995). Therefor e this model explains how fluctuating asymmetry is generated, not how it is regulat ed. Second, this m odel only considers cell proliferation as influencing size. It is known that both cell size and programmed cell death, or apoptosis, are also important in regulating body size (Raff 1992, Conlin and Raff 1999). Both of these may play a more pr ominent role in vertebrate development, than they do in insect wings, where apparently growth is terminated during its exponential phase. Nevertheless, this simple model seems to replicate very well, certain behavioral aspects of the di stribution of fluctuating asym metry in the insect wing. The Sources of Scaling The basic process generating power-law s caling effects illustrated here (Reed 2001) offers an alternative perspective to the of ten narrow explanations of power-laws caused by self-organized criticality in the interactions among syst em components (Gisiger 2000). The power law that results from self-organiz ed criticality is created from the multiplicity

PAGE 70

57 of interaction paths in the network. As th e distance between two in teracting objects is increased in a network or multidimensional la ttice, the number of potential interaction pathways increases exponentially, while the correlation between such paths decreases exponentially (Stanley 1995). These opposing exponential relationshi ps create the power law scaling observed in simulati ons of self-organized critical systems. While there is no such interaction in our model, there is a power-law generated by opposing exponential functions. The constant degree of change represented by the power law in both the statistical physics of critical systems and the mathematics of both Reed and Jorgensens process and the model given here is the resu lt of the combined battle between both the exponentially increasing and decreasing rates of change (Reed 2001). While the natural processes are quite different, the underlying mathematical be havior is very similar. Potential Application to Cancer Screening Just as changes in the shape of the di stribution of fluctuating asymmetry is normalized across a population of genetically redundant individuals genetic redundancy in a population of cells may also help mainta in normal cell size and appearance. The loss of genetic redundancy in a tissu e is a hallmark of cancer. The abnormal gene expression and consequent genetic instab ility that characterizes cancerous tissue often results in asymmetric morphology in cells, tissues a nd tissue borders. Baish and Jain (2000) review the many studies connect ing fractal (scale free) ge ometry to the morphology of cancer. Cancer cells also are typically pleo morphic or more variable in size and shape than normal cells and this pleomorphy is associ ated with intercellula r differences in the amount of genetic material (Ruddon 1995). Frig yesyi et al. (2003) have demonstrated a power-law distribution of chromosomal aberra tions in cancer. Currently the type of distributional shape of the pleomorphic variabil ity in cell size is not known, or at least not

PAGE 71

58 published. However, Mendes et al. (2001) demo nstrated that cluster size distributions of HN-5 (cancer) cell aggregates in culture fo llowed a power-law scaled distribution. Furthermore these authors also demonstrated that in MDCK (normal) cells and Hep-2 (cancer) cells, cluster size distributions tr ansitioned from short-tailed exponential distributions to long-tailed power-law di stributions over time. The transition is irreversible and is likely an adaptive re sponse to high density and long permanence in culture due to changes in either control of replication or perhaps cell signaling. Taken collectively, these studies may suggest that scaling at higher levels of biological organization observed in cancer is due to incr eased relative differences in length of cell cycling rates of highly pleomorphic cell populations that have relatively larger intercellular differences in amount of geneti c material. The stochastic growth model I have proposed as the basis for higher vari ability in population level developmental instability or FA may also pr ovide a possible explanation for higher variability in the cell sizes of cancerous tissue. If genetic redundancy in growing tissue has the same distributional effect as gene tic redundancy in populations of organisms and tends to normalize the observed statistical distribution, then one might predict that the genetic instability of cancerou s tissue would create a scaling effect that causes pleomorphy in cells and scaling in cell cluster aggregati ons. Statistical comparison of cell and cell cluster size distributions in normal and can cerous tissues may provide a useful and general screening technique for detecting when genetic redundancy is compromised by cancer in normal tissues. Conclusion Until now, the basis of fluctuating asymmetry has been addressed only with abstract models of hypothetical cell signaling, or elsewhere, at the level of selection

PAGE 72

59 working on the organism with potential mechanism remaining in the black box. However, fluctuating asymmetry must first a nd foremost be envisioned as a stochastic process occurring during tissue growth, or in other words, occurring in an exponentially expanding population of cells. As demonstr ated in the model presented here, this expansion process can be represented by stocha stic proportional (geometric) growth that is terminated or observed randomly over time. These results imply that the fluctuating asymmetry observed in populations is not only related to poten tial environmental stressors, but also to a large degree, the unde rlying genetic variabili ty in those molecular processes that control the termination of growth. Therefore, fluctuating asymmetry responses to stress may be hard to interpre t without controlling fo r genetic redundancy in the population. Both the simulation and experi mental results suggest that measures of distributional shape like kurto sis, scaling exponent and tail weight may actually be a strong signal of variability in the underlying process that causes developmental instability. Therefore the kurtosis paramete r of the fluctuating asymmetry distribution may provide more information about fluc tuating asymmetry response than does a populations average or mean fluctuating asymmetry. This may provide a novel method by which to resolve conflicts in previous under-sampled research without the collection of more data.

PAGE 73

60 CHAPTER 4 TEMPERATURE RESPONSE OF FLUCTUAT ING ASYMMETRY TO IN AN APHID CLONE: A PROPOSAL FOR DETE CTING SEXUAL SELECTION ON DEVELOPMENTAL INSTABILITY Introduction Developmental instability is a potentially maladaptive component of individual phenotypic variation with some unknown basi s in both gene and environment (Mller and Swaddle 1997, Fuller and Houle 2003). De velopmental instability is most often measured by the manifestation of fluctuating asymmetry (FA), the ri ght minus left side difference in size or shape in a single trait across the population (Palmer and Strobek 1986, 2003; Parsons 1992, Klingenberg and McIntyre 1998). Because FA is thought to indicate stress during development, the primar y interest in the study of FA has been its potential utility as an indi cator of good genes in mate choice (Mller 1990, Mller and Pomiankowski 1993) or its util ity as a general bioindicato r of environmental health (Parsons 1992). The Genetic Basis of FA For FA to become a sexually selected trait, it must be assumed that it has a significant genetic basis, and can ther efore evolve (Mller 1990, Mller and Pomiankowski 1993). However, researcher s who use FA as a bioindicator of environmental health often assume that FA is a phenotypic response that is mostly environmental in origin (Parsons 1992, Lens et al. 2002). Studies of the heritability of FA seem to indicate low heri tability for FA exists (Whitlock and Fowler 1997, Gangestad

PAGE 74

61 and Thornhill 1999). However, because FA is essentially a variance that is often measured with only two data points per indi vidual, FA may have a stronger but less easily detectable genetic ba sis (Whitlock 1996, Fuller and Houle 2003). Additive genetic variation in FA in most studi es has been found to be minimal, but several quantitative trait loci studies suggest signi ficant dominance and character specific epistatic influences on FA (Leamy 2003, Leamy and Klingenberg 2005). Babbitt (chapter 3) has demonstrated that a populations genetic variab ility affects the distributional shape of FA. So while studies investigating mean FA may be inconclusive, changes in the populations distributional shape seem to indicate potential genetic in fluence on FA. However, no studies have observed FA in a clonal orga nism for the express purpose of assessing developmental instability that is purely environmental (i.e., nongenetic) in its origin (i.e., developmental noise). The Environmental Basis of FA It has long been assumed that FA is the re sult of some level of genetically-based buffering of additive independent molecular noise during development. Because of the difference in scale between the size of mol ecules and growing cells it would be unlikely that molecular noise would comprise an im portant source of variation in functioning cells. However, Leamy and Klingenberg ( 2005) rationalize that mo lecular noise could only scale to the level of tissue when developmentally important molecules exist in very small quantity (e.g., DNA or protein) and th erefore FA may repr esent a stochastic component of gene expression. This is some what similar in spirit to the Emlen et al. (1993) explanation of FA th at invokes non-linear dynamics of signaling and supply that may also occur during growth. Here FA is thought to be the result of the scaling up of

PAGE 75

62 compounding temporal asymmetries in signali ng between cells during growth. In this model, hypothetical levels of signaling compounds (morphogens) and or growth precursors used in the construction of cells vary randomly over time. When growth suffers less interruption, thus when it occurs faster, there is also less complexity (or fractal dimension) in the dynami cs of signaling and supply. Graham et al. (1993) suggest that nonlinear dynamics of hormonal signaling across the whole body may also play a similarly important role in the manifestation of FA. However, while the levels of FA are certainly influenced by the regul ation of the growth process, both Graham et al. (2003) and Babbitt (chapter 3) also suggest that FA le vels reflect noise during cell cycling that is amplified by exponentially expanding populat ions of growing cells. Although the proximate basis of FA is not well understood, its ultimate evolutionary basis, while heavily debate d, is easier to unde rstand. Mller and Pomiankowski (1993) first suggested that str ong natural or sexual selection can remove regulatory steps controlling the symmetri c development of certain traits (e.g., morphology used in sexual display). They sugge st that with respect to these traits (and assuming that they are somehow costly to pr oduce), individuals may vary in their ability to buffer against environmental stress and de velopmental noise in relation to the size of their individual energetic reserves; which are in turn often indicative of individual genetic quality. Therefore high genetic quality is associated with low FA. Most existing proximate or growth mechan ical explanations of FA assume that rapid growth is less stressful in the sense that fewer interruptions of growth by various types of noise should result in lower FA (Emlen et al. 1993, Graham et al. 1993). However, ultimate evolutionary explanations of FA assume that rapid growth is

PAGE 76

63 potentially more stressful because it is ener getically costly and therefore rapid growth should increase the level of FA when energy supply is limiting (Mller and Pomiankowski 1993). This high FA is relative and so should be especially prominent in individuals of lower genetic qu ality who can least afford to pay this additional energy cost. This fundamental difference between the predictions of proximate (or mechanistic) level and ultimate evolutionary level effects of temperature on FA is shown in Figure 1. The theoretical difference in the correlation of temper ature and growth rate to FA in both the presence and absence of energetic limitation could be used to poten tially detect sexual selection on FA. However it first should be confirmed th at FA should decrease with more rapid growth in the absence of en ergetic limitation to growth and genetic differences between individuals in a population. This later objective is the primary goal of this study. Temperature and FA in and Aphid Clone At a very basic level, entropy or noise in physical and chemical systems has a direct relationship with the physical ener gy present in the system. This energy is measured by temperature. Because FA is specu lated to tap into biol ogical variation that is somewhat free of direct genetic control, it may therefore res pond to temperature in simple ways. First, increased temperature may increas e molecular entropy which may in turn increase developmental noise during deve lopment thereby increasing FA. Second, increased temperature may act as a cue to shorten development time (as in many aphids where higher temperature reduces both body size and development time), thereby reducing the total time in which developmental errors may occur. This should reduce

PAGE 77

64 FA. A third possibility is that a species spec ific optimal temperature exists. If so, FA should increase while approaching both the upper and lower thermal tolerance limits of organisms. PROXIMATE LEVEL EFFECT ULTIMATE LEVEL EFFECTTEMPERATURE DEVELOPMENTAL TIME GROWTH RATE FLUCTUATING ASYMMETRY +-+-NON-OPTIMAL TEMPERATURE DEVELOPMENTAL TIME GROWTH RATE FLUCTUATING ASYMMETRY --+-ENERGETIC RESERVES STRESS + + +-Figure 4-1. Predicted proximate and ultimat e level correlations of temperature and growth rate to fluctuating asymmetry are different. Ultimate level (evolutionary) effects assume energetic limitation of individuals in the system. Proximate level (growth mechanical) e ffects do not. Notice that temperature and fluctuating asymmetry are negatively correlated in the proximate model while in the ultimate model they are positively correlated. Only a few studies have di rectly investigated the relationship between FA and temperature. The results are conflicting. FA is either found to increase on both sides of an optimal temperature (Trotta et al. 2005, Zakarov and Shchepotkin 1995), to be highest at low temperature (Chapman and Goulson 2000), to simply increase with increasing temperature (Savage and Hogarth 1999, Mpho et al. 2002) or not to respond

PAGE 78

65 (Hogg et al. 2001). None of these studies investigate the relationship between developmental noise (FA in a clonal line) and temperature. The characterization of developmental noise in response to temperature was investigated in this study using the cotton aphid, Aphis gossipyii. These aphids reproduce parthenogenetically, are not energetically lim ited in their diet (because they excrete excess water and sugar as honeydew) and pro duce wings that are easily measured using multiple landmarks. They demonstrate larg e visible variation in body size, wing size and even wing FA. The visible levels of wing as ymmetry in cotton aphids reflect levels of FA that are about four times higher than that ob served in other insect wings (Babbitt et al. 2006, Babbitt in press). Because parthenoge netic aphids cannot purge deleterious mutations each generation and because Florida clones often never use sexual reproduction to produce over-wintering eggs, this remarkably high FA may be the result of Mullers ratchet. Cotton aphids are al so phenotypically plastic in response to temperature, producing smaller lighter morphs at high temperatures and larger dark morphs at low temperatures. This feat ure allows observation of two genetically homogeneous groups in which different gene expression patterns (causing the color morphs) exist. The central prediction is that of the proximate effect model: that in the absence of energetic limitation and genetic variation, temperature and environmentally induced FA (or developmental noise) should be negatively correlated in a more or less monotonic relationship. Methods In March 2003, a monoclonal population of Cotton Aphids ( Aphis gossipyii Glover) was obtained from Dr. J. P. Michaud in Lake Alfred Florida and was brought to

PAGE 79

66 the Department of Entomology and Nematology at the University of Florida. The culture was maintained on cotton seedlings ( Gossipium ) grown at different temperatures (12.5C, 15C, 17C, 19C, 22.5C and 25C with n = 677 total or about 100+ per treatment) under artificial grow lights ( 14L:10D cycle). Because of potential under-sampling caused by a non-normal distribution of FA (see Ba bbitt et al. 2006), a second monoclonal population collected from Gainesville, FL in June 2004 was reared similarly but in much larger numbers at 12.5C, 15C, 17C, 19C, and 25C (n = 1677 or about 300+ per treatment). Development time for individual apterous co tton aphids (Lake Alfred clone) were determined on excised cotton leaf discs usi ng the method Kersting et al. (1999). Twenty randomly selected females were placed upon twenty leaf discs (5 cm diameter) per temperature treatment. Discs were set upon we t cotton wool in petri dishes and any first instar nymphs (usually 3-5) appearing in 24 hours were then left on the discs. Development time was taken as the average num ber of days taken to reach adult stage and compared across temperatures. Presence of shed exoskeleton was used to determine instar stages. Cotton was wetted daily and leaf discs were changed every 5 days. Humidity was maintained at 50%. In each temperature treatment, single clona l populations were allowed to increase on plants until crowded in order to stimulat e alate (winged individuals) production. Temperature treatments above 17C produced li ght colored morphs that were smaller and tended to feed on the undersides of leaves of cotton seedlings. Temperature treatments below 17C produced larger dark morphs that tended to feed on the stems of cotton seedlings. Alatae were collected using small brushes dipped in alcoho l and stored in 80%

PAGE 80

67 ethanol. Wings were dissected using fine in sect mounting pins and dry mounted as pairs on microscope slides. Species identificati on was by Dr. Susan Halb ert at the State of Florida Department of Plan t Industry in Gainesville. Specimens were dried in 85% ethanol, and then pairs of wings were dissected (in ethanol) and air-dried to the gl ass slides while ethanol evaporated. Permount was used to attach cover slips. This technique prev ented wings from floating up during mounting, which might slightly distort the landmark configuration. Dry mounts were digitally photographed. Six landmarks were identified as the two wing vein intersections and four termination points for the third subcostal. See Appendix A for landmark locations. Wing vein intersections were di gitized using TPSDIG version 1.31 (Rohlf, 1999). Specimens damaged at or near any landmarks were discarded. Fluctuating asymmetry was calculated using a multivaria te geometric morphometric landmark-based method. All landmarks are shown in Appendi x A. FA (FA 1 in Palmer and Strobek 2003) was calculated as absolute value of (R L) where R and L are the centroid sizes of each wing (i.e., the sum of the distances of each landmark to their combined center of mass or centroid location). In addition, a mu ltivariate shape-based measure of FA known as the Procrustes distance was calculated as the square root of the sum of all squared Euclidean distances between each left a nd right landmark after two-dimensional Procrustes fitting of the data (Bookstein 1991; Klingenberg and McIntyre 1998; FA 18 in Palmer and Strobeck 2003; Smith et al. 1997) This removed any difference due to size alone. Centroid size calculation, Euclidea n distance calculation and Procrustes fitting were performed using yvind Hammers Paleontological Statistics program PAST version 0.98 (Hammer 2002). Percent measur ement error was computed as (ME/average

PAGE 81

68 FA) x 100 where 3 / ) 3 1 3 2 2 1 ( FA FA FA FA FA FA ME in a smaller subset (200 wings each measured 3 times = FA 1, FA 2 and FA 3) of the total sample. All subsequent statistical analys es were performed using SPSS Ba se 8.0 statistical software (SPSS Inc.). Unsigned multivariate size and sh ape FA as well as the kurtosis of signed FA were then compared at various temperatures using one-way ANOVA. Results Development time (Figure 4-2) was very similar to previously published data (Kersting et al. 1999, Xia et al. 1999) decreasing monotonically at a much steeper rate in dark morphs than in light morphs. The distri butional pattern of centroid size, size FA and shape FA appear similar exhibiting right log skew distributions (Figure 4-3). Similar distributional patterns are obs erved within temperatures ( not shown) as that observed across temperatures (Figure 4-3). 8 10 12 14 16 18 20 22 24 26 5 10 15 20 25 30 DEVELOPMENT TIME (days) TEMPERATURE (C) Light Morph Dark Morph Figure 4-2. Cotton aphid mean developmen t time SE in days in relation to temperature (n= 531).

PAGE 82

69 2 80 .0 260.0 2 40. 0 2 20 .0 200.0 1 80. 0 1 60 .0 140.0 1 20. 0 1 00. 0 8 0 .0 6 0.0 4 0. 0 2 0 .0500 400 300 200 100 0 Std. Dev = 36.40 Mean = 104.3 N = 3644.00 CENTROID SIZE CENTROID SIZE FLUCTUATING ASYMMETRY 37.6 34.4 31.3 28.1 24.9 21. 8 18. 6 15. 4 12.2 9.1 5.9 2.7 -.5300 200 100 0 Std. Dev = 5.96 Mean = 7.7 N = 1822.00 .163 .150 .138 .125 .113 .100 .087 .075 .063 .050 .038 .025 .013200 100 0 Std. Dev = .02 Mean = .056 N = 1219.00 SHAPE FLUCTUATING ASYMMETRYFREQUENCY Figure 4-3. Distribution of isogenic size, size based and sh ape based FA in monoclonal cotton aphids grown in controlled en vironment at different temperatures. Distributions within each temperature treatment are similar to overall distributions shown here.

PAGE 83

70 Coefficient of variation for FA was sli ghtly higher for dark morphs (12.5 C = 92.59%, 15 C = 93.02%, 17 C = 88.95%, 19 C = 79.02%, 22.5 C = 80.00% and 25 C = 85.35%). Mean isogenic FA (both size and shape) was highly si gnificantly different across temperatures (ANOVA F = 6.691, df between group = 4, df within group = 1673, p < 0.001) in the Gainesville FL clone (Figur e 4-4) but not in the Lake Alfred clone (ANOVA F = 1.992, df between group = 5, df within group = 672, p < 0.078). This is an indication of undersampling in the Lake Alfr ed clone. In the Gainesville clone, mean centroid size FA (Figure 4-4A) and devel opment time (Figure 4-2) follow a nearly identical pattern, decreasing rapidly at first then slowing with increased temperature. TEMPERATURE (C) mean isogenic fluctuating asymmetry 10 12 14 16 18 20 22 24 26 0.052 0.054 0.056 0.058 0.06 0.062 0.064 0.066 0.068 10 12 14 16 18 20 22 24 26 6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11 A B Light Morph Light Morph Dark Morph Dark Morph Figure 4-4. Mean isogenic FA for (A.) centr oid sizebased and (B .) Procustes shapebased) in monoclonal cotton aphids (c ollected in Gainesville FL) grown on isogenic cotton seedlings at different temperatures.

PAGE 84

71 Mean shape FA was also significantly different across temperature classes (ANOVA F = 4.863, df within group = 4, df between group = 1673, p = 0.001) but this difference is due solely to elevated FA in the 12.5C group (Figure 4-4B). Less than one percent of the variation in FA was due to variation in body size (r = -0.101 for shape FA; r = 0.088 for size FA). Kurtosis in the shape of the distribution of size based FA (Figure 4-5) was significantly higher in dark morphs th an in light morphs (t = -2.21 p = 0.027). Within each morph (light or dark), kurtosis in the distribution of FA appears to increase with temperature slightly (Figure 4-5). Measurement error for shape FA was estimated at2.6% (Lake Alfred clone) and 2.2% (Gainesville clone). For size FA these estimates were 6.1% (Lake Alfred) and 5.7% (Gainesville). TEMPERATURE (C) KURTOSIS 10 12 14 16 18 20 22 24 26 -1 0 1 2 3 4 5 Light Morph Dark Morph Figure 4-5. Kurtosis of size-based FA in monoclonal cotton aphids grown on isogenic cotton seedlings at different temperatures.

PAGE 85

72 Discussion It appears that the prediction of the proximate effect model holds in the case of this population of cotton aphid, which, in general, is not energetically limited and in this study, is not genetically vari able. Temperature and growth rate (which are positively correlated in insects) are negatively associat ed with purely environmentally derived FA (i.e., developmental noise). This confirms the predictions of several proximate models of the basis of FA. Furthermore, it appears that centroid size-based FA is a simple function of development time. Because individual ge netic differences in the capacity to buffer against developmental noise are, in a sense, controlled for in this study by the use of natural clones, the response of FA to temp erature in this stud y represents a purely environmental response of FA. The fact that aphids excr ete large amount of water and sugar in the form of honeydew, as well as th e lack of a strong co rrelation between FA and size also suggests that there is no real energeti c cost to being large in the aphids in this study. An important next step will be to comp are this result to the association of growth rate to FA in genetically di verse and energetically limited sexually selected traits where the predicted association between growth ra te and FA would be the opposite of this study. It also appears that because the enviro nmental component of size-based isogenic FA is largely a function of developmental time and temperat ure, dark morphs of cotton aphids, which have much longer developm ent time than light morphs, also have significantly higher FA. The temperature trend in mean FA within both light and dark morphs, where development times are similar, is not consistent although it appears that the kurtosis of FA increases slightly with temperatur e within each morph.

PAGE 86

73 It is very interesting th at there is a strong differen ce in kurtosis between the two temperature morphs in this study. Previous work suggests that kurtosis is related to genetic variability in a populat ion (Babbitt in press). The difference observed here in light of genetic homogeneity in the monoclonal aphid cultures, suggests that there may be a difference in developmental stability of li ght and dark aphid mo rphotypes that is due primarily to the differential expression of genes in each phenotype. It is surprising that temperature trends in mean developmental noise are slightly different regarding whether a size or shapebased approach was used. In populations where individuals are genetically diverse, bot h size and shape-based measures of FA are often correlated to some degr ee (as they are here too). However, size and shape are regulated somewhat differently in that cell pr oliferation is mediated both extrinsically via cyclin E acting at the G1/S checkpoint of the cell cycle, predominantly affecting size, and intrinsically at via cdc25/string at the G2/M checkpoint, pred ominantly affecting pattern or shape (Day and Lawrence 2000). Extrin sic mechanisms regulate size through the insulin pathway and its associated horm ones (Nijhout 2003) provi ding a link between size and the nutritional envi ronment. This may explain why size FA follows development time more closely than shape FA in this study. Size is usually less canalized than is shape and therefore more variable. While this makes the size-based measure of FA generally attractive, it has been found in previous work (Babbitt et al. 2006) that population averag es of size-based FA are often so variable that they are frequently under-sampled b ecause of the broad and often long tailed distribution of FA. Shape-base d FA does not suffer as much from this problem and so its estimation is much better although it may be le ss indicative of environmental influences

PAGE 87

74 and perhaps therefore better suited for genetic studies of FA. Because size FA has been adequately sampled in this study and because size is more heavily influenced by environmental factors like temperature, I find that size-based FA is the more interesting measure of developmental inst ability in this study. In general, it is clear that developmen tal noise is not constant in genetically identical individuals cultured under near sim ilar environments. Th e overall distribution pattern in both size and shape FA is log skew ed even within temperature classes. This suggests that sampling error due solely to random noise can play a very significant role in FA studies. Furthermore, it appears that th e response of FA to the environment is potentially quite strong. This suggests that FA may indeed be a re sponsive bioindicator, however its response will not be very generalizable in enviro nments with fluctuation in temperature. In conclusion, the environmental respons e of developmental noise to temperature in absence of genetic variabi lity and nutrient limitation suppo rts an important prediction of theoretical explanations of the proximate basis of FA. This prediction is that FA should decrease with growth rate (and temper ature) in ectothermic organisms. Because it is only in sexually selected tr aits that the opposite predicti on should hold, that FA should increase and not decrease with growth rate, th e results presented here may offer a useful method for discriminating between FA that is under sexual selection an d FA that is not.

PAGE 88

75 CHAPTER 5 CONCLUDING REMARKS AND RECOMMENDATIONS In this dissertation, I confirm some exis ting hypotheses concerning FA and provide a new explanatory framework th at explains alteration in th e distribution of fluctuating asymmetry (FA) and its subsequent effect upon mean FA. The basis of FA and the influences that shape its response to gene s and the environment in populations can be suitably modeled by stochastic proportional growth in expanding populations of cells on both sides of the body that are terminated with a small degree of genetically-based random error. I model stochastic growth with geometric Brownian motion, a random walk on a log scale. And I also model er ror in terminating growth with a normal distribution. The resulting di stribution of FA is a lognorma l distribution characterized by power-law scaled tails. Because under a pow er-law distribution, increased sample size increases the chance of sampling rare events under these long tail s, convergence to the mean is potentially slowed or even stopped depending on the scaling exponents of the power-law describing the tails. I demonstrated that the effect of reduced convergence to the mean is substantial and has probably cause d much of the previous work on FA to be under-sampled (Chapter 1). I have also dem onstrated that the s caling effect in the distribution of FA is directly related to genetic variation in the population. Therefore a low scaling exponent and also high kurtosis is associated with a la rge degree of genetic variation between individuals in their ability to precisely terminate the growth process (Chapter 2). I also demonstrate that when genetic differences do not exist (where FA is comprised of only developmental noise) and when development is not limited

PAGE 89

76 energetically, such as in a parthenogenetic aphid clone, FA depends upon developmental time (Chapter 3). How Many Samples Are Enough? The answer to this clearly depends upon wh ether researchers chose to study FA in size or FA in shape. The required sample size depends upon the di stributional type and parameters of the distribution of FA. As pr eviously discussed, FA distributions differ depending upon whether FA is based on size or shape differences between sides. As demonstrated in chapter 1, multivariate m easures of FA based upon shape, such as Procrustes Distances, tend to ha ve distributional parameters that allow convergence to the mean that is about five times more rapid th an either univariate or multivariate measures of FA based upon size (Euclidian distance or ce ntroid size). While this information might seem to favor a shape-based appro ach, there are some ot her very important considerations given below. In the end, sa mple size must be independently evaluated in each study depending on the magnitude of asymmetry that one wishes to detect between treatments or populations. As a general ru le, if shape FA is used, 100-200 samples may suffice, but if size FA is used then many hundreds or even a thousand samples may be required to detect a similar magnitude difference. What Measure of FA is best? The handicap of sampling requirements of size-based FA aside, it appears that measures of the distributional shape of FA like kurtosis, scaling exponent in the upper tail and skewness are strongly related to si milar but smaller changes in mean size FA (chapter 2). Because size FA is most co mmonly used in past studies of FA, these measures of distribution shape might allo w clearer conclusions to be drawn from literature reviews and meta-a nalyses as well as individual studies of FA where the

PAGE 90

77 collection of more data is not possible. My work linking shape of the FA distribution to the genetic variability of the population will re quire that future studies of FA which are focused on the environmental component of FA be controlled for genetic differences between populations. Where FA is sought as a potential bioindicator of environmental stress, the potentially differing genetic structur e of populations will need to be considered in order to make meaningful conclusions about levels of FA. Additi onally, because in the absence of genetic variation in monoclonal aphi ds, size FA is directly related to the total developmental times of individuals (chapter 3), it may a better choice than shape FA for studies of environmental influences on FA. In most organisms, body size is less canalized than body shape. Body size is highly polygenic and depends upon many environmenta lly linked character traits. However body shape or patterning is determined by a sequential progression of the activity of far fewer genes. Body size is also regulated at a different checkpoint during the cell cycle than is body pattern formation or shape (Day and Lawrence 2000). Patterning or shape is regulated at G2/M checkpoint while size is regulated at G1 /S checkpoint. The latter checkpoint is associated with the insulin pa thway, linking size to nutrition and hence to the environment. Because of this the st udy of environmental responses of FA may ultimately be best served by using multivariate size-based measures of FA (i.e., centroid size FA) while at the same time making sure to collect enough samples to accurately estimate the mean. In conclusion, multivariate size FA may be the better metric for large studies interested in the eff ects of environmental factors upon FA. Where data is harder to come by (e.g., vertebrate field studies), multivariate shape FA may perform better with the

PAGE 91

78 caveat that shape and pattern are less likely to vary in a population than does size. Because centroid size must be calculated in order to derive Procrustes distance, both measures can easily and should be examined together. Does rapid growth stabilize or destabilize development? The effects of temperature and growth ra te on FA appear to support the proximate model which predicts that FA declines with increasing growth rates (Figure 4-3). Aphids from the Gainesville clone decrease mean FA in response to incr eased temperature and growth rate. It is likely that the Lake Alfred clone is under-sampled and therefore estimates of mean FA are not accurate in th at sample. The results of the Gainesville clone, which was sampled adequately, suggests that the prediction of proximate models of the basis of FA (Emlen et al. 1993, Gr aham et al. 1993), that rapid growth should decrease FA holds true. Mllers hypothesis that rapid growth is stressful because of incurred energetic costs does not appear to hold in the case of aphids. Of course aphids, being phloem feeders, generally have access to more water and carbon than they ever need. This is evidenced by analysis of honeydew composition in this and many other aphid species. Growth and reproduction in aphids is probably more limited by nitrogen than by water or energy (carbon). So it would seem that this system may not be a very good one in which to assess Mllers hypothe sis directly. It should be further investigated whether FA of sexually selected traits in energetically limited, genetically diverse populations is positively associated with growth rate as is predicted by the ultimate or evolutionary model presented in Chapter 4. Can fluctuating asymmetry be a sexually selected trait? Given that a genetically a ltered population (chapter 3) demonstrates a consistent response in the shape of the FA distribution does suggest that a poten tially strong genetic

PAGE 92

79 basis for FA exists, but is not easily observabl e through levels of mean FA. This implies some heritability exists and possibly opens th e door for selection to act upon FA in the context of mate choice. However, in the futu re it must be demonstr ated, with appropriate sample sizes, that the level of FA in th e laboratory can be altered by selection for increased or decreased FA. The opposing pred ictions of the proximate or mechanistic and ultimate (evolution through sexual select ion) explanations of FA, regarding the relationship of temperature, growth rate a nd FA may offer a method for detecting sexual selection upon FA (chapter 4) If FA has evolved as an indicator of good genes, and hence has related energetic costs, the expect ation that FA should increase with growth rate and temperature (in ectotherms) is plausi ble. Otherwise, if FA is caused by random accumulation of error during development, then the expectation is reversed. FA should decrease with growth rate and temperature as I have demonstrated in a clonal population of aphids that are not energe tically limited (chapter 4). Is fluctuating asymmetry a valu able environmental bioindicator? Given that the genetic structure of a population (chapter 3) has potentially strong effects upon the shape and location (mean) of th e distribution of FA, the usefulness of FA as a bioindicator of environmental stress has to be questioned. Average levels of FA of populations would be difficult to interpret or compare without additional information regarding genetic heterogeneity However, provided that the study of FA and population genetics were undertaken simultaneously, this problem could be avoided. So FA could still be of use in this regard, however its expense and ease of use compared to other bioindicators would need to be reassessed.

PAGE 93

80 Scaling Effects in Statistical Distributions: The Bigger Picture This dissertation demonstrates that underlying distributions of some biological data can contain partial self-similarity or power-l aw scaling. The proper model for describing data such as this sits at a midpoint between those models used in cl assical statistics and those of statistical physics: Le vy statistics (Bardou et al. 2003 ). To my knowledge, this work represents the first application of such a model in biology. The sources of power-law scaling in the natural sciences are diverse. Sornette (2003) outlines 14 different ways that power-law s can be created, some of which are very simple. The hypothesis that all power-law scaling in nature is due to a single phenomena such as self-organized criti cality (SOC) (Bak 1996, Gisiger 2000) or highly optimized tolerance (HOT) (Newman 2000) is unlikely. Reed (2001) suggests that the model I have used here, stochastic proportional growth th at is observed randomly, may explain a great deal of power-law scaled size distributions formerly speculated to have a single cause like SOC or HOT. These phenomena include di stributions of city size (Zipfs law), personal income (Paretos Law), sand particle size, species per genus in flowering plants, frequencies of words in sequences of text, size s of areas burnt in forest fires, and species body sizes, just to name a few examples. One of the most common ways that power-laws can be obtained is by combining exponential functions. This is effect is also observed when positive exponential/ geometric/proportional growth is observed w ith a likelihood described by to a negative exponential function such as in Reed and Jo rgensens (2004) model for generating size distributions. Power laws generated by combi ng exponential functions are also present in the statistical mechanics of highly interactiv e systems (e.g., SOC) where the correlation in behavior between two nodes or objects in an interaction network or lattice decreases

PAGE 94

81 exponentially with the distance between them while the number of potential interaction paths increases exponentially with the distance between th em (Stanley 1995). In Laplacian fractals, or diffusion limited aggregations observed in chemical electrodeposition and bacterial colony growth (Viscek 2001) there may be a similar interplay between exponential growth (doubling) and the diffusion of nutrients supporting growth which are governed by the no rmal distribution, of the form2xe y. The exponential function hol ds a special place in th e natural sciences. Any frequency dependent rate of cha nge in nature, or in other wo rds, any rate of change of something that is dependent on the proportion of that something present at that time, is described by the exponential f unction. Therefore it finds application to many natural phenomena including behavior of populations, chemical reactions, ra dioactive decay, and diffusion just to name a few. It seems only fitting that many of the power-laws we observe in the natural sciences probably owe their existence to the interplay of exponential functions, one of the most comm on mathematical relationships observed in nature. As far as we can ascertain from the record ing of ancient civili zations, human beings have been using numbers for at least 5000 year s if not longer. And yet the concepts of probability are a relatively recent human inventi on. The idea first appears in 1545 in the writings of Girolamo Cardano and is late r adopted by the mathematicians, Galileo, Fermat, Pascal, Huygens, Bernoul li and de Moivre in disc ussions of gambling over the next several hundred years. The concep ts of odds and of random chance are not generalized until the early tw entieth century by Andreyevic h Markov and not formalized into mathematics until the work of Andrei Kolmogorov in 1946.

PAGE 95

82 The role of uncertainty in nature is yet to be resolved. Does uncertainty lie only with us or does it underlie the very fabric of the cosmos? Most 19th century scientists (excepting perhaps Darwin) believed that th e universe was governed by deterministic laws and that uncertainty is solely due to human error. The first app lication of statistical distributions by Carl Frederick Gauss and Pierre Simon Laplace were concerned only with the problem of accounting for measuremen t error in astronomical calculations. The more recent view, that uncertainty is some thing real, was largely the work of early twentieth century quantum theorists Werner Heisenberg and Erwin Schrodinger and the statistical theory embodied in the work of Sir Ronald Fi sher. The basic conceptual revolution in modern physics was that observa tions cannot be made at an atomic level without some disturbance, therefore while one might observe something exactly in time, one cannot predict how the act of observing wi ll affect the observed in the future with any degree of certainty (at least at very small scales). This is Heisenbergs Uncertainty or Indeterminancy Principle. Scientists observe natural laws only through emergent properties of many atoms observed at vastly larger scal es where individual behavior is averaged into a collective whole. Jakob Bernoullis Law of Large Nu mbers, Abraham de Moivres bell shaped curve, and Sir Ronald Fishers estimation of th e mean are also examples of how scientists rely upon emergent properties of large system s of randomly behaving things in order to make sense of what is thought to be fundame ntally uncertain world. However, this probabilistic view of the natu ral world has been challenged in recent decades by some mathematicians who are again championing a dete rministic view of nature. Observations by Edward Lorenz, Benoit Mandelbrot and ot hers, of simple and purely deterministic

PAGE 96

83 equations that behave in complex unpredic table ways, has led to a revival in the deterministic view among mathematicians; in this case, uncertainty is a product of human inability to perceive systems that are extr emely sensitive to initia l starting conditions. For this reason, this new view is often calle d deterministic chaos. And so the question as to whether uncertainty is real or imagined is yet to be resolved by modern science. One remarkable observation of this latest revolution in mathematics is the frequent occurrence of scale invariance or power-law scaling in system s that exhibit this sort of complex and unpredictable behavior. And so ju st as the normal distribution or bell curve is an emergent property governing the random behavior of independent objects, the power-law appears to be an emergent propert y of nature as well; one that seems not only to often to appear in the beha vior of interacting objects (i .e., critical systems) but in systems where growth in randomly observed as well. Statistical physics now recognizes two classes of stable laws, one that leads to the Gaussi an or normal distribution and another that is Levy or power-law distributed. In the former class, emergent behavior is governed by commonly occurring random events wh ile in the latter cl ass the behavior of the group is governed by a few rarely occu rring random events. As I have already reviewed earlier, we find that convergence to the mean under these two frameworks can be radically different. Yet bot h are present in the natural world, and as I have shown in my work, both of these classes of behavior can underlie naturally occurring distributions, causing partial scale inva riance in real data. It is my convicti on that the biological sciences must in the future adopt the statistical methods of working under both of these frameworks and not simply make assumpti ons of normality whenever and wherever random events are found to occur.

PAGE 97

84 APPENDIX A LANDMARK WING VEIN INTERSECTIONS CHOSEN FOR ANALYSISOF FLUCTUATING ASYMMETRY Figure A1. Six landmark locations digitized for Aphis gossipyii Figure A2. Six landmark locations digitized for Apis mellifera

PAGE 98

85 Figure A3. Six landmark locations digitized for Chrysosoma crinitus Figure A4. Eight landmark locations digitized for Drosophila simulans

PAGE 99

86 APPENDIX B USEFUL MATHEMATICAL FUNCTIONS The following is a list of the probability density functions for candidate models of the distribution of fluctuating asymmetry. L og likelihood forms of these functions were maximized to obtain best fitting parameters of each model for our data. Asymmetric Laplace Distribution (see Kotz et al. 2001) )) ( ( 1 1) )( ( ) (2 xe x f for x )) ( ( 1 11 2) )( ( ) ( xe x f for < x where = location, = scale and = skew index (Laplace when = 1) Half-Normal Distribution 2 ( 2 1) )( ( 1 2) )( ( ) ( xe x f where = minimum data value and = dispersion Lognormal Distribution (see Evans 2000 or Limpert et al. 2001) ) 2 / ) (log ( 2 / 12 2) 2 ( 1 ) ( xe x x f where = location and = shape or multiplicative standard deviation Double Pareto Lognormal Distribution (see Reed and Jorgensen 2004) )) (log( ) (1x g x fx given the normal-Lap lace distribution ) ( ) ( ) ( ) ( y y yR R y g where the Mill s ratio R(z) is

PAGE 100

87 ) ( ) ( 1 ) ( z z z R and where is the cumulative density function and is probability density function for standard normal dist ribution N(0,1), where and are parameters that control power-law scaling in the tails of th e lognormal distribution. The limiting forms of the double Pareto logno rmal are the left Pareto lognormal ( ), right Pareto lognormal distributions ( ), and lognormal distributions ( ) with Pareto tails on only the left side, only the right side, or on neither side, respectively. A description of Reed and Jorgensens ge nerative model of double Pareto lognormal size distribution. Reed and Jorgensons (2004) generative model begins with the Ito stochastic differential equation representing a geom etric Brownian motion given below. Xdw Xdt dX with initial state X(0) = X0 distributed lognormally, log X0 N( 2). After T time units the state X( T ) is also distributed lognormally with log X( T ) N( + (2/2) T 2 + 2 T ). The time T at which the process is observ ed, is distributed with density f T(t) = et where is a constant rate. The double Pareto lognormal distribution is generated when geometric Brownian motion is sampled repeat edly at time t with a negative exponential probability.

PAGE 101

88 LIST OF REFERENCES Albert R. and A-L. Barabasi. 2002. Statisti cal mechanics of complex networks. Reviews of Modern Physics. 74: 47-97. Babbitt G .A., R.A. Kiltie and B. Bolker. 2006 Are fluctuating asymmetry studies adequately sampled? Implications of a ne w model for size distribution. American Naturalist 167(2):230-245. Babbitt G.A. in press. Inbreeding reduces power-law scaling in the distribution of fluctuating asymmetry: an explanation of the basis of developmental instability. Heredity Baish J.W and R.K. Jain. 2000. Fractals a nd cancer. Cancer Research 60: 3683-3688 Bak P. 1996. How Nature Works, the Science of Self Organized Criti cality. Copernicus. New York. Balanda K.P. and H.L. MacGillivray. 1988. Ku rtosis: a critical review. The American Statistician 42(2):111-119. Bardou F., J.P. Bouchard, A. Aspect and A. Cohen-Tannoudji. 2003. Levy statistics and laser cooling: How rare events bring atom s to rest. Cambridge University Press. Bjorksten T.A., K. Fowler and A. Pomia nkowski. 2000. What does sexual trait FA tell us about stress? Trends in Ec ology and Evolution 15(4) 163-166. Bookstein F.L. 1991 Morphometric tools fo r landmark data: geometry and biology. Cambridge University Press Burnham K.P. and D.A. Anderson. 1998. M odel Selection and Multimodel Inference: a Practical Information Theoretical Approach. Springer. New York, NY. Carchini G., F. Chiarotti M.D. Domenico M. Mattoccia and G. Paganotti. 2001. Fluctuating asymmetry, mating succe ss, body size and heterozygosity in Coenagrion scitulum Animal Behavior 61:661-669. Chapman J.W. and D. Goulson. 2000. E nvironmental versus genetic influences on fluctuating asymmetry in the house fly, Musca domestica Biological Journal of the Linnean Society 70(3):403-413.

