Juan Sebastian Chahda

Home , Hans Spemann, Ontogeny

ANALYSIS OF SCALING PROPERTIES OF EMBRYONIC

MORPHOGEN GRADIENTS DURING DROSOPHILA

EVOLUTION

JUAN SEBASTIAN CHAHDA

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Biology Department

CASE WESTERN RESERVE UNIVERSTIY

August, 2015

CASE WESTERN RESERVE UNIVERSITY

SCHOOL OF GRADUATE STUDIES

We hereby approve the thesis/dissertation of

JUAN SEBASTIAN CHAHDA

candidate for the degree of Doctor of Philosophy *.

Committee Chair

Michael Bernard

Committee Member

Claudia Mieko Mizutani

Committee Member

Peter Harte

Committee Member

Jocelyn McDonald

Committee Member

Brian McDermott

Date of Defense

July 2nd 2015

*We also certify that written approval has been obtained

for any proprietary material contained therein.

TABLE OF CONTENTS

Table of Contents ………...…………………………………………………...… 1

List of Tables ...…… ……………………………………………..…………...… 4

List of Figures ….… .……………………………………………..…………...… 5

Acknowledgements …… ...……………………..………………..…………...… 6

List of Abbreviations … .…………………..………………..………...……...… 7

Abstract ………………… …..………………..…………………...…...……...… 8

Chapter I: Background and Significance …………………………………….. 10

1.1 Morphogenetic gradients specify cell fate in a concentration-dependent manner ……………………………………………………………………….. 10 1.2 The Dorsal/NF-κB gradient initially subdivides the Drosophila dorsal-ventral axis into mesoderm, neuroectoderm and ectoderm …………………….……. 16 1.3 The Dorsal/NF-κB and Dpp/BMP-4 morphogen gradients work together to pattern the neuroectoderm …………………………………………..….…… 21 1.4 The neuroblast map is heavily conserved across arthropods despite great differences in embryo size …………………………………………… ……. 24 1.5 Investigating the scaling of morphogenetic gradients ………… ...... … 27 1.6 Dorso-ventral scaling of germ layers in related Drosophila species ………..31 1.7 Mathematical modeling as a tool to test predictions of Dl gradient scaling during evolution …………………...… ...... …….. 34 1.8 Research aims……………………………………………………….… …… 36

Chapter 2: Variation in the Dorsal Gradient Distribution is a Source for Modified Scaling of Germ Layers in Drosophila ……………………… ……. 37

Published in Chahda JS, Sousa-Neves R, Mizutani CM. Curr Biol. 2013 Apr 22;23(8):710-6. doi: 10.1016/j.cub.2013.03.031

2.1 Abstract ………………………………………………… ………………….. 38 2.2 Results and Discussion ……………………………..…… ………………… 38 2.2.1 Cross-species comparison of nuclear Dl protein levels reveals gradients of different shapes …………………………...………… …………………. 40 2.2.2 Mesodermal expansion does not rely on altered sna or twi sensitivity to Dl levels in sibling species ……………………………… ……………….... 41 2.2.3 Different gradients in the same scale of threshold levels reveal unique properties of scaling ………………………………… …………………. 42

2.2.4 The Dl gradient shape in D. melanogaster is sensitive to changes in nuclear size and packing …………………………… …………..……… 43 2.2.5 Fast evolution of the Dl gradient and maintenance of the neuroectoderm …………………………………………………………………………... 46 2.3 Materials and Methods ………………………………… …………………... 48 2.3.1 Fly stocks and genetic crosses ……………………… ……………….… 48 2.3.2 Measurements of egg size, nuclear number, size, and packing densities ………………………………………………………………………...… 49 2.3.3 Immunohistochemistry …………………………… ………………….... 50 2.3.4 Quantification of the Dl gradient ………………… ……………………. 52 2.4 Supplemental Information …………………………… ...... …… 52 2.4.1 Normalization of Dl gradient allows cross-species comparison …… ..… 52 2.4.2 Scoring of hybrid crosses and embryos with altered ploidy ………… … 53 2.4.3 Measurements of egg size, nuclear numbers, size, and packing densities …………………………………………………………………………... 54 2.4.4 Method for Dl gradient graph transformation based on data from hybrids ………………………………………………………………...………… 55 2.5 Tables and Figures ………………………………………………… .……… 57 2.6 Conclusions and future directions ………………………………… ……….. 71 2.6.1 Changes to the Dl Gradient Directly Alter Mesoderm Specification … .. 71 2.6.2 Conservation of the Neuroectoderm is Unaffected by Changes to the Dl Gradient ……………………………………………………………….... 72 Chapter 3: Modeling of the Dorsal Gradient across Species Reveals Interaction between Embryo Morphology and Toll Signaling Pathway during Evolution ………………………………………………………… ….... 74

Published in Ambrosi P*, Chahda JS*, Koslen HR, Chiel HJ, Mizutani CM. PLoS Comput Biol. 2014 Aug 28;10(8):e1003807. doi: 0.1371/journal.pcbi.1003807. * First co-authors

3.1 Abstract ……………………………………………………………… …….. 75 3.2 Introduction ………………………………………………………… ……… 76 3.3 Results ……………………………………………………………… …….... 80 3.3.1 Reconstruction of the Kanodia model reproduces the Dl gradient shape in some mutant conditions ………………………………………… ...…… 80 3.3.2 Simulation of nuclei numbers and size can reproduce ssm gradient, but not the gyn gradient ……………………………………… ………………… 82 3.3.3 Refinement of the parameter values reveals the embryo geometry plus Dl diffusion and export rates play major roles in the model reproduction of the gyn gradient ……………………………………… ………………… 85

3.3.4 Embryonic Morphology alone does not fully explain species-specific Dl gradient shapes ……………………………………… ………………..... 88 3.3.5 Modulating a small subset of parameters affecting the Toll signaling pathway can reproduce species-specific Dl gradients … ………………. 89 3.3.6 Analyses of another pair of closely related sibling species suggest evolutionarily shared mechanisms for Dl gradient formation … ….…… 91 3.3.7 Dl and Cact protein sequence comparisons of melanogaster subgroup species support predictions made by the model ……………… ……..…. 93 3.3.8 Model robustness and sensitivity analysis reveals non-linear interaction between species morphology modifications and other relevant parameters ………………………………………………………………………..…. 95 3.4 Discussion ………………………………………………………… …..…… 96 3.4.1 Dorsal scaling within and across species ……………………… ….…… 96 3.4.2 Model sensitivity analysis support evolution of Dl gradient by small additive changes in Tl regulation pathway …………………… ……….. 99 3.4.3 The Dl gradient model predicts changes in the Tl pathway in Drosophila species that are consistent to their phylogenetic relationships ……...….. 99 3.5 Material and Methods ……………………………………………………… 101 3.5.1 Fly stocks ……………………………………………………………..... 101 3.5.2 Gradient quantification and measurements of nuclear size …………..... 102 3.5.3 Reproduction and modification of the Kanodia model in Mathematica ..102 3.5.4 Dimensionalized model of the last nuclear cycle ……………………… 103 3.5.5 Model validation ………………………………………………………. 103 3.5.6 Fit calculation and confidence intervals …………………………..…… 104 3.5.7 Sequence comparison of the Dl and Cact in melanogaster subgroup species ………..………………………………………………………… 104 3.6 Acknowledgements ………………………………………………………… 105 3.7 Tables and Figures ………………………………………………………..... 106 3.8 Conclusions and future directions …………………………...……… ……. 118 3.8.1 The Dl import, export and diffusion rates are greater than previously thought ………………………………………………….……… …….. 118 3.8.2 Evolution within the Toll signaling pathway can account for species- specific Dl gradient distributions ……………………………… …..…. 119 Chapter 4: Discussion ………… ...... ………………...… 121

4.1 Dorsal gradient amplitude and distribution can explain species-specific tissue allocation along the dorsal-ventral axis ……………… ..…………….…… 121 4.2 Morphological traits and fast evolution in Toll signaling pathway reshape the Dorsal gradient ……………………...………………… ..………………… 124

Bibliography …………………………………………………… …………..... 129

LIST OF TABLES

Table 1: Cross-Species Comparison of Embryo Size and DV Nuclei … …….… 57 Table 2: DV Nuclei in D. melanogaster WT, Haploid, and Triploid Mutants .....58 Table 3: Parameter values used in model simulations for D. melanogaster wild type and mutant conditions shown in Figures 10 and 11 ……………………... .106 Table 4: Selected parameter sets used in model simulations of Dl gradient for three different Drosophila species, D. busckii, D. simulans and D. sechellia ....107 Table 5: Selected parameter sets used in model simulations of Dl gradient for Drosophila yakuba and Drosophila santomea ……………………… ...... 108

LIST OF FIGURES

Figure 1: The French Flag Model for Positional Specification in the Embryo .....12 Figure 2: The morphogenetic Dl gradient activates target genes that cross regulate each other to specify the DV axis ...... 18 Figure 3: The Bicoid gradient scales along the AP axis and establishes a proportional segmentation pattern in embryos of different sizes ...... 30 Figure 4: Divergence in embryo size in related Drosophila species ...... 33 Figure 5: The Distributions of Nuclei, Mesodermal Domains, and Dl Levels Changed across Species ...... 59 Figure 6: Sensitivity of Mesodermal Gene Activation Is Identical among D. melanogaster Sibling Species ...... 61 Figure 7: Drosophila Species Vary in Nuclei Size and Densities ...... 62 Figure 8: The Dl Gradient Is Modified by Nuclear Size and Packing Density .....64 Figure 9: Dl gradient model rationale ...... 104 Figure 10: The Dl gradient is modulated by changes in nuclear size and density110 Figure 11: Comparison between experimental data and model output...... 111 Figure 12: Changes in kDeg allow the reproduction of the Dl gradient in embryos derived from dl+/dl- mothers ...... 112 Figure 13: Simulations of wild type and mutant gradients suggest increased diffusion and Dl export rates ...... 113 Figure 14: The model predicts similar adjustment in parameters consistent to the phylogenetic relationship of species ...... 114 Figure 15: Simulations of Dl gradient in an additional pair of closely related species with varying egg size suggest a shared Dl-Cact binding rates or Cact degradation rates ...... 115 Figure 16: Sensitivity analysis for main parameters tested ...... 116 Figure 17: Comparison of the Dorsal and Cactus proteins across species supports model findings ...... 117 Figure 18: Summary of Chapter 2 findings: differences in morphological traits, relative range of peak toll signaling and DV tissue specification across Drosophila species ...... 123 Figure 19: Successful Parameter sets that include changes to Toll signaling pathway are in agreement with phylogenetic relationships ...... 127 Figure S1: Dorsal gradient normalization and data from individual embryos ...... 65 Figure S2: The subdomains of the neuroectoderm are conserved across species ..67 Figure S3: Quantification of Differences in Nuclear Packing Density across Species ...... 69 Figure S4: Genotyping of gyn Triploids by Counting the Number of sna Nuclear Transcripts...... 70

ACKNOWLEDGEMENTS

First and foremost, I would like to thank my advisor/mentor Dr. Claudia Mizutani for supporting me throughout these last five years, pushing me to be ambitious in my scientific endeavors, and spending countless hours correcting my writings and presentations; Dr. Rui Sousa-Neves, who I had the fortune to collaborate with and was never short on advice or good stories; my current and former lab mates who provided constructive feedback and kept me company in the lab; Priscilla Ambrosi and Dr. Hillel Chiel for their tremendous work on the mathematical modeling of the Dorsal gradient; Hannah Koslen for her work on Dl gradient quantification; Julia Brown and the rest of the Biology department office for helping me navigate the Case Western system; and my family for their unconditional support. I would also like to thank my thesis committee members: Dr. Jocelyn McDonald, Dr. Brian McDermott, and Dr. Peter Harte. Finally, I would like to acknowledge the funding that made this work possible: National Science Foundation grant number IOS-1051662 to Claudia M. Mizutani., and the GAANN fellowship from the US Department of Education (grant number P200A090191).

LIST OF ABBREVIATIONS

AP Antero-posterior Bcd Bicoid BMP-4- Bone Morphogenetic Protein-4- Cact Cactus CRMs cis-regulatory modules Dl Dorsal Dbusc Drosophila busckii Dmel Drosophila melanogaster Dsan Drosophila santomea Dsech Drosophila sechellia Dsim Drosophila simulans Dyak Drosophila yakuba Dpp Decapentaplegic DV Dorso-ventral FISH Fluorescent in situ Hybridization gd gastrulation defective GFP Green Fluorescent Protein gyn-2; gyn-3 or gyn gynogenetic-2; gynogenetic-3 Г Diffusion rate of Dl and Dl complexes Ind intermediate neuroblast defective Kb Association rate of Dorsal to Cactus kD Dissociation rate of Dorsal-Cactus complex kDeg Degradation rate of Cactus ke Dorsal nuclear export rate ki Dorsal nuclear import rate min Minutes Msh Drop/Muscle segment homeobox MYA Million years ago NES Nuclear export signal PCact Cactus production rate R A parameter of space-dependence of kD Ssm Sesame Sna Snail Sog Short gastrulation Spz Spätzle S A paramenter of space-dependence of kD Tkv Thickveins Tl Toll Twi Twist µm micrometer Vkg Viking Vnd Ventral nervous system defective Wt Wild type ξ A parameter of space-dependence of kD

Analysis of Scaling Properties of Embryonic Morphogen Gradients during Drosophila Evolution

Abstract

JUAN SEBASTIAN CHAHDA

The Drosophila blastoderm is an excellent model to determine the

molecular mechanisms that allow tissues to acquire new proportions in different

species. The embryonic dorso-ventral (DV) axis is patterned by two opposing

morphogen gradients formed by the transcription factor Dorsal (Dl)/NF-κB and the secreted molecule Decapentaplegic (Dpp)/Bone Morphogenetic Protein-4.