PAGE 102

89 Clarke G.M. 1993. The genetic basis of deve lopmental instability: relationships between stability, heterozygosity and genomic co-adaptation. Genetica 89: 15-23. Clarke G.M. G.W. Brand and M.J. Whitten. 1986. Fluctuating asymmetry a technique for measuring developmental stress caused by inbreeding. Australian Journal of Biological Sciences 39(2): 145-153. Clipsham R. Y.H. Zhang, B.L. Huang and E.R.B. McCabe. 2002. Genetic network identification by high density multiplexed reverse transcriptional (HD-MRT) analysis in steroidogenic axis model cell lines. Molecular Genetics and Metabolism 77 : 159-178. Conlin I. and M. Raff. 1999. Size control in animal development. Cell 96:235-244. Crow J.F. and M. Kimura 1970. An Introducti on to Population Genetic s Theory. Harper and Row. New York. Day S J and P.A. Lawrence. 2000. Measuring dimensions: the regulation of size and shape. Development 127:2977-2987 Debat V. and P. David. 2001. Mapping phenotypes: canalizati on, plasticity and developmental stability. Trends in Ecology and Evolution 16(10) 555-561. Diaz B. and E. Moreno. 2005. The competitive nature of cells. Experimental Cell Research 306:317-322. Ditchkoff S.S., R.L. Lochmiller, R.E. Masters, W.R. Starry and D.M. Leslie jr. 2001. Does fluctuating asymmetry of antlers in White tailed deer ( Odocoileus virginianus ) follow patterns predicted for sexually selected traits? Proc. R. Soc. London Ser. B 268:891-898. Emlen J.M Freeman D.C. and J.H. Graham. 1993. Nonlinear growth dynamics and the origin of fluctuating asymmetry. Genetica 89: 77-96. Evans, M, N. Hastings, B. Peacock. 2000. St atistical Distributions John Wiley and Sons NY. Fowler K. and M.C. Whitlock. 1994. Fluctu ating asymmetry does not increase with moderate inbreeding in Drosophila melanogaster Heredity 73: 373-376. Fuller R.C. and D. Houle. 2003. Inheri tance of developmental instability. In Developmental Instability: Causes and Consequences Edited by M. Polak 2003. Oxford University Press. Frigesyi A., D. Gisselssonn., F. Mitelm an and M. Hoglund. 2003. Power law distribution of chromosomal aberrations in cancer. Cancer Research 63: 70947097.

PAGE 103

90 Gangestad S.W. and R. Thornhill. 1999. Individual differences in developmental precision and fluctuating asymmetry: a m odel and its implications. Journal of Evolutionary Biology 12(2): 402-416. Gisiger T. 2001. Scale invari ance in biology: coincidence or footprint of a universal mechanism? Biological Reviews 76:161-209. Graham J.H. 1992. Genomic coadaptation and de velopmental stability in hybrid zones. Acta Zoologica Fennica 191:121-131. Graham J.H., D.C. Freeman and J.M. Emlen 1993. Antisymmetry, directional asymmetry and dynamic morphogene sis. Genetica 89:121-137. Graham J.H., K. Shimizu, J.C. Emlen, D.C. Freeman. and J. Merkel. 2003. Growth models and the expected distribution of fluctuating asymmetry. Biological Journal of the Linnean Society 80: 57-65. Hammer 2002. PAST version 0.98 statisti cal software Palaontol ogiches Institut und Museum, Zurich http://folk.uio.no/ohammer/past Last accessed in April 2004. Hardersen S. and C. Frampton. 2003. The in fluence of differen tial survival on the distribution of fluctuating asymmetry a modeling approach. Journal of Theoretical Biology 224: 479-482. Hilborn D. and M. Mangel. 1998. The Ecologica l Detective. Princet on University Press, Princeton, NJ. Hoffman A.A. and R.E. Woods. 2003. a ssociating environmental stress with developmental stability: patterns and pr oblems. In Developmental Instability: Causes and Consequences. Edited by M. Polak 2003. Oxford University Press. Hogg I.D., Eadie J.M., D.D. Williams and D. Turner. 2001. Evaluating fluctuating asymmetry in a stream dwelli ng insect as an indicator of low level thermal stress: a large scale field experiment. Jour nal of Applied Ecology 38(6): 1326-1339. Houle D. 1989. Allozyme associated heterosis in Drosophila melanogaster Genetics 123:789-801. Houle D. 2000. A simple model of th e relationship between asymmetry and developmental stability. Journal of Evolutionary Biology 13: 720-730. Kersting U., S. Satar and N. Uygun. 1999. Eff ect of temperature on development rate and fecundity of apterous Aphis gossipyii Glover reared on Gossipium hirsutum L Journal of Applied Entomology 123:23-27. Klingenberg C.P. 2003. A developmental pe rspective on developmental instability: theory, models and mechanisms In De velopmental Instabil ity: Causes and Consequences. Edited by M. Polak 2003. Oxford University Press.

PAGE 104

91 Klingenberg C.P. and G.S. McIntyre. 1998. Geometric morphometrics of developmental instability: analyzing patterns of fluctu ating asymmetry with Procrustes methods. Evolution 52(5) 1363-1375 Klingenberg C.P. and H.F. Nijhout. 1999. Genetics of fluctuating asymmetry: a developmental model of developmental inst ability. Evolutio n 53(2) : 358-375. Koehn R.K. and B.L. Bayne. 1989. Towards and physiological and genetical understanding of the energetics of the st ress response. Biologi cal Journal of the Linnean Society 37: 157-171. Kotz S., T.J. Kozubowski and K. Podgorski. 2001. The Laplace Distribution and Generalizations: a Revisit with Applications to Communications, Economics, Engineering and Finance. Birkhauser (c/o Springer-verlag), Boston. Kozubowski T. J., and K. Podgorski. 2001. Asymmetric Laplace laws and modeling financial data. Mathematical a nd computer modeling 34:1003-1021. Kozubowski T. J., and K. Podgorski. 2002. Log-Laplace distributions, technical report no. 60, Department of Mathematics, University of Nevada at Reno. Leamy L. J. 2003. Dominance, epistasis and fluctuating asymmetry. In Developmental Instability: Causes and Consequences. Ed ited by M. Polak 2003. Oxford university Press. Leamy L. J. and C. P. Klingenberg. 2005. The genetics and evolution of fluctuating asymmetry. Annual Review of Ecology, Evolution and Systematics. 36:1-21. Leary R.F., F.W. Allendorf, K.L. Knudsen. 1983. Developmental stability and enzyme heterozygosity in rainbow trout. Nature 301: 71-72. Leary R.F., F.W. Allendorf, K.L. Knudsen. 1984. Superior developmental stability of heterozygotes at enzyme loci in Salmoni d fishes. American Naturalist 124: 540551. Lens L., S. Van Dongen and E. Mattthysen. 2002. Fluctuating asymmetry as an indicator of fitness: can we bridge th e gap between studies? Biological Reviews 77:27-38. Lerner I.M. 1954. Genetic Homeos tasis, Oliver and Boyd, London. Limpert E. W.A. Stahel and M. Abbot 2001. Lognormal distributions across the sciences: keys and clues. Bioscience 51(5) 341-352. Ludwig 1932. Das rechts-links problem im tierreich und beim Menschen Springer Berlin. MathCad 2001 Mathsoft Engineering and E ducation Inc. 101Main Street, Cambridge MA 02142 USA

PAGE 105

92 Mather K. 1953. Genetical control of stabil ity in development. Heredity 7: 297-336. McCoy K. A. and R.N. Harris. 2003. Integr ating developmental st ability analysis and current amphibian monitoring techniques: an experimental evaluation with the salamander Ambystoma maculatum Herpetologia 59: 22-36. Mendes R.L., Santos A.A. Martins M.L. and V ilela M.J. 2001. Cl uster size distribution of cell aggregates in culture. Physica A 298:471-487 Merila J. and M. Bjorklund. 1995. Fluctu ating asymmetry and measurement error. Systematic Biology 44: 97-101. Messier S. and J. Mitton. 1996. Heterozygos ity at the malate dehydrogenase locus and developmental homeostasis in Apis mellifera Heredity 76:616-622. Milan M, S. Campuzano and A. Garcia-Bel lido 1995. Cell cycli ng and patterned cell proliferation in the wing primordium of Drosophila Proceedings of the National Academy of Sciences 93:640-645. Milton C.C., B. Huynh, P. Batterham S. L. Rutherford and A.A. Hoffman. 2003. Quantitative trait symmetry independent of Hsp90 buffering: Distinct modes of genetic canalization and developmental st ability. Proceedings of the National Academy of Sciences 100(23) 13396-13401. Mitton J.B. and M.C. Grant. 1984. Associa tions among protein heterozygosity, growth rate and developmental homeostasis. Annual Reviews of Ecology and Systematics 15:479-499. Mitzenmacher M. (Draft manuscript) A brief history of generative models for power-law and lognormal distribution. Harvard university. Moller A.P. 1990. Fluctuating asymmetry in ma le sexual ornaments ma y reliably reveal male quality. Animal Behavior 40:185-1187. Moller A.P. and A. Pomiankowski 1993. Fl uctuating asymmetry a nd sexual selection. Genetica 89: 267-279. Moller A. P. and J. P. Swaddle. 1997. Asymmetry, Developmental Stability and Evolution. Oxford University Press. Montroll E.V. and M.F. Shlesinger. 1982. On 1/f noise and other di stributions with long tails. Proceedings of National A cademy of Sciences 79:3380-3383. Montroll E.V. and M.F. Shlesinger M.F. 1983. Maximum entropy formalism, fractals, scaling phenomena and 1/f noise: a tale of tails. Journal of Statistical Physics 32(2)209-230.

PAGE 106

93 Mosteller F and J.W. Tukey. 1977. Data Anal ysis and Regression. Addison-Wesley. Reading MA. Mpho M, A. Callaghan and G.J. Holloway. 2002. Temperature and genotypic effects on the life history and fluctuating asymmetry in a field strain of Culex pipiens Heredity 88:307-312. Newman N. 2000. The power of design. Nature 405:412-413. Nijhout H.F. 2003 The control of body size in insects. Developmental Biology 261:1-9 Oltvai Z.N. and A-L Baraba si. 2002. Lifes complexity pyramid. Science 298:763-764. Palmer A.R. and C. Strobeck. 1986. Fluctu ating asymmetry: measurement, analysis patterns. Annual reviews in Ecol ogy and Systematics 17:391-421. Palmer A.R. and C. Strobeck. 2003. Fluc tuating asymmetry studies revisited. In Developmental Instability: Causes and Consequences. Edited by M. Polak (2003) Oxford University Press. Parsons P. A. 1992. Fluctuating asymmetry, a biological monitor of environmental and genomic stress. Heredity 68:361-364. Perfectti F. and J.P.M. Camachi. 1999. Analysis of genotypic differences in developmental stability in Annona cherimola Evolution 53(5) 1396-1405. Pinheiro J. and D.M. Bates. 2000. Mixed-e ffects Models in S and S-PLUS. Springer, New York. Polak M, A.P. Moller, S.W. Gangestad, D.E. Kroeger, J.T. Manning and R. Thornhill. 2003. Does an individual asymmetry parameter exist? A meta-analysis. In Developmental Instability: Causes and Consequences. Edited by M. Polak (2003) Oxford University Press. Quandt R.E. 1966. Old and new methods of estimation and the Pareto distribution. Proceedings of Royal Statistical Society 1966:55-75. Radwan J. 2003. Inbreeding not stress increase s fluctuating asymmetry in the bulb mite. Evolutionary Ecology Research 5(2) 287-295. Raff M. 1992. Social controls on cell surv ival and cell death. Nature 356 April 2 1992 p397-400 Rao G.Y., S. Anderson and B. Widen. 2002. Flower and cotyledon asymmetry in Brassica cretica : genetic variation and relationships with fitness. Evolution 56 : 690-698.

PAGE 107

94 Rasmuson M. 2002. Fluctuating asymmetry indicator of what? Hereditas 136(3) : 177183. Reed W. J. 2001. The Pareto, Zipf and othe r power-laws. Economics Letters, 74: 15-19. Reed W. J. and M. Jorgensen. 2004. The double Paretolognormal distribution: a new parametric model for size distributions. Co mmunications in Statistics Theory and Methods 33:1733-1753. Reeve E.C.R. 1960. Some genetic tests of the asymmetry of sternopleural chaetae number in Drosophila Genetical Research 1:151-172. Rohlf J. 1999. TPSDIG version 1.31 distributed at S UNY morphometrics http://life.bio.sunysb.edu/morph/ Last accessed on June 2003. Roman H. E. and M. Porto. 2001. Self generated power-law tails in probability distributions. Physical Review E. 63 036128. Romeo M., V. Da costa and F. Bardou. 2003. Broad distribution effects in sums of lognormal random variables. Europ ean Journal of Physics B 32:513-525. Ruddon R.W. 1995. Cancer Biology 3rd ed. Oxford University Press. Rutherford S.L. and S. Linquist. 1998. Hsp90 as a capacitor for morphological evolution. Nature 396: 336-342. Savage A. and P.J. Hogarth. 1999. An an alysis of temperature induced fluctuating asymmetry in Asellus aquaticus Hydrobiologia 411:139-143. Smith D.R., B.J. Crespi and F.L Bookstei n. 1997. Fluctuating asymmetry in the honey bee, Apis mellifera : effects of ploidy and hybridiza tion. Journal of Evolutionary Biology 10:551-574. Sornette D. 2003. Critical Phenomena in the Natural Sciences: Chao s, Fractals, Self Organization and Disorder: Concepts and Tools. 2nd ed. Springer, Berlin. Stanley H. E. 1995. Power laws and universality. Nature 378: p554. Thornhill R. and A.P. Moller. 1998. The re lative importance of size and asymmetry in sexual selection. Behavior al Ecology 9(6):546-551 Trotta V, F.C.F. Calboli, F. Garoia, D. Grifoni and S Cavicchi. 2005. Fluctuating asymmetry as a measure of ecological stress in Drosophila melanogaster European journal of Entomology 102(2):195-200. Van Dongen S., W. Talloen and L. Lens 2005. High variation in developmental instability under non-no rmal developmental error: a Ba yesian perspective. Journal of Theoretical Biology 236:263-275.

PAGE 108

95 Van Valen L. 1962. A study of fluctuat ing asymmetry. Evolution 16 : 125-142. Vicsek T. 2001. Fluctuations and Scaling in Biology. Oxford University Press. Vollestad L A, K. Hindar and A.P. Moller. 1999. A meta-analy sis of fluctuating asymmetry in relation to hetero zygosity. Heredity 83:206-218. Waddington C.H. 1942. Canalization of deve lopment and the inheritance of acquired characters. Nature 150: 563-565. Watson P.J. and R. Thornhill. 1994. Fluctua ting asymmetry and sexual selection. Trends in Ecology and Evolution 9: 21-25. West-Eberhard M.J. 2003. Forward. In Developmental Instability: Causes and Consequences. Edited by M. Polak 2003. Oxford University Press Whitlock M. 1996. The heritability of fluctuat ing asymmetry and the genetic control of developmental stability. Proceedings of the Royal Society of London Series B 263:849-854. Whitlock M. 1998. The repeatability of fluctua ting asymmetry: a revision and extension. Proceeding of the Royal society of London Series B Biology 265:1429-1431. Wilson K.G. 1979. Problems in physics with many scales of length. Scientific American 241:158-179. Woolf C.M. and T. A. Markow. 2003. Genetic models for developmental homeostasis. In Developmental Instability: Causes and Consequences Edited by M. Polak 2003. Oxford University Press. Xia J.Y., W. Van der Werf and R. Rabbi nge. 1999. Influence of temperature on bionomics of cotton aphid, Aphis gossipyii on cotton. Entomologia Experimentalis et Applicata 90:25-35. Zakharov V.M. and D.V. Shchepotkin 1995. Effect of temperature on developmental stability of silkworm ( Bombyx-mori ) strains differing in he terozygosity level. Genetika 31(9):1254-1260

PAGE 109

96 BIOGRAPHICAL SKETCH Gregory Alan Babbitt received a Bachelor of Arts in zoology from Ohio Wesleyan University in 1989. While working with Dr. Edward H Burtt, he co-authored Occurrence and Demography of Mites in Ea stern Bluebird and Tree Swallow Nests, Chapter 6 in Bird-Parasite Interactions, edited by J.E. Loye and M.Zuk. He worked for nine years as a zookeeper and aviculturi st at the Columbus Zoological Gardens, designing a successful breeding pr ogram for tropical storks, and, during that time, served on the Ciconiiformes Taxon Advisory Group for the American Zoo and Aquarium Association. He received his Master of Sc ience in 2000 from the University of Florida Department of Wildlife Ecology and Conservati on. He worked with Dr. Peter Frederick in designing baseline captive reproductiv e studies of Scarlet Ibises on Disneys Discovery Island Park that would co mplement the research groups ongoing investigations of reproductive fa ilure in ibises in and near Everglades National Park. In 2001, Greg began investigations regard ing the underlying basis and proper characterization of developmental instabilit y, the subject of this dissertation work presented here. The second chapter is also published in the February 2006 American Naturalist 167(2) pp230-245. The third chapter ha s been accepted for publication in a future volume of Heredity. After receiving his PhD in z oology from the University of Florida, Greg will be starting post-doctoral work with Dr. Yuseob Kim at Arizona State Universitys Biodesign Institute in the Cent er for Evolutionary Functional Genomics.


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110218_AAAACK INGEST_TIME 2011-02-18T15:09:58Z PACKAGE UFE0013688_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 71608 DFID F20110218_AABMEO ORIGIN DEPOSITOR PATH babbitt_g_Page_060.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
bb2aebe2008ed7de34853d24aabe3e33
SHA-1
51a83f7627603a58e8c3e5e2d93049d1d55867d7
22106 F20110218_AABLZI babbitt_g_Page_066.pro
294e84989a2b7fff2091596f6c3ea8e2
41e5350296ded0039727b380dfc2aa1219164e93
51802 F20110218_AABLYU babbitt_g_Page_050.pro
3e9830b4d4bbdf9458bef411cab717ac
9bfb8b99424c38e271c88c197457a0a5b8380d8d
32323 F20110218_AABMFD babbitt_g_Page_068.QC.jpg
e22425a39e69baeadd1edab7e45809b6
d50cf1008d6d9b04bf66c3c81929081f24480003
21899 F20110218_AABMEP babbitt_g_Page_060.QC.jpg
20e0ba5bc82e2b0ec3f3a1abaebeeeab
ba50b9b987b820e735fb7a303b4cffab72178321
46213 F20110218_AABLZJ babbitt_g_Page_067.pro
363a6ead97843867c36ddef1d2385c6c
b97ac3d2d9fe770f1a08073f90493d4cb97241dd
53291 F20110218_AABLYV babbitt_g_Page_051.pro
6a570134db73482c628867f51e408c3b
fe53ce6414087792e76f1a949420f7348755f5d2
99319 F20110218_AABMFE babbitt_g_Page_069.jpg
c8835d49e25f27f565d74cc5e688f2a9
95cd7f5231af0507b64414bb42c434786a5a6cfc
102391 F20110218_AABMEQ babbitt_g_Page_061.jpg
52aed01f38b5ad0dadd8f765d590d933
57f38391ae468eaf6911b7ad776cba16ddbb14e9
51030 F20110218_AABLZK babbitt_g_Page_068.pro
bff4ea002debe3d743fb61462b9fa047
bf6286b8405eb6e2033006302b9f6d856204eee5
50330 F20110218_AABLYW babbitt_g_Page_052.pro
2d2258028060678d4b0b9baacde937e8
946457ab6ce81f22d8f1fc6f46040e088017a084
33195 F20110218_AABMFF babbitt_g_Page_069.QC.jpg
2df12ab297bbcfd960fc0b557ae2602e
51f580a9f9a9abc3a195ebd4c9cc431495d7c746
49532 F20110218_AABLZL babbitt_g_Page_069.pro
240bca0de61f004921b06c67ff4adfb9
0d47181954fdf08fb1371ca62207148f6c808bec
30225 F20110218_AABLYX babbitt_g_Page_054.pro
6420e2b2e4ed27951beb15da89acabb1
f5d1145ea3b4f51f728c1b65096b836f69171792
110323 F20110218_AABMFG babbitt_g_Page_070.jpg
832a2d66fd960aaf50b0610b5a1827a1
b35e42fa7d51df1ae12a246736b7a7860e3cd1d9
33129 F20110218_AABMER babbitt_g_Page_061.QC.jpg
e716e3bcdf3df0368508b1e0ea60230d
5896ca9ad189d4ec853b7f9d559988478ab5e441
50975 F20110218_AABLZM babbitt_g_Page_071.pro
3c5c90a7ea1069f0d1dc706dd611807f
011313748cef8a8c3b147a7a238bc54296b8e950
27807 F20110218_AABLYY babbitt_g_Page_055.pro
f3400e4ac4e3c241ced3153374a219b2
941e2e891e0e583d5f77ce958a26a1c150ff96c3
36414 F20110218_AABMFH babbitt_g_Page_070.QC.jpg
7653a5d863595731a2f99741dcebca74
fecb00ed3d5d092dc8e0d69f5029536e846cb2c1
96135 F20110218_AABMES babbitt_g_Page_062.jpg
a5bf2ab10f5267955e48c7de6504e8a6
ecd28068a6c373d8c7d04b1044244e3337e6da48
38503 F20110218_AABLZN babbitt_g_Page_072.pro
19a01dc093dd01ed2a0f46a1f95ad9a7
7dc1a5551f6b0c7a1a99f88e6babe76e8aebadd3
17350 F20110218_AABLYZ babbitt_g_Page_056.pro
03ff3bf0c9a4c41b5882935566c3f595
8d78f486c458f8e35eecc734e92d6440ae8c2bda
32146 F20110218_AABMET babbitt_g_Page_062.QC.jpg
f8e4f683b21ca6eae1fbb94c1cad692b
d64911c067381a8da98ed06ee850cb3745c36eea
38950 F20110218_AABLZO babbitt_g_Page_073.pro
21cec2dab5406f18dade02b40903d8d2
f52ccb1ef62f7b2a1f0c51c5025bdeb9672f1f07
102623 F20110218_AABMFI babbitt_g_Page_071.jpg
d3ebd99d70de15ad6e55e77644802576
3a93f5a48d8add99a75be5458de80c9f9c3cd315
62606 F20110218_AABMEU babbitt_g_Page_063.jpg
f800d4a5d5c1c46718f3066ab342c532
f7cfaf7567d17592411cc69d8c2b3da95d3e313a
50991 F20110218_AABLZP babbitt_g_Page_075.pro
4571c557d5ba018b09ebc5f6e1f75484
858bfbc32121a223c7ea774684cb34e5cc243692
34005 F20110218_AABMFJ babbitt_g_Page_071.QC.jpg
01f7aeb7c2557eacb208f16f937ac86f
63c51368b7075af834970a373b1626a8a44f898f
21041 F20110218_AABMEV babbitt_g_Page_063.QC.jpg
633e7fce9158496164a4fba0d61cb4f0
ffb269306bfd7dc963b8d19450437899d3ec9b91
46983 F20110218_AABLZQ babbitt_g_Page_076.pro
3c42a201c6b9ac25370b191b0139922b
563984cc53a2342bf75a73c1048751032dd83f4c
79496 F20110218_AABMFK babbitt_g_Page_072.jpg
3971413e91e477cf51a0fa69313d9399
35e318561e9e4f7fadba7bdd13b84e3cad5479ec
116568 F20110218_AABMEW babbitt_g_Page_064.jpg
f28a1705bbf9416569b046b84c962c69
eb2037ba97c0ecfc3dc16d10b3bf48ca3ac74326
37875 F20110218_AABLZR babbitt_g_Page_077.pro
6ed384a40fffa1c198342bea4a744fe7
694e8a4bd202cd65e6374ae4c9865eb4c0566e5c
77445 F20110218_AABMGA babbitt_g_Page_081.jpg
a106cf4fff5587765259365886f7a1ce
21d5b76a051ebf6057b24ad3e83bb4e2451c4208
25985 F20110218_AABMFL babbitt_g_Page_072.QC.jpg
21cc6102212631b38de2595ec7e79761
d59210fec28ab0c56e9f9fe538cd01d6ae8afe95
100186 F20110218_AABMEX babbitt_g_Page_065.jpg
25b2ffdd5831ed20545a3c736eea9c2a
5706ceb1617c63c20826f33a3e43014464a3de39
46964 F20110218_AABLZS babbitt_g_Page_078.pro
07cdbda65571fc38a974834fb2fe2b2e
234eb3ca13ef76d2102029fcd6bd851cb792c15b
26400 F20110218_AABMGB babbitt_g_Page_081.QC.jpg
38f721b7a76e4eb6e7a8fab5e3512450
b284d6ab84d3adfe162cce43a0ce5b8ab4b1046c
85545 F20110218_AABMFM babbitt_g_Page_073.jpg
c9a40f8470bf7926909f73fbbf7ce6a8
bc1d33b57b3ffe6438359b62f28b499fba8c7e10
32757 F20110218_AABMEY babbitt_g_Page_065.QC.jpg
ba60a1377d84dbded9a3c762cd93225a
e5fc31b4b665dea3c8f4d9986f0a855969f6fbe9
48571 F20110218_AABLZT babbitt_g_Page_079.pro
0c40a6f5877de38e4659e492cf681b78
a34557bc3fed47253d73bf3011173001a8ecfd33
38320 F20110218_AABMGC babbitt_g_Page_082.jpg
76922070d26d0bf00e1e49c2ccc1d791
2b5e4e7dcf992b48cdfc2cca49c45c21fadf45d9
26865 F20110218_AABMFN babbitt_g_Page_073.QC.jpg
cf7c3907cb660dfe3c4ec76b18db1eb5
8cff2cd7a43aeadeb8e9454575d4db9e6e576782
60421 F20110218_AABMEZ babbitt_g_Page_066.jpg
c1a1cab8b93aa57ea508c00eda1b3d4e
971da94c6469354a18723896137422ffccd4021a
50626 F20110218_AABLZU babbitt_g_Page_080.pro
7d26bdf2392ae59bdd89f4474eda7a57
3358e04083e83abd198a442594acb4df7a3a0338
12247 F20110218_AABMGD babbitt_g_Page_082.QC.jpg
e28d3d77c2a8fa56deb613f7d0e7fa5a
c48ab0afe4ed4368e6db945fb5d479a74157d719
102731 F20110218_AABMFO babbitt_g_Page_074.jpg
4709e4248643c4affb1670044714daca
0b53a621b9b3862627c912fe6ed3c674c192657b
35146 F20110218_AABLZV babbitt_g_Page_081.pro
83af21c5e74f0f99f80e88ff8331b57c
7e63659c2e0173b05b937381dd71a8144412b8c2
85649 F20110218_AABMGE babbitt_g_Page_083.jpg
94ab36bedec71fc9c1ea3f91da2bc1f4
238876673a4eead24fa81b0ff00d57a85b78a0bd
33386 F20110218_AABMFP babbitt_g_Page_074.QC.jpg
f0ec36aeb5a54653006cc99bc654e38d
a4342b72de2a9d583401743736c56f1105771932
15013 F20110218_AABLZW babbitt_g_Page_082.pro
179a129922ab66102c821b4bdaa3e484
b5891de52d75cf2a0e1215563e8df0f2e3f0bd48
27633 F20110218_AABMGF babbitt_g_Page_083.QC.jpg
4689c1203c7dda07ff44c3da32707e99
fb8d45a8cee92f2eaac7df1ebf59c8c0030a20cc
32819 F20110218_AABMFQ babbitt_g_Page_075.QC.jpg
a95418220da29a648ffb2bf50c0b9edb
a75d02c77b1978cb914944efe6581719b8267d94
34696 F20110218_AABLZX babbitt_g_Page_083.pro
5ec3031055b8131e22ad1992a9aec6db
38fcf61517dc7b409787ec18ebf19490c45e148f
24476 F20110218_AABMGG babbitt_g_Page_084.QC.jpg
859f793411daec34fe652e61967dbed2
9b1e02d83337be1ef82d33bacfa3631d4a37f494
97245 F20110218_AABMFR babbitt_g_Page_076.jpg
fb453bc4cd174fb317322432bcf585e2
6f979a8e1e08e034fd61078ebd565f5bf87bf4a3
32005 F20110218_AABLZY babbitt_g_Page_084.pro
f1ccc8137c508eb04a31e2961a0d88f2
7b929cf488d49dda717bb0370e38fcd2e59ee4ef
97764 F20110218_AABMGH babbitt_g_Page_085.jpg
65e605d3797d6f377e2b8c657d0a88b5
9a58beb0a3112d8e60f5a163eec1a9456f95322e
31468 F20110218_AABMFS babbitt_g_Page_076.QC.jpg
d6832a883487491adbedf8e9b18bdfdb
4919398d524e23312fd9d0fcf1871730cd600326
51676 F20110218_AABLZZ babbitt_g_Page_086.pro
bc0741ff245fbf92c681f97375e59c20
6d2a3543982641bd859619a205b6896fe91b8b82
31618 F20110218_AABMGI babbitt_g_Page_085.QC.jpg
d6b7c4c0e1f9109db8fd227f7cd55036
c6b4c913e903aef61d3465ad508cb8796ddd83a1
81624 F20110218_AABMFT babbitt_g_Page_077.jpg
37021ccdcca58e295e72d63495fb33e4
df8f78e895bdfafc0051a23612aa892e15ba432e
25809 F20110218_AABMFU babbitt_g_Page_077.QC.jpg
6608c9ac1bd283dbc632c56dbfd57ae4
b087e8d6a8b9c5d445470b6a7a94a0873f676c94
104542 F20110218_AABMGJ babbitt_g_Page_086.jpg
be71b045f680bf387362d610ff877db6
64a3edd9159edb175d037692aa709f1ec2d4455c
96969 F20110218_AABMFV babbitt_g_Page_078.jpg
e5dc349e9aa6ec28bb7e28a416509620
ce701bd2228b6b0ba0d7baca9c43cc050ee94c85
33638 F20110218_AABMGK babbitt_g_Page_086.QC.jpg
9e4c02833d1fc05eff919d00de02f623
066c2be0bb7f1efe1e75c906b44dde18741b3a8e
31853 F20110218_AABMFW babbitt_g_Page_078.QC.jpg
03e4b2823c8859951d31c5892c47c14d
05e4df8c4a41d1740237d1b4fde85f096e240956
83755 F20110218_AABMGL babbitt_g_Page_087.jpg
cda6263310dac2d9f3f21a50466d3dd2
19e6027decfee6091108af1a6fcb1b89b289ca77
99606 F20110218_AABMFX babbitt_g_Page_079.jpg
147b1f4aeb4ff8fd01606b80a91036af
627a5e078c1c406f949c5d1ff13e92ce8c7c3a01
33464 F20110218_AABMHA babbitt_g_Page_095.QC.jpg
61af4632594bd8f4bedade4f453e81c2
e07260c66594c9557c5419fccdc7816a1583affe
27433 F20110218_AABMGM babbitt_g_Page_087.QC.jpg
1911509bf43d2e898265eed5966f4238
237415f9ea6b5da712a5915cae1e309757e6dfaf
33053 F20110218_AABMFY babbitt_g_Page_079.QC.jpg
6053f5016026bcfc17dc5c7e53dc5365
48508e20311dd31977c918b1f39c6de6b11281c8
103536 F20110218_AABMHB babbitt_g_Page_096.jpg
979230513eea1c111a6ec9e4b325fb22
0fdea16846cf068508b0ba2ed3ef6ceb29ede5b9
97230 F20110218_AABMGN babbitt_g_Page_088.jpg
584b5cecefc0310cd74f33584ab21bfc
96d9a6c55ce02422dc1b6ed1593b2f3f499b2b8b
34253 F20110218_AABMFZ babbitt_g_Page_080.QC.jpg
08166827c6f7051e598ded78b3f52e74
ce73a3ae7cc0612e6fe7015f948852226105bbfb
33350 F20110218_AABMHC babbitt_g_Page_096.QC.jpg
1007c891a80c17381f63f7abc0cb3715
e1c4914e700f4b4b36dab60aebdfe07c5c2123c6
31368 F20110218_AABMGO babbitt_g_Page_088.QC.jpg
b5ab0b61e4c6ef98a2aa77e2cb97cc64
d70f52c76ece3c5293ce4c6846bf8e7cb70e5c26
40377 F20110218_AABMHD babbitt_g_Page_097.jpg
2cc31af4438a934913b26bae4575f955
ee8a2098b868f25565111b691e35cac9bcbf2212
99134 F20110218_AABMGP babbitt_g_Page_089.jpg
2e30650d531795c3dd07a28c4c8bd2e2
5a1e07a7ae5f2dfeb6c7bcebe1005b6f490dd0a3
12279 F20110218_AABMHE babbitt_g_Page_097.QC.jpg
77577fbb953d60d77bce45e0975bdd03
de99c06c8a8f84b640065b45e5b0421a2b99532d
32478 F20110218_AABMGQ babbitt_g_Page_089.QC.jpg
1cc5ae42a9f4f9a2ada8b8af82107403
107ff679665819d3cc3a5a7f05643f030f7eb447
33656 F20110218_AABMHF babbitt_g_Page_098.jpg
264aab3ce64bd0369dbd210c54d8f274
fc250bdbdb2cc322b3699a9fb4bf599620128d11
101474 F20110218_AABMGR babbitt_g_Page_090.jpg
6a720799f3d73bdf9858ccec31299799
2a91d934d1c308e6b12bf0a515cd80650d8fe2c9
9758 F20110218_AABMHG babbitt_g_Page_098.QC.jpg
b1377cd19a0c9cf70e57048bd700dae8
9382e5c5318e235eb0d92394fdc639fb574d1611
32741 F20110218_AABMGS babbitt_g_Page_090.QC.jpg
9435f9388fc752762e8df0988cb73189
7d9fb1522ce34df580b352b00663cd4fff6975da
54220 F20110218_AABMHH babbitt_g_Page_099.jpg
3e20fd0c467949813541ce70be38348f
d0a2caa8decb54a8dadfd6daf23bb71197d3bb4b
103630 F20110218_AABMGT babbitt_g_Page_091.jpg
1487f61bf388ce75c3ec3af6b9abd923
4fa8731d75a3ca703398c7e15885c59b66af297d
18514 F20110218_AABMHI babbitt_g_Page_099.QC.jpg
2177a285621661f7eaeaaae3787799e3
f456db6ccc8de1591d9dc2e3eac4a06441dca9e0
34548 F20110218_AABMGU babbitt_g_Page_091.QC.jpg
13b784f27946ad5a58aa10136b982388
55e394914f17ee6823b0dcca26eaacfcc81a32f0
23188 F20110218_AABMHJ babbitt_g_Page_100.QC.jpg
8c9e4e847234971f6c57c2aa3e93e714
7975cefec0a4ee4195e6a31b92a23153333e73b0
31498 F20110218_AABMGV babbitt_g_Page_092.QC.jpg
ae24639118e2e3981b6934a956cd5510
ea02c6e322dd91fa82853d5177f0101f998419cf
107558 F20110218_AABMGW babbitt_g_Page_093.jpg
cd090862764f93a55193837b19f76ce5
89df6bd1bec10853d77d139359333297b88db18e
97618 F20110218_AABMHK babbitt_g_Page_101.jpg
c2697d92591e979c963ad408229fbf7b
355fbc3d1cdd9d71ef037732f8f38ca289cd5d29
35090 F20110218_AABMGX babbitt_g_Page_093.QC.jpg
14592567a8af5a05a6d548626101cb4e
68e25e4f74a92b7382eb11d7985a6fa8c2907575
233934 F20110218_AABMIA babbitt_g_Page_001.jp2
348142fbca7739cea7d8404944a57948
484c54c0b1cfae21e175acf9b8a683ecc53a8abd
28630 F20110218_AABMHL babbitt_g_Page_101.QC.jpg
13cc03a28d964ba55ceebbff4644c4df
69481b90e260864e3210c8fb8819d99cd0b606f8
100633 F20110218_AABMGY babbitt_g_Page_094.jpg
69bb5f102b13f6730ee1a310b0880df9
34af53a00475c01adaa670fd4cfbe1791904feba
31196 F20110218_AABMIB babbitt_g_Page_002.jp2
4ea55e89eeb1e7660a02e3efaac1ef38
63aa024e466a39d40af158834be2a66f5f066f7c
116552 F20110218_AABMHM babbitt_g_Page_102.jpg
a212598307dedbedbdae5e0e10219e0d
e802617dc5e424aea361176a33661dcd8adfd330
31992 F20110218_AABMGZ babbitt_g_Page_094.QC.jpg
39092bc65033a59431dacb9cdda94b3b
bd6325bccd5ea3f64633139a189f271121b5acc3
67055 F20110218_AABMIC babbitt_g_Page_003.jp2
26013adf9232befe6b832ea8415e1278
436a858f7de251e33f8f2f96918e9b1c14474b1a
33381 F20110218_AABMHN babbitt_g_Page_102.QC.jpg
9376552ea62851568acb8f94dbfe4964
621dfcf98bf7f7e320d5bef65353163978a350a0
375035 F20110218_AABMID babbitt_g_Page_004.jp2
dc3dcd49e8de0efa8803b1052e98f53c
bcda736f5f95a7b73dda24f43f6a108f1642921f
131816 F20110218_AABMHO babbitt_g_Page_103.jpg
d3574452cb79b1d4b06fdc9c5ea1425f
a301589e9eb1a218cb5fa166924dc877b97db97d
860961 F20110218_AABMIE babbitt_g_Page_005.jp2
1e112c9639d0c2ce5d8683850a5c317c
88fb9c6e7e9067b13a257638a598fff2c9ac249a
35755 F20110218_AABMHP babbitt_g_Page_103.QC.jpg
bf58fcf1fa44a522ec5594d7994c2d86
6b61fb094911d5ad6a05845cd733bd3819e7920c
1051906 F20110218_AABMIF babbitt_g_Page_006.jp2
f7d3bf8980cd5624a75441d4651d7084
e2eb146b09bf601700f24fe338250fd2b3e5ecbe
127856 F20110218_AABMHQ babbitt_g_Page_104.jpg
18a4584d10df86aae10a22cde2f43b98
80e9024ae6fb95ab2360eb7d392300372771a372
1051975 F20110218_AABMIG babbitt_g_Page_009.jp2
bd7167d1c6ade5743ff8df9780016e77
7627667a17cb5473d53d61c0649b6bbc63e6b85c
35141 F20110218_AABMHR babbitt_g_Page_104.QC.jpg
565603761c986337b97ee1aceef6a7d0
ec3e128d13e60e99c465e6d803d690a83f6e8710
1051965 F20110218_AABMIH babbitt_g_Page_010.jp2
51fda109d4add63a780d3ef271caca98
d12df07f26f8eecabe05eb5e1ba04281b34a579d
112668 F20110218_AABMHS babbitt_g_Page_105.jpg
4744c04273e95737e7c31739c50718f3
8197898be42cb997acf8bbf7454c77397fd9efd2
83449 F20110218_AABMII babbitt_g_Page_011.jp2
73613013e51740caad0aaf337008f762
e26602eb69970ad3c957f7c25867e6073b78b6c7
33251 F20110218_AABMHT babbitt_g_Page_105.QC.jpg
8973ef0587339c884027d10c583ac3f9
86dd68adb4f6ce1bc02f8a92cc51a17892ac519d
876815 F20110218_AABMIJ babbitt_g_Page_012.jp2
5f1d5d1e2a9b9d731fbebd9d2227e18d
1601bdfe14d1b8501e955e9d02215438d3fd8d84
111997 F20110218_AABMHU babbitt_g_Page_106.jpg
95136cc2e1626caf5e93134975a13204
88ed3cef6d45101e6004828dfdd0398e08db0a8e
910446 F20110218_AABMIK babbitt_g_Page_013.jp2
f9ab7255030481ffcc915ca5dec700d0
2a7a01dc06b752a022a3c119a1dc1646454ba84c
32503 F20110218_AABMHV babbitt_g_Page_106.QC.jpg
728b0bcf760e54531af7cf9ae87637d0
be7be4eabbda5fd77c825ee30f889dcbf81165da
110633 F20110218_AABMHW babbitt_g_Page_107.jpg
b5ef654a00636d0018b95ce5f1c21d96
037246c450aff7e8b933999232017ffb4eb9a301
1051982 F20110218_AABMJA babbitt_g_Page_034.jp2
58d5ca88ceb5694469caa30393ca1ea9
6f882aa6f6f9a822efd793724d3097442822b172
1030336 F20110218_AABMIL babbitt_g_Page_014.jp2
3e1ea97bf01d5895496d8c01912b912d
3121c94d4056d0a22798023fda31b3f9652722e8
28019 F20110218_AABMHX babbitt_g_Page_108.QC.jpg
8d4d365a0f3f4a538276aa84a0274369
b1c579b4fc07ba539f889446e9f0019befcc32be
1051910 F20110218_AABMJB babbitt_g_Page_035.jp2
874452b217d7bf16be7b9b2d1eaf058c
6935735327a9e9e0ce39258eb4944681be9d3936
F20110218_AABMIM babbitt_g_Page_015.jp2
d72e3b4aea2d5221931867e695fdecd4
9795557d054720eb48bd1f84b1de27ee53307d12
89928 F20110218_AABMHY babbitt_g_Page_109.jpg
6eb333629847b2df0f50a1f06b1af6e9
b1367834d29076dd821814b211bf5ffec2bde341
1051979 F20110218_AABMJC babbitt_g_Page_036.jp2
618253ef0049251222d13a9de4e4a50a
50e6957b06238acb9e0f38c3678740ad5e603ab9
1051964 F20110218_AABMIN babbitt_g_Page_016.jp2
eccfa7056ce0a5a39754ad3daab909d3
588f429edacc48b9804f6070fc235fba597a0d0c
29818 F20110218_AABMHZ babbitt_g_Page_109.QC.jpg
a4a756f6591155472bd8bded9f40de35
6fb0a60f49eca8f9083e46d757fc00d0005214d2
897818 F20110218_AABMJD babbitt_g_Page_037.jp2
4980e5c6bd1794650febd5691e80f3bf
57c9da85091f537d414d66b7cf5dc55d66b71447
1051967 F20110218_AABMIO babbitt_g_Page_017.jp2
b1357b9ffe76709041010dd4892775c0
9371135a1f4360b6bcc88519a5a57643ade98902
584574 F20110218_AABMJE babbitt_g_Page_038.jp2
5e61e09ad411c0544c6562404bf570c7
b6d18b92eaf36acffb5e733c4ed0b4370037a83b
1051917 F20110218_AABMIP babbitt_g_Page_018.jp2
3eecc15fafbfb9bf9b0d2c4d5535437b
0155b090358be31175020bf2a801204a7ed0e689
654303 F20110218_AABMJF babbitt_g_Page_039.jp2
920962ae8c9563249d306b3eccbbf46b
3824cfb6ad0c6ae95e5db66d083e2cb35b360969
1051984 F20110218_AABMIQ babbitt_g_Page_019.jp2
a9d3b503095392e9fdb14e8010fbbba1
3accaa12d6de9feb91c0c7eecd4917ea9452e451
849977 F20110218_AABMJG babbitt_g_Page_040.jp2
ac1c4e7d3d0d6e40c6defc204fce47e1
230bb7b130f2dd6ac1dafb491add5885c07b5a5b
1051983 F20110218_AABMIR babbitt_g_Page_021.jp2
3d3238459169e7c081a970d7a847da2c
f75937169be6370f53c1b994b69976d88e904cde
735986 F20110218_AABMJH babbitt_g_Page_041.jp2
bd9837db2d7870aecc09fb477086b01c
b3ae5979befb570dab3206f5ecac2d92156bb66b
259396 F20110218_AABMIS babbitt_g_Page_022.jp2
a57ed41f6ffa34c1a8e9880e2a5927e5
e446489ca21d38511b3e533988a4e78819825a36
998381 F20110218_AABMJI babbitt_g_Page_042.jp2
e722fc95b4cd763e4df7ff50713a97c7
9be2e05932fc13d4f376f435a402e4dbc63ea75d
991043 F20110218_AABMIT babbitt_g_Page_023.jp2
051866922c9395c86b0872ea06b9793b
b2ad234c9162519ce1119d2d35e62f569451510f
852544 F20110218_AABMJJ babbitt_g_Page_043.jp2
200ae2c564524ff728b8d38066d21438
e3b8d9543275f9d6baac49a5e8ca0ab7160bf3ca
1051966 F20110218_AABMIU babbitt_g_Page_024.jp2
334aefcce3e87e04bc90475e33434b67
864515af3d01078e0868adf28159fc2f76b73725
1051930 F20110218_AABMJK babbitt_g_Page_044.jp2
7354de5ab8d52b1012ae5cf60076a3c0
9ce63dd8e374c3ae65d7614c1fbae3750c67cdcc
707337 F20110218_AABMIV babbitt_g_Page_027.jp2
2533cd2cec4d5b49c6c4d89593373496
a4a43edd9a5fcac27dbd2bf138c39ce2e8b6c919
1051921 F20110218_AABMJL babbitt_g_Page_045.jp2
48fa43f1523c7c1dada5d19e5c224c54
31813e01c3730f1f92b12c815172db9097ae468f
1051924 F20110218_AABMIW babbitt_g_Page_028.jp2
806cb7e290e2fbb9c98d8f994a3ef505
cddc4c50d8ecfe588b764b916a61b081ee266c83
1051946 F20110218_AABMIX babbitt_g_Page_029.jp2
fd60ece40fcf16c49d911390c9289dbe
6bf12d3364ffac6d87fc2d14f95b25815bd39826
1051931 F20110218_AABMKA babbitt_g_Page_062.jp2
b1c3bb0784fa3efb7e251fd1d9b52df3
1b422dba504671a48dc1afad4c39f4f8768a56b0
1051978 F20110218_AABMJM babbitt_g_Page_046.jp2
95d56ed8a6ed602dd6ce4e9a97e36a4e
c19e67d0542827371a646a6c2e1c7878702108c1
1051953 F20110218_AABMIY babbitt_g_Page_031.jp2
6e49735e878198bdfdb99e8f42447179
acb2c16dfa3b4719044b6fe4340d3dc74d15f6de
580855 F20110218_AABMKB babbitt_g_Page_063.jp2
c67554331b040a14a921ec448f590293
f2d7b8ee446fce535dfd15a6c07f4411edee51e2
F20110218_AABMJN babbitt_g_Page_047.jp2
59e4eaab562b3b8a29d6c0a5b9dc27f4
536094c784e09ee3d5d8c9c735902b9a0a645e8c
1051985 F20110218_AABMIZ babbitt_g_Page_032.jp2
45633ba3ee32b42e3b337de68cac369b
f1c7213cfcf27d2b38f1cedde24e45b014d6232d
F20110218_AABMKC babbitt_g_Page_064.jp2
15275592366eb443bf48bbd63dd7bd64
5fb8b89580ad1b03f6336437bf56af40e9bce952
F20110218_AABMKD babbitt_g_Page_065.jp2
ce88f2e61826640ef6f65c67cb32a19d
b4158a1d8a3fc02b56720cb4959928a8aaedd939
430124 F20110218_AABMJO babbitt_g_Page_048.jp2
a59ad9ee75158f2b46e331f62b3f9f95
fc188891cbd4341ee680e8ec656f9cdc0c4c9942
574942 F20110218_AABMKE babbitt_g_Page_066.jp2
a28f78484d5c7807a0a7abffa37ec82d
8852a9fc1587d92af120602c5866105a3993f4be
999189 F20110218_AABMJP babbitt_g_Page_049.jp2
7c0c1b9b46741525f106dc2c535e7c66
54691e2e067d5dba1f6818637fb5c90676596813
1048038 F20110218_AABMKF babbitt_g_Page_067.jp2
75f9f5f3821d48a38ded6d1dca62b89a
77350d778cc8fdd875c2de9d62e8bb54b4dc5efe
1051903 F20110218_AABMJQ babbitt_g_Page_050.jp2
37c38ecf23bf97839f6e575d1dfdb231
3c994eb947160fabfaedd45c67e3eedbeba39a1d
F20110218_AABMKG babbitt_g_Page_068.jp2
c1740cc068c6dd9d87cc6d95f35cad05
de9bf40e305cfceca7dab713f20809d2bbd3568d
1051986 F20110218_AABMJR babbitt_g_Page_051.jp2
75ac5da91461a145ba35de11d20957a1
0c41ad591105457dc0a742eec7a261ca23076ad6
1051962 F20110218_AABMKH babbitt_g_Page_069.jp2
58f1fe5995776ecc98aedfeb3ed10183
4009780412152eec61afae87743461e5d4f736d1
1051973 F20110218_AABMJS babbitt_g_Page_053.jp2
e01d7dffd447d393e7ff2bc950783cec
b6e22ee5bc02f40b31b30d63c8c20279f50f488f
F20110218_AABMKI babbitt_g_Page_070.jp2
bb3fe211cebff6fd28c0161fc9cad9a8
ec14cd464b7f7a4a3666f5955794433c8d4af2c1
688829 F20110218_AABMJT babbitt_g_Page_054.jp2
d327dd25de14ab3f64aca0d3d8e58175
06b6b9419b580e39faf47c1845c2478052363fbd
F20110218_AABMKJ babbitt_g_Page_071.jp2
06c42968cfd9c07f0b5d7d4c44ad81e8
7f072abba2fd2f98236cfd1c74cb79f80cc0e3d2
775334 F20110218_AABMJU babbitt_g_Page_055.jp2
10000853e00c4242be0d8eaeb9f17181
66633b01d8a0b772ae13ec213be673600e80a26b
909506 F20110218_AABMKK babbitt_g_Page_073.jp2
558b3888337734604bad6b1bf2e3be3b
964e83ca63f7030cb0f94022aeecf31977de3b00
471010 F20110218_AABMJV babbitt_g_Page_056.jp2
a3e175856442502e4d0e06e699f63122
28ee867dabbf7a8dc8ffe4476958dcfdf4c45e6d
1051893 F20110218_AABMKL babbitt_g_Page_074.jp2
c51e020891bc87d0860c659b48bf4614
1e0cbce724bd6539a066f099245949a57786ac38
613094 F20110218_AABMJW babbitt_g_Page_057.jp2
5e5707f4e64bbc2cdf4c0ffedae6b4fc
48d75a22bfa9eb331f7880c24cddbf8823b8b7b9
1051971 F20110218_AABMLA babbitt_g_Page_091.jp2
7a8d6b47f019d723bc646b4cdd568a9e
7200c290803dd2921b22dacde8463da7600fc139
1051916 F20110218_AABMKM babbitt_g_Page_075.jp2
d23ebc0f29222268ca569d74baf38c31
bb81077b40b4e813036ab7ce4a38ef5be0e76173
934044 F20110218_AABMJX babbitt_g_Page_059.jp2
2aee4126fa50fa1a32213afc272f3dc5
4108032f200e9374fcdba842c50256c80fdd8d1d
F20110218_AABMLB babbitt_g_Page_093.jp2
b01b6b6a6de9af25ed3685ded8a4c356
8e66fcfd98b42fda85aa7bd268f2ce5293f84d66
816009 F20110218_AABMJY babbitt_g_Page_060.jp2
0da4fb64ff5a0a25d1e53fa860f3dcda
1f9d0484fe9abf080153c3157f124e2803bca4bf
1051958 F20110218_AABMLC babbitt_g_Page_094.jp2
dc6e34b249c52887c98d511376f3f2ea
3d2b890a3b786800f635cab166f29f05fe6f1fdc
832983 F20110218_AABMKN babbitt_g_Page_077.jp2
30c8184441e5d21e9eaf9fedb4aa0ceb
f417b246958b55867da5accdb195d09592f1c6b5
1051980 F20110218_AABMJZ babbitt_g_Page_061.jp2
8762d30c53c8c3633b8d11a3a29838b7
da8187c23867fb550400c86533e6ba15662d68be
F20110218_AABMLD babbitt_g_Page_095.jp2
fc76799fd33a3a92f224991b33573414
6fbb290e7f174f27d524424b15e1c9bc8a443450
1051970 F20110218_AABMKO babbitt_g_Page_078.jp2
ba226c19a6bf13e3f0b160012e1b1515
efe838a197647c69885682e898e211589b06c57c
1051913 F20110218_AABMLE babbitt_g_Page_096.jp2
656e20275f7a1e42fa966ceb479a38b7
7c4e8e7df4b5371280bffc02129180f1622ec9e7
F20110218_AABMKP babbitt_g_Page_079.jp2
9f70b66750941d883a61ab7a7882797b
0c3257927caa032f44da5a5d1d7b444ed9afa665
496188 F20110218_AABMLF babbitt_g_Page_097.jp2
c67f869e136263f05749281f48da1b0a
8842768774ab5384ed7609b97e5721345908a1a1
F20110218_AABMKQ babbitt_g_Page_080.jp2
5ffdf564500afc3c54149d1bb576b3dd
eb24ab67a88e597401b6dcf9f90d3c584299fe74
452517 F20110218_AABMLG babbitt_g_Page_098.jp2
b780ed59396032c3e08a3f3a6ab84b5d
a575afb06ff8cfdac60e573d8fe5385ecf96f3e2
831380 F20110218_AABMKR babbitt_g_Page_081.jp2
ce9c8140b25e5c600a55a35bf4bcdd70
8be63f8f84aea9452fddceddfd1208329ea6c804
581177 F20110218_AABMLH babbitt_g_Page_099.jp2
c42cdb95aed442c60777f14183d8dda7
130784e75e33bce73949ee149a01620f580b0c4e
361431 F20110218_AABMKS babbitt_g_Page_082.jp2
dcfb28d5e47fc637fc05a5030f2e8084
9b2b98238fba6bfa0aa4974e536cb8c03e90d0e1
756334 F20110218_AABMLI babbitt_g_Page_100.jp2
2f2eaa0e6dc546db8f7c119deefa0d9a
8b14736eb47879e12c833e7f4c2b28c8617822ab
796523 F20110218_AABMKT babbitt_g_Page_084.jp2
a5d5f4136845586416b0ce65d3958a5e
0cd975d2a0c70deb33a3a5cbc04de42ba1f252ef
1049353 F20110218_AABMLJ babbitt_g_Page_101.jp2
73ca89aa58e47f7c7422b88f8a9793e6
d5718b0f49148ac7ca9ce11d0f26fee675bd746f
1051981 F20110218_AABMKU babbitt_g_Page_085.jp2
08f141883ba947e5aa386100d06c9943
025baa6367e07ae704c3dbddb3506a3b88067094
1051950 F20110218_AABMLK babbitt_g_Page_102.jp2
8990e1f2f23c6adbeb8ffe5f41608983
0df2f6d17afcc75438597507e9fc77eb81352f13
1051954 F20110218_AABMKV babbitt_g_Page_086.jp2
adb9c136e52f535f89e45e3de7967f34
273b1de13163a3692e2a486a9ad3083027245c97
1051892 F20110218_AABMLL babbitt_g_Page_103.jp2
29435d41b32dfcd2df2d033220b9ab73
16a07a46a06027f2990612aa20b4ac17c3f75bcb
898730 F20110218_AABMKW babbitt_g_Page_087.jp2
cde14a11a4b80b9660415468be12aa29
799804bc27da57854e43fa606a9e271634707da6
F20110218_AABMLM babbitt_g_Page_104.jp2
6dd05bfcb38981762063422416313f9a
af9b6bd4eed3b41d773a05dbb138f7d2a553b746
F20110218_AABMKX babbitt_g_Page_088.jp2
8bf2b8a8b96d69389ceee685df351170
64e1aa501594ff5be03066ae4cae00e385c312cc
6458 F20110218_AABMMA babbitt_g_Page_012thm.jpg
f75e273523923a2155b7466ccfdc8953
cfd4ab76ca7e8fb9dbf69e9e3ce45047fd1d717f
F20110218_AABMLN babbitt_g_Page_106.jp2
fac5939261d49049cc494f51f2d38c8d
c8079146bcd292aecd55651a5feb53c4368a7691
F20110218_AABMKY babbitt_g_Page_089.jp2
68d7f006f49bf7b8aae6d977c0d0bbe0
5f0a29ae139ddcd39d6a50c4d9841456bf59f9ac
6647 F20110218_AABMMB babbitt_g_Page_013thm.jpg
4c5069c0f771074eb530eefe4d3e22cc
f9fa063a4254296a0ed536713f4aff9da9fe2679
F20110218_AABMKZ babbitt_g_Page_090.jp2
58f3af8590bd419d7747f3d32a5bcb36
cdf0c9a0f483f6179c725d9f35e57409a6c7d63a
8436 F20110218_AABMMC babbitt_g_Page_015thm.jpg
31629851c87ecaa67c78ed4ecfd5943a
585967dd73bf8c7f244bf6c7b2c9c7ef33a385d8
F20110218_AABMLO babbitt_g_Page_107.jp2
d8b883db5560370d9a38f1f1aef570cf
11a3585eb38dcdf2b7852b465fcefe0a8a9dabbe
8308 F20110218_AABMMD babbitt_g_Page_016thm.jpg
a2f051aefeed298eb891c52c59f0a0b3
e6dedf2f30d561d66ae34f3aab446dcfed5ea226
996295 F20110218_AABMLP babbitt_g_Page_108.jp2
7ebbe453f2c8f8ae383b84d4fe5a634e
52dce5617a5592562219bad2124f858239bf4610
8285 F20110218_AABMME babbitt_g_Page_017thm.jpg
7fe9562d12eaf11c166c34172df793b2
67fec8a83823c59e582c089daccde41082bec3a3
972106 F20110218_AABMLQ babbitt_g_Page_109.jp2
cf1a22c1e7f2176bd93d480bbea88a48
bb8f5735e1c9d4ecbada03bf20faa6f10ae00ac7
8474 F20110218_AABMMF babbitt_g_Page_018thm.jpg
f0af1525943d23c31b5b450bfbd9c808
2370ffeeef49a9fe0766d52df3634183a150fbd5
2178 F20110218_AABMLR babbitt_g_Page_001thm.jpg
2fe61915b630b1ea2fb7082818b09f2b
6d73735d959e00661dffcc20150d4f4f69ae723d
8103 F20110218_AABMMG babbitt_g_Page_019thm.jpg
c382d8c53ff9091a295ea15279c09371
d2de05edb2c796cfda3dcbee29f94fa60706c435
637 F20110218_AABMLS babbitt_g_Page_002thm.jpg
af698b18ef33d8d3b8bff653df25e14b
01b73c8cac8385c11a9aad270b6b0f8d956178f8
7774 F20110218_AABMMH babbitt_g_Page_020thm.jpg
4a74f0adf72499408012b987cb718982
746a0be6d2606181bac8f4ed4e9edb2d20ff015c
3108 F20110218_AABMLT babbitt_g_Page_004thm.jpg
dbd61b3bc2dfc8a50d947048fc2c067e
4e1fc7d69751588fb9b21930c3462837fc67a10a
8510 F20110218_AABMMI babbitt_g_Page_021thm.jpg
bc8d663b0ac2ea5bfaef671476531958
492e27fceddde9252ca2713364ff0fd351c800f9
4895 F20110218_AABMLU babbitt_g_Page_005thm.jpg
fdaf3888b2dd3085ce83fcfae22bca56
594a87fd93a1ba3d24771f4d6c7b8e692ac78e03
2191 F20110218_AABMMJ babbitt_g_Page_022thm.jpg
a279c819a528dac3e35e887b7a3765e7
c57024b8c093440063e60a3f72b1a4b66e57c352
5769 F20110218_AABMLV babbitt_g_Page_006thm.jpg
ab17fbbb71ecf09a646b211ce89384b5
e0aa8c5aed648eab701fc85f1d6d36620c237f0c
7520 F20110218_AABMMK babbitt_g_Page_023thm.jpg
a440131c1f0f56755d6276bcc63308a6
62bf016214530048ad8f8c8f93dfc8e0e8296618
879 F20110218_AABMLW babbitt_g_Page_007thm.jpg
2acb0e95c46a9a682c0059bc5e6ef3ba
9695ee4c17f24c4df22977175c9777d69133a6e6
8340 F20110218_AABMML babbitt_g_Page_024thm.jpg
5f45e8da2e36139db582dea43c00b822
ff575f81eba357a78e427aa114ad762953ee6d28
6417 F20110218_AABMLX babbitt_g_Page_009thm.jpg
4c3ad540e4ce893de97bc2b7468eb88b
84a306ee057dab9aa158bc66ff5433be08f547cc
5940 F20110218_AABMNA babbitt_g_Page_040thm.jpg
7b16ec0a9fa0c7e08afbb84d4336145d
d7bf1b71e7d3c063dcb13bbd2d6f72035bc4e63d
F20110218_AABMMM babbitt_g_Page_025thm.jpg
2ab92e209078080e7b9c2b5d592dc6b0
e6ab12577b07875f184cc7cafde02c6bb05aaa9a
7562 F20110218_AABMLY babbitt_g_Page_010thm.jpg
c305c8e8b5e6d021b9c1db4035ed61f9
6aabd168dc4f149c78d5912c951c836360ee9a33
6239 F20110218_AABMNB babbitt_g_Page_041thm.jpg
2d86691539f11277898bdc03be1b0787
a0f8eebe550f7a0e1f349d0958c2088ac119d7ba
8457 F20110218_AABMMN babbitt_g_Page_026thm.jpg
947d93423ae6a04ac67dd940d2a71bcc
59ce728de292428549e6d43867536abb9669ea2a
863 F20110218_AABMLZ babbitt_g_Page_011thm.jpg
9a9ce1bbbbe1fda186830920e860159a
0e54de900abb617da75ad479f157ea99b77cf9fe
7257 F20110218_AABMNC babbitt_g_Page_042thm.jpg
17971544c75f5346bc5829191777707d
524ba892637a580a6acc51a7d153b703b5932dce
6204 F20110218_AABMMO babbitt_g_Page_027thm.jpg
b0725544ff67a559ceb637b961bb19a9
dbe8581b26ae3a8c24e349cc8b89ae73621e68b2
6900 F20110218_AABMND babbitt_g_Page_043thm.jpg
0cf73662cd2d1cb410d1d7cf1eda080d
a02bebdace8998dec9f3814031d11e00460f3133
8455 F20110218_AABMNE babbitt_g_Page_044thm.jpg
7d3e85f2dedab554840482a830b7025d
c13b9a68899244e91188796bd17079a90b685f89
8540 F20110218_AABMMP babbitt_g_Page_028thm.jpg
aa36f7e28c9914d481bebac40a8d64ae
06d5d6086821e3619e40a38cefdf4f2ccb0df976
8432 F20110218_AABMNF babbitt_g_Page_045thm.jpg
2bb1e7678615ff1227350b6efb836789
336b0f5370b3646b68b9bac702c5413d0bd60c7b
8583 F20110218_AABMMQ babbitt_g_Page_029thm.jpg
4d70b6d431bc63253f83cd60ba897ab6
acbdde542ab92ec4efa889d050704bf967f40552
8193 F20110218_AABMNG babbitt_g_Page_046thm.jpg
94b32e642387149dcaadd36d81dec5a1
179dc9bb081f4a2b064b216a46876855b9e51149
8094 F20110218_AABMMR babbitt_g_Page_030thm.jpg
2fdd0c0fdb9394dda223c3aa7d796799
97f35db641463c09bbf2db7e56ac12f25fb3d1fa
8368 F20110218_AABMNH babbitt_g_Page_047thm.jpg
d12e059daaee09ab388f78fae7c17d27
5166e6576dcd6069c1deebb267870971b0e7da79
8406 F20110218_AABMMS babbitt_g_Page_031thm.jpg
bc77b7e7bbb29b2c0860bf9d41fece33
c5ee7216f7a74cb16b0d5847c0f895cef131d17b
3507 F20110218_AABMNI babbitt_g_Page_048thm.jpg
2f4fd27a321a3be89ba60b15e11073d2
d68bfed40ecf12767322c430eff840076a56d87a
8530 F20110218_AABMMT babbitt_g_Page_033thm.jpg
34f31cb724373af5437ea7368c87506e
652bca4b10d3eb19a51c4d352182f683eaadd2ab
7285 F20110218_AABMNJ babbitt_g_Page_049thm.jpg
ba0bed606cce87e4b6e473411b61d2e2
7eb8f51c0535955de1e14f9a8658f8e4bf684fc5
8136 F20110218_AABMMU babbitt_g_Page_034thm.jpg
627d9e368986190c31277429e3f4416f
4a31b70a1ba3c785dc659798ae7b732c5ef2399e
8542 F20110218_AABMNK babbitt_g_Page_050thm.jpg
d8bc67a2d9324ca0061532770395f97f
b42b9b1b28415831a9b1df4046cfd3bb45db0bff
8113 F20110218_AABMMV babbitt_g_Page_035thm.jpg
299225c4c44a095017deef5266f358e3
062749d20cde7f7583f1bacdb13c7b1d3d197b07
8191 F20110218_AABMNL babbitt_g_Page_052thm.jpg
7e1e6645ea174f4e244f9137aa51644a
68dbe630c9afcec4dbab657fc457522474128ede
8758 F20110218_AABMMW babbitt_g_Page_036thm.jpg
c72b9714b037bb3b90268e9d7c5dec83
c3fb625e270beed5ca9a8fdd5fec5d040edc17c3
6378 F20110218_AABMOA babbitt_g_Page_072thm.jpg
d89303bd489c27f8cc247f5611306555
f9b77f4c942e137d8bb55378ad87713d735e7341