These gradients establish the mesodermal, neuroectodermal and ectodermal

domains. Previous work showed that insects that diverged over 300 MYA share a

conserved neuroectoderm with similar number and identity of neuroblasts, despite

significant changes in embryonic size. In contrast, the ventral mesoderm size

displays significant variations within sister Drosophila species that diverged 0.5

MYA. Here we investigated the mechanisms that allow the Dl gradient to modify

the mesoderm but conserve the neuroectoderm in Drosophilids. We asked if the

evolutionary variations of the mesoderm were caused by modifications in the Dl

gradient shape or cis-regulatory sequences of Dl-target genes. To that end, we

quantified Dl levels along the DV axis and found that the Dl gradient acquired

new shapes in these species. To investigate the Dl activation thresholds of

mesodermal genes, we created hybrid embryos between Drosophila sibling

8 species and showed that both copies of these genes from each species are activated by same Dl levels. We concluded that the evolutionary variation in the mesoderm is caused by Dl gradient changes, which repositions the borders of the lateral neuroectoderm, where Dl levels are conserved. Using a mathematical model, we investigated the mechanism behind Dl gradient modification. We show that Toll signaling pathway and divergent morphological traits interact non- linearly to create new Dl gradient distributions. We propose the neuroectoderm evolutionary conservation is achieved by robust patterning systems that reposition its borders as a single block, preserving the number of cells giving rise to a stereotyped ventral nerve cord.

Chapter I: Background and Significance

1.1 Morphogenetic gradients specify cell fate in a concentration-dependent

manner

Over the years, the field of developmental biology has relied on both experimental manipulations and theoretical models aimed to understand how cells acquire fates, organize tissues, and ultimately form a complex adult organism from an undifferentiated single cell, the oocyte. In the early 50s, the mathematician Alan Turing proposed a reaction-diffusion model in which the generation of patterns, such as repeating stripes or spots, could be generated from an initially uniform field in response to gradients of “form producer” chemical substances he called morphogens (Turing 1952). Inspired by this model, Lewis

Wolpert later used the French flag stripes as an analogy to describe the specification of cells in distinct fates (represented by the “blue”, “white” and

“red” colors) depending on their relative positions (Wolpert 1969). In the theoretical French Flag model, naïve and competent cells receive “positional information” that is conveyed by distinct concentration levels of a morphogen, and subsequently acquire distinct fates (Fig. 1).

Several years later, the existence of morphogens was validated for the first time in Drosophila with the discovery of the Bicoid (Bcd) morphogen gradient, which was shown to pattern the anterior-posterior (AP) axis of the embryo in a dosage-dependent fashion (Frohnhofer and Nusslein-volhard 1986; Driever and

Nüsslein-Volhard 1989). Shortly after the discovery that Bcd patterned the AP axis, another morphogen named Dorsal (Dl) was demonstrated to form a gradient

that specifies the coordinates along the dorsal-ventral (DV) axis (Roth, Stein, and

Nusslein-Volhard 1989). Together, the Bcd and Dl morphogen gradients establish the initial specification of both major axes of the Drosophila blastoderm, demonstrating that positional information represents a key mechanism to organize

the developing organism by creating cell diversity within distinct regions.

Figure 1: The French Flag Model. A space-dependent distribution of some chemical substance (the morphogen) provides positional information to equivalent, naïve cells that are competent to assume different fates. In the model, each cell is then able to read this morphogen concentration and compare it with a set of threshold concentrations in order to finally acquire a positional value dictated by the morphogen concentration at the cell's location. (Figure and legend recreated from Kerszberg and Wolpert 2007.)

The key molecular mechanism of action of morphogen gradients is the ability

to turn genes on and off in a dose-dependent fashion. The distinct gene regulatory

responses are achieved by stronger or weaker binding-affinities between the DNA cis-regulatory sequences of target genes and the morphogen itself, when it encodes a transcription factor as in the case of Bcd and Dl (Hanes and Brent

1989; Ip et al. 1991), or a downstream effector of the morphogen pathway

(reviewed in Ashe & Briscoe, 2006).

The logic governing differential gene regulation in response to morphogen gradients can be demonstrated by examining the mechanism by which the Bcd

gradient directs the expression of some of its initial downstream target genes that function in segmentation. These target genes were found in a zygotic screen in

1980 (Nusslein-Volhard and Wieschaus 1980) and include hunchback and

Krüppel, which are called as “gap genes” because their mutations cause loss of

several adjacent segments along the AP axis (Nusslein-Volhard and Wieschaus

1980). The most anteriorly expressing gene is hunchback, which specifies anterior

structures such as the head (Lehmann and Nüsslein-Volhard 1987; Tautz 1988).

Cells expressing hunchback experience the highest threshold levels of Bcd, while

more posterior cells that receive less Bcd signal do not express hunckback.

Consistent with a binding-affinity model, it was found that the enhancer of

hunchback contains low-affinity DNA binding sites for Bcd, which restricts its

expression to the anterior regions of the embryo only, where Bcd levels are the

highest (Driever and Nüsslein-Volhard 1989; Driever, Thoma, and Nusslein-

volhard 1989). In contrast, Krüppel is expressed in the mid-section of the embryo

(Knipple, Seifert, and Rosenberg 1985), where the concentration of Bcd is low

(Hoch, Seifert, and Jäckle 1991). Dissection of the cis-regulatory elements of the

Krüppel enhancer showed that it contains high-affinity binding sites for Bcd, consistent with the prediction that low levels of Bcd should be sufficient to drive its expression (Hoch et al. 1991).

The example above provides a simple molecular mechanism whereby different thresholds of a morphogen create distinct domains of spatial regulation.

However, the binding-affinity model alone is insufficient to explain the formation of precise borders between the gene expression domains they establish. As it will be explained later, additional cross-regulation between the morphogen target genes themselves is important to reinforce the borders initially created by a gradient. In the example above, Krüppel is not only activated by Bcd, but its enhancers also contains high affinity binding sites for Hunckback, which functions as a repressor in anterior regions and excludes Krüppel expression from this region (Hoch et al. 1991).

The earlier theoretical models of the 50’s and 60’s were fundamental to guide experimentation leading to the identification and understanding of morphogenetic gradients and their action during development. Remarkably, morphogenetic gradients and positional information were found to be fundamental principles of development not only in early embryonic development, but also in limb and organ formation; and not only in invertebrates such as Drosophila, but also in vertebrates including humans. With the advent of DNA cloning, a great advance was made in identifying target genes that respond to morphogens, which allowed

the investigation of their gene expression regulation, interactions and signaling pathways. Unexpectedly, the analysis of enhancer regulation soon revealed that even simple patterns of gene expression, such as the seven repeating AP stripes of even-skipped gene, can actually derive from a complex network of activating and repressive interactions that regulates each stripe independently (Small, Blair, and

Levine 1992), rather than having all stripes arising in direct response to the Bcd gradient. Due to an incomplete knowledge of these interacting networks, the more general theoretical models existing at the time fell out of use in developmental biology. It wasn’t until the early 2000’s that a more detailed knowledge of network interactions and biochemical relationships between signaling components became available and led to a resurgence of mathematical models that are now critical tools for understanding complex biological functions and making novel biologically relevant predictions in this field (Tomlin and Axelrod, 2007).

The present thesis addresses some of many unanswered questions regarding the establishment, dynamics, and modification of morphogen gradients during evolution, using a combination of experimentation and mathematical modeling.

We are mainly interested in understanding the mechanisms that can lead to changes in morphogen gradients to produce new gene expression patterns and tissue proportions during evolution. Here we take advantage of one of the best characterized patterning systems, the DV cell fate specification, as a starting point to address these questions. As it will be explained in more detail in the next section, the germ layers that give rise to the main tissue types (mesoderm, neuroectoderm and ectoderm) are established along the DV axis. One question

15 that only begun to be addressed more recently regards to the scaling properties of morphogens. This question is relevant to understand the formation of proportional limbs during growth and regeneration, and the adaptation of conserved morphogenetic gradients to evolution in embryo size in related species. The present work utilizes the Drosophila DV patterning system to investigate how embryonic morphogen gradient modification impacts tissue specification in species that produce embryos of similar and divergent sizes.

1.2 The Dorsal/NF-κB gradient initially subdivides the Drosophila dorsal-

ventral axis into mesoderm, neuroectoderm and ectoderm

To understand the specification of cell fates along the DV axis in Drosophila it is necessary to highlight some of the first steps of development, the biochemical reactions that lead to Dl gradient formation, and the key features of the DV genetic-regulatory network that control subdivision of the embryo.

Upon fertilization and pronuclear fusion, the Drosophila zygote undergoes 13 rapid, syncytial nuclear divisions. Nuclear divisions 1-6 occur in the interior of the egg. By nuclear cycle 7, the nuclei begin to migrate towards the periphery while continuing to divide. Finally, after nuclear cycle 13 the majority of nuclei have reached the periphery or cortical layer of the embryo. At cell cycle 14, also known as the blastoderm stage, the syncytial embryo is comprised of some 6,000 nuclei that begin the process of cellularizing into discrete cells (Edgar, Kiehle, and Schubiger 1986; Foe and Alberts 1983) .

Positional information and cell fate acquisition of blastoderm nuclei is initially

informed by maternally established morphogen gradients. As mentioned before,

nuclei receive positional information along the DV axis by a nuclear gradient of

the transcription factor Dl (Fig 2A). During oocyte development, Dl is maternally deposited as RNA in a uniform distribution throughout the cytoplasm of the egg

(Roth et al. 1989; Steward et al. 1988). Examination of pre-blastoderm Dl RNA levels using in situ hybridization revealed that Dl translation begins around nuclear cycle 10 (Steward et al. 1988). After translation of Dl RNA, the Dl protein forms a nuclear gradient by a regulated transport from the cytoplasm to the nucleus in response to the activation of the Toll (Tl) transmembrane receptor

(Hashimoto, Hudson, and Anderson 1988). Tl receptor activation leads to the degradation of the IκB protein Cactus, which binds to Dl in the cytoplasm and inhibits its nuclear translocation (Morisalo and Anderson 1995). Tl receptors are present throughout the perivitelline space of the embryo, but only highly activated in ventral regions due to localized maternal processing of the ligand Spätzle (Spz)

(Morisato and Anderson 1994). Peak activation of Tl in ventral regions and decreasing levels of Tl activation towards dorsal regions creates a nuclear Dl gradient with high nuclear levels in the ventral region and decreasing levels towards more dorsally located nuclei (Fig 2A). Recent studies using live-imaging of embryos expressing Dl fused to Green Fluorescent Protein (Dl-GFP) have shown that the nuclear Dl protein gradient begins to form as early as nuclear cycle

11, soon after its protein translation (DeLotto et al. 2007).

Figure 2: The morphogenetic Dl gradient activates target genes that cross- regulate each other to specify the DV axis. (A) Whole embryo stained with an antibody against Dl (magenta) and a nuclear dye (blue) at the blastoderm stage. Note higher levels of nuclear Dl in ventral regions. Left is anterior; down is ventral. (B) Dl-target genes that mark DV axis specification are activated or repressed by different concentration levels of nuclear Dl. (C) Dl-target genes cross-regulate each other to set up sharp boundaries between gene expression domains. Snail is expressed in the mesoderm in response to high levels of nuclear Dl. Snail is a transcriptional repressor that excludes the expression of neuroectodermal genes from the mesoderm. The gene product of the Dl-target gene vnd represses the expression of ind to create sharp boundaries of expression between neural identity genes. (D) Double RNA FISH for neural identity genes vnd (green) and ind (red). Note the non-overlapping domains of expression of vnd and ind. (E) Triple RNA FISH for twist (green), snail (red) and short gastrulation (white). Note that twist and snail are co-expressed in the mesoderm and do not overlap with short gastrulation expression. Nuclei are counterstained with DAPI in blue in D and E. Ventral is down and anterior is left for D and E.

Dl regulates the expression of several zygotic genes along the DV axis

(reviewed in (Stathopoulos and Levine 2005). High levels of nuclear Dl in ventral

regions specify the mesoderm by activating the expression of twist and snail (Fig

2A,B). Low levels of nuclear Dl in lateral regions specify the neuroectoderm by

activating the expression of neural genes. In dorsal regions of the blastoderm,

where little or no nuclear Dl is present, Dl levels are insufficient to repress

decapentaplegic (dpp) expression (Stathopoulos and Levine 2005). Dpp marks

ectoderm specification and forms its own zygotic gradient, as it will be explained

later.

Dl-target genes are activated and repressed at different points along the Dl

gradient due to a differential affinity binding for Dl in their enhancer sequences

(Fig 2B) (Stathopoulos and Levine 2005). An example of differential enhancer

affinity can be seen by the activation of twist and short gastrulation (sog) at

different nuclear concentration levels of the Dl gradient. twist requires high levels

of Dl to be activated due to its low-affinity Dl binding sites, and therefore it is

expressed in ventral regions where nuclear Dl concentrations are highest (Jiang

and Levine 1993). In contrast, neuroectodermal genes, such as sog (François et al.

1994), contain enhancers with clusters of high-affinity Dl binding sites (Markstein

et al. 2002) that can respond to the moderate levels of Dl within lateral regions of

the embryo.

A key feature of the Drosophila DV patterning system is the cross-regulation

among Dl-target genes, which helps establish the zygotic expression domains and reinforces the borders between the domains. For instance, high levels of Dl work

in concert with Twist to fully expand snail expression to encompass the ventral

domain of twist expression (Ip et al. 1992; Kosman et al. 1991; Ray et al. 1991;

Stathopoulos and Levine 2002). A sharp boundary between the mesoderm and

neuroectoderm is reinforced by the repression of neuroectodermal genes such as

sog, rho and vnd, by Snail, which encodes a transcriptional repressor that

excludes the expression of neuroectodermal genes from the mesoderm (Fig. 2C,

E) (Cowden and Levine 2002; Kosman et al. 1991; McHale et al. 2011; Mellerick and Nirenberg 1995; Stathopoulos and Levine 2002).

Additional cross-regulatory interactions among DV genes also occur within the neuroectoderm, leading to non-overlapping domains of the homeobox neural identity genes ventral nervous system defective (vnd), intermediate neuroblast defective (ind) and Drop/muscle segment homeobox (msh) (Chu et al. 1998;

D’Alessio and Frasch 1996; Isshiki, Takeichi, and Nose 1997; Jiménez et al.

1995; McDonald et al. 1998; Nirenberg et al. 1995; Weiss et al. 1998). As it will be explained in the next section, these genes subdivide the neuroectoderm into three DV domains that give rise to neuroblasts of distinct identity. Two of these genes, namely vnd and ind are directly activated by Dl, while msh does not require Dl to be activated (reviewed in (Cornell and Von Ohlen 2000).

The cross-regulation among DV genes has been coined “ventral dominance”

(Fig. 2C) (Cowden and Levine 2003), whereby genes that are more ventrally

expressed function as repressors of genes that are more dorsally expressed. In line

with the “ventral dominance” model, the regulatory relationships described above

can be summarized as follows. Sna is a ventral repressor of lateral

20 neuroectodermal genes. Within the neuroectoderm, the non-overlapping domains of neural identity genes are also dependent on cross-regulatory relationship between these genes. Vnd represses the lateral gene ind, excluding ind expression from the ventral most domain where vnd is expressed (McDonald et al. 1998)

(Fig. 2C,D). In addition, both Vnd and Ind repress the expression of msh, which excludes msh expression from ventral and lateral regions, and restricts it to the dorsal most region (Weiss et al. 1998), resulting in the non-overlapping, ventral- to-dorsal expression of vnd, ind and msh.

1.3 The Dorsal/NF-κB and Dpp/BMP-4 morphogen gradients work together

to pattern the neuroectoderm

As described above, the Dl gradient initiates the separation of the main presumptive germ layers along the DV axis, and also provides input to subdivide the neuroectoderm. The separation of the neural versus non-neural ectoderm is later reinforced by the activities of the Dl-target genes, sog and dpp. As mentioned before, Dpp is a secreted molecule belonging to the TGF-β superfamily and forms its own zygotic gradient along the DV axis (Wharton et al.

1993). Dpp plays a dual role as an inhibitor of neural fate development and as an inducer of amnioserosal and epidermal fates in the dorsal region of the embryo

(Biehs, Francois, and Bier 1996).

A striking discovery made in the early 90’s was the conservation of the molecular mechanisms that lead to the separation of neural versus non-neural ectoderm between invertebrates and vertebrates, a process referred to as neural

induction (Holley et al. 1995). This discovery unifies the role of the Dpp signaling

pathway, and its homolog BMP-4, in DV axial patterning in all branches of bilaterians. This ancestral role of Dpp/BMP-4 signaling pathway stands in contrast to the derived role of Dl in DV patterning, which is unique to insects but

not conserved in vertebrates (Mizutani and Bier 2008).

The discovery of neural induction and the birth of experimental neuroembryology can be attributed to transplantation experiments performed by Hilde

Mangold in 1921 under the guidance of Hans Spemann (Hamburger 1988). In these experiments, Mangold transplanted the upper lip of a salamander gastrula, into the flank region of another salamander gastrula and found that a secondary neural plate formed around the overlying ectoderm surrounding the transplanted upper lip (Spemann and Mangold 1924). This region with special organizing properties was called the Spemann-Mangold Organizer. For many years, it was generally assumed that the molecules secreted from the organizer were “inducers” of neural fate. However, the mechanism of neural induction turned out to occur through the inhibition or blocking of the Dpp/BMP-4 neural inhibitor signal.

One key discovery was the identification of Noggin, a secreted molecule by

Spemann’s Organizer with neural inducing properties. Experiments in Xenopus obtained Noggin RNA transcribed from cDNA isolated from Spemann’s

Organizer, and verified that its overexpression by RNA injection caused neural induction (Lamb et al. 1993). Noggin was shown to be a secreted factor that antagonizes the BMP anti-neural signal in the extracellular domain (Zimmerman,

De Jesús-Escobar, and Harland 1996). Soon after the discovery of Noggin, the

secreted antagonists Follistatin and Chordin were isolated and also shown to

participate in neural induction (Hemmati-Brivanlou, Kelly, and Melton 1994;

Sasai et al. 1994). Chordin is homolog of Drosophila Sog, and acts as a BMP-4

antagonist. As mentioned before, in Drosophila, sog is expressed in the lateral region of the blastoderm embryo and protects it from the Dpp signal, allowing this region to develop as neuroectoderm (Biehs et al. 1996; Ferguson and Anderson

1992b; François et al. 1994).

The finding that expressing BMP antagonists were sufficient to induce neural development led to the “default model of neural induction”, in which the embryonic ectodermal cells adopt a neuronal fate unless they receive the anti- neural signal provided by Dpp/BMP signaling (reviewed by Muñoz-Sanjuán and

Brivanlou 2002 & De Robertis 2006).

After neural induction, the Dpp/BMP4 is thought to form a low level gradient within the neuroectoderm, which helps define the dorsal borders of the expression domains of the DV neural identity genes vnd, ind and msh via a threshold- dependent repression mechanism (Mizutani et al. 2006). This role of BMP signaling in DV neural patterning is also conserved in vertebrates, where several studies have demonstrated that a gradient of BMP-4 is crucial for the development of DV fates in the vertebrate neural tube (Barth et al. 1999; Lee and Jessell 1999;

Timmer, Wang, and Niswander 2002). As described above for Drosophila, the vertebrate homologs of vnd (Nkx2.2 and Nkx2.1) (Barth and Wilson 1995), ind

(Gsh1 and Gsh2) (Hsieh-Li et al. 1995; Valerius et al. 1995), and msh (Msx1-3)

(Ekker et al. 1997) are expressed in corresponding non-overlapping DV domains.

Whereas the Drosophila neuroectoderm receives input from the Dl gradient from ventral regions, the vertebrate neural tube deploys a gradient of the morphogen Sonic Hedgehog that emanates from the ventral notochord (Dessaud,

McMahon, and Briscoe 2008). In conclusion, the specification of neuroectodermal cell fates along the DV axis is highly conserved during evolution and requires inputs from two opposing gradients; in the case of Drosophila, the Dl and Dpp gradients provide such crucial positional information cues.

1.4 The neuroblast map is heavily conserved across arthropods despite great differences in embryo size

One key observation that argues for a strong degree of conservation of the neuroectoderm patterning of insects is the similarity of their “neuroblast maps”,

These anatomical maps were built in an effort to trace highly stereotyped development of individual neural lineages, and depict the number, position and identity of all neuroblasts within a hemisegment of the ventral nerve cord. The neuroblasts are neural-stem cell-like progenitors that undergo a series of asymmetric divisions (reviewed in Kohwi & Doe, 2013) and give rise to

stereotyped neural lineages, which have been mapped in great detail (Schmid,

Chiba, and Doe 1999).

Each neuroblast is formed from a cluster of neuroectodermal cells expressing

proneural genes of the achaete-scute gene complex, which are regulated by the expression of neural identity genes, including vnd, ind and msh described above

(for review, see Gómez-Skarmeta, Campuzano, and Modolell 2003). During a

24 process called lateral inhibition, only a single cell from each proneural cluster will adopt the neuroblast fate and will delaminate from the neuroectoderm, while the remaining cells are inhibited by Delta-Notch signaling and develop as epidermal cells (Campos-Ortega 1995; Kunisch 1994).

The DV neural identity genes vnd, ind and msh play a particularly important role in the specification of neuroblasts because they establish the main types of neural cells (e.g. serotonergic, secretory and motorneurons), which are formed in distinct regions along the DV axis. In experimental manipulations that expanded, retracted or eliminated the expression domain of the neural identity genes, it was observed a duplication of neural lineages of a given DV identity at the expense of neural lineages of another DV identity from an adjacent location (reviewed in

(Cornell and Von Ohlen 2000). Thus, it is expected that a strong selective pressure maintains the sizes of the expression domains of these genes. Indeed, the high degree of conservation of the neuroblast maps of divergent insects supports this idea. Given that these species have significant variations in embryo size, the conservation of their neuroblast maps implies that the early morphogenetic gradients that define the neural identity domains must be somehow adjusted to size.

The first neuroblast map was made in the grasshopper Locusta migratoria by

C. M. Bate in 1976, where he described periodic rows of 30 neuroblasts per thoracic hemisegment (Bate 1976). Strikingly, the Drosophila neuroblast map contains 31 neuroblasts per hemisegment (Chris Q Doe 1992). As expected from a conserved neuroblast map with cells of homologous identity, the axonal

connections formed at later stages are strikingly conserved between grasshopper

and Drosophila (Thomas et al. 1984). The neuroblast map of basally branching silverfish insect was also found to be conserved (Whitington 1996). Finally, in the

crustacean Orchestia cavimana, the number of neuroblasts per thoracic

hemisegment is 26-30, similar to grasshopper and fruit fly (Ungerer and Scholtz

2008). Through cell-lineage tracing, it was demonstrated that some of these

corresponding neuroblasts also give rise to neurons of similar morphology and

location (Ungerer and Scholtz 2008). Therefore, a deep conservation can be seen

across arthropods in neuroblast map architecture, the neuronal cell lineages that

form the ventral nerve cord, and the axonal projection patterns.

The conservation in neuroblast map architecture is particularly impressive

when considering the disparity in embryo size between species. For example, a

grasshopper egg is about 10 times longer than a fruit fly egg (Patel 2000), yet the

embryo contained within them will give rise to about the same number of

neuroblasts per hemisegment. How can the same number of neuroblasts be

specified in a manner that does not scale with the overall size of the embryo? One

clue came from the work of Belu et al. (2011), which analyzed closely and

distantly related Drosophila species that produce embryos of varying sizes. The

authors found that the neuroectoderm is conserved in absolute size along the DV

axis across, while the ventral most region of the embryo where the mesoderm

forms is highly variable (Belu and Mizutani 2011).

Given the deep conservation of nervous system development across phylum, it

is of special interest to determine how this developmentally restricted tissue is

26 specified in embryos of different sizes and allow for proper development. This thesis aims to investigate the mechanisms by which the neuroectoderm conservation is achieved in Drosophila species with embryos of different sizes.

One plausible possibility that we explore here is that this conservation is likely to involve species-specific modifications in the Dl and Dpp morphogen gradients, given their prominent roles in the specification of the germ layers.

1.5 Investigating the scaling of morphogenetic gradients

The mechanisms that lead to a proportional tissue patterning scaling to match overall organism size is an ongoing field of investigation in developmental biology (Umulis and Othmer 2013). Here, we take advantage of the well-known

Drosophila DV patterning to study this question, which allows us to easily quantify the Dl morphogen distribution with the use of available antibodies and detection of the expression of transcriptional markers. The fact that Drosophila develops as a long-germ band insect provides another advantage to investigate tissue scaling. In this type of development, the entire body plan is established along the DV and AP primordial axes in the syncytial blastoderm, prior to major gastrulation movements and without growth of additional segments. (Kimelman and Martin 2012). Thus, it is possible to obtain quantifications of gradient and gene expression domains of the entire presumptive DV and AP axes, making

Drosophila an ideal candidate to study scaling of tissues and the overall body plan.

Finally, to study scaling it is necessary to understand the effects caused by

variations in size. In this case, a key advantage in using Drosophila is the possibility to analyze related species that share similar developmental mechanisms and that display significant variations in embryo size. In addition, closely related species that hybridize greatly expands our ability in performing genetic experimentation, as seen in Chapter 2. Our cross species comparison provides an evolutionary perspective of morphogen gradient modifications and allows us to test whether such changes are a common theme in tissue scaling, or if the evolution of cis-regulation of DV genes needs to be taken into account in early tissue patterning between species of the same genus.