5569 F20110218_AABMNM babbitt_g_Page_054thm.jpg
9442760f9970c341fd75e92e7afa6487
1a5d841723236a4ac19eba91bd4d93d810ef4002
7987 F20110218_AABMMX babbitt_g_Page_037thm.jpg
490c1fbb7e13c415e23157876d727507
50aee2e3534a11563340e8c815e33cc9f4f17937
6730 F20110218_AABMOB babbitt_g_Page_073thm.jpg
76d4dcbf11da4392aa48a89a623f7de6
75c0020951a19e3a5de60c33cc68820b666590b0
6042 F20110218_AABMNN babbitt_g_Page_055thm.jpg
cf9702aaa4122480e31d04528e45e4b7
fac3b6e2a66fbb6eb25fd6d64371634c6bb3b2f6
5631 F20110218_AABMMY babbitt_g_Page_038thm.jpg
c46c029ff9ab68b1493419eda88c7e66
fa63459b80377054102783bcbeecaab81a66a85e
8225 F20110218_AABMOC babbitt_g_Page_074thm.jpg
1bace505538efc51f5c55157357ed181
87de6f5a710a386b13e60ef7445f65d049130aac
4687 F20110218_AABMNO babbitt_g_Page_056thm.jpg
fcd49d1f8b793a63d53b4bc64580a167
c37232711ba570edd3f9aa7cb6b3ba2d15dee76f
5814 F20110218_AABMMZ babbitt_g_Page_039thm.jpg
c87134319208ecab0eb4d33323481164
f7fc720bff414d1b306cd2a775154a6eb1d59af9
8297 F20110218_AABMOD babbitt_g_Page_075thm.jpg
746bad190e7fb41382d21a6bd98e78e2
e3c1b5ace59884e27b15e8807b8ba998f1d37cdf
5623 F20110218_AABMNP babbitt_g_Page_057thm.jpg
dc21df1c39a4763c5d036a82c4ca5acb
e19d83e71aade1f8ad0850cf4e904d5e99039417
7998 F20110218_AABMOE babbitt_g_Page_076thm.jpg
d63a89a37b1590e5bfb6b861de67196b
46a779d130eff1f008f3fd0cebcdeebe39691ebf
6972 F20110218_AABMOF babbitt_g_Page_077thm.jpg
44205e2d0cb18dc569f7260e18ba153e
f9d3b14670560c1aa00b0df30efad8c820281f26
8058 F20110218_AABMNQ babbitt_g_Page_058thm.jpg
c58b0f1edc3ab4260d9c694cfdb4061f
c203f6707bd254a5558cf6f828735c3a5cc563d1
7817 F20110218_AABMOG babbitt_g_Page_078thm.jpg
4f394057f34b54564acb33c9598ce393
2a68e1da47db56328b04722cabf7ed00ae01b6cd
6388 F20110218_AABMNR babbitt_g_Page_059thm.jpg
f8bc043d4cb69dec7120495e79a0ca04
5468cb9aaffd68c9c63f76ef9158962994b6d1e9
8255 F20110218_AABMOH babbitt_g_Page_079thm.jpg
74c4c231dc850b66c5e31f21e667e96b
2682d0c5addcca3e1e3a0e4e1c3fc2ffb4a0e5c7
6262 F20110218_AABMNS babbitt_g_Page_060thm.jpg
6eccee2d6ec4d1152448f9d1301a1c2c
aef0bb84890a0fb3458e1b45049e53949c792d4b
8600 F20110218_AABMOI babbitt_g_Page_080thm.jpg
260500ddcb5edca6c6e95ae48d602ca9
7f34a0b6d51fdc97621ddeb4eea528ed66318bca
F20110218_AABMNT babbitt_g_Page_061thm.jpg
c02a609c09e59adc66de08f0c38cfe10
17db6a50e69eecb5ee0e0d815f57ffde862d5d35
3900 F20110218_AABMOJ babbitt_g_Page_082thm.jpg
2ace178d0381925d2ef73ac70daedfee
57f44d6c107b76bf7d2f8c677acd0199f10309fa
7772 F20110218_AABMNU babbitt_g_Page_062thm.jpg
8cd8dfdf5249bea8e99173aa3db96c1d
82c5735736cad7cc319179c34262c824b2840d80
6754 F20110218_AABMOK babbitt_g_Page_083thm.jpg
533027960ce2e908266de3b98c01892b
d31cc551ca47420b5aad581ac7e40467f71b05c8
8764 F20110218_AABMNV babbitt_g_Page_064thm.jpg
3ff62d5f0c5bb9715dbb52c6ded83a95
96ed1497741083e9b307fd5ab063625361a3d8e5
6832 F20110218_AABMOL babbitt_g_Page_084thm.jpg
a31142478ffc12f771bff339e32d17ea
fa2f2aeccd2a4870d0188eb3b107812682a86047
8252 F20110218_AABMNW babbitt_g_Page_065thm.jpg
a7e33ea58b404e9c7cb2bc502b385da8
8f8b9a89d08154d7a50a7e860068ac504e8f016a
5686 F20110218_AABMNX babbitt_g_Page_066thm.jpg
c5674d22d0a06defe21f32056eadf374
ad262392a593423f3449c5ea0150cc5fac4f37f8
8339 F20110218_AABMPA babbitt_g_Page_102thm.jpg
bc78937c13fb87f7e27853aab831b474
b20ca06caa2e212c3fc7073da7d87762731acc63
F20110218_AABMOM babbitt_g_Page_086thm.jpg
a45614e3217b808b91ed7defa459b6a2
39b5cdcc591ad6111c58054e2a7093704b85a295
8335 F20110218_AABMNY babbitt_g_Page_068thm.jpg
ae1519501d86fb26e20ce8d905f7da4b
f0f33799e87b423a31fedcf203c33ec281b475ca
8523 F20110218_AABMPB babbitt_g_Page_103thm.jpg
2bd70ccd1c00644eb0fa9e718c461a0b
90a2ea0d474a8ef9f3467a89fa4be3e19759946c
6849 F20110218_AABMON babbitt_g_Page_087thm.jpg
9095388d41f017c6d7cca156b90e7b48
9b111f5c15cd1e94339a8990729cba188fab845f
7914 F20110218_AABMNZ babbitt_g_Page_069thm.jpg
6aea625ccfecde3900660eb1c9ca8ec8
88355817ee48320f4d1b428699294eb35e7f0cf9
8097 F20110218_AABLMA babbitt_g_Page_071thm.jpg
f8f316c01ae51b0da2c9753e1ca08850
bd3fb3393d196892e0833bd5ae9303de455ad3a2
8654 F20110218_AABMPC babbitt_g_Page_104thm.jpg
ca3ee2e0e3b7f28148b9c19614fdd474
9439283c4e702d5038a8c9890ced4a3a5fa63564
7707 F20110218_AABMOO babbitt_g_Page_088thm.jpg
d218a6d85e2fe32daf3f74416021fc9d
506910522e2ea482246e2062d3bd6d0abd889a4a
803 F20110218_AABLMB babbitt_g_Page_003thm.jpg
36182a5d5467083938143db15bcd5a8a
a644ebf9d07f74a34e1817a04703e3f33123d1da
8208 F20110218_AABMPD babbitt_g_Page_105thm.jpg
c30c2d5d361ffd2506312064083124c1
b1332832e14907d6138f4ad7bae90d0d19012d97
8102 F20110218_AABMOP babbitt_g_Page_089thm.jpg
e20c2fe5c85d31f12c9856b90f13107c
1f87732d7ed1febfca056babfb461edeaa5eed78
F20110218_AABLMC babbitt_g_Page_105.jp2
b7f0fc9f27236c6e464e69a842430253
23ec3c6e8e7ce45d46888d8fbd6524062678f5ab
8537 F20110218_AABMPE babbitt_g_Page_107thm.jpg
96218519b5dd8c8c7f1dbfc0b152cc7d
b433780161da8410ff8e5d77a299871c7813a1bb
8404 F20110218_AABMOQ babbitt_g_Page_090thm.jpg
023813003513856f94d842b3e4517e7f
3004b0877bc27e2391f6d37faec9d1a54e380552
7891 F20110218_AABLMD babbitt_g_Page_085thm.jpg
280be5aaeca9ef942ba33ecd45520537
c3768b0ee72c9d1fd729b08359a3b975c53c6268
7222 F20110218_AABMPF babbitt_g_Page_109thm.jpg
0aed3845624c9b431d5239397ea3677a
3c744fd9741876b9d596ad71bf1bbb44678fc1a3
33678 F20110218_AABLME babbitt_g_Page_046.QC.jpg
747cb2cd838e7d18ed1c2ad45fa37eb8
d30669d70fc2c3675e6cd0dd414b67ec79f3a199
1468765 F20110218_AABMPG babbitt_g.pdf
afffe47d1d749e4e0fc349dc1b7c8513
e4331ca81f6161f0a3962c76da27e5ed6554b9fe
7736 F20110218_AABMOR babbitt_g_Page_092thm.jpg
5c9758359e812d0f4a12cc5111fb4b60
e9caf79abb8f5325d6b50faf7fa3823c827e0b4b
8423998 F20110218_AABLMF babbitt_g_Page_005.tif
87fc491fe9d86a13745390b0fe31526c
52f68d3cb842f86f93a5e31243b3378761913617
129098 F20110218_AABMPH UFE0013688_00001.mets FULL
9b831338f8c020fe91d898fad5751a7f
6567f4f4a4cadb752ff2fd548a0d503655da67fd
8422 F20110218_AABMOS babbitt_g_Page_093thm.jpg
b41c1eec575e1dac8eea5433c26fc303
ecb47e8d3d934a6084fef5690f0a2f90388330de
F20110218_AABLMG babbitt_g_Page_057.tif
20351f91185387417951a343093379dc
418001411471f69bd3e1ab72986429a49b675cac
8135 F20110218_AABMOT babbitt_g_Page_094thm.jpg
54b4a27de3ffcc9201bfb833bf40e5b8
ae557ad0081d0f3129c8f750348afcb430ca08ab
15228 F20110218_AABLMH babbitt_g_Page_008.QC.jpg
8fb62535a5be947512950d42191043ef
a9f5debdd5e9bdcdb5392da20bff3c7b6278a951
8330 F20110218_AABMOU babbitt_g_Page_095thm.jpg
5654e043e3d8b9edb066a37b574e6a5b
70300335f3714a8c783ab4e050409f67aeca4a8d
8179 F20110218_AABLMI babbitt_g_Page_106thm.jpg
d65ad7ffd2598a9adb725377701e6ff9
f70db9de03a1857ea0c35a682517922bf289e473
8451 F20110218_AABMOV babbitt_g_Page_096thm.jpg
f1a1b65b501e2eb5b02c0723e8a52ed5
5f68a780986f054b7ff086e876ac14658b8a5efd
7804 F20110218_AABLMJ babbitt_g_Page_067thm.jpg
5a5a79dc1bfe3c0c5393a5e625121d8c
3eb7c4433accab760b3004c187c68fea08b02855
93367 F20110218_AABLLU babbitt_g_Page_023.jpg
23295c855aa47a90abb706ee2cf9a7bb
e72cbf17066d1034149ef0397977251baafe478c
3458 F20110218_AABMOW babbitt_g_Page_097thm.jpg
b1c5cde503a2a425a81264e270533659
2670104a42f921d2c44e3e8a3e8e355994d735f7
15876 F20110218_AABLMK babbitt_g_Page_056.QC.jpg
df6f6ad3344d83ddee71939e5f907b10
09814f8d4d5c297b3918ea4b2ff80adff50db0af
F20110218_AABLLV babbitt_g_Page_058.jp2
89beb6b8e8428e8acc03527700f98617
4b44b6e5e45c2d44192400dde472154e013cb57a
3035 F20110218_AABMOX babbitt_g_Page_098thm.jpg
51cb1a5314e3f2ef36c588ef0207d59a
8913e461ed008888859f42396efad1579f1b7fbc
64086 F20110218_AABLML babbitt_g_Page_054.jpg
55ef0638bbb35097a1db363fdd2a5799
d654bcf161426e59cec2314288673791f97b8581
1850 F20110218_AABLLW babbitt_g_Page_062.txt
879c91384f199b32a3341b2729cd5168
d265a800a114e06c9b180ce313a1afae687b0d26
6281 F20110218_AABMOY babbitt_g_Page_100thm.jpg
49c0ecae249300617ba20ccc565330b9
d46f35dbede323a0721143627695c236d2e75c4a
1036683 F20110218_AABLNA babbitt_g_Page_030.jp2
458115820c548949f75b8a9f3f977558
69a5a8dbf594cc296b0b083c65a25bfe87393cd5
134448 F20110218_AABLMM babbitt_g_Page_010.jpg
60dd2ce96970253de2d7343fe0c301f4
c1d7af20d7dabd6f4e6c5ed421a8c4320a818b14
90146 F20110218_AABLLX babbitt_g_Page_037.jpg
7e4f7ceed848809a3d842510948854ac
d78f668a03d56734359cbe6dc413ce1614a94559
7184 F20110218_AABMOZ babbitt_g_Page_101thm.jpg
4dcaf06864161348d5b6a7c37753209c
36ca7a436f207aeb5ef8017d67524071dca5bc05
2009 F20110218_AABLNB babbitt_g_Page_068.txt
125c8c9d2bf7cae161fbe2c21a0d623d
6df300bcdbb4357f04322fa40163a4865a1117b8
F20110218_AABLMN babbitt_g_Page_027.tif
8e7003f8d2912cad9fb14dd42f5aa35d
09271430229063e59da54406b7d0b73443106ed1
75518 F20110218_AABLLY babbitt_g_Page_084.jpg
d5be1993845272ea1de48b04f2cda538
6b21dc692cef63f95816a2accc4faa5887d048d3
103604 F20110218_AABLNC babbitt_g_Page_095.jpg
8dfbfdac155605903861200c0b712463
81544a888fdb33fc677a8f21414485deb90ef7cd
69522 F20110218_AABLMO babbitt_g_Page_005.pro
4c9c8e36b7f39eb847354d1b8d3fae98
45ae779501989ba7be134ac143276de4308da6d8
70471 F20110218_AABLLZ babbitt_g_Page_100.jpg
667b08d1da350928b4b078ab41484b2f
dccac6a028fe4f2b0dff2939c270b1a141f289a9
4090 F20110218_AABLND babbitt_g_Page_006.txt
b6548283f021a98a43b9d3e71006eec7
ceeeda949234700d1dc91eeb4b592f5b6c2239df
1683 F20110218_AABLMP babbitt_g_Page_081.txt
67035787725c258d8ec2ce02ab0c774a
a484a479fae05edfcc3018e34a5350b7011d9876
54485 F20110218_AABLNE babbitt_g_Page_070.pro
cbe43d247a9759265fa0875186e70e47
54c6e5fb12d86af61a0d56ca6e93e4acf153655d
6927 F20110218_AABLNF babbitt_g_Page_081thm.jpg
35889f8aa9a9251e8a80e6f40bc0c4da
2eaa78fe44a34c2494ec91b7efc582ff6ffc8c48
F20110218_AABLMQ babbitt_g_Page_033.jp2
158a8a66d3616c9447bf8ce66ed2813e
c80ddb6e461d6290ecc6a37c089912f0fc3be817
F20110218_AABLNG babbitt_g_Page_074.tif
1b08863ffcb48c536beb1b6c6730119b
d1aad394785549a4feb1d8c4620cca91a9606556
F20110218_AABLMR babbitt_g_Page_087.tif
661e3d4707a6a9fa154df1a5fce1adfc
46456779ff569f090687d46085ee483cac0d27e8
35949 F20110218_AABLNH babbitt_g_Page_064.QC.jpg
d5ad7ae25046c6ed193c4cf6903b86f0
274fcb5ae83b7b85d43144fed60154949f864400
8504 F20110218_AABLMS babbitt_g_Page_070thm.jpg
9cf306390750e74b179ca6a9798a087d
5ec731f571edf01f8c3880e0a1ee9c8ed43aeb7c
8590 F20110218_AABLNI babbitt_g_Page_032thm.jpg
03e3dfe3ee0d30adf599a50b2e61066b
15144658d516bbedc93f0a5971625e202a0d5f8e
867128 F20110218_AABLMT babbitt_g_Page_072.jp2
a623b0b4b55ce3e8b728d17db765896e
ea2d75b91c6e3eecb1026bef78de2e4e2cd5159b
74229 F20110218_AABLNJ babbitt_g_Page_041.jpg
c8f6342c008fecf3716fc64cc93cb0fa
1af1ff57a14c16963ea943e69113d60a817c6ec3
48795 F20110218_AABLMU babbitt_g_Page_085.pro
0ec8cfec1cc5546af5d1be8f84af308f
ba1fa8ff397e4c12e3f845a427b3d7ead16c8b9e
2033 F20110218_AABLNK babbitt_g_Page_050.txt
d950b1fb0715e4387f1b5998b1a0ecc9
76fd5dd3de0a44bdfdf6827bdf2a6858528e70c7
6478 F20110218_AABLMV babbitt_g_Page_063thm.jpg
0ca6bd1f6489a65e2e92f5afe6ee8d48
c5f71c4c6243ed2d5a306c2afd5bda171e7502e3
1141 F20110218_AABLNL babbitt_g_Page_040.txt
98e13db282c8ca304131ea4fea662299
2a65ec0e90ebcf8129b81f9a01c016baa185f6bc
1051957 F20110218_AABLMW babbitt_g_Page_052.jp2
1ff3aec815b375a742654254cb1492d3
f2217ed5fc282ca3f972a687a1ca52ac4b45d6e7
926413 F20110218_AABLNM babbitt_g_Page_083.jp2
b18893c51897367f52a4afc1bc2e2638
ddd5a6919a34439f904c29bba3f109be396c87cf
F20110218_AABLMX babbitt_g_Page_080.tif
3305f6f31d9188376315879faf6690b2
f5db14527935690cceda284eb709b9520105f7a7
11630 F20110218_AABLOA babbitt_g_Page_022.pro
b40c6edf2267b96eca12623781bafe74
65645ee2bec05b6c8055a4d7f1e577b2e042fdc1
8251 F20110218_AABLNN babbitt_g_Page_053thm.jpg
25b30d386ab923454d35535188023901
df0f191d94aece3694a79620b04b37836aa0fe50
F20110218_AABLMY babbitt_g_Page_016.tif
b139e865160aac66d3dd34da2560375c
9521ada9ff5880cef0eef6e87a0cd3173dd8008b
49531 F20110218_AABLOB babbitt_g_Page_074.pro
e32c2f4a46f88c688d7cac7fa7761035
89d207b140281791bf70e7630e9fefea3f03567b
52919 F20110218_AABLNO babbitt_g_Page_053.pro
facbcad043c397e2f35f752d27ca97ee
21877338bb61f14bc5c98a4a326d167726f0ccae
F20110218_AABLMZ babbitt_g_Page_015.tif
d64a9f994abece0f25ba12ba6fbb5e1c
cf6177b0909af721cf1f87f2b0ec843082fc55ae
F20110218_AABLOC babbitt_g_Page_026.jp2
e7755e76cd8654d165832b3e7bff38da
6db5971982b249583fb2cc11f7db5ee09836a594
96467 F20110218_AABLNP babbitt_g_Page_108.jpg
6846ece04cccdc4491bcbbf594483bc1
40e9951f3c331422e8ec23012d78e2b338478780
F20110218_AABLOD babbitt_g_Page_072.tif
e46f067924ef63c319ceb32224a26d22
ff69d3df5efbf117658b1e27b59b4c4cd2534f4b
79738 F20110218_AABLNQ babbitt_g_Page_043.jpg
b8a0349f04328668b6114ad422745cc2
af9a94e6213189dd852c2e8cc8ebe9de5a35fb3f
F20110218_AABLOE babbitt_g_Page_084.tif
4f5946631e771b36cd71d02d4b373a2f
3b33cd355699ed6025ccc1e2765a1de44739a8d8
73888 F20110218_AABLOF babbitt_g_Page_059.jpg
b792593b818a79ba8c79f5a31b537eb4
948a0de4431495b5cfb911623294c050667044e1
F20110218_AABLNR babbitt_g_Page_060.tif
9a01fdc429d811f5c2b932fe981bb590
fa1cd0dd1afe33a81498dc5e4055964416540a48
51743 F20110218_AABLOG babbitt_g_Page_096.pro
0239c1ed25ecae32fea23657f28380bf
a4ad4d6a48c08b9f56842d0285748d0f7fa08616
219 F20110218_AABLNS babbitt_g_Page_011.txt
ec6129710abb018884681bd2687d893d
782eb2e4a0eae1a429bb83e5fbfe4a62bb133c76
4759 F20110218_AABLOH babbitt_g_Page_099thm.jpg
5284da4589a8074a6118aef6f0ce9593
ac9aef51c074386e05f54bb0e31da323afca65b6
102124 F20110218_AABLNT babbitt_g_Page_068.jpg
fc1a2b23095ddccaf24b33b3f2f97f9f
e4e446e9323ad86d8dad57b26f2028d98e46cc7f
1947 F20110218_AABLOI babbitt_g_Page_061.txt
6b7bbf7e13674ac602fb85940a72ce77
a05bb7de0e01702b9b5de08a5800c20ae5350961
29277 F20110218_AABLNU babbitt_g_Page_049.QC.jpg
ecdda5f6553e00881ac9e737163f0fe0
1cb3f21fcea2970f03add4e9cd1fa84acce90de4
15791 F20110218_AABLOJ babbitt_g_Page_038.pro
4d09aa58f0dbdd993622698a384b18e9
c08e791e35a22a86af0e0cde6ac0fc1f7f52fd02
53647 F20110218_AABLNV babbitt_g_Page_064.pro
178f36ad9b506bcc4a245e7de599e29f
fab3c7dad49fcfcbf36aedec91b935a2441c5477
33852 F20110218_AABLOK babbitt_g_Page_010.QC.jpg
334bd379ab58863541c1db265de47a89
c9907acec51d28722d58029c9476ee562df24fa8
F20110218_AABLNW babbitt_g_Page_028.tif
5c5ed34640b94b0d0479b7d60f378082
daa19a2dca6a862ee70e032847e1150bc4fd5990
1046876 F20110218_AABLOL babbitt_g_Page_076.jp2
212f105329f563038bda3bbbf48be6fb
f3cd839c335c923ba1bd2f0e820e2b4a192452d9
93311 F20110218_AABLNX babbitt_g_Page_049.jpg
bcec8b43ec880c7b8717f77a7a8fddb5
602af86af9a5f8f0b2717c0247015df7f2d02fa4
8332 F20110218_AABLPA babbitt_g_Page_091thm.jpg
19c0de0d30f68ac8a9efa77ba44a8bcc
aa422505ce1ce8cfe86f3f5907deb133f3012c92
44644 F20110218_AABLOM babbitt_g_Page_014.pro
3891cc66e543884128ddd30154fbebaa
0015957bc6630a10474cf71b53754e66a3559aef
52503 F20110218_AABLNY babbitt_g_Page_021.pro
c6aa1f61ebdd4c6bad9f2dccf3c46a5e
366a931f949f98c5fb4927b6cdd328a86ceddeff
97424 F20110218_AABLPB babbitt_g_Page_092.jpg
f16cb87052d533ddb0abd625dad98f8c
d62f28ef784b99ed03738a1f28b858ab403ce36b
F20110218_AABLON babbitt_g_Page_067.tif
92456d56f6146c665f8bba9cab0a4ff5
c4d3014295159f3aa380323266688c9d233f94cf
2040 F20110218_AABLPC babbitt_g_Page_086.txt
3aba64a6245a37760b6393116ba49f3c
588617efeb7ff6c1d0038c94e1968cafd699b1c5
103887 F20110218_AABLOO babbitt_g_Page_080.jpg
9580b49b905293bcee99ef49536c18d9
99b4ae83df2ca945b24dbc5e2cb44033e83dda8b
227 F20110218_AABLNZ babbitt_g_Page_098.txt
218e430c117a44c12f3cd8ce6a215113
f4f6a8599c1b4b1ba16bf60fb0db058de171a683
46576 F20110218_AABLPD babbitt_g_Page_030.pro
bdd100bccabe8e5bb2741cb898bb5387
2e3dcde5f23941ff73d45af376de1cb69ae88308
83404 F20110218_AABLOP babbitt_g_Page_013.jpg
c39408bc4785cbfd3c2d6a5d0517a302
b6326145467497ce976ea2d010111b31d51adcf1
3679 F20110218_AABLPE babbitt_g_Page_008thm.jpg
12f428442492972261bbda60d7a0830f
bc844fbc8b5bb57eaa0018a2892a676b7d2c7737
48550 F20110218_AABLOQ babbitt_g_Page_089.pro
273765e4ee8c9fd02a4158a192990a0f
eaf00ead2b0b6e97ab09281c214c6ebfcdd90a6d
2034 F20110218_AABLPF babbitt_g_Page_096.txt
3bf72873e697c7f7e0b144b755bd977b
3ab61026ec97c74fd5de9427bc744517ed212d4d
33026 F20110218_AABLOR babbitt_g_Page_107.QC.jpg
9d77c403c9cc5725a5201ccb7f6e71be
219ea17f5225c4ea2be0dfbf392adfea50896d3a
23272 F20110218_AABLPG babbitt_g_Page_040.QC.jpg
9b32b1eae43b5a14e54ef922f9d4d83a
0e5fc7f7d25ac067bddfc0d43c5e86c7776cc18f
F20110218_AABLPH babbitt_g_Page_066.tif
c58472837ee85757b10125c07c833954
aecaf75c092be55bb49e2f09505807b1777f55d1
107740 F20110218_AABLOS babbitt_g_Page_051.jpg
6c8778721464f7477b91792019661ba9
2896a645b1110655ba0a3c26525d04045f619f07
5398 F20110218_AABLPI babbitt_g_Page_011.pro
9dcb119e125631f3fe367e6c2d3e60a7
fef4ea6d2102d272ef5e2e66ff20b827d85cd830
65519 F20110218_AABLOT babbitt_g_Page_007.jp2
d1e1eea2f15a5f99a2356c2325eec164
741dcb09368d21eda8ef6d8c595011d267c30bbb
31773 F20110218_AABLPJ babbitt_g_Page_030.QC.jpg
45b665cbf002ad839289b092a26c3a66
9d9794034974e267810da894f9f931942d063c99
649827 F20110218_AABLOU babbitt_g_Page_008.jp2
9fcb7a9eb6d9fe41d1a2e2db871ae745
40e0a4ee0f9b893d17bfee0bf0b1559a948c5088
106367 F20110218_AABLPK babbitt_g_Page_026.jpg
ad931eb3267a1aab5f281b8a1ce96552
0a335024bb21ed3c5af264fa1e273faedbbcb439
F20110218_AABLOV babbitt_g_Page_029.tif
7cca77835b4b3498ea718ccf217984ed
9c25ac81a1dcef495ca9273ee1af4858fa591ff5
70497 F20110218_AABLPL babbitt_g_Page_027.jpg
e6188d738c71b913feb040807c51e51a
c75293c6ab78d9226e9e949ba5584335814a492c
1051932 F20110218_AABLOW babbitt_g_Page_020.jp2
4999df6e8d87ef437b160a7e26e52e9d
aaa947bb61bd0e5c64a66848b576a4ef84074e2c
34275 F20110218_AABLPM babbitt_g_Page_026.QC.jpg
c4ec5187f589a72c252f4fdc27594207
fb535ad475616b6649485280f43c00e24cd7377a
7376 F20110218_AABLOX babbitt_g_Page_014thm.jpg
db57d163aa5b49af602c52fba1751a7d
710bb14f9b2d268591692835fc9360f7fb36e99f
F20110218_AABLQA babbitt_g_Page_002.tif
aefd95d66bb50a8c5b74152be1b3ccbb
9d5f4e057dd9a500c5509215d1a0bcdd08dd9e66
42742 F20110218_AABLPN babbitt_g_Page_023.pro
ea373e0e7197f2ee6798fff13c925919
7b61b2f78b04fc1e4c848ee9e1c2112f9c12205f
8501 F20110218_AABLOY babbitt_g_Page_051thm.jpg
723d79692041c77283fd93e96d359b7f
f9975458f9d51d8cd0c123d2216137538a4e45a2
F20110218_AABLQB babbitt_g_Page_003.tif
c7bd1864552c58a23d206f62e01aa5c7
a9e487a7aa09990b1701e8bf610220f0394b7a2e
103026 F20110218_AABLPO babbitt_g_Page_075.jpg
b092ce49bec0b5a0b0657a28c884926c
09482a9ec543daea61da54eb92280534c737af2e
50175 F20110218_AABLOZ babbitt_g_Page_017.pro
c3d3edab64804e370bb1a1f195583dbc
d38e6a597a1c22885ceaab81b72dc4ee1ce9f833
F20110218_AABLQC babbitt_g_Page_004.tif
01875fe527dd8a48f8260674577ed511
37103b731c389dd866df05ed865fef3059d4d8c7
6862 F20110218_AABLPP babbitt_g_Page_108thm.jpg
4ce98d9bb1e20c812f36682d8a069b18
8c1827c10ba269555588dfbe9e2388d7596a4627
F20110218_AABLQD babbitt_g_Page_006.tif
15944a0a28eca360792e8a81f5d8a021
e3be0ac3a68857625b1419cd161d97c974a92be5
F20110218_AABLPQ babbitt_g_Page_092.jp2
d3ad6772a7e39cc4c786a80afb5d7107
4269b7531da3ec10b1dac563ba51df2d6a0abc28
F20110218_AABLQE babbitt_g_Page_007.tif
8b7d2e8f84c6aa5ab6e34dc25fbbedcb
7d8d1252c54f5b69351a57f02ce7be8d2d845023
1051919 F20110218_AABLPR babbitt_g_Page_025.jp2
78fc3f8444de3d89311c11ba5a9d954c
ddc724f2dd4f1b503bab705f0afe1f29af733525
F20110218_AABLQF babbitt_g_Page_008.tif
5ee467a50bde97f63e60665624fdf791
dedc290fa6dbbe0cd7d45d44e1f15fac25c437e5
33113 F20110218_AABLPS babbitt_g_Page_052.QC.jpg
f2715e7fdad07a5e7c85f70d94f1550e
58679c4e5db328e912ae60d1a8bb05524285124d
F20110218_AABLQG babbitt_g_Page_009.tif
b0531bcc5bec32bccf58ad7c0f85d1ee
35481041f78aaba335f63562ff7668e3d671b1f7
F20110218_AABLQH babbitt_g_Page_010.tif
6815f1549ef7f9e68222bf1fe77930bc
6db0ae7fbe02afcc8a63b4855dc00acbe07a32f4
1962 F20110218_AABLPT babbitt_g_Page_085.txt
7cddc01085fb2a4fee38ede6193d1aa1
b83dfa1bf074af57a368094fad701f5580c27a63
F20110218_AABLQI babbitt_g_Page_011.tif
fabda89585352767bc05406ddd513b0f
deffdb1f88135fed30197d721d55b8320116138b
F20110218_AABLPU babbitt_g_Page_038.tif
1d5814055288100171875f5e69b961f8
9ff8451ccf45329a054663fd152ae830a75e5351
F20110218_AABLQJ babbitt_g_Page_012.tif
773503bce6ab9d2d2eb4858bd6a0296c
16b7d5b7bbbf3401d66146a3d399a73568a8d290
50413 F20110218_AABLPV babbitt_g_Page_019.pro
26d73bb3cd68358f5270ea10852c600a
d811b8fef149d293e1e2e7d5ced339a12a5743f5
F20110218_AABLQK babbitt_g_Page_013.tif
5759dee5c52fea614070225c290d0d08
ef8fe489ed806d9af65db5c816514bfecd504f7d
177228 F20110218_AABLPW UFE0013688_00001.xml
a1d81cddc87d67bec977908711520c0b
9105331f0493191225bab2dd24af9aa3f59e6195
F20110218_AABLQL babbitt_g_Page_014.tif
0f8ee9c96fc99f37f10887022d1a8f2c
03c488dda776090bb3855e90ee94d668b26a3736
F20110218_AABLRA babbitt_g_Page_034.tif
df4ff2ee1ccecd155a14eb4d39df6841
3211c8e3d666387a622431b80196efc24322eac9
F20110218_AABLQM babbitt_g_Page_017.tif
4c39b89866adcc3c0cac91a340d754f1
23d8bbbbb2dd43b07d23b59d22120ea32c390f5c
F20110218_AABLRB babbitt_g_Page_035.tif
6467fae8c53e0bd2c70c4289ea1f8901
dc4c54f9a693476d8f64d65fc0bf20284a2339c3
F20110218_AABLQN babbitt_g_Page_018.tif
afa98ca224fa67fb3931d95247fa64b0
8a50921873e3e4b138c7c70c30b50107e4466eab
F20110218_AABLPZ babbitt_g_Page_001.tif
1a9c651b7d5708d1fd472d5d4554ce93
18f667565ff4b1f91ce971315f26f298b5aa66e3
F20110218_AABLRC babbitt_g_Page_036.tif
0303064ba09f5600eefd6e561f284233
4e9bbcb920053b47423a5dd963beea35a5b1cee4
F20110218_AABLQO babbitt_g_Page_019.tif
6f18a8d13f1438496875afc949653338
6a38d7c291149b3546ec932680f661ab801f935b
F20110218_AABLRD babbitt_g_Page_037.tif
a56f265474eaa462ac923f182293259c
4a121bc80aeeb743af88481c09438f08d557bab7
F20110218_AABLQP babbitt_g_Page_020.tif
5add4fe7514423e50c79777cf1c4a7e5
1a3e86c7002c66568764f21f9aa45cc6ed6d53d1
F20110218_AABLRE babbitt_g_Page_039.tif
471b8379083561c0bfe78888f07c3c8b
3fe23ee91612908ad7fbabf3c30d323493ac1fb8
F20110218_AABLQQ babbitt_g_Page_021.tif
4216f35528522f438c3c7b0f0820b79d
794d95822341b246050a33dd5c309f21d1ebc80f
F20110218_AABLRF babbitt_g_Page_040.tif
b38108bc38c2335c6da76b1746a6262b
94eb5efeb00739ab321e37fb092ccaaf6cfb924f
F20110218_AABLQR babbitt_g_Page_022.tif
46fb7544fcf33a6003ad30823181a039
e6d57db13178619c52f45ba8012a6a945bdcc21e
F20110218_AABLRG babbitt_g_Page_041.tif
38efa05fce8d5ec1ed86f658adfaa3ac
27a352d3b8795c5b9363a9a158a41df34f86be03
F20110218_AABLQS babbitt_g_Page_023.tif
ec2ff58662ded928cb31ac5cac5c664a
c922c721c9946957c6fc8e1ed7e70c7ae68ba4a6
F20110218_AABLRH babbitt_g_Page_042.tif
5a214896ce59ef5c6546526ecd2d4eb9
8c50da96500a9f2ec8f0290ae9f295c6d197606b
F20110218_AABLQT babbitt_g_Page_024.tif
140a0422cf41e892eac6aec4e86b486c
e9174c39e96889c73ed84427ffc20a329f9c16d3
F20110218_AABLRI babbitt_g_Page_043.tif
0d4e13e43be209daff71ee18c4231aa1
d54444448902d73c5725640dfc0e699cd8536164
F20110218_AABLRJ babbitt_g_Page_044.tif
7b320b2e214870365c1a8e1148b89ab8
2b67e3b5185d3fedf42a28c45cfb2f7f623b6b20
F20110218_AABLQU babbitt_g_Page_025.tif
fd7763c9691042cb270302fc072cbd5b
b5dd4aa5cb37ef4212d08916d8c7f84a06c3ef51
F20110218_AABLRK babbitt_g_Page_045.tif
90c9c11428a50e79b1e9a17bbc935d93
44fba2b974085b74d10f41f10d25ee0ba6e29c3e
F20110218_AABLQV babbitt_g_Page_026.tif
5cdb05e50e6addfeaacacf388f795214
4bafc3d2e5ecc6a91d7992260c5cc399c54d349b
F20110218_AABLRL babbitt_g_Page_046.tif
6c3c16a70d025e432a26cfe62d37413d
77ef46012be84de3f0112f9d069b01314b209e49
F20110218_AABLQW babbitt_g_Page_030.tif
e43aacf6c551e8603e0ae78854049383
acb89b8f53e2bbcc889452a9e06249d66e61c188
F20110218_AABLSA babbitt_g_Page_063.tif
afee25c9dd9dbb4144206926e4085905
166688f92596f63bf32214e659cdabf6efec7899
F20110218_AABLRM babbitt_g_Page_047.tif
15fa27369bedf038173ac9c2ad199bfb
e8f643704c763f30769218b245b010947c366128
F20110218_AABLQX babbitt_g_Page_031.tif
f86d0e5134bcca3e36c7d759cac46741
d7d01213d1c694663310ec5d387a56c62d29f7fc
F20110218_AABLSB babbitt_g_Page_064.tif
2afd2577fb6923c4f330dc67c80a5970
709959b94dacfe2057a4102570b52c46e49458ad
F20110218_AABLRN babbitt_g_Page_048.tif
85f99bec96ef7a48c69b89d41f998684
0f510df589d7fbcc634bc583e837201f0a0e179a
F20110218_AABLQY babbitt_g_Page_032.tif
897ad84e9369918d8f5baa1024853440
e3e58d078b7b6ef19f2bab05808f0b2a08694b92
F20110218_AABLSC babbitt_g_Page_065.tif
6155f073ab7ea88eaa500fd0d12a7a5d
08dfc5911875868143d09ed56dc2138e4d54b77c
F20110218_AABLRO babbitt_g_Page_049.tif
de960ab1835968e71add9dc78c8c5a20
d4b590febb6fc0c3926775adbd7da49e02e13b6d
F20110218_AABLQZ babbitt_g_Page_033.tif
2df77b0515ea274e544f9d7f0d252329
b0b9898f825ab9ce89f3243b03c105d980d2bdbb
F20110218_AABLSD babbitt_g_Page_068.tif
c3e9bb5d6901bf110a075facc9580658
3c2ee97160b67184e1f9c2653ee3e9971852549c
F20110218_AABLRP babbitt_g_Page_050.tif
59f10ad2057a7a56032d7c3a7c612701
578b339f57a65144bd7af7d218cee74b0c174719
F20110218_AABLSE babbitt_g_Page_069.tif
6a876144d1bf7d53820a4a50040b15c6
570abe0ad7340655d4577e300829369d0d69e277
F20110218_AABLRQ babbitt_g_Page_051.tif
13d122804c23796a52095292277b2634
edf30a59ef02583d35872d5587ec3b2d1198f33f
F20110218_AABLSF babbitt_g_Page_070.tif
d5b54e8eb77085de2bdda90db76d1e8a
e70179b5f0459e40ef1b8249c97946877f74c028
F20110218_AABLRR babbitt_g_Page_052.tif
032147cb2a0e660a24d23a2ff7ecadc7
202b63d0dd5cddc9bf31f7c7039dbfb1c612e579
F20110218_AABLSG babbitt_g_Page_071.tif
4cfc23da10b63d211814c9af24de6797
e131696770fcef165001eb7d14a19f4216ca749d
F20110218_AABLRS babbitt_g_Page_053.tif
89a7e8c83ab98eae9400b4758e26cca6
d3b09a5a34532efba2495b02896d76e71aac8b30
F20110218_AABLSH babbitt_g_Page_073.tif
ad29eedaef7f79241aae573c869cc53b
6ad6eeb348e668a89688dafc804edb094b8ef7e6
F20110218_AABLRT babbitt_g_Page_054.tif
f0930b56b06a7d06ff1e4bafe9a33014
31b4281b9ba2bdcdb104ec90a47c164610259e1b
F20110218_AABLSI babbitt_g_Page_075.tif
bf9dd9297f61fd35c92a8e8036e59b16
31b468b69e91e5271a8fddf0c9fc03d4651072bf
F20110218_AABLRU babbitt_g_Page_055.tif
36a40ad92472c270eb3c372f075bec02
f2caa5bf0fb23c938829fcc0388888dbfe1e9a4e
F20110218_AABLSJ babbitt_g_Page_076.tif
c76358ee6c7f88ee73e797216035c4d0
7b296b478c10206493905014569e6fd0cb843833
F20110218_AABLSK babbitt_g_Page_077.tif
60c908af379fe9e120ed771608193b84
a278c932f1e6deaec211545387ee4329c851b08a
F20110218_AABLRV babbitt_g_Page_056.tif
185a827b58ddf16dfe77294a89e41a8a
15c303002178b044a90c4cd1589e422dd66ab31e
F20110218_AABLSL babbitt_g_Page_078.tif
862835ee5c498db5aaeb3108fb05ae3b
600cd79cff7e78f28ffd347aa8b72d2be48476c8
F20110218_AABLRW babbitt_g_Page_058.tif
7029cdd086c60c513e97a58c35df405d
94a6b851d6562b0be9c769618b5151faae062c9c
F20110218_AABLSM babbitt_g_Page_079.tif
76d6b75676b240fd3a706ef361260e43
ac9c70f31a580d2cb56ee5780526966d84ad3856
F20110218_AABLRX babbitt_g_Page_059.tif
d7ab5d16b59d50f4579145ed060c16bf
3206d2714e3a6e3c94cca32d50c4ee364a6b9317
F20110218_AABLTA babbitt_g_Page_096.tif
7f6699638908d47ee586d7c32ec8508e
abe58f05c399106799529f2886628d701baf2a20
F20110218_AABLSN babbitt_g_Page_081.tif
c75acf90e925b76cc67d0f06675476b0
f4cbb32e895761eac6eb25f492bb8c84e4d6ddac
F20110218_AABLRY babbitt_g_Page_061.tif
9a2bb9483e8de8374533e24a3764c93e
2a8cc4c5b4358e13fef14974ba22c0b949a501fc
F20110218_AABLTB babbitt_g_Page_097.tif
fdf1fef446ed30b1c84da91209e98c43
93199313cf470b1b0e94311dc0179a03198febc9
F20110218_AABLSO babbitt_g_Page_082.tif
8fa45e8bd5fb51512d3501ff42232fbb
fc0913f565591f887c7b9789a53e2a65c576a4fe
F20110218_AABLRZ babbitt_g_Page_062.tif
7922dc3ff1e2d3bea0825b87cf2c62e6
eb59477005090c8bc8f09226232237b9313d8245
F20110218_AABLTC babbitt_g_Page_098.tif
8c9b93880cd9e73e54fd7690f08ba8d7
21fb6a66c000c9c8d6091899bbb01f6ab6152819
F20110218_AABLSP babbitt_g_Page_083.tif
86dd2b024d86f0f027b0c8ba32461461
01cd7e0bc4bb0d7813779b2ead17ed880fc64dfa
F20110218_AABLTD babbitt_g_Page_099.tif
bf79aa8b79b9cb501f0f2d884b73732f
3b36c1515bd315516cd624cf8661e38028d43cbd
F20110218_AABLSQ babbitt_g_Page_085.tif
11865c15a0cd368b0a27821f17ed6b07
43a5135f22added61609e374a3e47a034970623e
F20110218_AABLTE babbitt_g_Page_100.tif
ecc3fb1613047a660dfa75f86c04cb2c
8fb96e7a44b86a8d888e3b54f75dee423bf1aedc
F20110218_AABLSR babbitt_g_Page_086.tif
6fe3ac6b2f39e80b6e9237cfb33fda2e
27318937261017b4afb8f72365a1215d55cdb6c1
F20110218_AABLTF babbitt_g_Page_101.tif
876c22b637111519b465525277e4e368
b27600031c1b542f1307261e20b322c7c156f606
F20110218_AABLSS babbitt_g_Page_088.tif
e3a8fb529f05d013628e487b71cd3eb9
a7e386e42667184091a4c3768b4a4e2eedcf3f49
F20110218_AABLTG babbitt_g_Page_102.tif
926ac654a91a3e667b0e5443a2137e37
7a85c1490a860b6c0ddb405b691746ff916c6f05
F20110218_AABLST babbitt_g_Page_089.tif
f5241dcc1551f9493689c69f4be73d20
198640a42da19a525711e1d2647314144c89610f
F20110218_AABLTH babbitt_g_Page_103.tif
0de133b31996a42b184b004f4e85d98a
d80c1b489bc905a3daf45b05e2b61bb5df60d803
F20110218_AABLSU babbitt_g_Page_090.tif
5fa85b5e566881302dc31c8bc4fff463
e4939f91d4c770811f4eea71fc12e5b453dfefc0
F20110218_AABLTI babbitt_g_Page_104.tif
23a2dbcb04e75ce823bca3da2dcd1aef
91b144b968bbf3a721ad496c35cece72dc8d036a
F20110218_AABLSV babbitt_g_Page_091.tif
33990acc49dac60125e7929541d6cffa
493930357149b4f00588abeffdd48adafbceda65
F20110218_AABLTJ babbitt_g_Page_105.tif
8c13ca5f0694272c39a04a52046608c2
faad27b58ecb3714b5f7be85989735b8e1278f7d
F20110218_AABLTK babbitt_g_Page_106.tif
1abe02fcc0d0ae0052125a98520468e4
2f0f1dfe2d0a05fab24d6ce99966a46fa204c363
F20110218_AABLTL babbitt_g_Page_107.tif
7e11782c92628c3b43ef5c12e8a38e2e
2b2c9a934e3fd7daf06726999df606990ae2e223
F20110218_AABLSW babbitt_g_Page_092.tif
551348f957d7d6c658db71bf9a58317d
9e89c8d57971b8c84fb0a35aafdcd430fe5e78fe
2030 F20110218_AABLUA babbitt_g_Page_015.txt
a4732a6cb622e63149be19167114c01f
d42d1272ca88d2782b2332577eb457a2b7d201ba
F20110218_AABLTM babbitt_g_Page_108.tif
e6b5111c8811b9d4228eda08de40a932
7a36fc8435eb59e5ef5bc50002d4065aa7e146b8
F20110218_AABLSX babbitt_g_Page_093.tif
2ded28d6621dba9d2b064440a1786558
5de51820c77575ef95b3b6045055cdd31e38f9f0
1996 F20110218_AABLUB babbitt_g_Page_016.txt
251d74a06b3522008b2b666eaad59eae
4546ece9953a2e5e9c20d4b4b3a442791d55f4df
F20110218_AABLTN babbitt_g_Page_109.tif
bc7d9794e64c86dc52d33bc15264e538
93a4300d09c09d4b5f6f04796bef6c9589ee7966
F20110218_AABLSY babbitt_g_Page_094.tif
c83aca30f8f8cbad0cbb06c2a50cbe15
1d1d8fff37099faa89e53e6ee3e5ca909f2e849a
1974 F20110218_AABLUC babbitt_g_Page_017.txt
1574add6a3fe5868ff1b0fc25e7e73e7
cae3991959ef8ee69881797a7a4604e63e0e33f2
447 F20110218_AABLTO babbitt_g_Page_001.txt
1a2a9a24df7dc0671fa29ca6ff495fdf
07339edb3d226b9fac6cf2a1d9d99ddf964d51c2
F20110218_AABLSZ babbitt_g_Page_095.tif
2b1a3fa6871aa76cf43943fb6ea6f373
12eae7c34cd69c1a990050c287fb99265d55a5f2
2070 F20110218_AABLUD babbitt_g_Page_018.txt
1d4defd62ce97e3d431592b4edf2bd94
abfd98709af3b7b20354299fe134aa05d6bc74ba
120 F20110218_AABLTP babbitt_g_Page_002.txt
1e21b530e2c05a7b06926b6e67be2990
9c1fed7a924f37e647c6ae0e6a7988870915535f
1978 F20110218_AABLUE babbitt_g_Page_019.txt
941b106494f62aad66d384e904f4d560
f5941ce01effecc162feca71424697310b46ee18
186 F20110218_AABLTQ babbitt_g_Page_003.txt
8d26657c9fc29ee868c713ea1f872783
8a9562d91e28e404bc7148c237bcce3c3bb3493f
1883 F20110218_AABLUF babbitt_g_Page_020.txt
a9d79456bb7e8fb7d8fb1f8f50fa858d
bf9b68d04c50c5513cbfda8232e141c5c9a472fd
681 F20110218_AABLTR babbitt_g_Page_004.txt
54245b48e470b3354ffbae1547e31ac3
3e129a00fbd5661bd5bf36758ea91d949a9b8a49
40676 F20110218_AABMAA babbitt_g_Page_087.pro
2782d1e1451939aa175bbf7e2bf9674a
d72cfffdb4b2e5858456486a5d8bde41a8416d84
2058 F20110218_AABLUG babbitt_g_Page_021.txt
233dc30d4441b8d4a5fb30e65f23c52c
a6aa5b831bc646b2229ca44b19546ea176190b7b
2879 F20110218_AABLTS babbitt_g_Page_005.txt
4c2e6642c5d35c22f5355ded6a7a9a2f
e640ce6e386a241fc5c61a8649d7867f01472318
47088 F20110218_AABMAB babbitt_g_Page_088.pro
93f2f4881e8c7de6be090cb79040133c
f0196e5a806e093c8c2cd403f853642c08207874
505 F20110218_AABLUH babbitt_g_Page_022.txt
e53d37827198d92a7d9726b077b9ed07
248892504c426a51686e12b99d83d32db5bad69a
275 F20110218_AABLTT babbitt_g_Page_007.txt
7e6db1bddfeb3e04e17bd59a34b85c2a
8a78d5c7fd7160c296b921b520bb4378489c80ce
50253 F20110218_AABMAC babbitt_g_Page_090.pro
c9e751b4735e71c67b3ca9389a273091
8e36af0c4e28f7cfd7f5bca92f50669e02c844c5
1774 F20110218_AABLUI babbitt_g_Page_023.txt
1efdb4f2ff20697b7436ba34f10b93a0
2dc971b4552712fee75f27bb744e2092788712ee
1457 F20110218_AABLTU babbitt_g_Page_008.txt
42b28577fdd5feea611c1034758b66aa
b5ea3899c63d1b18e66436e8f27aaf2159bca214
1977 F20110218_AABLUJ babbitt_g_Page_024.txt
ef287e27562201707db888208c588e52
723eabf3185cb8abf069a0045521bd13322e2011
2847 F20110218_AABLTV babbitt_g_Page_009.txt
01eb4940b2e79f07ea0f30a9eee5253c
f288461ea2b100472332fc796031dbc64160645e
51445 F20110218_AABMAD babbitt_g_Page_091.pro
328b1267a19eb688c548b36314efa5b7
8ade95f6112d7721f9f0e14302b3713036eafb3a
2111 F20110218_AABLUK babbitt_g_Page_025.txt
f4c56757592d0baae7080eda1c65017f
05d7e19c4db0c4e6d3be9f666a61677aab0702a0
3447 F20110218_AABLTW babbitt_g_Page_010.txt
114b437f1a8f5613fc9b5091f44d3e8d
65b26d477cf24127d9a2d2c91731ffc4116fc3aa
48081 F20110218_AABMAE babbitt_g_Page_092.pro
46f1b4df15a35bb94f5a055420f68286
c2a1a6f3222e4600dd17fed75990f181a157c74f
2067 F20110218_AABLUL babbitt_g_Page_026.txt
87303fed82f5dc8c60318b39716ac154
6d93d2dbc0065d93e08cb82af32ce8e9e1ff1952
53373 F20110218_AABMAF babbitt_g_Page_093.pro
d2b0b37b40448889c013d0e5ab5ef7fc
99f88e201cc9f333c63ecc3f1325590b01d4bbd5
1236 F20110218_AABLUM babbitt_g_Page_027.txt
2f0887f554ce06285b9ee88da26b2a80
64e723a71b0b1dc3ed41231a8fd21c7804fad603
1731 F20110218_AABLTX babbitt_g_Page_012.txt
9bdc27e6d4b476033211785bb3e495ea
dc4f083081e26fe4e92a9d18e04dea0abca10d1d
48719 F20110218_AABMAG babbitt_g_Page_094.pro
236c01c61fffd69041e1ca732a132519
4d5c762bc6bb36f51ffd004bd9400a70de979ec0
2480 F20110218_AABLVA babbitt_g_Page_042.txt
6dce7f98b04c823ce02ff7ebcc76a0ae
36650146f81f2dd2d7e3eddfdfe1a6859e290e8a
2060 F20110218_AABLUN babbitt_g_Page_028.txt
c81becf327d18012beff36c750be0312
91c40ca4ec09097b01f68bd443d1a2d52c795b2b
1589 F20110218_AABLTY babbitt_g_Page_013.txt
9d34accbc74f117aad52fb3e5bdff12e
1f7e867368a80881a2543921ede6a64657668f85
50879 F20110218_AABMAH babbitt_g_Page_095.pro
4c1d8464e63669e114f9baabb0fb9d7a
2d87a3bba40042e6d19921c5f1764bfd66759fe8
2530 F20110218_AABLVB babbitt_g_Page_043.txt
24860b518c2b63079c092ca301764fb4
b87e0325a7c9c611c1ff5009a70b002eb34d7d71
2017 F20110218_AABLUO babbitt_g_Page_029.txt
db5fd9252cefefff1cb861a164323093
39240fcecf07f6a072ed0da9ad0f190267811dc4
1864 F20110218_AABLTZ babbitt_g_Page_014.txt
79f69f41e7edc38d48c0ae210930f63b
37dea8aed6ef69fad032da0094f3db551fe502ac
6037 F20110218_AABMAI babbitt_g_Page_097.pro
5f8fe96dca1c8c165e77ff53df706016
f14db34632efcdf92d7d1276d382fab18b66c997
2013 F20110218_AABLVC babbitt_g_Page_044.txt
735bd7c49171d523a1ec33870c2c0bda
8a56ead0af95aa1540b7e3bbff902ad785f21204
1890 F20110218_AABLUP babbitt_g_Page_030.txt
0321e28a1502a5b247db4f9a53f2a73c
72645afdeec861510548074ac8686d38d98ec5e3
4013 F20110218_AABMAJ babbitt_g_Page_098.pro
9fbf3583386c60219f0762ca484a4e33
0967ccb380e31ef037d97c2478d4141d2c69631c
2084 F20110218_AABLVD babbitt_g_Page_045.txt
aab8536a9b99d553b7f7d06bf2c1ce27
4bc433350c1746bcb879df7965f4f8af65316e73
1981 F20110218_AABLUQ babbitt_g_Page_031.txt
1f59a1ac98c2ee79a4a366c004fa4f11
0e7c7bbe02f270da8c0c729a04be7db1fc041da2
24987 F20110218_AABMAK babbitt_g_Page_099.pro
ecb3d8ffbf69763e04035e9e55be471c
1d8c7fc03f9518a255ec1a93b8c55887a7e2bfe8
2003 F20110218_AABLVE babbitt_g_Page_046.txt
711e937301f7a55c3b58f0269cd4f14a
3840d682784db3fa5cbebcceaa6112750cc74613
1908 F20110218_AABLUR babbitt_g_Page_032.txt
a86d5491e3637f148859856bb16b19b1
cadbbab61c8f8e1dce9e246d46b127bd1bbfe53e
2205 F20110218_AABMBA babbitt_g_Page_003.QC.jpg
1be41fb79ac93fba3824a8ad1337d0ec
46b790b43e5afd532ef40194a82fac0c076b415f
32969 F20110218_AABMAL babbitt_g_Page_100.pro
b009dfff8e108fd761623143363a99e2
86938f7b98a02ca4f8006bfb8951c0d63e93b43c
2028 F20110218_AABLVF babbitt_g_Page_047.txt
99ba7c16717270312d5c1971f16b33f6
47be8a2d3b23d620ebaf6f9931231beef671c91c
1989 F20110218_AABLUS babbitt_g_Page_033.txt
85c7584159086217a09fcc37bd30b4b3
50a74217af8859b3b9dd4bd80b74e6e947a350de
37033 F20110218_AABMBB babbitt_g_Page_004.jpg
905c9a631e28e58a6baffec1a1f69279
229b73f996d21a0258f044339392d973c08e2bba
46879 F20110218_AABMAM babbitt_g_Page_101.pro
b385f2245b12cacf7aa01ade3766afcb
fbf579ca9a59e91726d4e5285b7ff5bf4ab2c80b
751 F20110218_AABLVG babbitt_g_Page_048.txt
46589dcc4abe38a60d982a65366a751f
b6833d67de1d1ec8d0f6d199fe1a4f048bc7559a
1910 F20110218_AABLUT babbitt_g_Page_034.txt
bd94de20f1343226c63ac142a627edda
c3c29db389a6d88bbc9eb7d6dfb16e3b6e6fd8d8
11671 F20110218_AABMBC babbitt_g_Page_004.QC.jpg
38ccec8d45fed948da176d1903e12406
a042974620f6d645eef635b366479fbe63a30569
56990 F20110218_AABMAN babbitt_g_Page_102.pro
b483fa382802961371986d7e9ec72af2
cc9f5be34cc419db8c61e356688a589b986d4f1d
1810 F20110218_AABLVH babbitt_g_Page_049.txt
05854ba3c7c131198c361cb1d5c4a5f8
a3cbb1b23915e9c9a32d6b500da14947b95fcbef
1928 F20110218_AABLUU babbitt_g_Page_035.txt
774175664efc3417c02190da3c5db594
726a8894e829ee9fc42addae5655a1e6a0008d2a
84324 F20110218_AABMBD babbitt_g_Page_005.jpg
019492571fed8ba5886ab75eceed2746
fcdc5d41d8f57e3f048b108d9905bbecfb498490
62635 F20110218_AABMAO babbitt_g_Page_103.pro
96a9ae2ac76159d1cc31db81dc380de3
1b673cc38ef5c9669ba3012c7e0c0167246a5698
2086 F20110218_AABLVI babbitt_g_Page_051.txt
74c3d1400d01484b3eff36e9b0268113
3f5b83df3ec6a7630c1b29ef547f0a60a69cd934
2821 F20110218_AABLUV babbitt_g_Page_036.txt
d14fe408e9dcfa402f3d3cfb0c704768
5c17d4b19f362c0a066290450d6151227c2cd456
58987 F20110218_AABMAP babbitt_g_Page_104.pro
0dcdf660a9b57eed4944b357ea3c686f
b773938bf4716c541d65f7f06578e0f53eee3e02
1976 F20110218_AABLVJ babbitt_g_Page_052.txt
6edea83d1c765653e37b704a502aeee6
7b71c9a5a5479b535abd9c459b401ba507b4908a
1964 F20110218_AABLUW babbitt_g_Page_037.txt
56b08d3ac9c0a30825c613f02522c318
43da8b4776a6c7a2e897d1b83eaf6b09d4bd650b
20440 F20110218_AABMBE babbitt_g_Page_005.QC.jpg
32f5aee2c36a1116014f4df5e378d0cd
89e713b261926b381b55486c488b53acae6b153a
57355 F20110218_AABMAQ babbitt_g_Page_105.pro
a9ea8feeb601936d344fc1b8e92ffc3b
2dcd1fcbbf405083215bf9e8266b5b2369e2c7c4
2072 F20110218_AABLVK babbitt_g_Page_053.txt
ab1ae1e1238b01f9fab21423b66c8d3e
4f4e19db4502ee9540accb063286de795c851220
762 F20110218_AABLUX babbitt_g_Page_038.txt
73bd0cd3f0c7c4176a7fb9256cb0a4f2
32ca390d2208f3752e35d5fc34d11665e58e5353
107159 F20110218_AABMBF babbitt_g_Page_006.jpg
87dc5e2dcfe763cf74e327e1a3d5b744
5d6528e6b75525a1ca0ad8e799df9f0f648ec976
52604 F20110218_AABMAR babbitt_g_Page_106.pro
04bcebf30e4169729af506c8b5108efa
83e43ee6db7171d158789c466019c9d3eb2302b9
1404 F20110218_AABLVL babbitt_g_Page_054.txt
74c85152d16802285b5a03ae44a79980
6bf765884a993328b4599ddc5763127d4444932e
24847 F20110218_AABMBG babbitt_g_Page_006.QC.jpg
cc5e1506673769b51a0258005ad25244
a16fabeef5461d06dfc34ac0afaa96dd67212b8e
1528 F20110218_AABLWA babbitt_g_Page_072.txt
a2752ebbf547d83b5fae3859f55483d5
66948aea5b32064ec7727538c21ac4f1c2c6fc52
57698 F20110218_AABMAS babbitt_g_Page_107.pro
969ea587cf36a1d007ffca667548b053
b3469fd5e96370e22089fe9f94a4f4fd61cebe54
1328 F20110218_AABLVM babbitt_g_Page_055.txt
8995b83461efae0336261df68598a225
a7dd8576033317303168564af38281031650b9f9
449 F20110218_AABLUY babbitt_g_Page_039.txt
8d31421b90c60a5aa3556fcfac5994bb
c67dc21cabc53c597fb96cadef2889099a568474
9004 F20110218_AABMBH babbitt_g_Page_007.jpg
3d1188b851ff588396de9ca15f088157
73dde01d279194219338c9e8251560194f35bbe1
1645 F20110218_AABLWB babbitt_g_Page_073.txt
adc48f621cfc8b2281fe58b754ad333a
5ccaa7d511855138d06d0650227f7cd9e714424f
43771 F20110218_AABMAT babbitt_g_Page_108.pro
57aa18b8c00e87699a1ee0d2ab733953
b6ec05d57ef3be93c8374abd0362021bd8ea67c7
886 F20110218_AABLVN babbitt_g_Page_056.txt
5c05895ef0d271bbf12e31580f508fc2
4e354b3602cab49ff8ea1828d55b72ec2ccc8958
1346 F20110218_AABLUZ babbitt_g_Page_041.txt
b28262ed02e271f8ccb6d11e5c71978a
56810d22eac05c7e2ff3b64da448928aea4962ba
2582 F20110218_AABMBI babbitt_g_Page_007.QC.jpg
fb324f1f7d3cdd2c3b64bf8a77e0d91f
23eea272f23ad30c7526e59d3831a0e1adda80cc
1946 F20110218_AABLWC babbitt_g_Page_074.txt
31a00b6cdbe34d5eae76f239aaa5d398
b92dc7f3dc027484b8386e76bf65ded58725708e
42361 F20110218_AABMAU babbitt_g_Page_109.pro
d0ad058c600aa751c93d1cfde54bd09c
0517256bc5e97b5e5be8191e16e7bb746cfd2214
1629 F20110218_AABLVO babbitt_g_Page_057.txt
84a819ff7194336717ae0f97ab4202f9
15460199b5fe591a83c1c07224a64967c823f184
59681 F20110218_AABMBJ babbitt_g_Page_008.jpg
7a0c858673ca712f772855c07e31f0b9
2c9cd386db01bdc501cdcd7daeb94e7f67dbed0b
2008 F20110218_AABLWD babbitt_g_Page_075.txt
9cd0db5741dbce291688d3b7e6533d97
bf93549976175d9f83da1f9f6ffa7d0983294b9b
24819 F20110218_AABMAV babbitt_g_Page_001.jpg
74596212cafa77dba890fe1ccc1ebafd
adb601e48bb5381a04bd3332e88c1639587f6c64
1934 F20110218_AABLVP babbitt_g_Page_058.txt
8027293622ace213a5c688526c596b45
762468081f8defb6824fa92d6a6fb4da9af67d71
108779 F20110218_AABMBK babbitt_g_Page_009.jpg
f4d54d89d97531915ee241fc9382d4ae
89cf42e9ff18794e2c6dc61813b2bbb585e570f1
1855 F20110218_AABLWE babbitt_g_Page_076.txt
2f9d9721550bf9156b33d7e5440179da
542b310979677c749e7e918213d9174bc4fdbf3c
7716 F20110218_AABMAW babbitt_g_Page_001.QC.jpg
5cc07ec482ee702b20da42bd0b7543a5
6e27d6147ed6a4f814475d046b3ab42215123ff5
1489 F20110218_AABLVQ babbitt_g_Page_059.txt
9abaac46fe7dc3a0e708ad01f9ce4e38
89c52c9b6263e5e2ac651577af5da60aff145071
27910 F20110218_AABMBL babbitt_g_Page_009.QC.jpg
fe82b94d0f8a0893db012a6f6bccb4dd
b191aeaa0094e017fd2f05871b16baa71e7cbcc4
1957 F20110218_AABLWF babbitt_g_Page_077.txt
a74cf88c5e555a168ac6dc804b8b66de
bdfc153b22045e05e5d611ff8478f8605986272e
4992 F20110218_AABMAX babbitt_g_Page_002.jpg
ee8cfd4cb7208393ac02c5bdbd97ab11
8a5c0379c9eb3304351bd79e9d0f9ade8a55e335
1066 F20110218_AABLVR babbitt_g_Page_060.txt
3c6713b2f5adf57e52caf947d2aa67ec
7140ae81a3f7f1c02f230ccc8d03ca1501e6d30c
33801 F20110218_AABMCA babbitt_g_Page_018.QC.jpg
2273a602f006378c4242ffd877bab53b
169fa9b3517d8d5dd6c64715fad5d4546e2f7c24
10036 F20110218_AABMBM babbitt_g_Page_011.jpg
5ca84aee97c8a9b5c61d0d1ff1f2bbb6
d4f636258ec55269be50e91d80f616a469d863ec
1902 F20110218_AABLWG babbitt_g_Page_078.txt
689e2c9dca53090ce7f90a98b39502ea
850a80f74d248e0101a6b277ddb36daa67e75aa3
1704 F20110218_AABMAY babbitt_g_Page_002.QC.jpg
4832eab457243905e501e0283e13aa5f
cc65b6b5e04fe6e094afe45297d5be7d345e6f17
1009 F20110218_AABLVS babbitt_g_Page_063.txt
392560366bfe23a18dce33c88f0cd777
513399908a7229d36391b2d1f6c069dcdb65cf08
101733 F20110218_AABMCB babbitt_g_Page_019.jpg
189fdb5b5cf5a7aa340424c1a6c9c03f
6b6eb21180158f650e765cc5b8d9f2aa14d5f7d6
3215 F20110218_AABMBN babbitt_g_Page_011.QC.jpg
2a87619761f8f8c2a46856c95071894d
61c0d225048b3fd50e789207d7c946b4e1dc3db6
1915 F20110218_AABLWH babbitt_g_Page_079.txt
1ea91e8f5173aaf1eabb05c926058ae9
d2781c5b18ea431aef9eb0f1f6a8242876f7c9d6
8682 F20110218_AABMAZ babbitt_g_Page_003.jpg
998bac9214e08a278bdd05e540c8a190
0417e2940f616f187c23761e750538061bcc80df
2405 F20110218_AABLVT babbitt_g_Page_064.txt
2c4dca74dcf236c3af300d0431682fd1
c91a9b5bf52fa4e19857eacb510b24b178e1f7e8
32712 F20110218_AABMCC babbitt_g_Page_019.QC.jpg
1553c6ae00cf1a5c291963436732a531
97c4bcd18c9b369f6dfe1633e487af5666f887b9
79869 F20110218_AABMBO babbitt_g_Page_012.jpg
4a4cc65efefd5eff30655f6150cb70fc
3910d0057ada5fb3868c0e74a59aa594d5088688
F20110218_AABLWI babbitt_g_Page_080.txt
d75ffc14256d7f87b98987ae1cf40b83
392036a8e8ad968df10b49c25f4e470485123d05
1930 F20110218_AABLVU babbitt_g_Page_065.txt
fc56f8617b052f4a0171a440dce6acdc
a06a170b114234ad2e5c4a89579ab555c02be00b
97658 F20110218_AABMCD babbitt_g_Page_020.jpg
e759e0e9e6910f48210f7ade66566a92
b8cfee7a439092b8fef2c34ea30de3b9b40f7f82
25469 F20110218_AABMBP babbitt_g_Page_012.QC.jpg
b37a9a19be3ef31318742e26d0c9ce8a
7530c89036e75f63a8e2ee286fd2432f9848d110
940 F20110218_AABLWJ babbitt_g_Page_082.txt
71a4853dcfd599ef73abb8e7ce459dda
a9381558a8f7672ba22246c26e2424516eba97da
1048 F20110218_AABLVV babbitt_g_Page_066.txt
c3635984bc58e9fd3c2a9998d853d465
dd6b57068b3cef8275039218ccc7456e5b88891e
31311 F20110218_AABMCE babbitt_g_Page_020.QC.jpg
f81398281902e4c9f3b5966a3574a7d1
52d0947acf8e0f00198ebc45453cf5136916f360
26748 F20110218_AABMBQ babbitt_g_Page_013.QC.jpg
5e8f9ce8fceefd5f8666515ed78bc8b3
0491c5a848db250783ffc1b40599acea8f05d290
F20110218_AABLWK babbitt_g_Page_083.txt
096e5572b1806e2faaa4f4c8209c49d9
44d2c8cb0fd2bf385c8be5a0fba8b2045f08d87a
1866 F20110218_AABLVW babbitt_g_Page_067.txt
b7fb23598b4b66957b6ae0b910e39c82
fdc76fae7ee83ade352430272ddd13312d5e2fc2
92978 F20110218_AABMBR babbitt_g_Page_014.jpg
4a56bfe5aea5284afaa4915ad8ff1770
d1a5b3dd0d17afee74d0bf6be31ea8ac7f4cc29e
1316 F20110218_AABLWL babbitt_g_Page_084.txt
f2ccd713a7ceb29463a3a09da9b4788e
98374fa09919a43ba7df2a759b7eef1210d4fa5f
1950 F20110218_AABLVX babbitt_g_Page_069.txt
2a175ee8650bbcf4ecc0cc93a7819fcc
461d74edb1c4d4908d5ef49629bb7ebd7890d786
105333 F20110218_AABMCF babbitt_g_Page_021.jpg
c9a61567e9f07bfe48cb85b5d74ffe13
d9d4b6524b4dc9838e02d02844c6ceee31632510
2537 F20110218_AABLXA babbitt_g_Page_103.txt
4799bfb3c4884c91f3ba83da6f672e94
7d5264068db17cd68891ba475555162ec78983e5
30346 F20110218_AABMBS babbitt_g_Page_014.QC.jpg
8d3d627749a5fcb1f5bc4e4ebaba9f2f
23e45c370ce43e138ecfa17b6d224f7ecf3dca86
1628 F20110218_AABLWM babbitt_g_Page_087.txt
3eea5b3f82e66cab0739041d975acff2
e56d4695a69be9eb5755d88471fc84f0ac59ce7b
2131 F20110218_AABLVY babbitt_g_Page_070.txt
c5a5e0901e70c4a89023d69482708ef4
7107d96cfa9931c5acbad6886bc9a1d5de18d42f
34148 F20110218_AABMCG babbitt_g_Page_021.QC.jpg
7797e01828ff5cb71d88f3df9288f621
f996d2a411ae0be6d9aceeb601cefae08c1e6515
2385 F20110218_AABLXB babbitt_g_Page_104.txt
4ad7720d3e6d75173345461a61c081b7
85fcd0a0716ff0fe90b664c162f781ddfb8f442c
104123 F20110218_AABMBT babbitt_g_Page_015.jpg
06d688ba3c51563e1678157c1e1f5046
f11aef07ffea07b2cb3257f422ec688fee70ec3f
1900 F20110218_AABLWN babbitt_g_Page_088.txt
59ffa62c2dd9a8f99669ecfe72df8f57
c8d251fb69859f80de89755c425d75722bf9ee51
25840 F20110218_AABMCH babbitt_g_Page_022.jpg
afd1d2d8b3b3797774dd02f58d58c583
7ee14a3cd1b83c69a9780bda674eb2e5667020ac
2333 F20110218_AABLXC babbitt_g_Page_105.txt
51857c0e4c5572dc6d0ed4f629a7e8ef
e93d5e3c44858f5d380a7f8fd3d578d4dc523213
33489 F20110218_AABMBU babbitt_g_Page_015.QC.jpg
674a7e7f0d00b2974ca3129886cad51d
5cdd3049ab3a71d3117bffc17aca200a383778d8
1966 F20110218_AABLWO babbitt_g_Page_089.txt
ae8e66e0d41a8e20a415f38ea08b1911
ada3e5b709d0d26c1144bc2d76f8fb1e65973d73
2007 F20110218_AABLVZ babbitt_g_Page_071.txt
1f6a74bedc8d5199d1ba5ed1c69de531
78a97e3daeec95be8d97f9b8e49eb566a209f7f2
8688 F20110218_AABMCI babbitt_g_Page_022.QC.jpg
041ef2a47732b83bb160e17192e5d1cd
5723607b2c1cec597db723b65713f210c73ccdc2
2135 F20110218_AABLXD babbitt_g_Page_106.txt
392bc873d77f3095aefce4799a54a4af
c9bf7460dc9ccd609cfa66521714cf8492016e2e
102956 F20110218_AABMBV babbitt_g_Page_016.jpg
2e1ca7a32aa88e7b95a78f5f5cc18f0a
fd88de171755a0e1b799383c35d5610ae85d6010
1984 F20110218_AABLWP babbitt_g_Page_090.txt
5f23b505feb590c7f512ea604662aa45
ed77dc014b54484afa57d828a39a2b082e2d4d8e
28959 F20110218_AABMCJ babbitt_g_Page_023.QC.jpg
2737f7b1d4672655ddcd9e4213c0d5d4
3ac28663644c1589d1a32012d0a2f95429ba00b1
2324 F20110218_AABLXE babbitt_g_Page_107.txt
bdf3a379fea78173645c09fbcbb4eeba
ed6b6aeeaa92e682ea8383db1e402782976f88cd
32818 F20110218_AABMBW babbitt_g_Page_016.QC.jpg
7bec68c13b4b5465cc7051468334f595
88d194103a4975c2bd0b64b0d4b14b89009e8fd8
2059 F20110218_AABLWQ babbitt_g_Page_091.txt
ab4abf9442d01de6151598545ecd7731
88cd0920e1ec8186f13b2679c4769ff1cd167fae
102546 F20110218_AABMCK babbitt_g_Page_024.jpg
859b1a62a6d1fa3518a971147e2df205
b2fe42ce6aceaaacc2722895ccb9d76288b6bcdf
1796 F20110218_AABLXF babbitt_g_Page_108.txt
20284a3dee426c64be3f9db01b9c773e
499622cf85aa1f9017e679442db14c580296a0ef
102513 F20110218_AABMBX babbitt_g_Page_017.jpg
5c49bcc8fc7e733d0cd43cf3e16a233b
edb3b9302775e07ab50bafecf4fc9d459398dcb9
F20110218_AABLWR babbitt_g_Page_092.txt
5e0bac5dee236c99f5392e5d156c1872
d8d099f949d564163ee5555f43c4ed290c63eccc
100637 F20110218_AABMDA babbitt_g_Page_034.jpg
97dd50830b2c61b706dccd80998528e3
fd63b360d09b384782268aaa9501b6a37fed94db
32674 F20110218_AABMCL babbitt_g_Page_024.QC.jpg
9beed4eecab351af377263f1c5c5a2aa
eec66e717b56e0b747e347d16c622e52baf85e60
F20110218_AABLXG babbitt_g_Page_109.txt
e8487121d4921e31c6b1643287d179ba
40b69b9158f40e8a542ca8a761a8cc37471aee87
32744 F20110218_AABMBY babbitt_g_Page_017.QC.jpg
6c9b87bc2b79b292d2ac18232b014be3
3d926c4c790bc18c02d070048a4553b5e42a492d
2114 F20110218_AABLWS babbitt_g_Page_093.txt
9c36d5ea5bc95ded848aaa463742262f
440fb8c8b797d62c87c13a21d1be52e48c78cbbb
32450 F20110218_AABMDB babbitt_g_Page_034.QC.jpg
0aba282e7c2bf9f9366407b8b7b80508
148fe9c73333824652a81dcf3d42500a406ea736
107745 F20110218_AABMCM babbitt_g_Page_025.jpg
77c0750317311ad18d5a678411b417c9
e29270c99de94581e2401fb4f7237113cf7dd6a6
8074 F20110218_AABLXH babbitt_g_Page_001.pro
8a8da41179f96b54889db4852a3f66d6
34f2099c3848277d21e7259880f8a9ed12cd0826
105357 F20110218_AABMBZ babbitt_g_Page_018.jpg
60c9a1d74229edbf45f939596305f574
43b9220ac4b14201f6650def8b9a95c72870fc22
1924 F20110218_AABLWT babbitt_g_Page_094.txt
24680435d51145f232cc4f21c13d2a14
966a79561f1f9ceed67e29ff64a3cb549dbb2c73
97470 F20110218_AABMDC babbitt_g_Page_035.jpg
f84ad63dcdf766d1b7c5660804b3edbd
fe3ec8582d34b0bb6d36b557344a07fea8e93ad8
35589 F20110218_AABMCN babbitt_g_Page_025.QC.jpg
39ec41683dfa7531bb3088ee9ecf0fcc
4b31dc8a25b49bb90c836c9cfc512d7edf808f84
1305 F20110218_AABLXI babbitt_g_Page_002.pro
696f43df9d3622b16850fb7d2b87fa5e
6d698e5d15bbb88e0c7e0e93fc47c5db710f3b94
2004 F20110218_AABLWU babbitt_g_Page_095.txt
66aedd4d1f8c44bcfa8499817922ab6b
337671a5b1c90ba7c26acf3b913cab19b4a44c00
31846 F20110218_AABMDD babbitt_g_Page_035.QC.jpg
ed66f0f7a0abee666fc2bed0b988f556
f138612d88f4b02aae61b91d32fd939d7770444c
22049 F20110218_AABMCO babbitt_g_Page_027.QC.jpg
b97af19dd41df3fc61d50071de9d4d03
78dd726b4566b3f37b73bacec18b24a961fdc438
2954 F20110218_AABLXJ babbitt_g_Page_003.pro
33e89ef00aa4fd083ec8c9bd7f42b6ff
3b14ee11a012af690a40e0756e3c09da60383aa4
297 F20110218_AABLWV babbitt_g_Page_097.txt
9d3f13667ad42aa0e9bca08c8f4065e5
c91e366a7ea9f1c892cc8f27a925b47402ee6394
119549 F20110218_AABMDE babbitt_g_Page_036.jpg
8c7d129c2dab7d777139f35f3928b6c4
2aa7a502a8b3d31cc895d1ad77d79a2011d49185
105851 F20110218_AABMCP babbitt_g_Page_028.jpg
8a1e7e64437f7340e496522db47d2fd8
62c42824148e1247daa5d7b134581df8f6f1f154
15940 F20110218_AABLXK babbitt_g_Page_004.pro
cecd3aab713104a86431bc5421dee382
b3ae715a123a1d8cb550b3f60e66b7474f748fc9
1177 F20110218_AABLWW babbitt_g_Page_099.txt
3bcceb708f62d74dd8e04f64330fccc8
012dbf1d7329a02824632ec814453211ca00dcff
35460 F20110218_AABMDF babbitt_g_Page_036.QC.jpg
f2de41d0237c62d2605acbb35d19fb30
8c9d7112458d7009a8b0c15478adaa9e4522a249
34308 F20110218_AABMCQ babbitt_g_Page_028.QC.jpg
f59e17eab0f6530822c430000abb6937
8080b1915c1d7cfe33590420ae30eba15b98d50f
101713 F20110218_AABLXL babbitt_g_Page_006.pro
cf4360863d305ee5a8cc516bba88e8e0
099c8beb75334e4dfe8d65c2605342d0aedb3f18
1372 F20110218_AABLWX babbitt_g_Page_100.txt
a156cb4a7ea7f6b19181586f1377cb0f
6524f67f654490e5664407d99d6c10dd9912c7d8
104282 F20110218_AABMCR babbitt_g_Page_029.jpg
6b1d62aca2fb4c477d7967af7dea9d6e
71022bad2b190a6a249e8d763cf4c73213523ac6
6886 F20110218_AABLXM babbitt_g_Page_007.pro
e86c8b84d440dc51c93d450a34d8922e
4157bca516a0c9ab53efa03716047de685251d12
F20110218_AABLWY babbitt_g_Page_101.txt
68674387fe0e33f2266c45f09f185776
4fe5a228d4a1a49aa292bc21d1415de39e7636b9
28634 F20110218_AABMDG babbitt_g_Page_037.QC.jpg
f38919093fca647f6240563008ec4875
3c1b538a54ea3a99884d85c77c719ff4eee8785e
52389 F20110218_AABLYA babbitt_g_Page_028.pro
88fae03058eade756d5a66e5cedc6ac7
9d71878ee9b8c98736d44813da4d8edf95f596ad
34564 F20110218_AABMCS babbitt_g_Page_029.QC.jpg
1c3294150b769e6b37a53ae48a0028fb
2cc05aea2702128205581993c904afb79fae733f
35372 F20110218_AABLXN babbitt_g_Page_008.pro
7448e55222554363708af74f406022c3
1da70fbd75e582b02733379578e51f04d7f2f699
2322 F20110218_AABLWZ babbitt_g_Page_102.txt
da8ae07365ac684d046d2efaf6170a74
2df0df4f2d9f3c798581323d4141bb310cae654e
59232 F20110218_AABMDH babbitt_g_Page_038.jpg
e13243c3b74254dc31de6ffba59fcb3d
4978676cf14aba5fa6de4baa0e89e71ae549eecc
51453 F20110218_AABLYB babbitt_g_Page_029.pro
a5a06f2762bc25d2ba5d3ffd64769e71
d8387f263a3cedee7e4ca91b41afd3227ba371f0
95141 F20110218_AABMCT babbitt_g_Page_030.jpg
921f2725798e065d18950b2f9baee55f
bdcec8e36cfbd2bc99f94ed085efae4953b81c2c
69250 F20110218_AABLXO babbitt_g_Page_009.pro
5e6d37121cea73b641bd1700d9d31d91
9d06f78980e282f3b3d8c453498dd4e28a5af152
19973 F20110218_AABMDI babbitt_g_Page_038.QC.jpg
c9470af95701d6a2cb45e1f6d5853146
d1cd294570bdce26d4ea99ab2e55191e3040c23f
50098 F20110218_AABLYC babbitt_g_Page_031.pro
0237dd025415cc3419e9314599d1f30d
eba79b0ed96fa0e2c638c474d459bb4adaa8a932
102486 F20110218_AABMCU babbitt_g_Page_031.jpg
7ea9476ae976b99c51ecf776b4082344
36150b079f4fdc1816bd6fbae9ac999a4b2f477e
86178 F20110218_AABLXP babbitt_g_Page_010.pro
b51f1ccb13a9d650db318e06fbaab5f6
20edd32bf784e887176e0edae11b120ebc578cde
64790 F20110218_AABMDJ babbitt_g_Page_039.jpg
ba73c161ee62a0a4d645a410c3afb683
479d25e0e71f7c2a0a07be0ba6e6def049f35c22
48465 F20110218_AABLYD babbitt_g_Page_032.pro
b15601a6f7a0c878b08dbf10d5a4182d
3287525db14e354efa9afb6484c7a4ddd2d96fde
33573 F20110218_AABMCV babbitt_g_Page_031.QC.jpg
5fc7044f75e77b548c332f9b438a9180
03e1877373c8cbb5e2ec1d3a252c890829690873
38888 F20110218_AABLXQ babbitt_g_Page_012.pro
2aa839cccdfc2d14c08b13d8511b2589
9ee16be6d86ed46f66a1d15b044ced31d632e549
21222 F20110218_AABMDK babbitt_g_Page_039.QC.jpg
57cfe3818ac998879ddb1b280ee42dfe
fb11f8298fb70f474c3838785e07ec5a7bcdc332
50687 F20110218_AABLYE babbitt_g_Page_033.pro
bd5436e145c7c3e887433200c209280e
f8ea0d456d42aae5765bd294d22fd430452d27bb
99721 F20110218_AABMCW babbitt_g_Page_032.jpg
f3e62573e2ed1461bc5e71e20d3aca0b
6619265e987a9860dd44a77108ef2e4545b84014
39933 F20110218_AABLXR babbitt_g_Page_013.pro
23460040b284fc64abe0707ea889c6e0
54814d86902518c4d28a61aed55fccf533f4b103
34072 F20110218_AABMEA babbitt_g_Page_050.QC.jpg
609eaf323147a13aab6aa1b60c71fb66
481abc1105e050706c574664d6cd4d9fec18088b
75399 F20110218_AABMDL babbitt_g_Page_040.jpg
781e4fc67337dc66e7bbfb190a7a45dd
8974b8d163cccc7376ea4c524b2053da270ac094
48554 F20110218_AABLYF babbitt_g_Page_034.pro
0d3ecea44ec310262a0422aaba473939
be6f2e0a078f16aa81704c34e9df4adb3eac49fb
32458 F20110218_AABMCX babbitt_g_Page_032.QC.jpg
bfc39b5559f47c1abd6cac0c9b147047
5c605a0fdd539622cdb9777f352e919f9d2ecdb6
51601 F20110218_AABLXS babbitt_g_Page_015.pro
a6976019425c02e2c3e165d41a50e5bf
5a3278cb62c944dc3d5956bc4b151f4ec200bdf8
35780 F20110218_AABMEB babbitt_g_Page_051.QC.jpg
d2d3d290edff00dd48ecac00a723489c
52d2cad90ccc4781f90bfe2f379911f0f7059f56
23579 F20110218_AABMDM babbitt_g_Page_041.QC.jpg
ddf83fefa5ccd1985bbfc22e902bfa55
4909d0bcfeac582f7a82a07380728e60281062ed
47785 F20110218_AABLYG babbitt_g_Page_035.pro
c90196262a8a76c8f5374801f5633671
d7de21103aff395718376685c3b982767bcd00e2
105246 F20110218_AABMCY babbitt_g_Page_033.jpg
46bfc1554531d52c39859327b5a67efb
0056476c59560a353b70c89fd1f9b43faadac0ce
50748 F20110218_AABLXT babbitt_g_Page_016.pro
3401e6f1217669d553a3a0ccd403f692
d1fb62b4d6c186b5feba5646a9e4d764ae80cc9e
101782 F20110218_AABMEC babbitt_g_Page_052.jpg
ad8ec3302fc25bc6acbe5357bbca1955
e9c7396869e758285bf32bcd79b37efa2e4407f5
90365 F20110218_AABMDN babbitt_g_Page_042.jpg
d07e7ce3c287b04a860223706b987775
f9b9f05a220d05487b029c5f3eec2eb1e1adeb67
67397 F20110218_AABLYH babbitt_g_Page_036.pro
4f53e45f8e4e9269c87aafb4f6511d15
1ec6ef35e816a6ae85074e76f13d30e7521cb846
33425 F20110218_AABMCZ babbitt_g_Page_033.QC.jpg
ba9982a995b814305dce717e0272dd76
75b166b4e8362b0890f3003b92d76ba32ecfbbca
106359 F20110218_AABMED babbitt_g_Page_053.jpg
32fab01dec7d503409d24ca0e835b1d1
7a6a422679821241db607c4132f9a8f5d2da30c1
28583 F20110218_AABMDO babbitt_g_Page_042.QC.jpg
8be7009e4b18e3ef98d4c559edd162c9
b9dd3ac7dade6d72a00b76a284287e6beb0c89f0
45746 F20110218_AABLYI babbitt_g_Page_037.pro
5680203ee28f04bd00cbb013c9d86a0c
f28cc2065f9a19686d2d32974df9986530e04a3b
52642 F20110218_AABLXU babbitt_g_Page_018.pro
45d1ae5e97f564a589c0414603a5fa06
c05b7d70887aa361c1c38dce4669fddbc942ce4a
35172 F20110218_AABMEE babbitt_g_Page_053.QC.jpg
a4c25dd132fed07340a396f708fcfe50
f240f3f73b16d9fc5023998fe740c7894f1b9fc2
25715 F20110218_AABMDP babbitt_g_Page_043.QC.jpg
c10fe49207bf4fde8c2279ba6aaa1f76
c833a624a99608918bccf6522b74de0a82f86170
9072 F20110218_AABLYJ babbitt_g_Page_039.pro
099c4b1e5e9439cc9a4e5cfd0241b99b
39ccfcfb6fd9fe2249f63987fd6e26ed2c85ecd5
47649 F20110218_AABLXV babbitt_g_Page_020.pro
49cdd272ff6bfebdaefce7c4e6806eea
0dbbd9cca418e946ef747436f3e02fe480467044
20952 F20110218_AABMEF babbitt_g_Page_054.QC.jpg
ef82759c288cbb3f04e22837e57a0ade
30a879c0384d8df416bc5d16848c8ef1d8f36f17
104269 F20110218_AABMDQ babbitt_g_Page_044.jpg
f0572b74cd7f543340b53a681cdc8849
eef0bfae10920b1b0af1143e4093f48dcd1dd4de
21554 F20110218_AABLYK babbitt_g_Page_040.pro
ec6308b916c11ba5d375383fad61c42d
df921b648df7b8ed4db8882e0f057482955c04a0
50077 F20110218_AABLXW babbitt_g_Page_024.pro
805369cfc0c41ad118a79148c1df9105
b62d1fd360be2ae03f8a6b2e6eaf82f7f950cad5
71093 F20110218_AABMEG babbitt_g_Page_055.jpg
15575fef70106acde1f337ceb5038fdf
c9c3c06d2d09f15513bd9e82f4840f8a96a52542
33847 F20110218_AABMDR babbitt_g_Page_044.QC.jpg
a9700dcc818afe220fc45cee20db8bf5
7b510b52f7268fbbd56d4ec378d102df23a699f4
31461 F20110218_AABLYL babbitt_g_Page_041.pro
cdedadd917fc0ab355e4e0d54a5f638c
c99b648e4c99ec56d20348de8a393e950f815ae0
53994 F20110218_AABLXX babbitt_g_Page_025.pro
38b5cab42719400bc7d70e481c8f2b54
303b9b1574cf89408fb7ee09a62eaf0224356fc4
27548 F20110218_AABLZA babbitt_g_Page_057.pro
de327d51c3c70f61cd1ab5b807c2de98
d6b0aced3c60961d1103ef6811c1f750ddd2f0a4
108426 F20110218_AABMDS babbitt_g_Page_045.jpg
7a7f1de345d328499b7810973ae2814e
b74e4746886365f543a6e5e8fc644e227e37fa5d
51560 F20110218_AABLYM babbitt_g_Page_042.pro
dd734dad042dfbba2713958800456f78
a55384c2f70dfc83d0522ea8b2482f2f53da1f0f
52707 F20110218_AABLXY babbitt_g_Page_026.pro
25319e53fc4602032ae55ad353980231
157ea9079bc2ed20c1cc5fe12c6aa3b51f2c7ceb
23503 F20110218_AABMEH babbitt_g_Page_055.QC.jpg
af9da86195121b7fde2945a5944646a3
3f6ec2c65107cfbadd542e3d6c927f3cc369a468
48967 F20110218_AABLZB babbitt_g_Page_058.pro
7cabe58e659ce87eb800170104335832
ebc8dadd12b0d8374dc30808b60c26afd6721733
35626 F20110218_AABMDT babbitt_g_Page_045.QC.jpg
1dd37cee46bd28f044e459279dbb8dfa
41e332b9e525e3c2017893d35ff20365eeed910c
45363 F20110218_AABLYN babbitt_g_Page_043.pro
76cf62a2402277e1dc2e0a12dc794a9b
d66ea9abb29f13f0fee93a27dad9b6819b8b096a
27588 F20110218_AABLXZ babbitt_g_Page_027.pro
44944cb80e59016c1c52427a44792da4
2a870a91e91805fcfa759145b0394b4eaae5343c
48156 F20110218_AABMEI babbitt_g_Page_056.jpg
a16089e1750417fd7b1e8235d0724361
9efb1c7471d5b52478095c23cdc1c3b1feebea12
29549 F20110218_AABLZC babbitt_g_Page_059.pro
78f1fb47e1bf6ec6f7246538649caeae
17c9233ef973d93e1d6faceb3940e61d91d7ed7b
103679 F20110218_AABMDU babbitt_g_Page_046.jpg
8a512cb896bf9a82a694780eabe07ed0
04deaaa79d33c7d85f9eb2949bd41559f719704d
51329 F20110218_AABLYO babbitt_g_Page_044.pro
929ebf89e483a88fef4ae67757818587
bd150a6887c73f09a5bf157f857c5dd99d18f19e
64014 F20110218_AABMEJ babbitt_g_Page_057.jpg
ec8a6a4652f4745fd1c4d21afb1e58bc
c583507d84a32bdb47e5db3a62ebb1f696f76d9f
23879 F20110218_AABLZD babbitt_g_Page_060.pro
03ae214d2988b4b488435c2bd3100ac6
0db7d3333f78881e4f883b9bace354ddebe731fd
105633 F20110218_AABMDV babbitt_g_Page_047.jpg
624505cc5d43c1ab57e3085a16e87f01
9361f50f623bb0d17cb888df071bdf743dc4f49b
53261 F20110218_AABLYP babbitt_g_Page_045.pro
4f8ee4860130773b258ff79539dfe769
5e98c97356e6a26446fdff969ba4d2cb162422c0
19311 F20110218_AABMEK babbitt_g_Page_057.QC.jpg
77ff338a1079537afb1c384d3bbe8404
8d5d7fd74a18220315faf8b4cc0e7b5f04a2dec1
49615 F20110218_AABLZE babbitt_g_Page_061.pro
8ebc02ffba8627a5225a7eca997bf57a
4c84d3ac937a2a3ae5e0c4e0e7543214ac47535a
34048 F20110218_AABMDW babbitt_g_Page_047.QC.jpg
29fb2cf35d08e9bff45ab0af9db66738
c74d4997723910fd0179b6b7280cd0da86d7865b
51126 F20110218_AABLYQ babbitt_g_Page_046.pro
a562193125989ae6da906c7aae28d655
73714dd6c856dfa2a4effd1b7ebe7ce7c411a46b
100241 F20110218_AABMEL babbitt_g_Page_058.jpg
79e95ccc2f19ebad025dec3aea10b666
72c7259c068f6a5452ff2eb85c337cc93fea8cf2
45999 F20110218_AABLZF babbitt_g_Page_062.pro
17a1a3efb57cfdca814100ba8def8f20
bcc510a81bf395688dad3e9ff1cf83fc5b67894b
41268 F20110218_AABMDX babbitt_g_Page_048.jpg
03d721a36b039c292d451a0d6f4bc00e
0e3bafccf946d7449452f0c7c8af8e17d7806580
51770 F20110218_AABLYR babbitt_g_Page_047.pro
8b3fc5c4b64b3cb6a94e2d1cfa9478a5
38efec2000f060c15145441e372b7bc6ed4da82e
18969 F20110218_AABMFA babbitt_g_Page_066.QC.jpg
99bfec28d2047cb93ca76b97247fab91
33eccadf58c6415dbf1d2907f8f274e299bc1e9b
33190 F20110218_AABMEM babbitt_g_Page_058.QC.jpg
6ff0bc70a3110e60d47507d69b22ee3f
02c3a72ed632178d5a8404c4a7edcc647537812f
22576 F20110218_AABLZG babbitt_g_Page_063.pro
2f669580ac1cf773fc3101f517d0d6d9
f8fa62343ec9a884a5631d183cb919f16bb6c665
13825 F20110218_AABMDY babbitt_g_Page_048.QC.jpg
f2295f9c02b4a8ebe595f4cb47a710cd
fb84ec836fbf2ec185af075f8cda6599121ffa08
18867 F20110218_AABLYS babbitt_g_Page_048.pro
5c5ccc60d3f2f96a91ad961e36e3f4d0
0efccefa6df64c1842e95d2074088ffd682c4e09
95632 F20110218_AABMFB babbitt_g_Page_067.jpg
29b979bf9a1eb210b15cc4cd4830694c
6326e86ef5eaa5182dce6fe18cd5064635d75e56
23488 F20110218_AABMEN babbitt_g_Page_059.QC.jpg
9f9f825a302f63c7a16d543b8c4ef798
c0866cfe6cf7329afec61880ac1c8dd05764f3dc
49127 F20110218_AABLZH babbitt_g_Page_065.pro
2617e1f1e57374e2a27d566370e55271
808a5c67bf3f63eacad7a9695f234d004cac180c
105810 F20110218_AABMDZ babbitt_g_Page_050.jpg
8d3655de162808db2a6625384b83dbac
f1c3c9cc1bf962934469db67d840fe0e728c8d75
43046 F20110218_AABLYT babbitt_g_Page_049.pro
d24eaf8688abe626c32e6bbd61da46da
d1e504b0eb5f15db77c9b2ad73921f9f50824e48
31305 F20110218_AABMFC babbitt_g_Page_067.QC.jpg
a4c9db6b780798824b6f8fcbed50105f
707568794f01e7682206ce94b8e4bdc1c021d79e