Gregor et al. (2005) provides an excellent example of morphogen and tissue scaling along the AP axis across Diptera (Gregor et al. 2005). As explained before, the Bcd morphogen gradient patterns the AP segments along the AP axis by establishing proper expression of segmentation genes, such as gap genes and pair-rule genes (Kraut and Levine 1991; Small et al. 1991). Gregor et al. showed that the expression stripes of gap and pair-rule gene are located in the same relative position along the AP axis of embryos of species that vary up to fivefold in embryo length: D. busckii, D. melanogaster and L. sericata (Gregor et al.

2005). The authors show that the Bcd gradient extends further in the large embryos of L. sericata than in the smaller embryos D. melanogaster and D. busckii (Fig. 3). However, when Bcd levels are plotted against relative position along the AP axis, its distribution was shown to be equivalent across species.

Thus, the scaling of AP segmentation in embryos that vary up to fivefold in length

28 was demonstrated to be primarily caused by Bcd morphogen scaling along the embryo length, rather than the alternative possibility that cis-regulatory sequences of Bcd-target genes had been modified (Gregor et al. 2005) (Fig. 3).

The observations above raise the question about the mechanisms behind the

Bcd scaling and led to the proposal that species-specific adjustments in protein lifetime might contribute to gradient scaling, i.e., Bcd would have an extended lifetime in large L. sericata eggs allowing Bcd to reach further distances (Gregor et al. 2005). However, follow-up transgenic experiments revealed that the stability of L. sericata Bcd is similar compared to D. melanogaster, when tested in a D. melanogaster environment (Gregor, McGregor, and Wieschaus 2008). It remains to be tested whether or not the stability of the protein is increased in its own native L. sericate environment.

Expanding on work done on Bcd gradient scaling, Cheung et al. (2011) showed that in D. melanogaster lines inbred to produce small and large embryos

(C. M. Miles et al. 2011), the Bcd gradient also exhibits proper scaling (Cheung et al. 2011). They found that the amount of maternally loaded bcd mRNA scales to embryo volume (Cheung et al. 2011) (Fig. 3), and as a result, larger embryos will have a higher Bcd production rate allowing for a scaling of the Bcd gradient to the length of the embryo (Cheung et al. 2011). However, it remains untested whether the Bcd production rate correlates with embryo size across other species of

Diptera.

Figure 3: The Bicoid gradient scales along the AP axis and establishes a proportional segmentation pattern in embryos of different sizes. The diagram depicts the scaling of segment formation across embryos of different sizes from different Diptera species, and in distinct D. melanogaster lines selected for embryo size. The Bcd gradient (orange) scales in shape along the length of the embryo from small (D. busckii) to large (L. sericata) embryos. The Bcd gradient initially establishes the segmentation process along the AP axis, which is marked here by expression of the pair-rule gene runt (blue). Note that the AP axis is segmented in the same relative position and maintain same number of runt stripes, regardless of embryo size. Findings of Bcd gradient and pair-rule scaling across species taken from (Gregor et al. 2005). The Bcd gradient and segment formation also exhibit scaling properties across D. melanogaster lines inbred to produce small or large embryos (Cheung et al. 2011; C. Miles, Lott, and Hendriks 2011). Across these D. melanogaster lines, it was found that the amount of maternal bcd mRNA maternally loaded into the embryo correlates with the size of the embryo, allowing the Bcd production rate and Bcd gradient shape to scale with embryo volume (Cheung et al. 2011).

1.6 Scaling of DV germ layers in related Drosophila species

While there has been considerable work done on scaling of AP axis specification with respect to the Bcd gradient, less is known about the scaling mechanisms along the DV axis . This question is particularly important because the DV axis generates the main tissues types of the adult fly, including nervous system that is a highly conserved to the single cell level. As briefly mentioned above, Belu et al. showed that the mesoderm varies in size across related species of Drosophila, which might compensate for their variation in embryo size and allow for the specification of a neuroectoderm that is conserved in size (Belu and

Mizutani 2011).

Here we are interested in testing whether or not the Dl and Dpp gradient distributions have been modified in these species, and what would be the cellular and molecular mechanisms that drive such changes. Noteworthy, the Dl and the

Dpp gradients are each formed by distinct molecular mechanisms that differ from the Bcd gradient. Thus, our line of investigation opened the potential to reveal novel properties of gradient formation and scaling.

We selected six Drosophila species that encompasses roughly 50 million years of independent evolution. Specifically, the species investigated here are three sibling species D. melanogaster, D. simulans and D. sechellia, an additional pair of sibling species of the melanogaster subgroup that underwent a separate speciation event, D. santomea, D. yakuba, and a more divergent species D. busckii, (Fig. 4). D. busckii produces the smallest embryos studied here, and

diverged from D. melanogaster some 50 MYA. D. simulans/D. sechellia and D.

yakuba/D. santomea represent two pairs of sister-species that underwent separate speciation (David et al. 2007; Lachaise et al. 1986, 2000). These species diverge from the common D. melanogaster ancestor by roughly 5-6 MYA (Tamura,

Subramanian, and Kumar 2004). Notably, within a short period of only 0.3-0.5

MYA since the separation from their sister species, D. sechellia and D. santomea have evolved much larger embryos than D. simulans and D. yakuba, and than D. melanogaster as well (Fig. 4) (Lott et al. 2007).

In chapter 2, I investigate the body axis specification along the DV axis of D. melanogaster, D. simulans, D. sechellia and D. busckii, and compared the Dl morphogen distributions across these species. We find within short divergence times, the Dl gradient acquired species-specific shapes varying from sharp to flattened nuclear level distributions. I also directly implicate these changes in the

Dl gradient shape to the previously observed changes of mesodermal size, and further demonstrate that the cis-regulatory sequences of mesodermal genes are activated by same Dl threshold levels. In addition, I also test and confirm the hypothesis that differences in nuclear size and density can drive modifications in the distribution of Dl levels. These findings motivated us to use a mathematical model to pinpoint the causal factors behind Dl gradient evolution, which is presented in Chapter 3.

Figure 4: Divergence in embryo size in related Drosophila species. Phylogenetic tree of Drosophila sibling species (melanogaster subgroup) and D. busckii. Phylogenetic tree includes the relative size of relative of embryos produced by these species.

1.7 Mathematical modeling as a tool to test predictions of Dl gradient scaling

during evolution

The experimental data presented in Chapter 2 suggests multiple avenues for

modifying the Dl gradient within a very short divergence time that separates some

of the species of Drosophila described above. In order to test predictive causes

behind species-specific Dl gradient alterations, in Chapter 3, we employed a

mathematical model of the Dl gradient to narrow down key factors that drive the

Dl gradient alterations, which could not be easily tested experimentally in all

species. Specifically, we hypothesized that the fast-evolving morphological traits and differences in the ranges of Toll signaling could explain species-specific Dl gradients.

We modified a previously published model of the Dl gradient based on differential equations that describe the biochemical interactions between components of the Tl signaling pathway and that takes into consideration the numbers and size of cell compartments along the DV axis (Kanodia et al. 2009).

This model was originally created to model the dynamics of the gradient formation from early cleavage cycles to late blastoderm stages in D. melanogaster. Some parameter values used by Kanodia et al. could be directly

measured, while other values were determined by a stochastic evolutionary

optimization technique that was partly constrained by Dl-GFP data (Kanodia et al.

2009). Their model successfully reproduced the dynamics of the Dl gradient formation during the last nuclear cycles and during the cellularization stage in normal wildtype D. melanogaster embryos. However, this model was neither

34 validated in mutant conditions nor in other species. To validate and extend the applications of this model, we made modifications to allow the independent adjustment of 19 parameters in order to make the model more amenable to test hypotheses for mutant conditions or other species. We ensured that the representative parameter set obtained from Kanodia et al. was biologically relevant by further constraining the parameter values with experimental data from wild type (WT) and mutant Dl gradients that were determined using antibodies against endogenous Dl.

Our modified model made important predictions that were supported by experimental data. First, it revealed that closely related species share similar changes in Toll signaling pathway, despite the fact they may have very different

Dl gradient shapes. Second, the model indicates the existence of an interaction between morphological traits and Toll signaling pathway regulation. If parameters affecting certain morphological traits are changed together with parameters affecting the regulation of Toll pathway, the impact over the gradient shape will be more significant than changing these parameters in isolation. Finally, another prediction made by the model was that the changes of the Dl gradient are more likely to occur through additive small changes that modify the biochemical interactions in more than one component of the Tl pathway, rather than a large change affecting a single component. We further tested and confirmed the existence of several amino acid changes in various specific Dl and Cact protein domains, which are consistent with this last prediction.

1.8 Research aims

1.) In Chapter 2, we investigate the lack of scaling of the mesodermal domain

previously observed in Drosophila species that vary in embryo size. Our

goal was to quantify the distribution of Dorsal protein levels and devise an

experiment to test whether the mesodermal genes responded to same Dl

threshold levels across closely related species. Finally, we also aimed to

test if nuclear size and density could influence the shape of the Dorsal

gradient.

2.) In Chapter 3, we sought to determine the contribution of morphological

traits and Toll signaling components on the formation of the Dl

morphogen gradient in different species of Drosophila using mathematical

modeling.

Chapter 2: Variation in the Dorsal Gradient Distribution is a Source for

Modified Scaling of Germ Layers in Drosophila

In order to study how morphogen gradients and cell fates scale in embryos

of different sizes, we investigated DV axis patterning across Drosophila species.

It is known that the size of the mesoderm and neuroectoderm respond differently

to changes in embryo size (Belu and Mizutani 2011). However, the underlying

mechanism that leads to this unequal scaling is unknown. Due to its prominent

role in patterning the mesoderm and neuroectoderm, we investigated whether the

Dl gradient was modified in these species. The main goals of this chapter were to

quantify the Dl gradient distribution across Drosophila species, test whether the

Dl-target genes from these species had modified sensitivity to Dl levels, and finally test if the gradient could be altered by modified numbers and packing of blastoderm nuclei.

The material below was previously published in the Journal Current Biology: Chahda JS, Sousa-Neves R, Mizutani CM. Curr Biol. 2013 Apr 22;23(8):710-6. doi: 10.1016/j.cub.2013.03.031

2.1 Abstract

Specification of germ layers along the DV axis by morphogenetic gradients is an ideal model to study scaling properties of gradients and cell fate changes during evolution. Classical anatomical studies in divergent insects (e.g. flies and grasshopper) revealed that the neuroectodermal size is conserved and originates similar numbers of neuroblasts of homologous identity (C Q Doe 1992; Thomas

et al. 1984; Whitington 1996). In contrast, mesodermal domains vary significantly

in closely related Drosophila species (Belu and Mizutani 2011). To further

investigate the underlying mechanisms of scaling of germ layers across

Drosophila species, we quantified the Dorsal (Dl)/NFk-B gradient, the main

morphogenetic gradient that initiates separation of the mesoderm, neuroectoderm

and ectoderm (Hong et al. 2008; Lynch and Roth 2011; Reeves and Stathopoulos

2009). We discovered a variable range of Toll activation across species and that

Dl activates mesodermal genes at same threshold levels in melanogaster sibling

species. We also show that the Dl gradient distribution can be modulated by

nuclear size and packing densities. We propose that variation in mesodermal size

occurs at a fast evolutionary rate and is an important mechanism to define the

ventral boundary of the neuroectoderm.

2.2 Results and Discussion

In Drosophila, a ventral-to-dorsal nuclear concentration gradient of maternal origin is established by the transport of Dl into the nuclei upon activation of Toll

receptor (Anderson, Bokla, and Nusslein-Volhard 1985; Roth, Stein, and

Nüsslein-Volhard 1989; Rushlow et al. 1989; Steward et al. 1988). Different Dl

concentration levels turn on or off several target genes depending on their cis-

regulatory sequences, which bind to Dl with different affinities (reviewed in

(Stathopoulos and Levine 2002)). Although the Dl regulatory network and

characterization of cis-regulatory elements of target genes have been extensively

studied (Reeves and Stathopoulos 2009), currently it is not known whether the

shape and range of the Dl gradient itself vary across species and contribute to

novel expression patterns. Different Drosophila species can have variations in

egg size, total numbers of nuclei and packing densities (Fowlkes et al. 2011;

Gregor et al. 2005; Lott et al. 2007; Markow, Beall, and Matzkin 2009), which are

predicted to impact the formation of the Dl gradient.

To investigate these variables, we measured the embryonic DV diameter and

total nuclei numbers distributed along the DV axis in D. busckii and D. sechellia,

which have small and large egg sizes, respectively, and D. melanogaster and D.

simulans, which have similar intermediate-sized eggs. The DV diameter increases

35% from small to intermediate eggs, and 15% from intermediate to large, while

the nuclei vary from 84 to 101 (Table 1; Fig. 4A). Since cleavage cycles are

evolutionarily conserved (Baker, Theurkauf, and Schubiger 1993; Gregor et al.

2008), the variation in nuclei numbers likely arises from failed nuclei divisions,

migration to the cortex and asymmetric packing distribution along the axes

(Fowlkes et al. 2011; Keranen et al. 2006).

We next determined a set of measurements of the mesoderm in these species

that included net numbers of DV nuclei expressing the mesodermal marker snail

(sna) (“mesodermal nuclei”), the percentage of mesodermal nuclei in relation to

all DV nuclei, and arc length distance. The latter measurement corresponds to the

region occupied by the mesoderm in relation to the embryonic circumference and

reports the range of peak Toll activation with highest Dl levels that activate sna.

We found that the mesodermal nuclei in these species deviate significantly from

the average 19 in D. melanogaster (Fig. 5A; Table 1) (Fowlkes et al. 2008; Ip et al. 1992; McHale et al. 2011). In D. melanogaster, 21% of its DV nuclei are allocated to the mesoderm and occupy 21% arc length of the embryonic circumference. The percentages of mesodermal nuclei and arc length also match in D. busckii (17%). In contrast, D. simulans and D. sechellia have about 24% and

22% of mesodermal nuclei, respectively, but a mesodermal arc length of 27%.

These results confirm and extend our previous results that the range of Toll signaling modify the absolute number of nuclei committed to the mesoderm in different species (Belu and Mizutani 2011). The discrepancy in percentages of

mesodermal nuclei and arc length also corroborates previous findings that nuclei

packing densities vary along the DV axis (Fowlkes et al. 2011; Keranen et al.

2006).

2.2.1 Cross-species comparison of nuclear Dl protein levels reveals

gradients of different shapes

Next we quantified the Dl gradients in these species (Fig. 4B-E; (Kanodia et al. 2009; L M Liberman, Reeves, and Stathopoulos 2009)). These data reveal

striking variations in the distribution of Dl (Fig. 5F-J. Individual graphs in Figure

S1). D. busckii has the smallest mesoderm and sharpest gradient among all

species, with highest Dl peak levels and steepest slope of the gradient. By fitting

the normalized data to a Gaussian curve, we note a 19.3% decrease in full width

at half maximum in the D. busckii curve in comparison to D. melanogaster (Fig.

5J). In contrast, D. simulans, the species with highest percentage of mesodermal

nuclei, also has the broadest gradient, with a shallow distribution of Dl levels

corresponding to an increase of 22.7% in width compared to D. melanogaster

(Fig. 5J). Finally, D. melanogaster and D. sechellia have nearly identical gradient

shapes (Fig. 5G, I, J).

2.2.2 Mesodermal expansion does not rely on altered sna or twi sensitivity

to Dl levels in sibling species

Since the normalization of the gradients is dimensionless, we could not

distinguish whether the mesoderm is specified at similar or different Dl

thresholds. To test whether the mesodermal span is influenced by either the Dl

gradient shape or modified sensitivity of Dl target genes, we compared the Dl

threshold levels required to activate the target genes sna and twist (twi) of sibling

species in a single organism. We took advantage that D. melanogaster, D.

simulans and D. sechellia hybridize to create hybrid embryos that receive maternal information solely from one species to establish the Dl gradient, and carry one autosomal copy of sna and twi genes from both species (Fig. 6). We

then visualized sna or twi nuclear nascent transcripts in hybrid embryos at the

border of the mesoderm and neuroectoderm, and asked whether these nuclei

responded to the same Dl threshold (i.e. presence of two nascent transcription

dots) or different thresholds (i.e. presence of one dot).

Hybrid embryos from D. melanogaster mothers have a mesodermal size similar to the maternal species (Fig. 6A, C) and two sna transcription dots at the boundary of the neuroectoderm (Fig. 6E). Thus, the sna copy from D. simulans does not elicit a broader expression due to a higher sensitivity to Dl. To exclude the possibility of differential sna activation due to divergence of Dl sequence, we analyzed hybrid embryos from the reverse cross. In hybrid embryos with Dl gradient inherited from D. simulans, the same results were obtained, i.e. the mesodermal size is similar to the maternal species, and both sna nascent transcripts are activated in all mesodermal cells (Fig. 6B, D, F). The same result was obtained for twi, a direct Dl target (Fig. S2A,B). Finally, forward and reverse hybrids between D. melanogaster and D. sechellia reveal same results (data not shown).

2.2.3 Different gradients in the same scale of threshold levels reveal unique

properties of scaling

When we transformed the Dl graphs to report the actual levels required for sna activation in the sibling species (Fig. 6G, see supplemental methods), two important features become apparent. First, the mesodermal expansion in D. sechellia is achieved by an absolute increase in Dl levels in comparison to D. melanogaster, which explains why the two species have different mesodermal domains despite their identical Dl gradient distribution. Second, the broadest mesodermal domain seen in D. simulans is consistent with its broader gradient

compared to D. melanogaster. Thus, within a short divergence of 0.5 to 4.5 MYA

that separate these three species (Tamura et al. 2004), the Dl gradient acquired

novel shapes and levels. These changes primarily affect the mesoderm, while in

the neuroectoderm, Dl levels are very low and appear to equalize in these species

(Fig. 6G) without altering the expression domains of sog (Belu and Mizutani

2011) or columnar neural identity genes (Fig. S2, C-F). Thus, in all three sibling species, sna and twi have equal sensitivity to Dl, and the mesodermal size increase is exclusively caused by changes in the Dl gradient. Regarding the more divergent species D. busckii, which does not hybridize with melanogaster subgroup, the sna sensitivity to Dl remains to be tested.

2.2.4 The Dl gradient shape in D. melanogaster is sensitive to changes in

nuclear size and packing

Although our cross-species comparisons and hybrid analyses show that species-specific ranges of Toll activation alone can explain the range of the Dl gradient, embryos of different species exhibit differences in nuclear size and packing (Fowlkes et al. 2011), which might contribute to the final shape of the gradient. The nuclear diameters vary from 4 to 7 µm (Fig. 7), significantly expanding the nuclear surface area from 50.2 µm2 (D. busckii), 95 µm2 (D.

simulans), 113 µm2 (D. melanogaster) to 153.9 µm2 (D. sechellia). Additionally,

D. melanogaster and D. sechellia have densely packed nuclei compared to D.

busckii, while D. simulans has an intermediate packing density (Fig. S3).

To isolate the effect of nuclear size and density over the Dl gradient

formation, we analyzed D. melanogaster embryos with unaltered Toll signaling,

but with altered nuclear size and densities. We used sesame (ssm) and gynogenetic-2; gynogenetic-3 (gyn-2; gyn-3) mutants that generate haploid embryos (i.e. undergo one more nuclear division) and triploids (i.e. one less division), to change nuclei numbers, size and packing (Fig. 8A-C) (Edgar et al.

1986; Erickson and Quintero 2007; Grosshans, Muller, and Wieschaus 2003).

These zygotic mutations do not affect the maternal Toll pathway.

The net numbers of sna+ mesodermal nuclei in haploids and triploids

changes significantly. Haploids have on average 25 mesodermal nuclei, which are

statistically greater than the wild type (wt) D. melanogaster average, but similar to D. sechellia and D. simulans. Triploids have 15 mesodermal nuclei, similar to

D. busckii (Fig. 6A-C, tables 1-2). Despite net variations in mesodermal nuclei,

the percentage of mesodermal nuclei remains at 21%, which is characteristic of

the D. melanogaster species. Similarly, the percent arc length of mesodermal

domain in haploids and triploids is equal to wild type D. melanogaster (data not

shown), which is expected and consistent with the unaltered maternal Toll

signaling pathway in these zygotic mutants.

The quantification of Dl gradients in ploidy mutants reveals significant

alterations in the way Dl is distributed (Fig. 8; Fig. S1). In haploids, the Dl

gradient becomes broader and with lower peak levels in the ventral midline,

following a distribution analogous to that of D. simulans. In contrast, triploids

show a similar profile to D. busckii, with sharper Dl distribution and higher peak

levels than in the wt. Thus, physical changes in nuclei size and packing can

reshape the Dl gradient and consequently modify the number of nuclei allocated

44 to the mesoderm, even in the presence of invariable Toll signaling levels.

Previous live imaging with Dl-GFP excludes the possibility that the Dl concentration is modulated by chromatin binding (e.g. addition of one genome copy from haploid, to diploid, to triploid), since Dl never accumulates in nuclei but instead transiently binds to and dissociates from chromatin in short intervals

(DeLotto et al. 2007).

The effect of nuclei over gradient formation has been modeled as reversible traps and as localized sites for morphogen degradation (Coppey et al. 2008;

Gregor et al. 2007). A high density of nuclei could divert Dl transport into the nucleus and flatten the gradient (e.g. D. simulans and haploids), whereas a low density would sharpen it (e.g D. busckii and triploids). A caveat is the constant gradient shape seen throughout the last nuclear divisions of D. melanogaster

(DeLotto et al. 2007; Kanodia et al. 2009; L M Liberman et al. 2009), when nuclear size decreases and density doubles at each cycle (Gregor et al. 2007). Two hypotheses could explain our findings. One possibility is that the Dl gradient shapes of haploids and triploids are altered from the onset of gradient formation, since these mutants start out with smaller or larger nuclei than the wild type.

Alternatively, it is possible that subtle changes in the Dl gradient distribution in the wild type throughout nuclear divisions do exist, but are undetectable with current measurement methods. The behavior of the Dl gradient seen here in the mutants could be used in the future to test a computational model proposed for the

Dl gradient, which relied on restricted parameters that best fit the final shape of

the gradient, but discarded several possibilities for the dynamics of the gradient at

early stages (Kanodia et al. 2009).

2.2.5 Fast evolution of the Dl gradient and maintenance of the

neuroectoderm

Unlike the AP gradient of Bicoid that is scaled to size in divergent flies

(Gregor et al. 2005, 2008), the Dl gradient does not intrinsically scale. Indeed,

distortions in mesodermal size are significantly higher than minor changes in the

positioning of stripes of segmentation genes (Fowlkes et al. 2011; Lott et al.

2007). From an evolutionary standpoint, it is not entirely surprising the lack of

co-evolution of cis-regulatory sequences of Dl targets, because the species studied

here diverged very recently and did not have time to accumulate differences as

noticeable as those of more distant lineages, such as D. virilis and D.

pseudoobscura (Crocker, Potter, and Erives 2010; Crocker, Tamori, and Erives

2008). What is surprising is the change in several traits of DV diameter, nuclear

size and density in a relatively short time. Our experiments with ploidy mutants

indicate that nuclear size and density traits can effectively generate diverse shapes

and intensities of the Dl gradient. Interestingly, these physical traits evolved fast

in parallel to a second group of fast-evolving genes dedicated to immune response

(Clark et al. 2007; Jiggins and Kim 2007; Obbard et al. 2009; Sousa-Neves and

Rosas 2010) that are also shared by the Toll DV pathway. The changes in the Toll pathway and effect of nuclear size and density over Dl nuclear import can easily explain the variations in the range of Toll activation observed (Fig. 5) and the

46 diverse shapes and intensities of the Dl gradient in each Drosophila lineage (Fig.

5).

We previously showed that the evolutionary expansions and retractions of the mesoderm do not modify the stereotyped array of somatic muscles (Belu and

Mizutani 2011), and as such these variations could be considered a neutral or non- adaptive trait. However, the present results indicate that the DV patterning system evolved to allow shifts in the neuroectodermal borders to new DV positions that preserve the width of the neuroectodermal domain, which is adaptive since the constancy of this domain is absolutely crucial for correct specification of neuronal lineages (Jiménez et al. 1995; McDonald et al. 1998; Weiss et al. 1998).

Therefore, the observed Dl gradient shapes and DV repositioning of neuroectodermal borders in Drosophilids are likely to have been selected over generations from a pool of individuals with modified range of Toll signaling, nuclear size and density.

The experiments with hybrids reveal that the underlying mechanism that controls mesodermal size and shifts the ventral neuroectodermal border involves exclusively a variation in the range of Toll activation and Dl gradient shape, and is not due to differential gene response. Thus, whenever the range of Dl distribution is changed, the mesodermal/neuroectodermal border acquires a new position. The shift in the ventral neuroectodermal border concomitantly repositions the dorsal neuroectodermal border in relation to the ventral midline, beyond which the Dl levels are insufficient to repress decapentaplegic

(dpp)/BMP-4. The acquisition of a new upper limit of the neuroectoderm is

supported by three independent lines of evidence. First, the hybrid experiments

show that the sibling species have equal Dl levels at the

mesodermal/neuroectodermal border, which are likely to have similar decay to

low background levels within the neuroectoderm, as suggested in the transformed

Dl graphs (Fig. 6G). Second, consistent with comparative anatomical studies in

insects (C Q Doe 1992; Thomas et al. 1984; Whitington 1996), the width of the

neuroectoderm remains constant in the species tested here, as shown previously

(Belu and Mizutani 2011) and in greater detail here (Fig. S2). Finally, the dpp+

nuclei numbers and subdomains of gene expression within the ectoderm vary

across species (Ambrosi and Chahda, unpublished data).