Permanent Link: http://ufdc.ufl.edu/UFE0013688/00001

Material Information

Title: Characterization of the Distribution of Developmental Instability
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013688:00001

Permanent Link: http://ufdc.ufl.edu/UFE0013688/00001

Material Information

Title: Characterization of the Distribution of Developmental Instability
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013688:00001


This item has the following downloads:


Full Text












CHARACTERIZATION OF THE DISTRIBUTION OF DEVELOPMENTAL
INSTABILITY













By

GREGORY ALAN BABBITT


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2006

































Copyright 2006

by

Gregory Alan Babbitt



























This document is dedicated to my grandmother, Sarah Miller, who taught me to admire
the natural world.















ACKNOWLEDGMENTS

I would like to thank Rebecca Kimball, Susan Halbert, Bernie Hauser, Jane

Brockmann, and Christian Klingenberg for helpful discussions and comments; Susan

Halbert and Gary Steck (Division of Plant Industry, State of Florida) for specimen ID;

Marta Wayne for use of her microscope and digital camera; Glenn Hall (Bee Lab,

University of Florida) for a plentiful supply of bees; and the Department of Entomology

and Nematology (University of Florida) for use of its environmental chambers. I thank

Callie Babbitt for help in manuscript preparation and much additional love and support.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ................................................................................................. iv

LIST OF TABLES .................................................... ....... .. .............. viii

LIST OF FIGURES ......... ......................... ...... ........ ............ ix

ABSTRACT ........ ........................... .. ...... .......... .......... xii

CHAPTER

1 SEARCHING FOR A CONSISTENT INTERPRETATION OF
DEVELOPMENTAL INSTABILITY? A GENERAL INTRODUCTION ................1

2 ARE FLUCTUATING ASYMMETRY STUDIES ADEQUATELY SAMPLED?
IMPLICATIONS OF A NEW MODEL FOR SIZE DISTRIBUTION..................... 10

Intro du action .......................... ................... ... ... ................. ............ 10
Developmental Stability: Definition, Measurement, and Current Debate ..........10
The Distribution of Fluctuating Asymmetry .................................................12
M e th o d s ........................................................................... 1 7
R e su lts ...................................... .................................................... 2 2
D isc u ssio n ......................................................... .............. ................ 3 0
T he D distribution of F A ................................................................. .....................30
Sample Size and the Estimation of M ean FA................... .................................32
The Basis of Fluctuating Asymmetry..... .......... ...................................... 33
C o n c lu sio n .................................................. ................ 3 4

3 INBREEDING REDUCES POWER-LAW SCALING IN THE DISTRIBUTION
OF FLUCTUATING ASYMMETRY: AN EXPLANATION OF THE BASIS OF
DEVELOPMENTAL INSTABILITY .............................. .....................36

In tro d u ctio n ......................................................... ............. ................ 3 6
W hat is the B asis of FA ? .................................. ........... .................. 36
Exponential Growth and Non-Normal Distribution of FA..............................38
Testing a M odel for the Basis of FA ............................ .............. ............. 40
M e th o d s ..............................................................................4 1
M odel D evelopm ent ....... .......... ........ .................... .............. 41
Simulation of geometric Brownian motion................................ ............... 41









Simulation of fluctuating asymmetry .........................................................44
Inbreeding E xperim ent ............................................................. .....................4 5
M orphom etric analyses ............................................... ............................ 48
M odel selection and inference....................................... .......................... 49
R e su lts ...................................... .................................................... 4 9
M odel Sim ulation ...................... .................. ................... ..... .... 49
E x p erim ental R esu lts......................................... ............................................50
M odel Selection and Inference................................ ......................... ........ 52
D iscu ssion ................................................................................. 54
Revealing the Genetic Component of FA .............. .....................................54
Lim stations of the M odel ......................................................... .............. 56
T he Sources of Scaling .......................................................................... .... ... 56
Potential Application to Cancer Screening........................................................57
C o n c lu sio n .................................................. ................ 5 8

4 TEMPERATURE RESPONSE OF FLUCTUATING ASYMMETRY TO IN AN
APHID CLONE: A PROPOSAL FOR DETECTING SEXUAL SELECTION ON
DEVELOPM EN TAL IN STABILITY ........................................................................60

In tro d u ctio n ......................................................... ............. ................ 6 0
T he G enetic B asis of F A ......................................................................... ..... 60
The Environm ental Basis of FA ....................................................................... 61
Temperature and FA in and Aphid Clone ................................. ............... 63
M eth o d s ..............................................................................6 5
R e su lts ...................................... .................................................... 6 8
D iscu ssio n ...................................... ................................................. 7 2

5 CONCLUDING REMARKS AND RECOMMENDATIONS .............................. 75

H ow M any Sam ples Are Enough? ................................ .. ............................... 76
W hat M measure of FA is best?................ ......................................... ............... 76
Does rapid growth stabilize or destabilize development? .......................................78
Can fluctuating asymmetry be a sexually selected trait?................. .................78
Is fluctuating asymmetry a valuable environmental bioindicator?.............................79
Scaling Effects in Statistical Distributions: The Bigger Picture..............................80

APPENDIX

A LANDMARK WING VEIN INTERSECTIONS CHOSEN FOR ANALYSISOF
FLUCTUATING A SYM M ETRY ..................................... ......................... .......... 84

B USEFUL MATHEMATICAL FUNCTIONS ...................................... ...............86

A sym m etric Laplace D istribution......................................... .......................... 86
H alf-N orm al D distribution ........................................ ............................................86
L ognorm al D distribution ............................................................. ...........................86
Double Pareto Lognormal Distribution ............................. .................... 86









L IST O F R E F E R E N C E S ........................................................................ .. ....................88

B IO G R A PH IC A L SK E TCH ...................................................................... ..................96















LIST OF TABLES


Tablege

2-1. Maximized log-likelihood (MLL), number of model parameters (P) and Akaike
Information Criterion differences (A AIC) for all distributional models tested.
Winning models have AIC difference of zero. Models with nearly equivalent
goodness-of-fit to winners are underscored (A AIC <3.0). 4.0 indicates some support for specified model. A AIC >10.0 indicates no support
(Burnham and Anderson 1998). Distances between landmarks (LM) used for
first univariate size FA, aphids LM 1-2, bees LM 1-4 and long-legged fly LM 3-
6 and for second univariate FA aphids LM 2-3, bees LM 2-6 and long-legged fly
L M 4 -5 ............ .......... .... ................. .................................................2 3

2-2. Best fit parameters for models in Table 2-1. Parameters for univariate size FA
are very similar to multivariate size FA and are not shown. Skew indexes for
asymmetric Laplace are also not shown............... ................... ....... ........... 24

3-1. Distribution parameters and model fit for multivariate FA in two wild
populations and four inbred lines of Drosophila simulans and one isogenic line
of Drosophila melanogaster. Model fits are A AIC for unsigned centroid size
FA (zero is best fit, lowest number is next best fit). .............................................. 51















LIST OF FIGURES


Figure page

2-1. Schematic representation of mathematical relationships between candidate
models for the distribution of fluctuating asymmetry. Mixtures here are
continue ou s. ........................................................ ................. 14

2-2. Distribution of multivariate shape FA of A) cotton aphid (Aphis gossipyii) B)
domestic honeybee (Apis mellifera) and C) long-legged fly (Chrysosoma
crinitus). Best fitting lognormal (dashed line, lower inset), and double Pareto
lognormal (solid line, upper inset) are indicated................................................25

2-3. Distribution of multivariate centroid size FA of A) cotton aphid (Aphis gossipyii)
B) domestic honeybee (Apis mellifera) and C) long-legged fly (Chrysosoma
crinitus). Best fitting half-normal (dashed line, lower inset) and double Pareto
lognormal distribution (solid line, upper inset) are indicated. ................................26

2-4. Distribution of univariate unsigned size FA of A) cotton aphid (Aphis gossipyii)
B) domestic honeybee (Apis mellifera) and C) long-legged fly (Chrysosoma
crinitus). Best fitting half-normal (dashed line, lower inset) and double Pareto
lognormal distribution (solid line, upper inset) are indicated. ................................27

2-5. Distribution of sample sizes (n) from 229 fluctuating asymmetry studies reported
in three recent meta-analyses (Vollestad et al. 1999, Thornhill and Moller 1998
and Polak et al. 2003). Only five studies had sample sizes greater than 500 (not
show n). ...............................................................................28

2-6. Relationship between sample size and % error for estimates of mean FA drawn
from best fitting size (dashed line) and shape (solid line) distributions using
1000 draws per sample size. All runs use typical winning double Pareto
lognormal parameters (shape FA v = -3.7, T = 0.2, a = 1000, 0 = 9; for size FA v
= 1.2 = 0 .7 a = 4 .0 = 4 .0 ) .......................................................................... .....2 9

2-7. The proportion and percentage (inset) of individuals with visible developmental
errors on wings (phenodeviants) are shown for cotton aphids (A) and honeybees
(B) in relation to distribution of shape FA (Procrustes distance). Average FA for
both normal and phenodeviant aphids (C) and honeybees (D) are also given.........30

3-1. Ordinary Brownian motion (lower panel) in N simulated by summing
independent uniform random variables (W) (upper panel) ....................................42









3-2. Geometric Brownian motion in N and log N simulated by multiplying
independent uniform random variables. This was generated using Equation 1.5
w ith C = 0 .5 4 ...................................................................... 4 3

3-3. Geometric Brownian motion in N and log N with upward drift. This was
generated using Equation 1.5 with C = 0.60. ................................ ..................44

3-4. Model representation of Reed and Jorgensen's (2004) physical size distribution
model. Variable negative exponentially distributed stopping Times of random
proportional growth (GBM with C = 0.5) create double Pareto lognormal
distribution of size .............................. ................................... 46

3-5. A model representation of developmental instability. Normally variable
stopping times of random proportional growth (GBM with C > 0.5) create
double Pareto lognormal distribution of size. .................................. .................47

3-6. Simulated distributions of cell population size and FA for different amounts of
variation in the termination of growth (variance in normally distributed growth
stop time). The fit of simulated data to the normal distribution can determined
by how closely the plotted points follow the horizontal line (a good fit is
h horizontal ........................................................ ................ 50

3-7. Distribution of fluctuating asymmetry and detrended fit to normal for two
samples of wild population collected in Gainesville, FL in summers of 2004 and
2005 and four inbred lines of Drosophila simulans derived from eight
generations of full-sib crossing of the wild population of 2004. Also included is
one isogenic line of Drosophila melanogaster (me175). All n = 1000. The fit of
data to the normal distribution can determined by how closely the plotted points
follow the horizontal line (a good fit is horizontal). .............................................. 53

4-1. Predicted proximate and ultimate level correlations of temperature and growth
rate to fluctuating asymmetry are different. Ultimate level (evolutionary) effects
assume energetic limitation of individuals in the system. Proximate level
(growth mechanical) effects do not. Notice that temperature and fluctuating
asymmetry are negatively correlated in the proximate model while in the
ultimate model they are positively correlated. ................................. ............... 64

4-2. Cotton aphid mean development time +1 SE in days in relation to temperature
(n= 53 1). ...............................................................................68

4-3. Distribution of isogenic size, size based and shape based FA in monoclonal
cotton aphids grown in controlled environment at different temperatures.
Distributions within each temperature treatment are similar to overall
distributions show n here................................................. .............................. 69

4-4. Mean isogenic FA for (A.) centroid size- based and (B.) Procustes shape-based)
in monoclonal cotton aphids (collected in Gainesville FL) grown on isogenic
cotton seedlings at different temperatures.................... ........... .............. 70









4-5. Kurtosis of size-based FA in monoclonal cotton aphids grown on isogenic cotton
seedlings at different tem peratures ........................................................................ 71














Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

CHARACTERIZATION OF THE DISTRIBUTION OF DEVELOPMENTAL
INSTABILITY

By

Gregory Alan Babbitt

May 2006

Chair: Rebecca Kimball
Cochair: Benjamin Bolker
Major Department: Zoology

Previous work on fluctuating asymmetry (FA), a measure of developmental

instability, has highlighted its controversial relationship with environmental stress and

genetic architecture. I suggest that conflict may derive from the fact that the basis of FA

is poorly understood and, as a consequence, the methodology for FA studies may be

flawed. While size-based measures of FA have been assumed to have half-normal

distributions within populations, developmental modeling studies have suggested other

plausible distributions for FA. Support for a non-normal distribution of FA is further

supported by empirical studies that often record leptokurtic (i.e., fat or long-tailed)

distributions of FA as well. In this dissertation, I investigate a series of questions

regarding the both the basis and distribution of FA in populations. Is FA normally

distributed and therefore likely to be properly sampled in FA studies? If not normal,

what candidate model distribution best fits the distribution of FA? Is the shape of the

distribution of FA similar to a simple and specific growth model (geometric Brownian









motion)? Does reducing individual variation in populations through inbreeding affect

follow the prediction of this model? How does this shape respond to environmental

factors such as temperature when genetic variation is controlled?

In three species of insects (cotton aphid, Aphis gossipyii Glover; honeybee, Apis

mellifera; and long-legged fly, Chrysosoma crinitus (Dolichipodidae)), I find that FA was

best described by a double Pareto lognormal distribution (DPLN), a lognormal

distribution with power-law tails. The large variance in FA under this distribution and

the scaling in the tails both act to slow convergence to the mean, suggesting that many

past FA studies are under-sampled when the distribution of FA is assumed to be normal.

Because DPLN can be generated by geometric Brownian motion, it is ideal for describing

behavior of cell populations in growing tissue. I demonstrated through both a

mathematical growth model and an inbreeding experiment in Drosophila simulans that

the shape of the distribution of FA is highly dependent on the level of genetic redundancy

or heterogeneity in a population. In monoclonal lines of cotton aphids, I also demonstrate

that FA decreases with temperature and that a shift in kurtosis is associated with

temperature induced phenotypic plasticity. This supports the prediction of a proximate

model for the basis of FA and also suggests shape of the distribution of FA responds to

environmentally induced changes in gene expression on the same genetic background.














CHAPTER 1
SEARCHING FOR A CONSISTENT INTERPRETATION OF DEVELOPMENTAL
INSTABILITY? A GENERAL INTRODUCTION

Everywhere, nature works true to scale and everything has its proper size.
-- D'Arcy Thompson

There have been many incarnations of the idea that stability and symmetry are

somehow related. Hippocrates (460-377 B.C.) was the first to postulate internal

corrective properties that work in the presence of disease. Waddington (1942) suggested

existence of similar homeostatic buffering against random and presumably additive errors

occurring during development. The term "fluctuating asymmetry," first introduced by

Ludwig (1932), was later adopted by Mather (1953), Reeve (1960) and Van Valen (1962)

to describe a measurable form of morphological noise representing a hypothetical lack of

buffering that is always present during development of organisms. Recently, biologists

have become very interested both in fluctuating asymmetry's potential usefulness as a

universal bioindicator of environmental health (Parsons 1992) and in its potential as an

indicator or even an overt signal of an individual's overall genetic quality (Moller 1990).

However, over a decade of work has left the field with no clear relationship between

increased fluctuating asymmetry and either environmental or genetic stress (Bjorksten et

al. 2000, Lens et al. 2002). Despite this fact, fluctuating asymmetry is still often assumed

to indicate of developmental instability. Recently, Debat and David (2001) page 560

define developmental stability as "a set of mechanisms historically selected to keep the

phenotype constant in spite of small random developmental irregularities potentially









inducing slight differences among homologous parts within individuals." (I have

italicized the aspects of this definition that I think are left undefined.)

Fluctuating asymmetry is defined and measured as the average right minus left

difference in size or shape of morphological characters in a population and has been

generally accepted as an indicator of developmental instability because both sides of a

bilaterally symmetric organism have been developed by the same genetic program in the

same environment (Moller and Swaddle 1997). Fluctuating asymmetry is measured as

left-right side differences in the size or shape of paired bilaterally symmetric biological

structures of organisms. In a character trait that demonstrates fluctuating asymmetry, it is

assumed that the distribution of signed left-right differences is near zero and that there is

no selection for asymmetry (Palmer and Strobek 1986, 2003). Other types of asymmetry

do exist and are thought to indicate selection against symmetry. Directional asymmetry

denotes a bias towards left or right sidedness that causes the population mean to move

away from zero. In antisymmetry, left or right side biases occur equally at the individual

level creating a population that is bimodal or platykurtic.

While much attention has been directed toward the possible genetic basis of

fluctuating asymmetry (reviewed by Leamy and Klingenberg 2005, Woolf and Markow

2003), response of fluctuating asymmetry to stress (reviewed by Hoffman and Woods

2003) and correlation of fluctuating asymmetry with mate choice (Moller and Swaddle

1997); little scientific effort has been directed toward investigating its basis or origin at

levels of organization lower than the individual. A few theoretical explanations for the

basis of fluctuating asymmetry have been developed (reviewed by Klingenberg 2003).

Only two of these offer a causal explanation for the increased levels of fluctuating









asymmetry that are sometimes observed under periods of environmental stress. While

not mutually exclusive, these two explanations differ at the organizational level at which

they are thought to act. Moller and Pomiankowski (1993) first suggested that strong

natural or sexual selection can remove regulatory steps controlling the symmetric

development of certain traits (e.g., morphology used in sexual display). They suggest that

with respect to sexually selected traits (and assuming that they are somehow costly to

produce), individuals may vary in their ability to buffer against environmental stress in

proportion to the size of their own energetic reserves. These reserves are often indicative

of the genetic quality of individuals. Therefore, high genetic quality is expected to be

associated with low fluctuating asymmetry. Emlen et al. (1993) present another and

more proximate explanation of the basis of fluctuating asymmetry. They do not invoke

sexual selection but hypothesize that fluctuating asymmetry is due largely to the non-

linear dynamics of signaling and supply that may occur during growth. Here fluctuating

asymmetry is thought to result from the scaling up of compounding temporal

asymmetries in signaling between cells during growth. In their model, hypothetical

levels of signaling compounds (morphogens) and or growth precursors used in the

construction of cells vary randomly over time. When growth suffers less interruption, in

other words, when it occurs faster and under less stress, there is also less complexity (and

fractal dimension) in the dynamics of signaling and supply. This is should reduce

fluctuating asymmetry.

Only a few other models have since been proposed. A model by Graham et al.

(1993) suggests that fluctuating asymmetry in the individual can also be the net result of

compounding time lags and chaotic behavior between hormonally controlled growth rates









on both sides of an axis of bilateral symmetry. More recently Klingenberg and Nijhout

(1999) present a model of morphogen diffusion and threshold response that includes

genetic control of each component. They demonstrate that fluctuating asymmetry can

result from genetically modulated expression of variation that is entirely non-genetic in

origin. In other words, even without specific genes for fluctuating asymmetry,

interaction between genetic and non-genetic sources of variation (G x E) can cause

fluctuating asymmetry (Klingenberg 2003).

While all these theories for the basis of fluctuating asymmetry have proven

useful in making some predictions about fluctuating asymmetry in relation to sexual

selection and growth rate/trait size, none are grounded in any known molecular

mechanisms. The search for any single molecular mechanism that stabilizes the

developmental process has proven elusive. A recent candidate was the heat shock

protein, Hsp90, which normally target conformationally plastic proteins that act as signal

tranducers (i.e., molecular switches) in many developmental pathways (Rutherford and

Lundquist 1998). Because Hsp90 recognizes protein folding, it can also be diverted to

misfolded proteins that are denatured during environmental stress. Therefore, Hsp90 can

potentially link the developmental process to the environment and these authors suggest it

may also capacitate the evolution of novel morphology during times of stress by

revealing genetic variation previously hidden to selection in non-stressful environments.

However, additional research by Milton et al. (2003) shows that while Hsp90 does buffer

against a wide range of morphologic changes and does mask the effect of much hidden

genetic variation in Drosophila, it does not appear to affect average levels of fluctuating

asymmetry through any single Hsp90 dependent pathway or process.









It is said that wisdom begins with the naming of things. The case of fluctuating

asymmetry reminds us that the act of giving names to things in science can, in fact, lend a

false impression that we have achieved a true understanding of that which has been

named. While fluctuating asymmetry has had several specific descriptive definitions, we

do not really know how to define it at a fundamental level because we do not understand

exactly how development is destabilized under certain conditions of both gene and

environment. All we can say for now is that fluctuating asymmetry is a mysterious form

of morphological variation. Mary Jane West-Eberhard (2003) has dubbed it the "dark

side" of variation because it may represent that noisy fraction of the physical-biological

interface that is still free of selection, and not under direct control of the gene.

Any study of natural variation would do well to begin by simply observing its

shape or its distribution in full. In nature, statistical distributions come mostly in two

flavors: those generated by large systems of independent additive components and those

generated by large systems of interacting multiplicative components (Vicsek 2001,

Sornette 2003). When large systems are composed of independent subunits, random

processes result in the normal distribution, the cornerstone of classical statistics. The

normal distribution is both unique and extreme in its rapidly decaying tails, its very

strong central tendency and its sufficient description by just two parameters, the mean

and variance, making it, in this sense, the most parsimonious description of random

variation. However, the normal distribution does not describe all kinds of stochastic or

random behavior commonly observed in the natural sciences. When subunits comprising

large systems interact, random processes are best described by models that underlie

statistical physics (sometimes called Levy statistics) (Bardou et al. 2003. Sornette 2003).









Fat-tailed distributions are often the result of propagation of error in the presence of

strong interaction. These types of distributions include the power function, Pareto, Zipf

and the double Pareto or log-Laplace distributions, all characterized by a power law and

an independence of scale. Collective effects in complex interacting systems are also

often characterized by these power laws (Wilson 1979, Stanley 1995). Examples include

higher order phase transitions, self-organized criticality and percolation. During second

order phase transition at the critical temperature between physical phases, external

perturbation of the network of microscopic interactions between molecules results in

system reorganization at a macroscopic level far above that of interacting molecules (e.g.,

the change from water to ice). This results in collective imitation that propagates among

neighboring molecules over long distances. Exactly at these critical temperatures,

imitation between neighbors can be observed at all scales creating regions of similar

behavior that occur at all sizes (Wilson 1979). Thus, a self-similar power-law manifests

itself in the interacting system's susceptibility to perturbation and results, in this case,

from the multiplicity of interaction paths in the interaction network (Stanley 1995). As

the distance between two objects in a network increases, the number of potential

interaction pathways increases exponentially and the correlation between such paths

decreases exponentially. The constant continuous degree of change represented by the

power law is the result of a combined effect between both an exponentially increasing

and decreasing rate of change. This highlights the fact that power laws can be easily

manifested from combinations of exponential functions which are very common to

patterns of change in many natural populations (both living and non-living).









Therefore, it is important to note that power laws in statistical distributions do not

have to always indicate strong interaction in a system and that many other simple

mechanisms can create them (Sornette 2003), especially where the behavior of natural

populations are concerned. For example, an apparently common method by which power

laws are generated in nature is when stochastic proportional (geometric) growth is

observed randomly in time (Reed 2001). Power law size distributions in particle size,

human population size, and economic factors are all potentially explained by this process

(Reed and Jorgensen 2004). Here we also have exponential increase in size opposing an

exponential decrease in the probability of observation or termination that results in a

gently decaying power law. This process can also explain why power law scaling can

occur through the mixing of certain distributions where locally exponentially increasing

and decreasing distributions overlap. For example, superimposing lognormal

distributions results in a lognormal distribution with power law tails (Montroll and

Shlesinger 1982,1983).

Given that exponential relationships are so common in the natural world, we should

assume that in observing any large natural population outside of an experiment, there is

probably some potential for a power-law scaling effect to occur. Therefore some degree

of non-normal behavior may be likely to be observed, often in the underlying

distribution's tail. If we assume an underlying normal distribution, and sample it

accordingly, we are likely to under-sample this tail. And so we rarely ever present

ourselves with enough data to challenge our assumption of normality, and we risk

missing the chance to observe a potentially important aspect of natural variation.









Until now, the basis of fluctuating asymmetry has been addressed only with very

abstract models of hypothetical cell signaling, or at the level of selection working on the

organism with potential mechanism remaining hypothetical. In this dissertation, the

underlying common theme is that fluctuating asymmetry must first and foremost be

envisioned as a stochastic process occurring during tissue growth, or in other words,

occurring within an exponentially expanding population of cells. This expansion process

can be modeled by stochastic proportional (geometric) growth that is terminated or

observed randomly over time. As will be explained in subsequent chapters, this

generative process can naturally lead to variation that distributes according to a

lognormal distribution with power laws in both tails (Reed 2001). In chapter 2, I

examine the distribution of fluctuating asymmetry in the wings of three species of insects

(cotton aphid, Aphis gossipyi Glover, honeybee, Apis mellifera, and long-legged flies,

Chrysosoma crinitus) and test various candidate models that might describe the statistical

distribution of fluctuating asymmetry. I then address whether, given the best candidate

model, fluctuating asymmetry studies have been appropriately sampled. I suggest that

much of the current controversy over fluctuating asymmetry may be due to the fact that

past studies have been under-sampled. In chapter 3, I extend and test a

phenomenological model for fluctuating asymmetry that is introduced in chapter 2. I

present the model and then examine some of its unique predictions concerning the effects

of inbreeding on the shape of the distribution of fluctuating asymmetry in Drosophila. I

present evidence that the genetic structure of a population can have a profound effect on

the scaling and shape of the observed distribution of fluctuating asymmetry. In chapter 4,

I characterize the pattern of developmental noise (fluctuating asymmetry in the absence






9


of genetic variation) in two monoclonal populations of cotton aphid cultured under

graded environmental temperatures. I investigate how developmental noise is altered by

this simple change in the environment. I present evidence that the environmental

response of size-based FA is directly related to developmental time. Lastly, I conclude

with a review of my major findings in the context of the introduction I have presented

here.














CHAPTER 2
ARE FLUCTUATING ASYMMETRY STUDIES ADEQUATELY SAMPLED?
IMPLICATIONS OF A NEW MODEL FOR SIZE DISTRIBUTION

Introduction

Developmental Stability: Definition, Measurement, and Current Debate

Developmental stability is maintained by an unknown set of mechanisms that

buffer the phenotype against small random perturbations during development (Debat and

David 2001). Fluctuating asymmetry (FA), the most commonly used assay of

developmental instability, is defined either as the average deviation of multiple traits

within a single individual (Van Valen 1962) or the deviation of a single trait within a

population (Palmer and Strobek 1986, 2003; Parsons 1992) from perfect bilateral

symmetry. Ultimately, an individual's developmental stability is the collective result of

random noise, environmental influences, and the exact genetic architecture underlying the

developmental processes in that individual (Klingenberg 2003; Palmer and Strobek

1986). Extending this to a population, developmental stability is the result of individual

variation within each of these three components.

Currently, there is conflict in the literature regarding the effect of both environment

and genes on the developmental stability of populations. The development of bilateral

symmetry appears to be destabilized to various degrees by both environmental stressors

(review in Moller and Swaddle 1997) and certain genetic architectures (usually created

by inbreeding: Graham 1992; Lerner 1954; Mather 1953; Messier and Mitton 1996;

review by Mitton and Grant 1984). While the influence of inbreeding on FA is not









consistent (Radwan 2003; Carchini et al. 2001; Fowler and Whitlock 1994; Leary et al.

1983,1984; Lens et al. 2002; Perfectti and Camachi 1999; Rao et al. 2002; Vollestad et al.

1999), it has led biologists to use terminology such as "genetic stress" or "developmental

stress" when describing inbred populations (Clarke et al. 1986 and 1993; Koehn and

Bayne 1989; Palmer and Strobek 1986).

While genetic and environmental stressors have been shown to contribute to

developmental instability and FA, the full picture is still unclear (Bjorksten et al. 2000;

Lens et al. 2002). While FA has been proposed as a universal indicator of stress within

individual organisms (Parsons 1992), its utility as a general indicator of environmental

stress has been contentious (Bjorksten et al. 2000; Ditchkoff et al. 2001; McCoy and

Harris 2003; Merila and Bjorklund 1995; Moller 1990; Rasmuson 2002; Thornhill and

Moller 1998; Watson and Thornhill 1994; Whitlock 1998). Despite many studies, no

clear general relationship between environmental stress and FA has been demonstrated or

replicated through experimentation across different taxa (Bjorksten et al. 2000; Lens et al.

2002). Furthermore, the effects of stress on FA appear to be not only species-specific but

also trait-specific and stress-specific (Bjorksten et al. 2000). Several meta-analyses have

attempted to unify individual studies on the relation of sexual selection, heterozygosity,

and trait specificity to FA (Polak et al. 2003; Thornhill and Moller 1998; Vollestad et al.

1999); while some weak general effects have been found, their biological significance is

still unresolved.

Taken together, the ambiguity of the results from FA studies suggests unresolved

problems regarding the definition and/or measurement of FA. The distribution and

overall variability of FA are sometimes discussed with regards to repeatability in FA









studies (Whitlock 1998; Palmer and Strobeck 2003), but is seldom a primary target of

investigation. Until we can quantify FA more reliably and understand its statistical

properties, the potential for misinterpretation of FA is likely to persist.

The Distribution of Fluctuating Asymmetry

Although it is always risky to infer underlying processes from observed patterns,

careful examination of the distribution of FA in large samples may help distinguish

between possible scenarios driving FA. For instance, a good fit to a single statistical

distribution may imply that the same process operates to create FA in all individuals in a

population. In contrast, a good fit to a discrete mixture of several different density

functions might suggest that highly asymmetric individuals suffer from fundamentally

different developmental processes than their more symmetric counterparts. Thin-tailed

distributions (e.g., normal or exponential) may indicate relative independence in the

accumulation of small random developmental errors, whereas heavy-tailed distributions

may implicate non-independent cascades in the propagation of such error. Despite much

interest in the relationship between environmental stress and levels of FA, the basic

patterns of its distribution in populations remain largely unexplored.

One common distributional attribute of FA, leptokurtosis, has been discussed in the

literature. Leptokurtosis denotes a distribution that has many small and many extreme

values, relative to the normal distribution. Two primary causes of this kind of departure

from the normal distribution are the mixing of distributions and/or scaling effects in data.

For example, the Laplace or double exponential distribution is leptokurtic (but not heavy-

tailed) and can be represented as a continuous mixture of normal distributions (Kotz et al.

2001; Kozubowski and Podgorski 2001). Just as log scaling in the normal distribution

results in the lognormal distribution, log scaling in the Laplace leads to log-Laplace (also









called double Pareto) distributions (Kozubowski and Podgorski 2002; Reed 2001), which

are both leptokurtic and heavy-tailed (see Figure 2-1). Several explanations for

leptokurtosis in the distribution of FA have been proposed. Both individual differences

in developmental stability within a population (Gangestad and Thornhill 1999) and

differences in FA between subpopulations (Houle 2000) have been suggested to lead to

continuous or discrete mixtures of normal distributions with different developmental

variances, which in turn would cause leptokurtosis (e.g., a Laplace distribution) in the

overall distribution of FA. Mixtures of non-normal distributions may also cause either

leptokurtosis or platykurtosis (more intermediate values than the normal distribution:

Palmer and Strobek 2003). A potential example of this is illustrated by Hardersen and

Frampton's (2003) demonstration that a positive relationship between mortality and

asymmetry can cause leptokurtosis. Alternatively, Graham et al. (2003) have argued that

developmental error should behave multiplicatively in actively growing tissues, creating

a lognormal size distribution in most traits rather than the normal distribution that is

usually assumed. They argue that this ultimately results in leptokurtosis (but not fat tails)

and size dependent expression of FA. Because simple growth models are often geometric,

we should not be surprised if distributions of size-based FA followed the lognormal

distribution (see Limpert et al. 2001 for a review of lognormal distributions in sciences).

Not well recognized within biology is the fact that close interaction of many

components can result in power-law scaling distributionall tails that decrease

proportional to x-a rather than to some exponential function of x such as exp(-ax)

[exponential] or exp(-ax2) [normal] and hence to heavy-tailed distributions (Sornette

2003). Power- law scaling is often associated with the tail of the lognormal distribution,






14


especially when log standard deviation is large ( Mitzenmacher unpublished ms.;

Montroll and Shlesinger 1982, 1983; Roman and Porto 2001; Romeo et al. 2003).


additive/linear


geometric/logarithmic


NORMAL


Normal-
Laplace


gA


Double Pareto
lognormal (DPLN)

:t


LAPLACE


LOGNORMAL


M
x
X
T
U
R
E


LOG-LAPLACE


Figure 2-1. Schematic representation of mathematical relationships between candidate
models for the distribution of fluctuating asymmetry. Mixtures here are
continuous.

KEY:
Solid line scale of variable (x *- ln(x))
Dashed line random walk observed at constant stopping rate (i.e., negative
exponentially distributed stop times) Note: random walk on log scale exhibits geometric
Brownian motion
Dotted line convolution of two distributions (one of each type)
Block arrow a continuous mixture of distributions with stochastic (exponentially
distributed) variance
(Note Log-Laplace is also called double Pareto by Reed 2001, Reed and Jorgensen
2004)


* *









Because FA in a population may reflect fundamental developmental differences

between different classes or groups of individuals, for example stressed and non-stressed,

or between different subpopulations as suggested by Houle (2000), we might expect

discrete mixtures of different distributions to best describe FA. For instance, extreme

individuals falling in a heavy upper tail may be those who have exceeded some

developmental threshold. Major disruption of development, resulting in high FA, may

also reveal the scaling that exists in the underlying gene regulatory network (Albert and

Barabasi 2002; Clipsham et al. 2002; Olvai and Barabasi 2002). Alternatively, ifFA is

produced by a single process, but to various degrees in different individuals, then one

might expect a continuous mixture model to best describe the distribution of FA.

The possibility of non-normal distribution of FA opens the door to several potential

sampling problems. For instance, if the lognormal shape, or multiplicative variance

parameter, is large, then broad distribution effects may slow the convergence of the

sample mean to the population mean as sample sizes are increased (Romeo et al. 2003).

An additional, thornier, problem is caused by power-law scaling in the tails of

distributions. Many lognormally distributed datasets exhibit power-law scaling (or

amplification) in the tail region (Montroll and Shlesinger 1982, 1983; Romeo et al. 2003;

Sornette 2003), sometimes called Pareto-Levy tails or just Levy tails. As sample sizes

grow infinitely large, power-law and Pareto distributions may approach infinite mean (if

the scaling exponent is less than three) and infinite variance (if the scaling exponent is

less than two), and therefore will not obey the Law of Large Numbers (that sample means

approach the population mean as sample sizes increase). Increased sampling actually

increases the likelihood of sampling a larger value in the tail of a Pareto distribution









(Bardou et al. 2003; Quandt 1966), creating more uncertainty in estimates of the mean as

sample size increases. The presence of power-law tails can slow overall convergence

considerably even in distributions that are otherwise lognormal with low variance (which

may not look very different from well-behaved lognormal distributions unless a large

amount of data is accumulated).

This discussion points to two effects that need to be assessed broad distribution

effects, controlled by the shape (lognormal variance) parameter of the body of the FA

distribution, and power-law tails, controlled by the scaling exponents of the tails of the

FA distribution. To assess these effects, I apply a new statistical model, the double

Pareto lognormal (DPLN) distribution (Reed and Jorgensen 2004). The DPLN

distribution is a lognormal distribution with power-law behavior in both tails (for values

near zero and large positive values). Similar to the log-Laplace distributions, the DPLN

distribution can be represented as a continuous mixture of lognormal distributions with

different variances. It can also be derived from a geometric Brownian motion (a

multiplicative random walk) that is stopped or "killed" at a constant rate (i.e., the

distribution of stop times is exponentially distributed: Reed and Jorgensen 2004; Sornette

2003). The parameters of the DPLN distribution include a lognormal mean (v) and

variance ('2) parameter which control the location and spread of the body of the

distribution, and power-law scaling exponents for the left (3) and right (c) tails. Special

cases of the DPLN include the right Pareto lognormal (RPLN) distribution, with a power-

law tail on the right but not the left side (->oo); the left Pareto lognormal (LPLN)

distribution, with a power-law tail only near zero (ca-c); and the lognormal distribution,

with no power-law tails (a->c,3->co). For comparison, I also fit normal and half-normal









distributions as well as the asymmetric Laplace distribution to the data on FA. See

Figure 2-1 for schematic representation of relationships between these candidate models

for FA.

In the following study, I directly test the fit of different distributions to large FA

datasets from three species of insects. I include a lab cultured monoclonal line of cotton

aphid, Aphis gossipyii Glover, in an attempt to isolate the distribution of developmental

noise for the first time. I also analyze data from a semi-wild population of domestic

honeybee, Apis mellifera, taken from a single inseminated single queen colony, and from

a large sample of unrelated wild-trapped Long-legged flies (Chrysosoma crinitus:

Dolichopididae).

I address two primary groups of goals in this study. First, I investigate what

distributions fit FA data best and how the parameters of these distributions vary across

species, rearing conditions, and levels of genetic relatedness. I also address whether

"outliers" (individuals with visible developmental errors) appear to result from discrete or

continuous processes. Secondly, I determine how accurate the estimates of population

mean FA are at various sample sizes to determine whether past studies of FA been

adequately sampled to accurately estimate mean or average FA in populations. In

addition to these two primary goals, I also compare the best-fitting distributions and level

of sampling error for three of the most common methods of measuring FA: a univariate

and a multivariate size-based metric of asymmetry, and a multivariate shape-based

method.

Methods

Wings were collected and removed from three populations of insects and dry

mounted on microscope slides. These populations included a monoclonal population of









1022 Cotton Aphid (Aphis gossipyii Glover) started from a single individual collected

from citrus in Lake Alfred, Florida, 1001 honeybees Apis mellifera, maintained at

University of Florida and 889 long-legged flies (Chrysosoma crinitus :Dolichopodidae).

All species identifications were made through the State of Florida Department of Plant

Industry in Gainesville and voucher specimens remain available in their collections.

Aphid cultures were maintained on potted plants in reach-in environmental

chambers at 150C with constant 14/10 hour LD cycle generated by 4 20 watt

fluorescent Grolux brand bulbs. Aphid cultures were cultured on approximately 10 day

old cotton seedlings (Gossipium) and allowed to propagate until crowded. Crowding

stimulated alate formation (winged forms) in later generations which were collected

every twenty days with a fine camel hair brush wetted in ethanol. New plants were added

every ten days and alates were allowed to move freely from plant to plant starting new

clones until they were collected. The temperature at which the colony was maintained

created a low temperature "dark morph" Cotton Aphid which still propagated on host

plants quickly but was larger than high temperature "light morphs" that form at

temperatures greater than 17C. Dark morphs colonize stems on cotton whereas light

morphs colonized the undersides of leaves.

The bees were collected in June 2004 by Dr. Glenn Hall at the University of

Florida's Bee Lab from a single inseminated single queen colony. They were presumably

all foragers and haplodiploid sisters collected as they exited the hive into a collection bag.

The bag was frozen for three hours and then the bees were placed in 85% ethanol.

Long-legged flies were trapped from a wild population using 14 yellow plastic

water pan traps in southwest Gainesville, Florida, during May 2003 and May-June 2004.









A very small amount of dishwashing detergent was added to the water to eliminate

surface tension and enhance trapping. Traps were checked every three hours during

daylight and set up fresh every day of trapping.

All insect specimens were dried in 85% ethanol, and then pairs of wings were

dissected (in ethanol) and air-dried to the glass slides while the ethanol evaporated.

Permount was used to attach cover slips. This technique prevented wings from floating

up during mounting, which might slightly distort the landmark configuration. Dry

mounts were digitally photographed. All landmarks were identified as wing vein

intersections on the digital images (six landmarks on each wing for aphids, eight for

honeybees and Dolichopodid flies). See Appendix A for landmark locations on wings for

each species.

Wing vein intersections were digitized three times each on all specimens using

TPSDIG version 1.31 (Rohlf, 1999). All measures of FA were taken as the average FA

value of the three replicate measurements for each specimen. Specimens damaged at or

near any landmarks were discarded. Fluctuating asymmetry was measured in three ways

on all specimens. First, a common univariate metric of absolute unsigned asymmetry

was taken for two different landmarks: FA = abs(R L) where R and L are the Euclidean

distances between the same two landmarks on either wing. In aphids, landmarks 1-2 and

2-3 were used; in bees, landmarks 1-4 and 2-6 were used; and in long-legged flies,

landmarks 3-6 and 4-5 were used. Two multivariate geometric morphometrics using

landmark-based methods were performed using all landmarks shown in Appendix A. A

multivariate size-based FA (FA 1 in Palmer and Strobek 2003) was calculated as absolute

value of (R L) where R and L are the centroid sizes of each wing (i.e., the average of









the distances of each landmark to their combined center of mass or centroid location). In

addition, a multivariate shape-based measure of FA known as the Procrustes distance was

calculated as the square root of the sum of all squared Euclidean distances between each

left and right landmark after two-dimensional Procrustes fitting of the data (Bookstein

1991; Klingenberg and McIntyre 1998; FA 18 in Palmer and Strobeck 2003; Smith et al.

1997). Procrustes fitting is a three step process including a normalization for centroid

size followed by superimposition of two sets of landmarks (right and left) and rotation

until all distances between each landmark set is minimized. Centroid size calculation,

Euclidean distance calculation and Procrustes fitting were performed using 0yvind

Hammer's Paleontological Statistics program PAST version 0.98 (Hammer 2002). For

assessing measurement error (ME) of FA (or more specifically, the digitizing error), we

conducted a Procrustes ANOVA (in Microsoft Excel) on all pairs of wing images

resampled three times each for every species (Klingenberg and McIntyre 1998). Percent

measurement error was computed as (ME/average FA) x 100 where

ME = (FA- FA2 + FA2 FA3 + FA FA3)/ 3 All subsequent statistical analyses

were performed using SPSS Base 8.0 statistical software (SPSS Inc.).

The fits of all measures of FA to eight distributional models (normal, half-normal,

lognormal, asymmetric Laplace, double Pareto lognormal (DPLN), two limiting forms of

DPLN, the right Pareto lognormal (RPLN) and the left Pareto lognormal (LPLN) and a

discrete mixture of lognormal and Pareto) were compared by calculating negative log

likelihood and Likelihood Ratio Test (LRT) if models were nested and Akaike

Information Criteria (AIC) if not nested (Burnham and Anderson 1998; Hilborn and

Mangel 1998). Both of these approaches penalize more complex models (those with









more parameters) when selecting the best-fit distributional model for a given dataset.

(The Likelihood Ratio Test does not technically apply when the nesting parameter is at

the boundary of its allowed region, e.g., when c->oc for the DPLN, but Pinheiro and

Bates (2000) suggest that the LRT is conservative, favoring simpler models, under these

conditions.) Best fitting parameters were obtained by maximizing the log-likelihood

function for each model (Appendix B). The maximization was performed using the

conjugate gradient method within unconstrained solve blocks in the program MathCad by

MathSoft Engineering and Education Inc (2001), and was also confirmed using Nelder-

Mead simplex algorithm or quasi-Newton methods in R version 2.0.1 (2003), a

programming environment for data analysis and graphics.

Phenodeviants were defined as individuals demonstrating missing wing veins, extra

wing veins or partial wing veins on either one or both wings. All phenodeviants in

honeybees involved absence of the vein at landmark 6 (LM 6). Phenodeviants in aphids

were more variable but mostly involved absence of wing vein intersections at LM 2 or

LM 3. Procrustes distances were estimated for phenodeviants by omitting the missing

landmarks (caused by the phenodeviance) and controlling for the effect of this removal

on the sums of squares. I added an average of the remaining sums of squares in place of

the missing sums of squares so that the calculated Procrustes distance is comparable to

normal specimens (i.e., six landmarks). In almost all phenodeviants, this involved

omission of only one set of landmark values. The frequency of phenodeviants was

examined across the range of the FA distribution (i.e., Procrustes distance), and mean

values of the FA for phenodeviants were compared to normal individuals in order to









assess whether phenodeviants tended to show higher than normal levels of FA in

characters that were not affected by the missing, partial, or extra wing veins.

The best fitting parameters of the best fitting models were used to build a

distributional model under which repeated sampling was simulated at various sample

sizes. Average error in estimation of the mean FA was calculated as a coefficient of

variation ((s / x) 100) for 1,000 randomly generated datasets. Lastly, comparison were

made of the estimation errors given the best fitting distributions of FA to a distribution of

sample sizes from 229 FA studies published in three recent meta-analyses (Polak et al.

2003; Thornhill and Moller 1998; Vollestad et al. 1999).

Results

In the distributions of shape-based FA in monoclonal cotton aphids (n = 1022),

domesticated honeybees (n = 1001), and wild trapped long-legged flies (n = 889), AIC

and LRTs always favored DPLN or RPLN models by a large margin. All size-based FA

distributions favored DPLN or LPLN by a large margin (Table 2-1). All variants of

discrete mixture models we tried had very poor results (data not shown). Figures 2-2

through 2-4 demonstrate best fitting models for multivariate shape FA (DPLN and

lognormal), multivariate centroid size FA (DPLN and half-normal), and univariate size

FA (DPLN and half-normal) for all three species. FA was often visually noticeable in

aphids, where the mean shape FA (Procrustes distance) was three times higher (0.062 +

0.00050) than in bees (0.023 0.00026) or flies (0.019 0.00028). I note that

distribution of size FA in aphids and bees fit half-normal distribution in the upper tails

fairly well but fit relatively poorly among individuals with low FA. Long-legged flies

exhibit poor fit to half-normal in both tails.












Table 2-1. Maximized log-likelihood (MLL), number of model parameters (P) and
Akaike Information Criterion differences (AAIC) for distributional models
tested. Winning models have AAIC zero. Models with goodness-of-fit nearly
equal to winners are underscored (AAIC <3.0). 4.0 some support for specified model. AAIC >10.0 indicates no support (Burnham
and Anderson 1998). Distances between landmarks for first univariate size
FA, aphids LM 1-2, bees LM 1-4 and long-legged fly LM 3-6 and for second
univariate FA aphids LM 2-3, bees LM 2-6 and long-legged fly LM 4-5.