An interesting feature of Dpp/BMP signaling is its role in repressing neural

gene expression in the ectoderm and forming an opposing dorsal-to-ventral gradient that helps pattern the neuroectoderm (Mizutani et al. 2006). We speculate

that the interaction of Dl and Dpp/BMP gradients represents a larger self-

organizing system capable of responding to the rapid evolution of nuclear size,

density and embryo size, by modifying the mesoderm while correctly assigning

neuroectodermal DV fates.

2.3 Materials and Methods

2.3.1 Fly stocks and genetic crosses

y w D. melanogaster was used as wild type. The following stocks from

Drosophila Species Center (UCSD) were used: D. busckii (wildtype, 300-0081-

23), D. simulans (wild type, 14021-0251-199) and D. sechellia (zn[1] v[1] f[1],

14021-0248.19). To obtain hybrid embryos, y w D. melanogaster females were

crossed to either D. simulans or D. sechellia males. The reverse cross was done using the isogenized mutant line Santa Maria D. melanogaster, collected from

natural population (Sousa-Neves, unpublished data). Santa Maria males can

bypass sexual rejection of D. simulans and D. sechellia females. Scoring and

confirmation of hybrid progeny are described in Supplemental data. Haploid and

triploid embryos were generated using w, ssm (Erickson and Quintero 2007;

Loppin et al. 2000) (a gift from James Erickson) and gynogenetic-2; gynogenetic-

3 (gyn) ((Fuyama 1986) Bloomington Stock Center), respectively. Genetic

schemes and genotyping of embryos (Fig. S4) are explained in Supplemental data.

2.3.2 Measurements of egg size, nuclear number, size, and packing densities

Measurements of embryo size were obtained from intact (Belu et al. 2010) and

sectioned embryos using a Confocal microscope (Zeiss LSM700, see

supplemental data). For DV measurements of DV perimeter, nuclear counts and

mesodermal arc-length, cross-sections of trunk regions of stained embryos for sna

RNA and DAPI nuclear dye (Kosman et al. 2004) were analyzed using Image J

software. For nuclear size and packing calculation, early blastoderm stage

embryos stained for sna mRNA and anti-Lamin were mounted longitudinally, and

confocal optical slices were taken across the entire width of sna+ ventral nuclei.

Images of the optical slice corresponding to the center of nuclei were analyzed

using Photoshop to calculate pixel densities of the space between nuclei (Fig. S3).

Statistical analyses were performed using the PAST software (version 2.09,

http://folk.uio.no/ohammer/past/). The data was compared using one-way

ANOVA, followed by Tukey’s test for pairwise comparison. The cutoff used for

statistical significance was p<0.05.

2.3.3 Immunohistochemistry

Embryos were collected for 5-6hrs at 250 C in grape juice agar plates supplemented with yeast, or Noni fruit leather (for D. sechellia), fixed and processed for in situ and protein staining, as described in(Kosman et al. 2004).

Probes against sna and twi were labeled with digoxigenin (DIG) (Roche). Primary antibodies and dilutions used were: Sheep anti-DIG (1: 1,000, Roche), mouse anti-Laminin (1:1,000, Iowa Hybridoma Bank), mouse anti-Dorsal (1:1,000, Iowa

Hybridoma Bank, used for D. melanogaster, D. simulans and D. sechellia species) and Rabbit anti-Dorsal (1:2,000, a gift from Steve Wasserman, UCSD, used for D. busckii). Rabbit anti-Dorsal and mouse-anti-Dorsal antibodies provided identical results for yw D. melanogaster (data not shown). Secondary antibodies were used at 1:500 concentration: Donkey anti-Sheep Alexa 488,

Donkey anti-Rabbit Alexa 555, Donkey anti-mouse Alexa 647 (Invitrogen).

Nuclei were stained with DAPI (Invitrogen) at 300 nM for 15 min.

2.3.4 Quantification of the Dl gradient

Cross sections from trunk regions of stained embryos were cut using a micro knife (Roboz) or a 26 gauge 3/8” needle (Grosshans and Wieschaus 2000).

Embryo slices were mounted in ProLong Gold antifade (Invitrogen) and cured for

24 hrs at room temperature prior to imaging in a Zeiss LSM700 Confocal. Gain and offset settings were adjusted to non-saturating levels spanning entire 12-bit

dynamic range [19]. Images were exported to AxioVision 4.8 (Zeiss) for data

analysis. Average fluorescent intensity levels were obtained from circles of 10

µm2 centered on the 30 most ventral nuclei stained with DAPI and sna. To

normalize the gradient, we used a modified version of a previously described

Dorsal normalization method (Kanodia et al. 2009), in which the lowest fluorescent intensity was subtracted from each data point, and then each data point was divided by the sum of all the data points. To estimate width at half maximum values, Dl concentration graphs were fitted to a Gaussian curve, using curve fitting feature from Matlab. The curve fitting also confirmed the location of ventral midline at the ventral-most cell expressing sna. For details on methods used for normalization and transformation of Dl graphs based on threshold levels that activate sna and twi, see supplemental information.

2.4 Supplemental Information

2.4.1 Normalization of Dl gradient allows cross-species comparison

Antibodies raised against several D. melanogaster proteins have been widely used with success in other species to visualize expression patterns as well as to quantify morphogenetic gradients (e.g. Bcd gradient in (Gregor et al. 2008)).

Our quantification method accurately extracts species-specific Dl gradient shapes by normalizing raw fluorescent data obtained from antibody stainings (see methods) to compensate for experimental variation and possible differences in antibody affinities across species. It is important to emphasize that even if different affinities for the antibody exist in other species (which formally it is not known), because the kinetics of antibody-protein binding relies on constants of association (Ka) and dissociation (Kd), the readings along the entire curve would increase or decrease after normalization is made (see fig. S1). To normalize Dl levels in the 30 most ventral nuclei, the data point (nuclei; i) with the lowest fluorescent intensity was subtracted from each data point. Next, each data point was divided by the sum of all the data points. After normalization, each data point represents a percentage all fluorescent values obtained. Therefore, even if differences in antibody kinetics exist across species, the shape of the Dl gradient can be assessed, because raw fluorescent values will be equally increased or reduced at every data point, but an individual data point’s value in terms percentage of all fluorescence obtained is unchanged (Fig. S1A). Figure S1B shows the average shape of the Dl gradient in different species (Fig. 5F-I) and in

D. melanogaster mutants (Fig. 8E-F) obtained from individual embryos.

To test if sna and twi are activated by same Dl threshold levels across

species, we detected nascent transcripts at the border between the mesoderm and

neuroectoderm of D. melanogaster/D. simulans hybrids. The in situ displayed in

Fig. 6A-F shows that the lowest Dl levels required to activate sna expression in

both D. melanogaster and D. simulans are identical. Figures S2A and B show that

the threshold level of Dl needed to activate twi expression is also the same across

these species.

With the shift of the mesodermal boundary in these species (as well as in

D. busckii), there is a corresponding shift in the dorsal most border of the

neuroectoderm, since the neuroectodermal domain contains similar numbers of

sog expressing cells in the species D. melanogaster, D. busckii, D. simulans, D.

sechellia (Belu and Mizutani 2011). Figures S2C-F show that the columnar neuroectodermal genes maintain same domain and borders within the neuroectoderm.

In order to investigate a possible cause of Dl gradient modification within species, we analyzed the size and packing density of blastoderm nuclei using anti-

Lamin staining. Figure 6 shows that the diameter of blastoderm nuclei is different among D. busckii, D. melanogaster, D. simulans and D. sechellia. To quantify the

apparent differences in nuclear packing density, we used image analysis tools to

extract the areas between nuclei and calculated the number of pixels of these areas

(Fig. S4).

2.4.2 Scoring of hybrid crosses and embryos with altered ploidy

The hybrid progeny was confirmed by scoring classical unisexual progeny. When D. melanogaster are used as mothers, only females hatch in the

F1 and those have defects in dorsoventral bristles and are sterile. The F1 of the

reverse cross is composed exclusively by sterile males, which have genital arcs of

intermediate size between D. melanogaster and D. simulans or D. sechellia. Over

200 adult hybrids were recovered, sexed and confirmed to be infertile (atrophied

testis and ovary).

Haploid embryos were generated by mating homozygous females w, ssm

(a gift from James Erickson) (Erickson and Quintero 2007; Loppin et al. 2000) to

wild-type males. 100% of the progeny of this cross develop as haploids. Triploid

embryos were generated by mating gynogenetic-2; gynogenetic-3 (gyn) (Fuyama

1986) double homozygous females to wild-type males (stock obtained from

Bloomington Stock Center). About 12% of the embryos derived from this cross develop as triploids. To maintain the 12% triploidy rate, gyn2; gyn3 females were mated to sterile males (ms(3)K81) and F1 females that developed parthenogetically (Fuyama 1986) were used to generate triploid embryos, as described above. Haploid and triploid embryos were identified by their nuclear density and size at the onset of cellularization, and the number of nuclear nascent transcripts for autosomal genes (e.g. sna) as one copy in haploids, two copies in diploids and three copies in triploids (See Fig. S4, below).

2.4.3 Measurements of egg size, nuclear numbers, size, and packing

densities

For measuring the anteroposterior axis, eggs with chorion were mounted in 70% glycerol/PBS with a small amount of glass beads (150-210 μm size,

Polysciences) to prevent flattening caused by coverslip. Eggs were imaged on a

Confocal microscope (Zeiss LSM700) at an approximate mid-section focal plane.

Zeiss AxioVision 4.8 software was used to obtain distances from anterior to

posterior poles of eggs. For DV measurement of D. simulans and D.

melanogaster, additional intact embryos of were analyzed using upright imaging

in glycerin jelly (Belu et al. 2010) after a treatment in bleach to remove the

chorion, 15 min fixation, de-vitellinization in heptane/methanol, post-fixation for

15 min and staining with DAPI. Confocal images were obtained for 10 D.

melanogaster and 15 D. simulans embryos at the focal plane with largest DV perimeter, and analyzed using Image J. These additional measurements confirmed that these two species have statistically identical DV diameters.

2.4.4 Method for Dl gradient graph transformation based on data from

hybrids

After normalization of Dl levels was performed as described above, we were able to transform graphs representing Dl gradient shape to directly compare relative Dl levels. This transformation was possible after determining that the threshold levels of Dl required to activate transcription of snail or twist in D. melanogaster, D. simulans and D. sechellia is identical (Fig. 8 in paper).

We arbitrarily set a value of “one” to the data point representing the last nucleus to express snail/twist. These Dl threshold levels correspond to the dorsal mesodermal border, or the last cells with lowest amount of Dl that still express snail or twist. Next, we corrected the gradient of each species by multiplying every point on the curve by a factor X, where X=1/Y, and Y equals to the normalized fluorescent value of the last nucleus, on average, expressing

55 snail/twist. For example, if in the D. melanogaster gradient the fluorescent intensity of the most ventral nucleus is 3x more intense than the sixth nucleus located on the dorsal mesodermal border, then after the transformation, the sixth nucleus will have a value of 1 and the most ventral nucleus will have a value of 3.

2.5 Tables and Figures

Table 1. Cross-Species Comparison of Embryo Size and DV Nuclei. Means are given followed by standard deviations. n, sample size.

Species Length Diameter Total DV Mesoderma Mesodermal nuclei l nuclei percentage D. busckii 378 µm 186.8 µm 84.3 ± 4.24 15 ± 1.35 18% (n=6) (n=10) (n=12) (n=13)

D. 483 µm 204.8 µm 91.3 ± 2.95 19 ± .82 21% melanogaster (n=6) (n=7) (n=13) (n=8) D. simulans 472 µm 205 µm (n=9) 97.6 ± 6.34 23.64 ± 1.63 24% (n=6) (n=17) (n=11) D. sechellia 573 µm 269.7 µm 101.3 ± 4.19 22.9 ± 1.73 23% (n=6) (n=9) (n=13) (n=10)

Table 2. DV Nuclei in D. melanogaster WT, Haploid, and Triploid Mutants. Means are given followed by standard deviations. n, sample size.

Strain Total nuclei mesodermal nuclei wt 91.3 ± 2.95 (n=13) 19 ± .82 (n=8) ssm 115.92 ± 11.96 (n=12) 25.46 ± 3.73 (n=13) gyn 67.67± 2.06 (n=6) 14.67 ± 1.86 (n=6)

Figure 5. The Distributions of Nuclei, Mesodermal Domains, and Dl Levels Changed across Species. (A) Histogram of the average number of total nuclei (blue) and mesodermal nuclei (red) along the DV axis of blastoderm embryos. Bottom panel: pie charts with average percentage of nuclei that are mesodermal (red) and average percent of arc length corresponding to the mesoderm(red). Sample sizes for total nuclei counts are 12 D. busckii, 13 D. melanogaster, 17 D. simulans, and 13 D. sechellia. Sample sizes for mesodermal nuclei are 13 D. busckii, 8 D. melanogaster, 11 D. simulans, and 10 D. sechellia. Error bars are one SD in both directions. *p < 0.05, **p < 0.01, ***p < 0.001. (B–E) Blastoderm cross-sections used for Dl gradient quantification, stained for Dl protein (magenta), sna mRNA (green, C–E), and DAPI nuclear dye (blue). D. busckii (Dbusc; B) has the smallest embryo, followed by D. melanogaster (Dmel; C), D. simulans (Dsim; D), and D. sechellia (Dsech; E). Ventral side is down. Scale bar represents 100 mm. (F–J) Normalized graphs of average intensity levels of nuclear Dl protein (y axis) per individual nucleus (x axis). Graphs are centered on the ventral midline (x = 15) based on sna expression domain and extend dorsally from the center to the left (x = 0) and right (x = 30). (F–H) Average Dl distribution in D. busckii embryos (F). Note a sharper gradient with higher peak levels than D. melanogaster (G). In contrast, D. simulans Dl gradient (H) has a shallow profile, with lower peak levels and broader amplitude than D.

59 melanogaster. (I) D. sechellia gradient distribution is similar to D. melanogaster. (J) Average distributions from all species combined onto one graph. Arrows indicate the Dl threshold levels for sna activation for the dorsal most sna+ nuclei at the border of mesoderm and neuroectoderm. Error bars are one SD in both directions. See also Figure S1.

Figure 6. Sensitivity of Mesodermal Gene Activation Is Identical among D. melanogaster Sibling Species. (A–D) Ventral view of whole-mount blastoderm embryos stained for sna mRNA (red) and DAPI nuclear stain (blue). (A) D. melanogaster. (B) D. simulans. (C) Hybrid embryo from D. melanogaster mother and D. simulans father. (D) Hybrid embryo from D. simulans mother and D. melanogaster father. (E and F) High magnification of boxed areas in (C) and (D), respectively. Note the presence of two nuclear transcription dots per nucleus in cells along the border of sna expression, indicating that both copies of the sna gene from each species are activated. The abutting cells outside the mesoderm have both sna copies turned off. Similar results were obtained for twi (Figures S2A and S2B) and hybrids between D. sechellia and D. melanogaster (data not shown). (G) Transformed graphs of Drosophila sibling species adjusted for sna activation levels (arrow). Note the higher Dl levels in D. sechellia compared to D. melanogaster, and the lower levels and modified shape in D. simulans. Error bars are one SD in both directions. See also Figure S2.

Figure 7. Drosophila Species Vary in Nuclei Size and Densities. Anti-lamin stainings of D. busckii (A and B), D. melanogaster (C and D), D. simulans (E and F), and D. sechellia (G and H) embryos. (A, C, E, and G) Sagittal view of embryos showing start of membrane growth using differential interference contrast transmitted light merged to lamin staining (orange). (B, D, F, and H) Images of a single confocal plane corresponding to the center of nucleus were used to calculate average nuclear diameter, nuclear packing, and nuclear surface area (see Figure S1 and Experimental Procedures). Ventral mesodermal nuclei of

D. busckii (B) have the smallest size compared to the other species and the lowest density packing. D. melanogaster (D) has nuclei slightly larger than D. simulans (F). Nuclei of D. sechellia (H) have the largest size compared to the other species and exhibit highest density packing along with D. melanogaster. Embryos were double stained for sna (data not shown) to localize the ventral region from where the images were taken. See also Figure S3.

Figure 8. The Dl Gradient Is Modified by Nuclear Size and Packing Density. (A–C) Left: whole-mount blastoderm embryos (left) and corresponding cross- sections (middle) stained with DAPI nuclear dye (cyan, left), anti-Dl (magenta), and sna mRNA (green) of WT D. melanogaster (A), ssm (B), and gyn (C) mutants. Right: increasing size in nuclei and density packing from haploid embryos (B) to diploids (A) to triploids (C) in anti-lamin staining preparations (magenta). (D–G) Normalized graphs of average intensity levels of nuclear Dl protein (y axis) per individual nucleus (x axis). Graphs are centered on the ventral midline (x = 15) and extend dorsally from the center to the left (x = 0) and right (x = 30). (D–F) Average Dl distribution of WT D. melanogaster (D), haploid ssm mutants (E), and triploid gyn mutants (F). (G) Average distributions were combined onto one graph: WT D. melanogaster (blue line), haploid ssm (green), and triploid gyn (purple). Error bars are one SD in both directions. See also Figure S4.

Supplemental Figure 1. Related to Figures 5 and 8. Dorsal gradient normalization and data from individual embryos. (A) Hypothetical normalization from two species with exact Dorsal gradient shapes but different antibody affinities. (B) Normalized intensity levels of nuclear Dl protein (y-axis) per individual nucleus (x-axis) obtained from individual embryos. Graphs are centered on the ventral midline (x=15) based on sna expression domain, and extend dorsally from the center to the left (x=0) and right (x=30). D. busckii embryos (n=5), D. melanogaster (n=12), D. simulans (n=10), D. sechellia (n=12), D. melanogaster ssm haploids (n=9) and gyn-2; gyn-3 triploids (n=8).

Supplemental Figure 2. Related to Figures 6. The subdomains of the neuroectoderm are conserved across species. (A and B) Sensitivity of twi

67 activation is identical between D. melanogaster and D. simulans. Ventral view of whole mount blastoderm embryos stained for twi mRNA (red) and DAPI nuclear stain (blue). (A) Hybrid embryo from D. melanogaster mother and D. simulans father. (B) Hybrid embryo from D. simulans mother and D. melanogaster father. Note that there are two nuclear transcription dots present in all cells along the border of twi expression, indicating that both copies of the twi gene from each species are activated at same Dl levels. The abutting cells outside the mesoderm have both twi copies turned off. (C–F) Domains of columnar neural identity genes are conserved in D. busckii, D. simulans, D. melanogaster, D. sechellia, with similar numbers of cells within the vnd (green), ind (red) and msh (white) domains. For D. busckii (C) we were able to determine the size of vnd and ind domains by doing a double staining with an antibody anti-Ind (red), and an in situ for sog (green). At the stage when ind expression is observed, sog mRNA levels demarcates the ventral border of the neuroectoderm, but it is decreased from the dorsal most part of the neuroectoderm.

Supplemental Figure 3. Related to Figures 7: Quantification of Differences in Nuclear Packing Density across Species. (A) Confocal images of anti-Lamin staining from a single plane at the center of nuclei in different Drosophila species, as indicated. (B) Image segmentation from (A) with colored space between nuclei shown in black, and space occupied by nuclei shown in white. (C) Histogram of pixel numbers from white areas corresponding to the nuclei for each species. The higher the pixel value (y-axis), the greater is the nuclear packing density.

Supplemental Figure 4. Related to Experimental Procedures: Genotyping of gyn Triploids by Counting the Number of sna Nuclear Transcripts. The identification of gyn triploids was verified by the presence of three nuclear dots in embryos stained for the autosomal gene sna mRNA, instead of two copies in wild-type. (A) Cross-section from a diploid gyn. Arrow indicates a nucleus with two nuclear dots on the same plane. (B) Maximum confocal projection showing surface view of embryo from which (A) was cut from; note the presence of two nuclear dots in all nuclei shown. (C) Cross-section from a triploid gyn. Arrow indicates a nucleus with three nuclear dots. (D) Embryo from which (C) was cut from; note the presence of three nuclear dots in a maximum projection.

2.6 Conclusions and future directions

2.6.1 Changes to the Dl Gradient Directly Alter Mesoderm Specification

The mesoderm, as measured by sna expression, is species-specific in absolute and relative size and therefore does not scale with embryo size (Fig. 5A). We find that changes in the shape and amplitude of the Dl gradient are directly responsible for species-specific alterations in mesoderm specification, and exclude the possibility of alteration to sna and twi cis-regulation as contributing to differences in mesoderm specification across Drosophila sibling species (Fig. 6A-G).

Our findings are supported by a more recent publication by Garcia et al.

(2013) that used inbred lines of D. melanogaster that were artificially selected to produce eggs of small or large sizes (C. M. Miles et al. 2011). These authors reported that the Dl gradient increases in width and the mesoderm increases in size in the lines with larger eggs (Garcia et al. 2013). Their results demonstrate that existing natural variation in the Dl gradient and the Toll pathway can explain changes to mesoderm specification (Garcia et al. 2013).

Another finding from Garcia et al. (2013) was that the mesoderm and neuroectoderm respond differently to changes in the Dl gradient and embryo size.

In the lines that produce small, normal and large embryos, the widths of sna, vnd and ind expression domains are regulated independently of each other in response to changes in the Dl gradient width. For instance, sna was found to expand in step with the Dl gradient, while the domain of ind expression remained constant

(Garcia et al. 2013). This intra-species observation is similar to our inter-species

71 findings that the sub-domains of the neuroectoderm are conserved despite changes to mesoderm specification and Dl gradient distribution (Chahda, Sousa-Neves, and Mizutani 2013). Garcia et al. (2013) propose that changes to the range of peak

Toll receptor activation are responsible for scaling of the Dl gradient width in small to large D. melanogaster lines, which is also supported by the correlation we observe in the range of peak Toll receptor activation and Dl gradient width across species (Fig. 5A and Fig. 6G).

Additionally, we showed that variation in morphological traits shape the Dl gradient and lead to changes in mesoderm specification. The initial observation that the size and density of blastoderm nuclei is species-specific (Fig. 7), led to the investigation of Dl gradient formation in mutants within D. melanogaster that alter nuclear size and density. ssm and gyn mutations alter the diameter and density of nuclei without altering the range of Toll signaling, which would presumably also alter the Dl gradient. Comparison of ssm and gyn Dl gradient shapes (Fig. 8) clearly shows the impact of nuclear size and density on the final

Dl gradient distribution, which, as expected, also modified mesoderm specification in these mutations (Table 2). Experiments in ssm and gyn mutant prove that morphological traits alone, independent from the Toll pathway, can alter Dl gradient distribution and mesoderm specification.

2.6.2 Conservation of the Neuroectoderm is Unaffected by Changes to the

Dl Gradient

Unlike the mesoderm, we found that the sub-domains of the neuroectoderm,

as measured by vnd, ind and msh expression, are absolutely conserved in size across species, in accordance with the previous findings (Belu and Mizutani

2011), confirming that tissues specified along DV axis respond differently to changes in embryo size.

How does the Dl patterning system simultaneously allow for the specification of one tissue that is seemingly plastic (the mesoderm) and another that is rigidly conserved (the neuroectoderm)? One possibility is that high concentration levels of Dl found in ventral regions are amplified in larger embryos to produce wider domains of sna expression, while lower Dl levels abutting sna expression are similar in concentration across species. Supporting this possibility, we observed that the concentration and slope of the Dl gradient beyond mesoderm specification was similar between D. melanogaster, D. simulans and D. sechellia

(Fig. 6G). In other words, ventral regions of the Dl gradient are changing in width and amplitude, while lateral regions of the gradient appear to be conserved in slope and amplitude.

Chapter 3: Modeling of the Dorsal Gradient across Species Reveals

Interaction between Embryo Morphology and Toll Signaling Pathway

during Evolution

In the previous chapter, we showed that the Dl gradient has been modified during evolution and specifies mesoderms that are species-specific in size.

Evidence from ssm and gyn experiments revealed that changes in nuclear size and

density can alter the Dl gradient distribution. Although we also reported the

existence of differences in nuclear size and density in non-melanogaster species,

it is not known whether these morphological changes are the main causes for their

modified Dl gradient shapes. To better understand the mechanism behind Dl gradient modification in species with embryos of different sizes, we employed a

mathematical model of Dl gradient formation to simulate the effects of

modifications in nuclear size and density, number of DV nuclei, range of peak

Toll signaling. Our expectation was to be able to predict which parameters would

be relevant for the evolution to reshape the Dl gradient. Ultimately, the goal of

this chapter is to investigate the mechanisms that can re-shape a morphogen

gradient during evolution.

The material below was published previously in the Journal of PloS

Computational Biology: Ambrosi P*, Chahda JS*, Koslen HR, Chiel

HJ, Mizutani CM. PLoS Comput Biol. 2014 Aug 28;10(8):e1003807. doi:

0.1371/journal.pcbi.1003807. * First co-authors

Additional supporting information available online can be found at:

http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003807#s5

3.1 Abstract

Morphogenetic gradients are essential to allocate cell fates in embryos of varying sizes within and across closely related species. We previously showed that the maternal NF-κB/Dorsal (Dl) gradient has acquired different shapes in

Drosophila species, which result in unequally scaled germ layers along the dorso-

ventral axis and the repositioning of the neuroectodermal borders. Here we

combined experimentation and mathematical modeling to investigate which

factors might have contributed to the fast evolutionary changes of this gradient.

To this end, we modified a previously developed model that employs differential

equations of the main biochemical interactions of the Toll (Tl) signaling pathway,

which regulates Dl nuclear transport. The original model simulations fit well the

D. melanogaster wild type, but not mutant conditions. To broaden the

applicability of this model and probe evolutionary changes in gradient

distributions, we adjusted a set of 19 independent parameters to reproduce three

quantified experimental conditions (i.e. Dl levels lowered, nuclear size and

density increased or decreased). We next searched for the most relevant

parameters that reproduce the species-specific Dl gradients. We show that

adjusting parameters relative to morphological traits (i.e. embryo diameter,

nuclear size and density) alone is not sufficient to reproduce the species Dl

gradients. Since components of the Tl pathway simulated by the model are fast-

evolving, we next asked which parameters related to Tl would most effectively

reproduce these gradients and identified a particular subset. A sensitivity analysis

reveals the existence of nonlinear interactions between the two fast-evolving traits

tested above, namely the embryonic morphological changes and Tl pathway

components. Our modeling further suggests that distinct Dl gradient shapes

observed in closely related melanogaster sub-group lineages may be caused by similar sequence modifications in Tl pathway components, which are in agreement with their phylogenetic relationships.