Multivariate shape FA Multivariate size FA First univariate size FA


MLL
255.117
264.72
256.723
734.726
2241
3523
637.044


132.908
130.224
136.361
523.667
3371
5319
1148


146.603
128.617
149.585
506.553
2985
5147
1041


A AIC
0.000
17.207
1.213
955.218
3968
6530
761.855


7.369
0.000
12.275
784.888
6480
10370
2036


37.971
0.000
41.935
753.871
5710
10030
1825


MLL
542.906
547.44
543.288
1261
2847
2370
1638


536.278
549.641
535.906
1262
2180
1775
1387


452.728
454.056
455.818
1055
1760
1084
940.241


A AIC
1.236
8.303
0.000
1433
4606
3650
2189


2.744
27.47
0.000
1451
3286
2475
1703


0.000
0.656
4.18
1201
2610
1256
973.02


MLL
545.725
549.789
545.317
1265
3352
2893
1883


454.337
460.51
455.483
1081
1709
2687
1162


453.958
460.261
456.237
1064
1806
1252
1016


A AIC
2.815
8.944
0.000
1438
5612
4691
2676


0.000
10.346
0.292
1249
2505
1771
1413


0.000
10.606
2.559
1216
2701
1590
1121


Second univariate size FA

MLL AAIC
614.123 2.333
617.498 7.083
613.957 0.000
1413 1596
3534 5838
3047 4862
1950 2671


Species/model
Cotton Aphid
DPLN
RPLN
LPLN
LNORM
NORM
HNORM
LAPLACE
Honey Bee
DPLN
RPLN
LPLN
LNORM
NORM
HNORM
LAPLACE
Long-legged Fly
DPLN
RPLN
LPLN
LNORM
NORM
HNORM
LAPLACE


6.329
23.82
0.000
1393
3574
2816
1863


0.000
20.182
0.212
1144
2508
1368
1169


For multivariate shape analysis, right tail power-law exponents (a) were very high


(thousands), left-tail exponents (3) from 3.9-9.9, while the dispersion parameter (') was


narrowly distributed from 0.310 to 0.356. Thus, shape FA exhibited little scaling in tails


(i.e., nearly lognormal). For size-based FA, dispersion was much larger (0.57-0.74), and


power-law exponents more variable for univariate and multivariate size-based FA (left


515.93
525.675
513.765
1211
2302
1924
1445


412.007
422.993
412.902
985.894
1668
1099
997.28










tail (P) and right tail (ca) were generally low, indicating moderate power-law scaling in

both tails) as shown in Table 2-2.

Table 2-2. Best fit parameters for models in Table 1. Parameters for univariate size FA
are very similar to multivariate size FA and are not shown. Skew indexes for
asymmetric Laplace are also not shown.


Multivariate shape FA


Multivariate size FA


Cotton Location Dispersion/ Right tail


Aphid
DPLN
RPLN
LPLN
LNORM
NORM
HNORM
LAPLACE
Honey Bee


DPLN
RPLN
LPLN


-2.62
-3.01


Shape
0.356
0.415


-2.62 0.353


-2.87
0.062
0.009
0.049


1160
7.03


Left
Tail
4.04
00


0o 3.90


0.434
0.027
0.059
0.058


-3.74 0.310 8380 9.80
-4.02 0.274 5.52 oo


-3.71 0.305


LNORM -3.84
NORM 0.023
HNORM 0.007
LAPLACE 0.021
Long-legged Fly
DPLN Q


RPLN
LPLN

LNORM
NORM
HNORM
LAPLACE


00 7.75


0.327
0.008
0.018
0.018


flnA AQnf -an ICW~


2
-4.28 0.231 3.73 oo


-3.92 0.341


-4.02
0.019
0.007
0.017


0.351
0.008
0.015
0.017


oo 9.91
2


Location Dispersion/ Right tail


1.45
1.07
1.73
1.28
4.92
0.163
0.657


1.32
0.571
1.52
0.777
2.95
0.078
0.550


Shape
0.735
0.800
0.699
0.824
3.94
6.31
2.19


6.24
4.58


Left
Tail
3.01
00


0o 2.23


0.565 10.4 1.57
0.829 4.82 oo


0.510
0.850
2.14
3.64
2.24


oo 1.35


-0.067 0.683 3.09 5.14


-0.302 0.698 2.75


0.319 0.743


- 0.060
- 2.00
- 0.075
- 0.346


oo 3.85


0.783
2.24
2.11
1.06


Figure 2-5 shows the distribution of FA sample sizes from 229 studies published in

three recent meta-analyses (Polak et al. 2003; Thornhill and Moller 1998; Vollestad et al.


1999).














A. Cc tton Aphid





.J. I


B. Domestic Honeybee


, DPLN DPLN

0 005 1 0 15 02 025
obsDATA

o Lognormal 9
0xLGN 2
o 1
S005 01 0 15 02 025









12 0 002 004 006 022 008 24
10 DPLNo
x pDPLN


ob-DATA


Lognormal
X..


Figure 2-2. Distribution of multivariate shape FA of A) cotton aphid (Aphis gossipyii) B)
domestic honeybee (Apis mellifera) and C) long-legged fly (Chrysosoma
crinitus). Best fitting lognormal (dashed line, lower inset), and double Pareto
lognormal (solid line, upper inset) are indicated.


Il*lr.mr:,~ --n- I I I "



















































Figure 2-3. Distribution of multivariate centroid size FA of A) cotton aphid (Aphis
gossipyii) B) domestic honeybee (Apis mellifera) and C) long-legged fly
(Chrysosoma crinitus). Best fitting half-normal (dashed line, lower inset) and
double Pareto lognormal distribution (solid line, upper inset) are indicated.












A Cotton Aphid (LM 2-3)


DPLN

o o 0 f-normal
oboATA



Half-normal
, i i i~


B Domestic Honeybee (LM 1-4)


B Domestic Honeybee (LM 2-6)1
xpDaFN


DPLN

obr ATA

Half-normal ; +
n5, i


Fly (LM 3-6) l
S,- DPLN




Half-normal

5 I


C Long-legged Fly (LM 4-5)


Nf


A Cotton Aphid (LM 1-2)


Figure 2-4. Distribution of univariate unsigned size FA of A) cotton aphid (Aphis
gossipyii) B) domestic honeybee (Apis mellifera) and C) long-legged fly
(Chrysosoma crinitus). Best fitting half-normal (dashed line, lower inset) and
double Pareto lognormal distribution (solid line, upper inset) are indicated.


'\1 h P


20 DPLN

o o 0 o o 4'o
obhDATA



Half-normal








0 5 0 10
DPLN




obsDATA
Half-normal











DPLN


Hfr
obDATA

Half-normal
i i i










40-



30



20



10



0


sample or treatment size

Figure 2-5. Distribution of sample sizes (n) from 229 fluctuating asymmetry studies
reported in three recent meta-analyses (Vollestad et al. 1999, Thornhill and
Moller 1998 and Polak et al. 2003). Only five studies had sample sizes greater
than 500 (not shown).

Nearly 70% of the 229 FA studies have sample or treatment sizes less than 100.

Figure 2-6 demonstrates the hypothetical error levels (coefficients of variations) in

estimated mean FA at various sample sizes. Approximate best fit parameters were used

to estimate the coefficient of variation (CV) under the DPLN distribution (for shape FA,

v = -3.7, T = 0.35, a = 1000, 0 = 9; for size FA, v = 1.2, T = 0.70, a = 4.0, 0 = 4.0). For

the same set of landmarks, multivariate shape FA measures lead to the least amount of

error in estimating mean FA under DPLN at any sample size. Both univariate and

multivariate size FA perform more poorly in terms of both convergence and overall

percentage error.

I found that while phenodeviants occurred in almost all regions of the distribution

range of FA, the percentage of phenodeviant individuals increased dramatically with










increasing FA (Figure 2-7; aphids, r = 0.625 p = 0.013; bees, r = 0.843 p = 0.001). I also

found that individuals with phenodeviant wings (both aphids and bees) showed

significantly higher levels of FA across those wing landmarks unaffected by the

phenodeviant traits (p < 0.002 in both aphids and bees). Only a single case of

phenodeviance was sampled in long-legged flies.

45
4 0 1 .. ...............
40
35
3 0 ...................... .. .. .. .... .... .. ..............................................
o30 Legend
0 ............................................................................... ................. heg epn d F ....................... ... .............

% ERROR IN ESTIMATION 25 shape FA
OF THE MEAN (CV) ----- size FA
20
2 0 .. ... .. ... .. ... .... .... ..... ... ... .... ... .... .... ... ... .... ................................................ ....... .....................

15

0
0 \ \. I, '. i ii

0 50 100 150 200 250 300 350 400 450 500

SAMPLE SIZE
Figure 2-6. Relationship between sample size and % error for estimates of mean FA
drawn from best fitting size (dashed line) and shape (solid line) distributions
using 1000 draws per sample size. All runs use typical winning double Pareto
lognormal parameters (shape FA v = -3.7, T = 0.2, a = 1000, 0 = 9; for size FA
v = 1.2, z = 0.7, a = 4.0, 0 = 4.0).

Percent measurement error for shape FA was 1.41% in aphids, 1.63% in bees, and

2.42 % in flies while for size FA, it was -4.5% in aphids, -5% in bees, and -6.5 % in

flies. In a Procrustes ANOVA (Klingenberg and McIntyre 1998) the mean squares for

the interaction term of the ANOVA (MSinteraction) was highly significant p<0.001 in all

three species indicating that FA variation was significantly larger than variation in

measurement error (ME). The distribution of signed ME was normal and exhibited











moderate platykurtosis for all types of FA in all species examined. Measurement error


was very weakly correlated to FA in all samples (0.01 < r2 <0.07).

300 300
A B
250 250 -

200 200
Frequency
150 150

100 100
Legend IIIILegend
50 n phenodeviants I phenodeviants
50 normal 50 normal

0 0
&'' '' & .O k '',:' .... ...... O *' o g"O" C o" O
o o. r t $ *0
Procrustes Distance Procrustes Distance
064 030

C 0 D
So 028
o60
Q 027
058


054 024

S052 023
NORMALWING PHENODEVIANT NORMALWING PHENODEVIANT
APHID WINGS BEE WINGS


Figure 2-7. The proportion and percentage (inset) of individuals with visible
developmental errors on wings (phenodeviants) are shown for cotton aphids
(A) and honeybees (B) in relation to distribution of shape FA (Procrustes
distance). Average FA for both normal and phenodeviant aphids (C) and
honeybees (D) are also given.


Discussion


The Distribution of FA


The data demonstrate a common pattern of distribution in the FA in wing size and


shape of three different species whose populations existed under very different


environmental conditions (lab culture, free-living domesticated, and wild) and genetic


structure (monoclonal, haplodiploid sisters, and unrelated). The similarities across the


very different species and rearing conditions used in this study suggest that the









distribution of size and shape FA may have universal parameters (e.g., T 0.35 for shape

FA and T 0.7 for size based FA). The data confirm that although size FA sometimes

exhibits reasonable fit to half-normal in the upper tail, and shape FA is reasonably well fit

by lognormal distributions, large datasets of FA in both size and shape are always best

described by a double Pareto lognormal distribution (DPLN) or one of its limiting forms,

LPLN and RPLN. Multivariate shape FA demonstrates narrow distribution with a large

right tail, including the top few percent of the most extremely asymmetric individuals,

that is best fit by DPLN or RPLN. Both univariate and multivariate size FA exhibit a

considerably broader distribution with moderate leptokurtosis that is best fit by DPLN or

LPLN. The data suggest that the DPLN distribution and its limiting forms are generally

the most appropriate models for the distribution of FA regardless of method of

measurement.

Evidence that distribution of FA closely follows DPLN, a continuous mixture

model, and appearance of phenodeviance across nearly the entire range of data suggested

that developmental errors may be caused by a similar process across the entire

distribution of FA in a population. In other words, variation in FA may have a single

cause in most of the data. Although phenodeviance is significantly related to increased

levels of FA and is more prevalent in the right tail region of the shape FA distribution, it

does not appear associated exclusively with the right tail, as a threshold model for high

FA might predict. The very poor fits to all variations on the discrete mixture model also

suggest a lack of distinct processes creating extreme FA in the three datasets. However, I

caution that use of maximum likelihood methods to fit data to discrete mixtures is often

technically challenging. I found no block effects (e.g., no differences in FA levels









between long-legged fly samples collected on different weeks or aphid samples collected

from different pots or growth chambers), so it appears there is no obvious sample

heterogeneity that could result in a discrete mixture. With the usual caveats about

inferring process from pattern, I do not find obvious thresholds in the distribution of

asymmetries at the population level that would suggest threshold effects at a genetic or

molecular level. Lastly, based on the appearance of only one phenodeviant among our

wild trapped long-legged fly population (as opposed to many in the bee and lab reared

aphid populations), I speculate that mortality related to phenodeviance (and perhaps high

FA) in wing morphology may be relaxed in lab culture and domestication. But this could

be confounded by other differences between the three datasets including genetic

redundancy and species differences. Further comparisons among populations of single

species under different conditions would be needed to test this idea.

Sample Size and the Estimation of Mean FA

In random sampling under DPLN, we found that broad distribution effects due to

the shape parameter were minimal in their effect of slowing convergence to the

population mean in multivariate shape-based FA ('z 0.35 for all three datasets).

However, these effects are considerable for univariate and multivariate size-based FA

(where T Z 0.70). The effects of scaling in the tails of the distribution, which cause

divergence from the mean, appear to have little effect in the right tail of the distribution

of shape-based FA. However, larger effects in the tails of the size FA create more

individuals with very low and very high asymmetry than expected under the assumption

of normality. The point estimates of the scaling exponents for size FA are close to the

range where very extreme values may be sampled under the distribution tails (if a < 3 or

p <3), greatly affecting confidence in the estimate of mean FA. With a sample size of 50









and the best fitting DPLN parameters typical for asymmetry in multivariate shape, we

found that the coefficient of variation for mean FA is about 5%, whereas size-based mean

FA fluctuates about 13% from sample to sample. At a sample size of 100, these

coefficients of variation are 3.2% and 7.5% respectively. Unless experimental treatment

effects in most FA studies are larger than this, which is unlikely in studies using size-

based measures of FA of more canalized traits, statistical power and repeatability will be

low. Given the sample size range of most previous studies (n = 30-100) and their

tendency to favor size-based measurement methods, our results suggest that many past

FA studies may be under-sampled. Furthermore, it is also likely that given the small

sample sizes in many FA studies, particularly involving vertebrates where n < 50, the tail

regions of natural FA distributions are often severely under sampled and sometimes

truncated by the removal of outliers. These factors may artificially cause non-normal

distributions to appear normal, also potentially resulting in inaccurate estimation of mean

FA.

The Basis of Fluctuating Asymmetry

The surprisingly good fit of FA distributions to the DPLN model in our study

suggests that the physical basis of FA may be created by the combination of random

effects in geometrically expanding populations of cells on either side of the axis of

symmetry (i.e., geometric Brownian motion). Studies in the Drosophila wing indicate

that cell lines generally compete to fill a prescribed space during development with more

rapidly dividing lines out-competing weaker ones (Day and Lawrence 2000). Because

regulation of the growth of such cell populations involves either nutrients and/or

signaling substances that stop the cell cycle when exhausted, it is likely that the

distribution of numbers of cells present at the completion of growth follows an









exponential distribution. Reed and Jorgensen (2004) demonstrate that when a population

of repeated geometric random walks is "killed" at such a constant rate, the DPLN

distribution is the natural result. There are many other examples of growth processes in

econometrics and physics where random proportional change combined with random

stopping/observation create size distributions of the kind described here (Reed 2001;

Kozubowski and Podgorski 2002). In the future, when applying this model to instability

during biological growth, it would be very interesting to investigate how genetic and

environmental stress might affect the parameters of this model. If scaling effects are

found to vary with stress, then leptokurtosis may potentially be a better candidate signal

of developmental instability than increased mean FA.

Conclusion

Although size-based FA distributions can sometimes appear to fit normal

distributions reasonably well as previous definitions of FA suppose, I demonstrate that

three large empirical datasets all support a new statistical model for the distribution of FA

(the double Pareto lognormal distribution), which potentially exhibits power-law scaling

in the tail regions and leading to uncertain estimation of true population mean at sample

sizes reported by most FA studies. The assumption of normality fails every time

candidate models are compared on large datasets. Failure of this assumption in many

datasets may have been a major source of discontinuity in results of past FA studies.

Future work should attempt to collect larger sample/treatment sizes (n = 500) unless the

magnitude of treatment effects on FA (and thus the statistical power of comparisons) is

very large. Our results demonstrate that multivariate shape-based methods (Klingenberg

and McIntyre 1998) result in more repeatable estimates of mean FA than either

multivariate or univariate size-based methods. I would also recommend that









methodology be re-examined even in large sample studies of FA. For example, because

Drosophila are usually reared in many replicates of small tubes with less than 50 larvae

per tube, many large studies may still be compromised by individual sizes of replicate

samples. I also suggest that authors of past meta-analyses and reviews of FA literature

reassess their conclusions after excluding studies in which under-sampling is found to be

problematic. Careful attention to distributional and sampling issues in FA studies has the

potential both to mitigate problems with repeatability and possibly to suggest some of the

underlying mechanisms driving variation in FA among individuals, populations, and

species.














CHAPTER 3
INBREEDING REDUCES POWER-LAW SCALING IN THE DISTRIBUTION OF
FLUCTUATING ASYMMETRY: AN EXPLANATION OF THE BASIS OF
DEVELOPMENTAL INSTABILITY

Introduction

Fluctuating asymmetry (FA) is the average difference in size or shape of paired or

bilaterally symmetric morphological trait sampled across a population. The study of FA,

thought to be a measure of developmental instability, has a controversial history.

Fluctuating asymmetry is hypothesized by some to universally represent a population's

response to environmental and/or genetic stress (Parsons 1992, Clarke 1993, Graham

1992). It is also generally accepted that FA may be co-opted as an indicator or even a

signal of individual genetic buffering capacity to environmental stress (Moller 1990,

Moller and Pomiankowski 1993). Recent literature reviews reveal that these conclusions

are perhaps premature and analyses of individual studies often demonstrate conflicting

results (Lens 2002, Bjorksten 2000). Babbitt et al. (2006) demonstrate that this conflict

may be caused by under-sampling due to a false assumption that FA always exhibits a

normal distribution. Also, FA may be responding to experimental treatment in a complex

and as yet unpredictable fashion. Until the basis of FA is better understood, general

interpretation of FA studies remains difficult.

What is the Basis of FA?

Earlier studies of sexual selection and FA conclude that FA is ultimately a result

of strong selection against the regulation of the development of a particular morphology

(e.g., morphology used in sexual display). Thus in some instances FA may increase and









therefore become a better signal and/or more honest indicator of good genes. Sexually

selected traits tend to have increased FA (Moller and Swaddle 1997), however the exact

mechanism by which FA increases remains unexplained (i.e., a black box). More

recently, theoretical attempts have been made to explain how FA may be generated but

none have been explicitly tested. Models for the phenomenological basis of FA fall into

two general categories: reaction-diffusion models and diffusion-threshold models

(reviewed in Klingenberg 2003). The former class of models involves the chaotic and

nonlinear dynamics in the regulation or negative feedback among neighboring cells

(Emlen et al. 1993) or adjacent bilateral morphology (Graham et al. 1993). The latter

class of models combines morphogen diffusion and a threshold response, the parameters

of which are controlled by hypothetical genes and a small amount of random

developmental noise (Klingenberg and Nijhout 1999). The result of this latter class of

model is that different genotypes respond differently to the same amount of noise,

providing an explanation for genetic variation in FA response to the same environments.

Traditionally, models for the basis of FA assume that variation in FA arises from

independent stochastic events that influence the regulation of growth through negative

feedback rather than processes that may fuel or promote growth. None of these models

explicitly or mathematically address the effect of stochastic behavior in cell cycling on

the exponential growth curve. More recently Graham et al. (2003) makes a compelling

argument that fluctuating asymmetry often results from multiplicative errors during

growth. This is consistent with one particular detail about how cells behave during

growth. For several decades there has been evidence that during development, cells

actually compete to fill prescribed space until limiting nutrients or growth signals are









depleted (Diaz and Moreno 2005, Day and Lawrence 2000). Cell populations effectively

double each generation until signaled or forced to stop. It has also been observed that in

Drosophila wing disc development, that synchrony in cell cycling does not occur across

large tissue fields but rather extends only to an average cluster of 4-8 neighboring cells

regardless of the size and stage of development of the imaginal disc (Milan et al. 1995).

The assumption of previous models, particularly the reaction-diffusion type, that cell

populations are collectively controlling their cell cycling rates across a whole

developmental compartment is probably unrealistic. Regulatory control of fluctuating

asymmetry almost certainly does occur, but probably at a higher level involving multiple

developmental compartments where competing cells are prevented from crossing

boundaries. However, given that individual cells are behaving more or less

autonomously during growth within a single developmental compartment, I suggest that

variation in fluctuating asymmetry can be easily generated at this level by a process

related to stochastic exponential expansion and its termination in addition to regulatory

interactions that probably act at higher levels in the organism. In this paper, I explore and

test simple model predictions regarding the generation of fluctuating asymmetry though

multiplicative error without regulatory feedback.

Exponential Growth and Non-Normal Distribution of FA

In previous work, Babbitt et al. (2006) demonstrate that the distribution of unsigned

FA best fits a lognormal distribution with scaled or power-law tails (double Pareto

lognormal distribution or DPLN). This distribution can be generated by random

proportional (exponential) growth (or geometric Brownian motion) that is stopped or

observed randomly according to a negative exponential probability (Reed and Jorgensen

2004). I suggest that this source of power-law scaling in the tails of the FA distribution is









also the cause of leptokurtosis that is often observed empirically in the distribution of FA.

Kurtosis is the value of the standardized fourth central moment. Like the other moments,

(location, scale, and skewness), kurtosis is best viewed as a concept that can be

formalized in multiple ways (Mosteller and Tukey 1977). Leptokurtosis is best

visualized as the location and scale-free movement of probability mass from the

shoulders of a symmetric distribution towards both its center and tail (Balanda and

MacGillivray 1988). Both the Pareto and power-function distributions have shapes

characterized by the power-law and a large tail and therefore exhibit a lack of

characteristic scale. Both because kurtosis is strongly affected by tail behavior, and

because leptokurtosis involves a diminishing of characteristic scale in the shape of a

distribution, the concepts of scaling and kurtosis in real data can be, but are not

necessarily always, inter-related.

Both in the past and very recently, leptokurtosis in the distribution of FA has been

attributed to a mixture of normal FA distributions caused by a mixing of individuals, all

with different genetically-based developmental buffering capacity, or in other words,

different propensity for expressing FA (Gangestad and Thornhill 1999, Palmer and

Strobek 2003, Van Dongen et al. 2005). Although not noted by these authors, continuous

mixtures of normal distributions generate the Laplace distribution (Kotz et al. 2001,

Kozubowski and Podgorski 2001) and can be distinguished from other potential

candidate distributions by using appropriate model selection techniques, such as the

Akaike Information Criterion technique (Burnham and Anderson 1998). Graham et al.

(2003) rejects the typical explanation of leptokurtosis through the mixing of normal

distributions by noting that differences in random lognormal variable can generate









leptokurtosis. Babbitt et al. (2006) also reject the explanation that leptokurtosis in the

distribution of FA is caused by a mixture of normal distributions because the double

Pareto-lognormal distribution, not the Laplace distribution, always appears the better fit

to large samples of FA. Therefore, leptokurtosis, often observed in the distribution of

FA, may not be due to a mixing process, but instead may be an artifact of scaling in the

distribution tail, which provides evidence of geometric Brownian motion during

exponential expansion of populations of cells.

Testing a Model for the Basis of FA

I propose that the proximate basis for variation of FA in a population of organisms

is due to the random termination of stochastic geometric growth. The combination of

opposing stochastic exponential functions results in the slow power-law decay that

describes the shape of the distribution's tail. In this paper, I present a model for FA and

through simulation, test the prediction that genetic variation in the ability to precisely

terminate growth will lead to increased kurtosis and decreased scaling exponent in the

upper tail of the distribution. Then, I assess the validity of this model by direct

comparison to the distribution of FA within large samples of wild and inbred populations

of Drosophila. Under the assumption that a less heterozygous population will have less

variance in the termination of growth, I would predict that inbreeding should act to

reduce power-law scaling effects in the distribution of FA in a population. Inbreeding

should also reduce the tail weight kurtosiss) and mean FA assuming inbred individuals

have lower variance in the times at which they terminate growth. I also assume that

inbreeding within specific lines does not act to amplify FA due to inbreeding depression.

It has been demonstrated that Drosophila melanogaster do not increase mean FA in

response to inbreeding (Fowler and Whitlock 1994) and it is suggested that large









panmictic populations typical of Drosophila melanogaster may not harbor as many

hidden deleterious recessive mutations as other species (Houle 1989), making them

resistant to much of the typical genetic stress of inbreeding. Therefore, the absence of

specific gene effects during inbreeding suggests that Drosophila may be a good model

for investigating the validity of our model as an explanation of the natural variation

occurring in population level FA.

Methods

Model Development

Simulation of geometric Brownian motion

Ordinary Brownian motion is most easily simulated by summing independent

Gaussian distributed random numbers or white noise (X,). See Figure 3-1.


(3.1) W(X)= X,
i=1

which simulated in discrete steps is


(3.2) N,= Nt, + 1

where N = cell population size, t = time step and W = a random Gaussian variable.

Exponential or geometric Brownian motion, a random walk on a natural log scale, can be

similarly simulated. Geometric Brownian motion is described by the stochastic

differential equation

(3.3) dY(t) = pY(t)dt + vY(t)dW(t)

or also as


(3.4) d =(t) udt + udW(t)
Y(t)










WHITE NOISE










ORDINARY BROWNIAN MOTION
-10050 If----------------------------

1000

600

N 200

47 70 93 116139162185 08 231 2542773003233466 43
-400
-600

time

Figure 3-1. Ordinary Brownian motion (lower panel) in N simulated by summing
independent uniform random variables (W) (upper panel).

where W(t) is a Brownian motion (or Weiner process) and / and v are constants that

represent drift and volatility respectively. Equation 3.3 has a lognormal analytic solution


(3.5) Y(t) = Y(O)e(-' 12)t+vW(t)

A simulation of geometric Brownian motion in discrete form follows as


(3.6) Nt = Nt-,1 t-W,


where N = cell population size, t = time step and W = a random Gaussian variable. See

Figure 3-2. Equation 3.6 is identical to the equation for multiplicative error in Graham et

al. (2003). I modify equation 3.6 slightly by letting Wrange uniformly from 0.0 to 1.0


with W= 0.5 and adding the drift constant C that allows for stochastic upward drift (at C

> 0.5) or downward drift (at C < 0.5).












(3.7) N -= CN1 + Nt-lW


= (C + W,)N,,


At C = 0.5, eqns. 3.6 and 3.7 behave identically. Geometric Brownian motion with


upward drift is shown in Figure 3-3.


0.00045

0.0004

0.00035

0.0003

0.00025

0.0002

0.00015

0.0001

0.00005

0


35 69 103 137 171 205 239 273 307 341 375 409 443 477
time step


Log N


1
0.1
0.01
0.001
0.0001
0.00001
0.000001
0.0000001


time step


time


Figure 3-2. Geometric Brownian motion in N and log N simulated by multiplying
independent uniform random variables. This was generated using Equation
1.5 with C = 0.54.


36 71 106 141 176211 246 281 316 351 386421 456


j___A__ War_
AA VW


-
-
-







44



4500000000
4000000000
3500000000
3000000000
2500000000
N 2000000000
N
1500000000
1000000000
500000000
0
1 37 73 109 145 181 217 253 289 325 361 397 433 469
time step

1E+11
1E+09
1000000
0
100000

Log N 1000
10 -
01 34 133 166 199 232 265 298 331 364 397 430 463
0 001
000001
time step


time


Figure 3-3. Geometric Brownian motion in N and log N with upward drift. This was
generated using Equation 1.5 with C = 0.60.



Simulation of fluctuating asymmetry

Using the MathCad 13 (Mathsoft Engineering and Education 2005), two


independent geometric random walks were performed and stopped randomly at mean


time t= 200 steps with some variable normal probability. The random walks result in cell


population size equal to N, (orNf and NI on left and right sides respectively).


Fluctuating asymmetry (FA) was defined as the difference in size resulting from this


random proportional growth on both sides of the bodies axis of symmetry plus a small


degree of random uniformly distributed noise or










(3.8) FA= NL -N +rv

where (rv) was uniformly distributed with a range of +0.1 ( N N ). Using a

MathCad-based simulation in VisSim LE, the generation of individual FA values was

repeated until a sample size of 5000 was reached. The random noise (rv) has no effect on

the shape of the FA distribution but fills empty bins (gaps) in distribution tails. Because

rv is small in comparison to N NJ, its effect is similar to that of measurement error

(which would be normally distributed rather than uniform). Schematic representation of

the simulation process for Reed and Jorgensen (2004) model and simulation of

fluctuating asymmetry are shown in Figure 4 and 5. The simulation of the distribution of

fluctuating asymmetry was compared at normal standard deviation of termination of

growth (t) ranging from a = 0.5, 0.8, 1.2,3 and 7 with a drift constant ofC = 0.7.

Inbreeding Experiment

In May 2004 in Gainesville, Florida, 320 free-living Drosophila simulans were

collected in banana baited traps and put into a large glass jar and cultured on instant rice

meal and brewer's yeast. After two generational cycles 1000 individuals were collected

in alcohol. This was repeated again in June 2005 with 200 wild trapped flies.

Lines of inbred flies were created from the May 2004 wild population through eight

generations of full sib crosses removing an estimated 75% of the preexisting

heterozygosity (after Crow and Kimura 1970). Initially, ten individual pairs were

isolated from the stock culture and mated in 1/2 pint mason jars with media and capped

with coffee filters. In each generation, and in each line, and to ensure that inbred lines

were not accidentally lost though an inviable pairing, four pairs of Fl sibs from each

cross were then mated in 12 pint jars. Offspring from one of these four crosses were











randomly selected to set up the next generation. Of the original ten lines, only four

remained viable after eight generations of full sib crossing. These remaining lines were

allowed to increase to 1000+ individuals in 1 quart mason jars and then were collected

for analysis in 85% ethanol. This generally took about 4 generations (8 weeks) of open

breeding. One completely isogenic (balancer) line of Drosophila melanogaster was

obtained from Dr. Marta Wayne, University of Florida and also propagated and collected.
















0 10 20 30 40 50 60 70 80 90 100
I001










1 DOUBLE PARETO LOGNORMAL DISTRIBUTIONITH











NEGATIVE EXPONENTI AL REPEAT
VARIATION IN GROWTH SAMPLING
0 10 20 30 40 50 60 70 80 90 100













STOP TIME
T SO DOUBLE PARETO LOGNORMAL DISTRIBUTION













'0 100
o VARIATION IN GO 00










Figure 3-4. Model representation of Reed and Jorgensen's (2004) physical size
distribution model. Variable negative exponentially distributed stopping
Times of random proportional growth (GBM with C = 0.5) create double
Pareto lognormal distribution of size.












Stochastic Geometric Growth


RANDOM WALKS WITH
GEOMETRIC BROWNIAN
MOTION


GROWTH STOP TIME


SIZE DISTRIBUTION


I_ \ 1

REPEAT
NORMAL VARIATION SAMPLING
IN GROWTH STOP TIME





0 50 100 150

Figure 3-5. A model representation of developmental instability. Normally variable
stopping times of random proportional growth (GBM with C > 0.5) create
double Pareto lognormal distribution of size.


All flies were cultured on rice meal and yeast at 290C with 12:12 LD cycle in

environmentally controlled chambers at the Department of Entomology and Nematology

at the University of Florida. Wings were collected and removed from 1000 flies from

each of the two samples of wild population and four samples of inbred lines and dry

mounted on microscope slides. Specimens were dried in 85% ethanol, and then pairs of









wings were dissected (in ethanol) and air-dried to the glass slides. Permount was used to

attach cover slips. This technique prevented wings from floating up during mounting,

which might slightly distort the landmark configuration. Dry mounts were digitally

photographed. All landmarks were identified as wing vein intersections on the digital

images (eight landmarks on each wing). See Appendix A for landmark locations.

Morphometric analyses

Wing vein intersections were digitized on all specimens using TPSDIG version

1.31 (Rohlf, 1999). Specimens damaged at or near any landmarks were discarded.

Fluctuating asymmetry was measured in two ways on all specimens using landmark-

based multivariate geometric morphometrics. A multivariate size-based FA (FA 1 in

Palmer and Strobek 2003) was calculated as absolute value of(R L) or just R-L in

signed FA distributions where R and L are the centroid sizes of each wing (i.e., the sum

of the distances of each landmark to their combined center of mass or centroid location).

In addition, a multivariate shape-based measure of FA known as the Procrustes distance

was calculated as the square root of the sum of all squared Euclidean distances between

each left and right landmark after two-dimensional Procrustes fitting of the data

(Bookstein 1991; Klingenberg and McIntyre 1998; FA 18 in Palmer and Strobeck 2003;

Smith et al. 1997). Centroid size calculation, Euclidean distance calculation and

Procrustes fitting were performed using 0yvind Hammer's Paleontological Statistics

program PAST version 0.98 (Hammer 2002). A sub-sample of 50 individuals from the

fourth inbred line (pp4B3) was digitized five times to estimate measurement error. In

these cases, measures of FA were taken as the average FA value of the five replicate

measurements for each specimen. Percent measurement error was also computed as









(ME/average FA) x 100 whereME = sd(FA, FA2, FA3, FA4, FA5) (as per Palmer and

Strobek 2003). For assessing whether measurement error (ME) interfered significantly

with FA, a Procrustes ANOVA (in Microsoft Excel) was performed on the five

replications of the 50 specimen sub-sample (Klingenberg and McIntyre 1998). Any

subsequent statistical analyses were performed using SPSS Base 8.0 statistical software

(SPSS Inc.).

Model selection and inference

The fits of unsigned size FA to three distributional models (half-normal, lognormal,

and double Pareto lognormal (DPLN)) were compared in the Drosophila lines, by

calculating negative log likelihood and Akaike Information Criteria (AIC) (Burnham

and Anderson 1998; Hilborn and Mangel 1998). This method penalizes more complex

models (those with more parameters) when selecting the best-fit distributional model for

a given dataset. Best fitting parameters were obtained by maximizing the log-likelihood

function for each model (Appendix B). The maximization was performed using the

conjugate gradient method within unconstrained solve blocks in the program MathCAD

by MathSoft Engineering and Education Inc (2001).

Results

Model Simulation

The amount of variance in termination times related directly to levels of simulated

FA (i.e., low variance in termination time (t) gives low FA and vice versa). I found that

not only does amount of FA increase with increased variance in (t), but so do both

kurtosis and the scaling effect in the distribution tails. In Figure 3-6, normal quantile

plots of signed FA are shown for different standard deviations in the normal variation of

the termination of growth of geometrically expanding cell populations. The degree of the







50


S or sigmoidal shape in the plot indicates level ofleptokurtosis. The leptokurtosis in the

quantile plot is reduced greatly with a decrease in the standard deviation of the normal

variation in growth termination times.


S= 0.5
kurtosis


o =0.8
kurtosis


- =1.2
kurtosis


0.634







0.933






=1.149


o =3.0
kurtosis = 11.812


r = 7.0
kurtosis


:27.266


SIGNED DISTRIBUTION OF
FLUCTUATING ASYMMETRY


4
7a-t


a









fl
th


DETRENDED FIT
TO NORMAL


Figure 3-6. Simulated distributions of cell population size and FA for different amounts
of variation in the termination of growth (variance in normally distributed
growth stop time). The fit of simulated data to the normal distribution can
determined by how closely the plotted points follow the horizontal line (a
good fit is horizontal).


I


I









Experimental Results

Both mean unsigned size FA and shape FA decreased with inbreeding in all lines.

I also observed that the kurtosis of signed size FA and the skewness of both unsigned size

and shape FA follow an identical trend. The trend was strongest in kurtosis, which

decreased rapidly with inbreeding, indicating that, as predicted, changes in mean FA are

influenced strongly by the shape and tail behavior of the distribution of FA (Table 3-1).

Table 3-1. Distribution parameters and model fit for multivariate FA in two wild
populations and four inbred lines of Drosophila simulans and one isogenic
line of Drosophila melanogaster. Model fits are A AIC for unsigned centroid
size FA (zero is best fit, lowest number is next best fit).

Kurtosis Mean Mean Scaling
(signed (unsigned (shape FA) exponent AAIC AAIC AAIC
size FA) size FA) in upper HNORM LGN DPLN
tail
Wild 2004 28.3 5.73 0.0223 1.49 90.6 324 0.000
+0.12 0.38 0.0003
Wild 2005 45.9 4.68 0.0237 2.87 66.2 261 0.000
+0.15 0.19 0.0003
Inbred line 2.43 3.27 0.0183 5.09 0.000 104 15.5
1 (OR18D) 0.11 0.09 0.0002
Inbred line 8.85 4.24 0.0212 3.10 11.50 223 0.000
2(OR18D3) 0.15 0.13 0.0002
Inbred line 2.90 3.31 0.0184 4.63 0.000 130 44.0
3 (PP4B2) 0.15 0.09 0.0002
Inbred line 3.30 3.72 0.0213 5.78 0.000 228 79.7
4 (PP4B3) 0.15 0.11 0.0003
Isogenic 0.550 3.90 0.0201 9.99 0.000 84.9 101
(me175) 0.11 0.10 0.0002

While wild populations on the whole, demonstrated increased mean, skew and

leptokurtosis of FA compared to inbred lines, significant differences between populations

were found in all distributional parameters, even between each of the inbred lines

(ANOVA mean size FA for all lines; F = 27.49, p < 0.001; ANOVA mean size FA for

inbred lines only; F = 14.38, p < 0.001; note shape FA also show same result). Overall,









inbred lines demonstrated lower kurtosis, just slightly above that expected from a normal

distribution (Table 1). They also had lower mean FA. No significant differences were

found with respect to these results according to sexes of flies. In Figure 3-7, I show the

distribution of FA and detrended normal quantile plots for the wild population, four

inbred lines and one isogenic line respectively.

As in the simulated data, the degree of the S shape in the plot indicated level of

leptokurtosis. The S shape in quantile plot is reduced greatly with inbreeding and nearly

disappears in the isogenic line.

Model Selection and Inference

The comparison of candidate distributional models of FA demonstrated normalization

associated with inbreeding in three of the four inbred lines (Table 3-1). In the remaining

inbred line, the half-normal candidate model was a close second to the double Pareto

lognormal distribution. In the wild population samples, the distribution of unsigned size-

based FA was best described by the double Pareto lognormal distribution (DPLN), a

lognormal distribution with scaling in both tails. In the best fitting parameters of this

distribution there was no observable trend in lognormal mean or variance across wild

populations and inbred lines. The scaling exponent of the lower tail (P in Reed and

Jorgensen 2004) was close to one in all lines while the scaling exponent in the upper tail

(a in Reed and Jorgensen 2004) increased with inbreeding (Table 3-1). Both samples of

the wild population demonstrated the lowest scaling exponent in the upper tail of the

distribution. The low scaling exponents here indicate divergence in variance (2004 and

2005 where a < 3) and in the mean (2004 only where a < 2). The inbred lines all show

higher scaling exponents that are consistent with converging mean and variance.













wild 2004
kurtosis = 28.25


wild 2005
kurtos =45.85





inbred -OR18D
kurtosis =2.43






inbred -OR18D3
kurtosis =8.85






inbred -PP4B2
kurtosis = 2.90


kurtosis =3.30






isogenic mel 75
kurtosis =0.55


aP


II

A




r







a
U,\


[r


eat ai


Figure 3-7. Distribution of fluctuating asymmetry and detrended fit to normal for two
samples of wild population collected in Gainesville, FL in summers of 2004
and 2005 and four inbred lines of Drosophila simulans derived from eight
generations of full-sib crossing of the wild population of 2004. Also included
is one isogenic line of Drosophila melanogaster (me175). All n = 1000. The
fit of data to the normal distribution can determined by how closely the
plotted points follow the horizontal line (a good fit is horizontal).


-a


I


~----~-
"









The isogenic Drosophila melanogaster line demonstrated the highest scaling exponent

and the most normalized distribution of unsigned FA.

Measurement error was 7.6% for shape-based FA and 13.0% for centroid size-

based FA. In a Procrustes ANOVA (Klingenberg and McIntyre 1998) the mean squares

for the interaction term of the ANOVA (MSinteraction) was highly significant p<0.001

indicating that FA variation was significantly larger than variation due to measurement

error (ME).

Discussion

Revealing the Genetic Component of FA

While we should be cautious about inferring process from pattern, the very similar

results of both the modeling and the inbreeding experiment in Drosophila seem to

suggest the presence of a scaling component in the distribution of fluctuating asymmetry

that is caused by a random multiplicative growth process as suggested previously by

Graham et al. 2003. This parameter appears to change with the genetic redundancy of the

population which is presumably increased by genetic drift and and reduction in

heterozygosity during inbreeding. Specifically, the scaling exponent(s) of the upper tail

(a) of the unsigned FA distribution, or outer tails of the signed FA distribution, are

increased with inbreeding, causing more rapid power-law decay in the shape of the tails.

This effect also reduces kurtosis and apparently normalizes the distribution of FA in more

inbred populations. This suggests that individual genetic differences in the capacity to

control variance in the termination of random proportional growth (i.e., geometric

Brownian motion) may be responsible for determining the shape and kurtosis of the

distribution of FA. In other words, leptokurtosis kurtosiss >3) in signed FA distribution









indicates genetic variability in the population while normality kurtosiss = 3) indicates

genetic redundancy.

Because leptokurtosis is very often observed in the distribution of FA, genetic

variability potentially underlies a large proportion of the variability observed in the FA of

a given population. Observed differences or changes in FA are therefore not only a

response of development to environmental stress, but clearly also can reflect inherent

differences in the genetic redundancy of populations. The significantly different levels of

mean FA among the inbred lines in this study, presumably caused by the random fixation

of certain alleles, also suggests that there is a strong genetic component to the ability to

buffer development against random noise. It is assumed that the differences observed in

FA between wild trapped and inbred populations in this study do not indicate an effect of

inbreeding depression in the study for two reasons. First, the four inbred lines analyzed

in this study were vigorous in culture so the fixation of random alleles was probably not

deleterious. Second, and more important, inbreeding reduced FA rather than increasing it

as would have been expected under genetic stress.

It is also important to note that kurtosis is potentially a much stronger indicator of

FA than the distribution mean. The low scaling exponents found in the non-normal

distributions of FA in the wild populations of Drosophila simulans are capable of

slowing and perhaps even stopping, the convergence on mean FA with increased sample

size. Because kurtosis is a fourth order moment, estimating it accurately also requires

larger sample sizes. However, if kurtosis can be demonstrated to respond as strongly to

environmental stress as it does here to inbreeding, its potential strength as a signal of FA

may allow new interpretation of past studies of FA without the collection of more data.









This may help resolve some of the current debate regarding FA as a universal indicator of

environmental health and as a potential sexual signal in "good genes" models of sexual

selection.

Limitations of the Model

There are certain aspects of the model presented here that may be

oversimplifications of the real developmental process. First of all, this model assumes

developmental instability is generated left-right growing tissue fields with no regulatory

feedback or control other than when growth is stopped. It is very likely that left-right

regulation is able to occur at higher levels of organization (e.g., across multiple

developmental compartments) even though there is no evidence of regulated cell cycling

rates beyond the distance of 6-8 cells on average within any given developmental

compartment (Milan et al. 1995). Therefore this model explains how fluctuating

asymmetry is generated, not how it is regulated. Second, this model only considers cell

proliferation as influencing size. It is known that both cell size and programmed cell

death, or apoptosis, are also important in regulating body size (Raff 1992, Conlin and

Raff 1999). Both of these may play a more prominent role in vertebrate development,

than they do in insect wings, where apparently growth is terminated during its

exponential phase. Nevertheless, this simple model seems to replicate very well, certain

behavioral aspects of the distribution of fluctuating asymmetry in the insect wing.

The Sources of Scaling

The basic process generating power-law scaling effects illustrated here (Reed 2001)

offers an alternative perspective to the often narrow explanations of power-laws caused

by self-organized criticality in the interactions among system components (Gisiger 2000).

The power law that results from self-organized criticality is created from the multiplicity









of interaction paths in the network. As the distance between two interacting objects is

increased in a network or multidimensional lattice, the number of potential interaction

pathways increases exponentially, while the correlation between such paths decreases

exponentially (Stanley 1995). These opposing exponential relationships create the power

law scaling observed in simulations of self-organized critical systems. While there is no

such "interaction" in our model, there is a power-law generated by opposing exponential

functions. The constant degree of change represented by the power law in both the

statistical physics of critical systems and the mathematics of both Reed and Jorgensen's

process and the model given here is the result of the combined battle between both the

exponentially increasing and decreasing rates of change (Reed 2001). While the natural

processes are quite different, the underlying mathematical behavior is very similar.

Potential Application to Cancer Screening

Just as changes in the shape of the distribution of fluctuating asymmetry is

normalized across a population of genetically redundant individuals, genetic redundancy

in a population of cells may also help maintain normal cell size and appearance. The loss

of genetic redundancy in a tissue is a hallmark of cancer. The abnormal gene expression

and consequent genetic instability that characterizes cancerous tissue often results in

asymmetric morphology in cells, tissues and tissue borders. Baish and Jain (2000)

review the many studies connecting fractal (scale free) geometry to the morphology of

cancer. Cancer cells also are typically pleomorphic or more variable in size and shape

than normal cells and this pleomorphy is associated with intercellular differences in the

amount of genetic material (Ruddon 1995). Frigyesyi et al. (2003) have demonstrated a

power-law distribution of chromosomal aberrations in cancer. Currently the type of

distributional shape of the pleomorphic variability in cell size is not known, or at least not









published. However, Mendes et al. (2001) demonstrated that cluster size distributions of

HN-5 (cancer) cell aggregates in culture followed a power-law scaled distribution.

Furthermore these authors also demonstrated that in MDCK (normal) cells and Hep-2

(cancer) cells, cluster size distributions transitioned from short-tailed exponential

distributions to long-tailed power-law distributions over time. The transition is

irreversible and is likely an adaptive response to high density and long permanence in

culture due to changes in either control of replication or perhaps cell signaling. Taken

collectively, these studies may suggest that scaling at higher levels of biological

organization observed in cancer is due to increased relative differences in length of cell

cycling rates of highly pleomorphic cell populations that have relatively larger

intercellular differences in amount of genetic material. The stochastic growth model I

have proposed as the basis for higher variability in population level developmental

instability or FA may also provide a possible explanation for higher variability in the cell

sizes of cancerous tissue. If genetic redundancy in growing tissue has the same

distributional effect as genetic redundancy in populations of organisms and tends to

normalize the observed statistical distribution, then one might predict that the genetic

instability of cancerous tissue would create a scaling effect that causes pleomorphy in

cells and scaling in cell cluster aggregations. Statistical comparison of cell and cell

cluster size distributions in normal and cancerous tissues may provide a useful and

general screening technique for detecting when genetic redundancy is compromised by

cancer in normal tissues.

Conclusion

Until now, the basis of fluctuating asymmetry has been addressed only with

abstract models of hypothetical cell signaling, or elsewhere, at the level of selection









working on the organism with potential mechanism remaining in the black box.

However, fluctuating asymmetry must first and foremost be envisioned as a stochastic

process occurring during tissue growth, or in other words, occurring in an exponentially

expanding population of cells. As demonstrated in the model presented here, this

expansion process can be represented by stochastic proportional (geometric) growth that

is terminated or observed randomly over time. These results imply that the fluctuating

asymmetry observed in populations is not only related to potential environmental

stressors, but also to a large degree, the underlying genetic variability in those molecular

processes that control the termination of growth. Therefore, fluctuating asymmetry

responses to stress may be hard to interpret without controlling for genetic redundancy in

the population. Both the simulation and experimental results suggest that measures of

distributional shape like kurtosis, scaling exponent and tail weight may actually be a

strong signal of variability in the underlying process that causes developmental

instability. Therefore the kurtosis parameter of the fluctuating asymmetry distribution

may provide more information about fluctuating asymmetry response than does a

populations' average or mean fluctuating asymmetry. This may provide a novel method

by which to resolve conflicts in previous under-sampled research without the collection

of more data.














CHAPTER 4
TEMPERATURE RESPONSE OF FLUCTUATING ASYMMETRY TO IN AN APHID
CLONE: A PROPOSAL FOR DETECTING SEXUAL SELECTION ON
DEVELOPMENTAL INSTABILITY

Introduction


Developmental instability is a potentially maladaptive component of individual

phenotypic variation with some unknown basis in both gene and environment (Moller

and Swaddle 1997, Fuller and Houle 2003). Developmental instability is most often

measured by the manifestation of fluctuating asymmetry (FA), the right minus left side

difference in size or shape in a single trait across the population (Palmer and Strobek

1986, 2003; Parsons 1992, Klingenberg and McIntyre 1998). Because FA is thought to

indicate stress during development, the primary interest in the study of FA has been its

potential utility as an indicator of good genes in mate choice (Moller 1990, Moller and

Pomiankowski 1993) or its utility as a general bioindicator of environmental health

(Parsons 1992).

The Genetic Basis of FA


For FA to become a sexually selected trait, it must be assumed that it has a

significant genetic basis, and can therefore evolve (Moller 1990, Moller and

Pomiankowski 1993). However, researchers who use FA as a bioindicator of

environmental health often assume that FA is a phenotypic response that is mostly

environmental in origin (Parsons 1992, Lens et al. 2002). Studies of the heritability of

FA seem to indicate low heritability for FA exists (Whitlock and Fowler 1997, Gangestad









and Thornhill 1999). However, because FA is essentially a variance that is often

measured with only two data points per individual, FA may have a stronger but less

easily detectable genetic basis (Whitlock 1996, Fuller and Houle 2003). Additive genetic

variation in FA in most studies has been found to be minimal, but several quantitative

trait loci studies suggest significant dominance and character specific epistatic influences

on FA (Leamy 2003, Leamy and Klingenberg 2005). Babbitt (chapter 3) has

demonstrated that a population's genetic variability affects the distributional shape of FA.

So while studies investigating mean FA may be inconclusive, changes in the population's

distributional shape seem to indicate potential genetic influence on FA. However, no

studies have observed FA in a clonal organism for the express purpose of assessing

developmental instability that is purely environmental (i.e., non-genetic) in its origin (i.e.,

developmental noise).

The Environmental Basis of FA


It has long been assumed that FA is the result of some level of genetically-based

buffering of additive independent molecular noise during development. Because of the

difference in scale between the size of molecules and growing cells it would be unlikely

that molecular noise would comprise an important source of variation in functioning

cells. However, Leamy and Klingenberg (2005) rationalize that molecular noise could

only scale to the level of tissue when developmentally important molecules exist in very

small quantity (e.g., DNA or protein) and therefore FA may represent a stochastic

component of gene expression. This is somewhat similar in spirit to the Emlen et al.

(1993) explanation of FA that invokes non-linear dynamics of signaling and supply that

may also occur during growth. Here FA is thought to be the result of the scaling up of









compounding temporal asymmetries in signaling between cells during growth. In this

model, hypothetical levels of signaling compounds (morphogens) and or growth

precursors used in the construction of cells vary randomly over time. When growth

suffers less interruption, thus when it occurs faster, there is also less complexity (or

fractal dimension) in the dynamics of signaling and supply. Graham et al. (1993) suggest

that nonlinear dynamics of hormonal signaling across the whole body may also play a

similarly important role in the manifestation of FA. However, while the levels of FA are

certainly influenced by the regulation of the growth process, both Graham et al. (2003)

and Babbitt (chapter 3) also suggest that FA levels reflect noise during cell cycling that is

amplified by exponentially expanding populations of growing cells.

Although the proximate basis of FA is not well understood, its ultimate

evolutionary basis, while heavily debated, is easier to understand. Moller and

Pomiankowski (1993) first suggested that strong natural or sexual selection can remove

regulatory steps controlling the symmetric development of certain traits (e.g.,

morphology used in sexual display). They suggest that with respect to these traits (and

assuming that they are somehow costly to produce), individuals may vary in their ability

to buffer against environmental stress and developmental noise in relation to the size of

their individual energetic reserves; which are in turn often indicative of individual genetic

quality. Therefore high genetic quality is associated with low FA.

Most existing proximate or growth mechanical explanations of FA assume that

rapid growth is less stressful in the sense that fewer interruptions of growth by various

types of noise should result in lower FA (Emlen et al. 1993, Graham et al. 1993).

However, ultimate evolutionary explanations of FA assume that rapid growth is









potentially more stressful because it is energetically costly and therefore rapid growth

should increase the level of FA when energy supply is limiting (Moller and

Pomiankowski 1993). This high FA is relative and so should be especially prominent in

individuals of lower genetic quality who can least afford to pay this additional energy

cost. This fundamental difference between the predictions of proximate (or mechanistic)

level and ultimate evolutionary level effects of temperature on FA is shown in Figure 1.

The theoretical difference in the correlation of temperature and growth rate to FA in both

the presence and absence of energetic limitation could be used to potentially detect sexual

selection on FA. However it first should be confirmed that FA should decrease with

more rapid growth in the absence of energetic limitation to growth and genetic

differences between individuals in a population. This later objective is the primary goal

of this study.

Temperature and FA in and Aphid Clone


At a very basic level, entropy or noise in physical and chemical systems has a

direct relationship with the physical energy present in the system. This energy is

measured by temperature. Because FA is speculated to tap into biological variation that

is somewhat free of direct genetic control, it may therefore respond to temperature in

simple ways.

First, increased temperature may increase molecular entropy which may in turn

increase developmental noise during development thereby increasing FA. Second,

increased temperature may act as a cue to shorten development time (as in many aphids

where higher temperature reduces both body size and development time), thereby

reducing the total time in which developmental errors may occur. This should reduce










FA. A third possibility is that a species specific optimal temperature exists. If so, FA

should increase while approaching both the upper and lower thermal tolerance limits of

organisms.

PROXIMATE LEVEL EFFECT


DEVELOPMENTAL +
TIME

TEMPERATURE FLUCTUATING
ASYMMETRY

GROWTH RATE


ULTIMATE LEVEL EFFECT

+
STRESS DEVELOPMENTAL
+ \^ / TIME

NON-OPTIMAL -- FLUCTUATING
TEMPERATURE ASYMMETRY
ENERGETIC G R
RESERVES GROWTH RATE
+
Figure 4-1. Predicted proximate and ultimate level correlations of temperature and
growth rate to fluctuating asymmetry are different. Ultimate level
(evolutionary) effects assume energetic limitation of individuals in the system.
Proximate level (growth mechanical) effects do not. Notice that temperature
and fluctuating asymmetry are negatively correlated in the proximate model
while in the ultimate model they are positively correlated.

Only a few studies have directly investigated the relationship between FA and

temperature. The results are conflicting. FA is either found to increase on both sides of

an "optimal" temperature (Trotta et al. 2005, Zakarov and Shchepotkin 1995), to be

highest at low temperature (Chapman and Goulson 2000), to simply increase with

increasing temperature (Savage and Hogarth 1999, Mpho et al. 2002) or not to respond









(Hogg et al. 2001). None of these studies investigate the relationship between

developmental noise (FA in a clonal line) and temperature.

The characterization of developmental noise in response to temperature was

investigated in this study using the cotton aphid, Aphis gossipyii. These aphids reproduce

parthenogenetically, are not energetically limited in their diet (because they excrete

excess water and sugar as "honeydew") and produce wings that are easily measured using

multiple landmarks. They demonstrate large visible variation in body size, wing size and

even wing FA. The visible levels of wing asymmetry in cotton aphids reflect levels of

FA that are about four times higher than that observed in other insect wings (Babbitt et al.

2006, Babbitt in press). Because parthenogenetic aphids cannot purge deleterious

mutations each generation and because Florida clones often never use sexual

reproduction to produce over-wintering eggs, this remarkably high FA may be the result

of Muller's ratchet. Cotton aphids are also phenotypically plastic in response to

temperature, producing smaller lighter morphs at high temperatures and larger dark

morphs at low temperatures. This feature allows observation of two genetically

homogeneous groups in which different gene expression patterns (causing the color

morphs) exist. The central prediction is that of the proximate effect model: that in the

absence of energetic limitation and genetic variation, temperature and environmentally

induced FA (or developmental noise) should be negatively correlated in a more or less

monotonic relationship.

Methods


In March 2003, a monoclonal population of Cotton Aphids (Aphis gossipyii

Glover) was obtained from Dr. J.P. Michaud in Lake Alfred, Florida and was brought to









the Department of Entomology and Nematology at the University of Florida. The culture

was maintained on cotton seedlings (Gossipium) grown at different temperatures (12.50C,

150C, 170C, 190C, 22.50C and 250C with n = 677 total or about 100+ per treatment)

under artificial grow lights (14L:10D cycle). Because of potential under-sampling caused

by a non-normal distribution of FA (see Babbitt et al. 2006), a second monoclonal

population collected from Gainesville, FL in June 2004 was reared similarly but in much

larger numbers at 12.50C, 150C, 170C, 190C, and 250C (n = 1677 or about 300+ per

treatment).

Development time for individual apterous cotton aphids (Lake Alfred clone) were

determined on excised cotton leaf discs using the method Kersting et al. (1999). Twenty

randomly selected females were placed upon twenty leaf discs (5 cm diameter) per

temperature treatment. Discs were set upon wet cotton wool in petri dishes and any first

instar nymphs (usually 3-5) appearing in 24 hours were then left on the discs.

Development time was taken as the average number of days taken to reach adult stage

and compared across temperatures. Presence of shed exoskeleton was used to determine

instar stages. Cotton was wetted daily and leaf discs were changed every 5 days.

Humidity was maintained at 50+5%.

In each temperature treatment, single clonal populations were allowed to increase

on plants until crowded in order to stimulate alate (winged individuals) production.

Temperature treatments above 170C produced light colored morphs that were smaller and

tended to feed on the undersides of leaves of cotton seedlings. Temperature treatments

below 17C produced larger dark morphs that tended to feed on the stems of cotton

seedlings. Alatae were collected using small brushes dipped in alcohol and stored in 80%









ethanol. Wings were dissected using fine insect mounting pins and dry mounted as pairs

on microscope slides. Species identification was by Dr. Susan Halbert at the State of

Florida Department of Plant Industry in Gainesville.

Specimens were dried in 85% ethanol, and then pairs of wings were dissected (in

ethanol) and air-dried to the glass slides while ethanol evaporated. Permount was used to

attach cover slips. This technique prevented wings from floating up during mounting,

which might slightly distort the landmark configuration. Dry mounts were digitally

photographed. Six landmarks were identified as the two wing vein intersections and four

termination points for the third subcostal. See Appendix A for landmark locations.

Wing vein intersections were digitized using TPSDIG version 1.31 (Rohlf,

1999). Specimens damaged at or near any landmarks were discarded. Fluctuating

asymmetry was calculated using a multivariate geometric morphometric landmark-based

method. All landmarks are shown in Appendix A. FA (FA 1 in Palmer and Strobek

2003) was calculated as absolute value of (R L) where R and L are the centroid sizes of

each wing (i.e., the sum of the distances of each landmark to their combined center of

mass or centroid location). In addition, a multivariate shape-based measure of FA known

as the Procrustes distance was calculated as the square root of the sum of all squared

Euclidean distances between each left and right landmark after two-dimensional

Procrustes fitting of the data (Bookstein 1991; Klingenberg and McIntyre 1998; FA 18 in

Palmer and Strobeck 2003; Smith et al. 1997). This removed any difference due to size

alone. Centroid size calculation, Euclidean distance calculation and Procrustes fitting

were performed using 0yvind Hammer's Paleontological Statistics program PAST

version 0.98 (Hammer 2002). Percent measurement error was computed as (ME/average










FA) x 100 where ME= ( FA FA2 + FA2 -FA3 + FA FA3)/3 in a smaller subset

(200 wings each measured 3 times = FA1, FA2 and FA3) of the total sample. All

subsequent statistical analyses were performed using SPSS Base 8.0 statistical software

(SPSS Inc.). Unsigned multivariate size and shape FA as well as the kurtosis of signed

FA were then compared at various temperatures using one-way ANOVA.

Results


Development time (Figure 4-2) was very similar to previously published data

(Kersting et al. 1999, Xia et al. 1999) decreasing monotonically at a much steeper rate in

dark morphs than in light morphs. The distributional pattern of centroid size, size FA and

shape FA appear similar exhibiting right log skew distributions (Figure 4-3). Similar

distributional patterns are observed within temperatures (not shown) as that observed

across temperatures (Figure 4-3).

30

Dark Morph Light Morph
2 5 ......................... . . .
25 -


20 -

DEVELOPMENT
TIME (days) 15 -



1 5 .. .... .... ... 1 ......... 1 ........ ............................................................

10
8 10 12 14 16 18 20 22 24 26

TEMPERATURE (C)

Figure 4-2. Cotton aphid mean development time +1 SE in days in relation to
temperature (n= 531).


































FREQUENCY


100


ND S364 00

CENTROID SIZE


St Dev = 596

T O N = 1822 00

CENTROID SIZE FLUCTUATING ASYMMETRY


J Std Dev= 02
Mean= 056
0 N = 121900
,% *% ,% ,% ,% ,% *o, *,% *% ,% *,%. *% *%

SHAPE FLUCTUATING ASYMMETRY


Figure 4-3. Distribution of isogenic size, size based and shape based FA in monoclonal
cotton aphids grown in controlled environment at different temperatures.
Distributions within each temperature treatment are similar to overall
distributions shown here.











Coefficient of variation for FA was slightly higher for dark morphs (12.5 C =


92.59%, 15 C = 93.02%, 17 C = 88.95%, 19 C = 79.02%, 22.5 C = 80.00% and 25 C =


85.35%). Mean isogenic FA (both size and shape) was highly significantly different


across temperatures (ANOVA F = 6.691, df between group = 4, df within group = 1673,


p < 0.001) in the Gainesville FL clone (Figure 4-4) but not in the Lake Alfred clone


(ANOVA F = 1.992, df between group = 5, df within group = 672, p < 0.078). This is an


indication ofundersampling in the Lake Alfred clone. In the Gainesville clone, mean


centroid size FA (Figure 4-4A) and development time (Figure 4-2) follow a nearly


identical pattern, decreasing rapidly at first then slowing with increased temperature.


1 1 1 1 1 1
10.5 Dark Morph Light Morph
10 A
9.5 -
9
8.5

7.5

6.5

mean isogenic 610 12 14 16 18 20 22 24 26
fluctuating asymmetry
0.068 -
Dark Morph Light Morph
0.066 -
0.064 -
0.062
0.06
0.058
0.056
0.054
0.052 .
10 12 14 16 18 20 22 24 26


TEMPERATURE (C)


Figure 4-4. Mean isogenic FA for (A.) centroid size- based and (B.) Procustes shape-
based) in monoclonal cotton aphids (collected in Gainesville FL) grown on
isogenic cotton seedlings at different temperatures.






71


Mean shape FA was also significantly different across temperature classes

(ANOVA F = 4.863, df within group = 4, df between group = 1673, p = 0.001) but this

difference is due solely to elevated FA in the 12.50C group (Figure 4-4B). Less than one

percent of the variation in FA was due to variation in body size (r = -0.101 for shape FA;

r = 0.088 for size FA). Kurtosis in the shape of the distribution of size based FA (Figure

4-5) was significantly higher in dark morphs than in light morphs (t = -2.21 p = 0.027).

Within each morph (light or dark), kurtosis in the distribution of FA appears to increase

with temperature slightly (Figure 4-5). Measurement error for shape FA was estimated

at2.6% (Lake Alfred clone) and 2.2% (Gainesville clone). For size FA these estimates

were 6.1% (Lake Alfred) and 5.7% (Gainesville).


KURTOSIS


-1


10
10


12 14 16 18 20

TEMPERATURE (C)


Figure 4-5. Kurtosis of size-based FA in monoclonal cotton aphids grown on isogenic
cotton seedlings at different temperatures.


Dark Morph Light Morph ...................................
. ..........


22 24









Discussion


It appears that the prediction of the proximate effect model holds in the case of this

population of cotton aphid, which, in general, is not energetically limited and in this

study, is not genetically variable. Temperature and growth rate (which are positively

correlated in insects) are negatively associated with purely environmentally derived FA

(i.e., developmental noise). This confirms the predictions of several proximate models of

the basis of FA. Furthermore, it appears that centroid size-based FA is a simple function

of development time. Because individual genetic differences in the capacity to buffer

against developmental noise are, in a sense, controlled for in this study by the use of

natural clones, the response of FA to temperature in this study represents a purely

environmental response of FA. The fact that aphids excrete large amount of water and

sugar in the form of honeydew, as well as the lack of a strong correlation between FA and

size also suggests that there is no real energetic "cost" to being large in the aphids in this

study. An important next step will be to compare this result to the association of growth

rate to FA in genetically diverse and energetically limited sexually selected traits where

the predicted association between growth rate and FA would be the opposite of this

study.

It also appears that because the environmental component of size-based isogenic

FA is largely a function of developmental time and temperature, dark morphs of cotton

aphids, which have much longer development time than light morphs, also have

significantly higher FA. The temperature trend in mean FA within both light and dark

morphs, where development times are similar, is not consistent although it appears that

the kurtosis of FA increases slightly with temperature within each morph.









It is very interesting that there is a strong difference in kurtosis between the two

temperature morphs in this study. Previous work suggests that kurtosis is related to

genetic variability in a population (Babbitt in press). The difference observed here in

light of genetic homogeneity in the monoclonal aphid cultures, suggests that there may be

a difference in developmental stability of light and dark aphid morphotypes that is due

primarily to the differential expression of genes in each phenotype.

It is surprising that temperature trends in mean developmental noise are slightly

different regarding whether a size or shape-based approach was used. In populations

where individuals are genetically diverse, both size and shape-based measures of FA are

often correlated to some degree (as they are here too). However, size and shape are

regulated somewhat differently in that cell proliferation is mediated both extrinsically via

cyclin E acting at the G1/S checkpoint of the cell cycle, predominantly affecting size, and

intrinsically at via cdc25/string at the G2/M checkpoint, predominantly affecting pattern

or shape (Day and Lawrence 2000). Extrinsic mechanisms regulate size through the

insulin pathway and its associated hormones (Nijhout 2003) providing a link between

size and the nutritional environment. This may explain why size FA follows

development time more closely than shape FA in this study.

Size is usually less canalized than is shape and therefore more variable. While this

makes the size-based measure of FA generally attractive, it has been found in previous

work (Babbitt et al. 2006) that population averages of size-based FA are often so variable

that they are frequently under-sampled because of the broad and often long tailed

distribution of FA. Shape-based FA does not suffer as much from this problem and so its

estimation is much better although it may be less indicative of environmental influences









and perhaps therefore better suited for genetic studies of FA. Because size FA has been

adequately sampled in this study and because size is more heavily influenced by

environmental factors like temperature, I find that size-based FA is the more interesting

measure of developmental instability in this study.

In general, it is clear that developmental noise is not constant in genetically

identical individuals cultured under near similar environments. The overall distribution

pattern in both size and shape FA is log skewed even within temperature classes. This

suggests that sampling error due solely to random noise can play a very significant role in

FA studies. Furthermore, it appears that the response of FA to the environment is

potentially quite strong. This suggests that FA may indeed be a responsive bioindicator,

however its response will not be very generalizable in environments with fluctuation in

temperature.

In conclusion, the environmental response of developmental noise to temperature

in absence of genetic variability and nutrient limitation supports an important prediction

of theoretical explanations of the proximate basis of FA. This prediction is that FA

should decrease with growth rate (and temperature) in ectothermic organisms. Because it

is only in sexually selected traits that the opposite prediction should hold, that FA should

increase and not decrease with growth rate, the results presented here may offer a useful

method for discriminating between FA that is under sexual selection and FA that is not.














CHAPTER 5
CONCLUDING REMARKS AND RECOMMENDATIONS

In this dissertation, I confirm some existing hypotheses concerning FA and provide

a new explanatory framework that explains alteration in the distribution of fluctuating

asymmetry (FA) and its subsequent effect upon mean FA. The basis of FA and the

influences that shape its response to genes and the environment in populations can be

suitably modeled by stochastic proportional growth in expanding populations of cells on

both sides of the body that are terminated with a small degree of genetically-based

random error. I model stochastic growth with geometric Brownian motion, a random

walk on a log scale. And I also model error in terminating growth with a normal

distribution. The resulting distribution of FA is a lognormal distribution characterized by

power-law scaled tails. Because under a power-law distribution, increased sample size

increases the chance of sampling rare events under these long tails, convergence to the

mean is potentially slowed or even stopped depending on the scaling exponents of the

power-law describing the tails. I demonstrated that the effect of reduced convergence to

the mean is substantial and has probably caused much of the previous work on FA to be

under-sampled (Chapter 1). I have also demonstrated that the scaling effect in the

distribution of FA is directly related to genetic variation in the population. Therefore a

low scaling exponent and also high kurtosis is associated with a large degree of genetic

variation between individuals in their ability to precisely terminate the growth process

(Chapter 2). I also demonstrate that when genetic differences do not exist (where FA is

comprised of only developmental noise) and when development is not limited









energetically, such as in a parthenogenetic aphid clone, FA depends upon developmental

time (Chapter 3).

How Many Samples Are Enough?

The answer to this clearly depends upon whether researchers chose to study FA in

size or FA in shape. The required sample size depends upon the distributional type and

parameters of the distribution of FA. As previously discussed, FA distributions differ

depending upon whether FA is based on size or shape differences between sides. As

demonstrated in chapter 1, multivariate measures of FA based upon shape, such as

Procrustes Distances, tend to have distributional parameters that allow convergence to the

mean that is about five times more rapid than either univariate or multivariate measures

of FA based upon size (Euclidian distance or centroid size). While this information

might seem to favor a shape-based approach, there are some other very important

considerations given below. In the end, sample size must be independently evaluated in

each study depending on the magnitude of asymmetry that one wishes to detect between

treatments or populations. As a general rule, if shape FA is used, 100-200 samples may

suffice, but if size FA is used then many hundreds or even a thousand samples may be

required to detect a similar magnitude difference.

What Measure of FA is best?

The handicap of sampling requirements of size-based FA aside, it appears that

measures of the distributional shape of FA, like kurtosis, scaling exponent in the upper

tail and skewness are strongly related to similar but smaller changes in mean size FA

(chapter 2). Because size FA is most commonly used in past studies of FA, these

measures of distribution shape might allow clearer conclusions to be drawn from

literature reviews and meta-analyses as well as individual studies of FA where the









collection of more data is not possible. My work linking shape of the FA distribution to

the genetic variability of the population will require that future studies of FA which are

focused on the environmental component of FA be controlled for genetic differences

between populations. Where FA is sought as a potential bioindicator of environmental

stress, the potentially differing genetic structure of populations will need to be considered

in order to make meaningful conclusions about levels of FA. Additionally, because in the

absence of genetic variation in monoclonal aphids, size FA is directly related to the total

developmental times of individuals (chapter 3), it may a better choice than shape FA for

studies of environmental influences on FA.

In most organisms, body size is less canalized than body shape. Body size is highly

polygenic and depends upon many environmentally linked character traits. However

body shape or patterning is determined by a sequential progression of the activity of far

fewer genes. Body size is also regulated at a different checkpoint during the cell cycle

than is body pattern formation or shape (Day and Lawrence 2000). Patterning or shape is

regulated at G2/M checkpoint while size is regulated at G1/S checkpoint. The latter

checkpoint is associated with the insulin pathway, linking size to nutrition and hence to

the environment. Because of this the study of environmental responses of FA may

ultimately be best served by using multivariate size-based measures of FA (i.e., centroid

size FA) while at the same time making sure to collect enough samples to accurately

estimate the mean.

In conclusion, multivariate size FA may be the better metric for large studies

interested in the effects of environmental factors upon FA. Where data is harder to come

by (e.g., vertebrate field studies), multivariate shape FA may perform better with the









caveat that shape and pattern are less likely to vary in a population than does size.

Because centroid size must be calculated in order to derive Procrustes distance, both

measures can easily and should be examined together.

Does rapid growth stabilize or destabilize development?

The effects of temperature and growth rate on FA appear to support the proximate

model which predicts that FA declines with increasing growth rates (Figure 4-3). Aphids

from the Gainesville clone decrease mean FA in response to increased temperature and

growth rate. It is likely that the Lake Alfred clone is under-sampled and therefore

estimates of mean FA are not accurate in that sample. The results of the Gainesville

clone, which was sampled adequately, suggests that the prediction of proximate models

of the basis ofFA (Emlen et al. 1993, Graham et al. 1993), that rapid growth should

decrease FA holds true. Moller's hypothesis that rapid growth is stressful because of

incurred energetic costs does not appear to hold in the case of aphids. Of course aphids,

being phloem feeders, generally have access to more water and carbon than they ever

need. This is evidenced by analysis of honeydew composition in this and many other

aphid species. Growth and reproduction in aphids is probably more limited by nitrogen

than by water or energy (carbon). So it would seem that this system may not be a very

good one in which to assess Moller's hypothesis directly. It should be further

investigated whether FA of sexually selected traits in energetically limited, genetically

diverse populations is positively associated with growth rate as is predicted by the

ultimate or evolutionary model presented in Chapter 4.

Can fluctuating asymmetry be a sexually selected trait?

Given that a genetically altered population (chapter 3) demonstrates a consistent

response in the shape of the FA distribution does suggest that a potentially strong genetic









basis for FA exists, but is not easily observable through levels of mean FA. This implies

some heritability exists and possibly opens the door for selection to act upon FA in the

context of mate choice. However, in the future it must be demonstrated, with appropriate

sample sizes, that the level of FA in the laboratory can be altered by selection for

increased or decreased FA. The opposing predictions of the proximate or mechanistic

and ultimate (evolution through sexual selection) explanations of FA, regarding the

relationship of temperature, growth rate and FA may offer a method for detecting sexual

selection upon FA (chapter 4). If FA has evolved as an indicator of good genes, and

hence has related energetic costs, the expectation that FA should increase with growth

rate and temperature (in ectotherms) is plausible. Otherwise, ifFA is caused by random

accumulation of error during development, then the expectation is reversed. FA should

decrease with growth rate and temperature as I have demonstrated in a clonal population

of aphids that are not energetically limited (chapter 4).

Is fluctuating asymmetry a valuable environmental bioindicator?

Given that the genetic structure of a population (chapter 3) has potentially strong

effects upon the shape and location (mean) of the distribution of FA, the usefulness of FA

as a bioindicator of environmental stress has to be questioned. Average levels of FA of

populations would be difficult to interpret or compare without additional information

regarding genetic heterogeneity. However, provided that the study of FA and population

genetics were undertaken simultaneously, this problem could be avoided. So FA could

still be of use in this regard, however its expense and ease of use compared to other

bioindicators would need to be reassessed.









Scaling Effects in Statistical Distributions: The Bigger Picture

This dissertation demonstrates that underlying distributions of some biological data

can contain partial self-similarity or power-law scaling. The proper model for describing

data such as this sits at a midpoint between those models used in classical statistics and

those of statistical physics: Levy statistics (Bardou et al. 2003). To my knowledge, this

work represents the first application of such a model in biology.

The sources of power-law scaling in the natural sciences are diverse. Sornette

(2003) outlines 14 different ways that power-laws can be created, some of which are very

simple. The hypothesis that all power-law scaling in nature is due to a single phenomena

such as self-organized criticality (SOC) (Bak 1996, Gisiger 2000) or highly optimized

tolerance (HOT) (Newman 2000) is unlikely. Reed (2001) suggests that the model I have

used here, stochastic proportional growth that is observed randomly, may explain a great

deal of power-law scaled size distributions formerly speculated to have a single cause

like SOC or HOT. These phenomena include distributions of city size (Zipf's law),

personal income (Pareto's Law), sand particle size, species per genus in flowering plants,

frequencies of words in sequences of text, sizes of areas burnt in forest fires, and species

body sizes, just to name a few examples.

One of the most common ways that power-laws can be obtained is by combining

exponential functions. This is effect is also observed when positive exponential/

geometric/proportional growth is observed with a likelihood described by to a negative

exponential function such as in Reed and Jorgensen's (2004) model for generating size

distributions. Power laws generated by combing exponential functions are also present in

the statistical mechanics of highly interactive systems (e.g., SOC) where the correlation

in behavior between two nodes or objects in an interaction network or lattice decreases









exponentially with the distance between them while the number of potential interaction

paths increases exponentially with the distance between them (Stanley 1995). In

Laplacian fractals, or diffusion limited aggregations observed in chemical

electrodeposition and bacterial colony growth (Viscek 2001) there may be a similar

interplay between exponential growth (doubling) and the diffusion of nutrients supporting


growth which are governed by the normal distribution, of the form y = e

The exponential function holds a special place in the natural sciences. Any

frequency dependent rate of change in nature, or in other words, any rate of change of

something that is dependent on the proportion of that something present at that time, is

described by the exponential function. Therefore it finds application to many natural

phenomena including behavior of populations, chemical reactions, radioactive decay, and

diffusion just to name a few. It seems only fitting that many of the power-laws we

observe in the natural sciences probably owe their existence to the interplay of

exponential functions, one of the most common mathematical relationships observed in

nature.

As far as we can ascertain from the recording of ancient civilizations, human beings

have been using numbers for at least 5000 years if not longer. And yet the concepts of

probability are a relatively recent human invention. The idea first appears in 1545 in the

writings of Girolamo Cardano and is later adopted by the mathematicians, Galileo,

Fermat, Pascal, Huygens, Bernoulli and de Moivre in discussions of gambling over the

next several hundred years. The concepts of odds and of random chance are not

generalized until the early twentieth century by Andreyevich Markov and not formalized

into mathematics until the work of Andrei Kolmogorov in 1946.









The role of uncertainty in nature is yet to be resolved. Does uncertainty lie only

with us or does it underlie the very fabric of the cosmos? Most 19th century scientists

(excepting perhaps Darwin) believed that the universe was governed by deterministic

laws and that uncertainty is solely due to human error. The first application of statistical

distributions by Carl Frederick Gauss and Pierre Simon Laplace were concerned only

with the problem of accounting for measurement error in astronomical calculations. The

more recent view, that uncertainty is something real, was largely the work of early

twentieth century quantum theorists Werner Heisenberg and Erwin Schrodinger and the

statistical theory embodied in the work of Sir Ronald Fisher. The basic conceptual

revolution in modem physics was that observations cannot be made at an atomic level

without some disturbance, therefore while one might observe something exactly in time,

one cannot predict how the act of observing will affect the observed in the future with

any degree of certainty (at least at very small scales). This is Heisenberg's Uncertainty or

Indeterminancy Principle.

Scientists observe natural "laws" only through emergent properties of many atoms

observed at vastly larger scales where individual behavior is averaged into a collective

whole. Jakob Bernoulli's Law of Large Numbers, Abraham de Moivre's bell shaped

curve, and Sir Ronald Fisher's estimation of the mean are also examples of how scientists

rely upon emergent properties of large systems of randomly behaving things in order to

make sense of what is thought to be fundamentally uncertain world. However, this

probabilistic view of the natural world has been challenged in recent decades by some

mathematicians who are again championing a deterministic view of nature. Observations

by Edward Lorenz, Benoit Mandelbrot and others, of simple and purely deterministic









equations that behave in complex unpredictable ways, has led to a revival in the

deterministic view among mathematicians; in this case, uncertainty is a product of human

inability to perceive systems that are extremely sensitive to initial starting conditions.

For this reason, this new view is often called deterministicc chaos". And so the question

as to whether uncertainty is real or imagined is yet to be resolved by modern science.

One remarkable observation of this latest revolution in mathematics is the frequent

occurrence of scale invariance or power-law scaling in systems that exhibit this sort of

complex and unpredictable behavior. And so just as the normal distribution or bell curve

is an emergent property governing the random behavior of independent objects, the

power-law appears to be an emergent property of nature as well; one that seems not only

to often to appear in the behavior of interacting objects (i.e., critical systems) but in

systems where growth in randomly observed as well. Statistical physics now recognizes

two classes of "stable laws", one that leads to the Gaussian or normal distribution and

another that is Levy or power-law distributed. In the former class, emergent behavior is

governed by commonly occurring random events while in the latter class the behavior of

the group is governed by a few rarely occurring random events. As I have already

reviewed earlier, we find that convergence to the mean under these two frameworks can

be radically different. Yet both are present in the natural world, and as I have shown in

my work, both of these classes of behavior can underlie naturally occurring distributions,

causing partial scale invariance in real data. It is my conviction that the biological

sciences must in the future adopt the statistical methods of working under both of these

frameworks and not simply make assumptions of normality whenever and wherever

random events are found to occur.














APPENDIX A
LANDMARK WING VEIN INTERSECTIONS CHOSEN FOR ANALYSISOF
FLUCTUATING ASYMMETRY


Figure Al. Six landmark locations digitized for Aphis gossipyii


Figure A2. Six landmark locations digitized for Apis mellifera






85















Figure A3. Six landmark locations digitized for Chrysosoma crinitus


Figure A4. Eight landmark locations digitized for Drosophila simulans













APPENDIX B
USEFUL MATHEMATICAL FUNCTIONS

The following is a list of the probability density functions for candidate models of
the distribution of fluctuating asymmetry. Log likelihood forms of these functions were
maximized to obtain best fitting parameters of each model for our data.
Asymmetric Laplace Distribution

(see Kotz et al. 2001)

f(x) = ( 1)e eo for x > 0

f(x) = ((1 )) for x < 0

where 0 = location, c = scale and K = skew index (Laplace when K = 1)
Half-Normal Distribution

f(x) = ( )( )e'

where [= minimum data value and 0 = dispersion
Lognormal Distribution

(see Evans 2000 or Limpert et al. 2001)

f(x) = 1 e( (logx v) 2-2)
xr(2r)1/2
where v = location and T = shape or multiplicative standard deviation
Double Pareto Lognormal Distribution

(see Reed and Jorgensen 2004)

f(x) = -- g(log(x))
given the normal-Laplace distribution

g(y) = -(At) [R(a -Y)+ R(P. ) + )]
where the Mill s ratio R(z) is









1- O(z)
R(z) =

and where D is the cumulative density function and q is probability density function for

standard normal distribution N(0,1), where a and 0 are parameters that control power-law

scaling in the tails of the lognormal distribution.

The limiting forms of the double Pareto lognormal are the left Pareto lognormal

(a = o ), right Pareto lognormal distributions (f = oo ), and lognormal distributions

(a = o, f = o) with Pareto tails on only the left side, only the right side, or on neither

side, respectively.

A description of Reed and Jorgensen's generative model of double Pareto lognormal size
distribution.

Reed and Jorgenson's (2004) generative model begins with the Ito stochastic

differential equation representing a geometric Brownian motion given below.

dX = uXdt + crXdw

with initial state X(0) = XO distributed lognormally, log XO N(v,'2). After Ttime units

the state X(T) is also distributed lognormally with log X(T) N(v + (La-C2/2)T,T2 + C2T).

The time T, at which the process is observed, is distributed with densityfT(t) = Xe-Xt

where X is a constant rate. The double Pareto lognormal distribution is generated when

geometric Brownian motion is sampled repeatedly at time t with a negative exponential

probability.