3.2 Introduction

The embryonic patterning and development of limbs rely on morphogenetic gradients that set up territories of gene expression in a dosage- dependent fashion (Rogers and Schier 2011; Zeller, López-Ríos, and Zuniga

2009). Rather than a static process, cell fate specification normally occurs under dynamically changing environments that involve cell divisions and tissue growth expansion. One important property of morphogenetic gradients is the ability to scale and accommodate tissue cell types despite fluctuations in organismal size, for instance, due to feeding conditions or mutations affecting growth. Scaling is also a fascinating problem in evolutionary biology and can be observed in related species that have dramatically changed in embryo size but kept fixed gene expression domains at relatively similar positions in relation to the whole body

(Gregor et al. 2005). Recent quantitative studies have begun to elucidate the scaling mechanisms of morphogenetic gradients during tissue growth

(Hamaratoglu et al. 2011), regeneration (Ben-Zvi et al. 2008), as well as in related species (Chahda et al. 2013; Lott et al. 2007; Umulis et al. 2010) or artificially selected strains of same species that vary in embryo size (Cheung et al. 2011,

2014; Garcia et al. 2013; C. M. Miles et al. 2011). In particular, studies in

Drosophila embryonic gradients stand out as being especially amenable to quantitative analysis and modeling (Umulis and Othmer 2013). The relatively simple syncytial organization of the embryo allows precise detection of target gene expression with single cell resolution, and models can be built based on the extensive biochemical data of signaling pathways responsible for gradient formation. Remarkably, the isolation of new closely related species to the

Drosophila melanogaster model (reviewed in (David et al. 2007)) provides a rich natural repertoire of genetic variations in egg size, cell numbers and gene divergence, which can be used to test the impact of these evolutionary changes on the scaling of gradients.

Here we address the question of gradient scaling across related Drosophila species using the embryonic dorso-ventral (DV) patterning as a model system.

The maternal nuclear concentration gradient of the NF-κB related transcription factor Dorsal (Dl) subdivides the embryo into three germ layers: the mesoderm, neuroectoderm and ectoderm. High levels of nuclear Dl in the ventral embryonic side activate expression of mesodermal genes, such as snail (sna), whereas moderate levels in lateral regions activate neuroectodermal genes. Low to

negligible levels of nuclear Dl in dorsal regions allow the expression of

ectodermal genes such as decapentaplegic (dpp) and zerknult (zen), due to the lack of repression that Dl exert on these genes (reviewed in (David et al. 2007)).

We recently reported that the Dl gradient has unique distribution profiles in related Drosophilids that vary in embryo size, which result in unequally scaled germ layers (Chahda et al. 2013). For instance, changes in mesodermal size serve as a mechanism to specify the border of the neuroectoderm and keep it at a constant size. Here we combined experimental approaches and mathematical modeling in order to identify parameters that might be responsible for the modified distributions of Dl gradient across species. Previously, Kanodia et al.

(2009) (Kanodia et al. 2009) developed a mathematical model for D. melanogaster that reproduces the dynamics of the Dl gradient formation during

cleavage cycles (Fig. 9A, B). Their model consists of differential equations

derived from mass balance equations of the main biochemical interactions of the

Toll (Tl) pathway that lead to Dl transport into the nucleus, which were

numerically solved using globally optimized parameters. Briefly, the model

simulates the graded nuclear translocation of Dl initiated by the space-dependent

dissociation of the cytoplasmic complex formed between Dl and Cactus/Ik-B

(Cact). This dissociation is modeled by a reaction rate constant kD and represents

the graded activation of Tl receptors along the embryonic DV axis. The Dl-Cact complex prevents Dl from entering the nucleus and its dissociation due to Tl activation frees Dl to enter the nucleus. The model also recreates the geometric arrangement of embryonic nuclei during cleavage cycles, as well as changes in

nuclear surface area, which affect Dl nuclear import and export rates. The

Kanodia model captures essential properties of the Dl gradient formation and

correctly reproduces the dynamics of the gradient formation during early

embryonic cycles. However, this model has not yet been formally validated in

conditions other than wild type D. melanogaster embryos, or used to simulate the

Dl gradient of other species.

Kanodia et al. (2009) (Kanodia et al. 2009) employed a genetic algorithm

to identify a cloud of dimensionless parameters that satisfied a small dataset of

experimental Dl gradient measurements from wild type embryos only. In this

work, we built upon this model, and attempted to validate its generality by fixing

free parameters using biological measurements, and manipulating only the subset

of parameters that were most likely to be biologically relevant. We manipulated a

single representative parameter set from this model in order to identify which

parameter changes are sufficient to reproduce the experimental Dl gradients from

three distinct experimental conditions in D. melanogaster: (1) embryos with

decreased Dl levels, (2) decreased nuclear size with high nuclear density, and (3)

increased nuclear size with low nuclear density. Once we obtained adjusted

parameters for D. melanogaster that also satisfied these extended conditions, we next asked which parameters from this representative set were most likely to be modified in Drosophila species that display distinct Dl gradient shapes. To this

end, we selected a divergent species with small embryos, Drosophila busckii, and

two additional pairs of species belonging to the melanogaster subgroup,

Drosophila simulans/Drosophila sechellia and Drosophila santomea/Drosophila

yakuba, which diverged from D. melanogaster between 5 to 6 MYA (Fig. 9C)

(David et al. 2007; Lachaise et al. 1986, 2000). These species give us the unique opportunity to assay the behavior of the Dl gradient in lineages that have undergone a separate speciation event, but share some commonalities. For example, D. sechellia and D. santomea diverged very recently from their ancestral siblings D. simulans and D. yakuba, respectively, at an estimated 0.3-0.5 MYA.

Despite such short divergence time, D. sechellia and D. santomea have much larger embryos than their siblings (Lott et al. 2007). The use of modeling gave us insights in the evolution of Dl gradient shapes that are in agreement with the phylogenetic relationships of the species analyzed. We show that although the modified embryonic anatomy of these species influence the Dl gradient distribution, the species-specific Dl gradient shapes also depend on genetic modifications in the Tl pathway, which are shared in closely related species pairs.

3.3 Results

3.3.1 Reconstruction of the Kanodia model reproduces the Dl gradient

shape in some mutant conditions

We are interested in understanding how the Dl gradient acquired distinct shapes in related Drosophila species. One notable phenotypic difference reported in several Drosophila species is the significant variability in egg size (Lott et al.

2007; Markow et al. 2009). In addition, the nuclear size and density also vary in these species (Fowlkes et al. 2011). We previously showed that manipulations in nuclei size and density in mutant D. melanogaster embryos can recreate Dl

80 gradient shapes that are found in nature, leading us to hypothesize that nuclei density and size changes might be sufficient to modify the Dl gradient shape.

The mutation sesame (ssm) generates haploid embryos that undergo an additional round of mitotic division, causing a high nuclear density and decreased nucleus size (Fig. 10A, B) (Loppin et al. 2000). In these mutants, the Dl gradient becomes flattened (Fig. 10D, E). The second mutation used, gynogenetic-2, gynogenetic-3 (referred to as gyn (Fuyama 1986)), generates triploid embryos that stop dividing one cycle earlier causing a lower nuclear density and larger nucleus size compared to the wild type diploid embryos (Fig. 10C). The Dl gradient of gyn embryos is sharper than the wild type (Fig. 10F). These mutations are known to affect ploidy, but are not expected to alter components in the Tl signaling pathway or embryo size. One way of explaining the altered Dl gradients of these mutants would be if the density of nuclei modifies the reading of Tl signal from one nucleus to the next, and consequently the rate of Tl signal decays (Fig. 10G-

J). In this scenario, a lower versus higher nuclei density would lead to a steeper and a flatter gradient, respectively (Fig. 10L). In addition, since these mutants have the same amount of maternal Dl protein, the increase in nuclei density would decrease the amount of Dl per cell compartment and flatten the gradient. Another consideration is the differences in nuclei size, which increases the surface area available for Dl transport. Thus, nuclear size may counterbalance the effects of nuclei density.

Since a qualitative analysis would not be sufficient to predict all of the combined effects described above, we employed a numerical approach using a

modified version of the Kanodia model (See Text S1 available online). We used the same values of a representative parameter set used in the original MATLAB code to reconstruct the original model and run simulations of the wild type gradient formation using Wolfram’s Mathematica, which successfully reproduced key features of the model (Text S1 and Tables S1-3 available online). In principle, any set within the restricted cloud of parameter sets identified by Kanodia et al.

(Kanodia et al. 2009) could be used to model the Dl gradient and investigate qualitative changes to simulate the mutant gradients. We then asked if the shape of ssm and gyn gradients were altered from the onset of the Dl gradient formation, at nuclear cycle 10 (nc10). One of the Kanodia model findings was that the wild type Dl gradient has a constant shape throughout the nuclear cycles, which matches experimental data (Kanodia et al. 2009). We initially tested the effect of changing nuclear radius in the wild type from nc10 to nc13 over the final gradient shape at the last stage (nc14) (Figure S2 available online). We found that altering

the size of nuclei modifies the shape of the Dl gradient at early stages, but does

not affect its final shape at nc14. Since we are most interested in the gradient

shape at the final cycle nc14, and not the dynamics of the mutant gradients, this

result indicates that the effect of incorrect assumptions about early cycles is

minimized.

3.3.2 Simulation of nuclei numbers and size can reproduce ssm gradient,

but not the gyn gradient

We next attempted to reproduce the Dl gradients from ssm and gyn mutants by

using the selected representative set of parameters from Kanodia et al.(Kanodia et

82 al. 2009) and adjusting it for nuclei size and density according to our experimental measurements (Fig. 11) (Table S4 available online). Few additional parameters were changed, especially related to early cycles (Text S2 available online), but given the model robustness these changes did not significantly affect our results.

We also normalized the model output to match our experimental data (see

Methods), which is restricted to the 30 most ventral cells instead of the entire embryonic cross section (Fig. 11A-C). This ventral region includes the entire mesoderm and few additional cells in wild type and mutant embryos, and encompasses reliably measurable levels of nuclear Dl with distinguishable signal from background noise. This also represents the region where significant variations in the gradient shape are present (Chahda et al. 2013). With our normalization, we represent the overall shape of the Dl gradient instead of absolute values of Dl concentration (Fig. 11D-F). Unless otherwise noted, the normalized gradient restricted to the 30 most ventral cells is referred to as “Dl gradient”. Given the graded levels of nuclear Dl, we also verified that the variations in the net numbers of mesodermal cells between wild type and mutant embryos do not alter the overall shape of the gradients after normalization.

In non-normalized graphs, our simulations show that ssm embryos have the highest peak of nuclear Dl concentration, while gyn embryos have the lowest peak

(Fig. 11B). Thus, even though ssm has a smaller amount of Dl per cell compartment and nuclear surface area available for Dl translocation, the model predicts that its smaller nuclear volume is the major determinant of the absolute concentration of nuclear Dl. In terms of Dl gradient shape seen in normalized

graphs, the model correctly reproduces the flattened ssm gradient, but not the

steep gyn gradient, which instead appears with the same shape as wild type (Fig.

11E-F).

The fact that the model can reproduce the ssm but not the gyn gradient points

to two non-exclusive deductions: (1) changes in nuclei density and size are sufficient to explain the ssm distorted gradient, but not gyn, i.e. our hypothesis is

only partially correct; and (2) the parameter set used creates a strongly artificial

robustness, buffering the effect of our manipulations. To investigate if our

manipulations were being buffered, we first tested the individual effects of nuclei

density and size on the Dl gradient shape. We found that either higher nuclei

density or larger nuclear size result in a flattened gradient (Fig. 11G,H), indicating

that the flattened ssm gradient is mostly determined by its higher nuclei density,

which overrides the effect of its smaller nuclei. In contrast, the effect of larger

nuclei in gyn was only slightly compensated by its reduced nuclei density,

resulting in a Dl gradient shape similar to wild type in our simulations, rather than

the steep gradient obtained experimentally.

The results above suggest that some of the assumptions that apply to wild type

and ssm may not apply to gyn. One possibility is that one or more general

parameters, such as Dl diffusion rates and Dl nuclear export rates are different

from the values employed in the model, but they have a more significant effect

under the gyn conditions. We also observe that the wild type simulation is not

completely satisfactory, suggesting that this representative parameter set used

84 could be further improved. We next modified the model to determine which parameter combinations could better reproduce our experimental Dl gradients.

3.3.3 Refinement of the parameter values reveals the embryo geometry plus

Dl diffusion and export rates play major roles in the model

reproduction of the gyn gradient

To increase the model flexibility and allow testing the effects of individual parameter changes, we used dimensionalized equations and focused on the simulation at the last nuclear cycle only (see Methods). In the original model, 9 dimensionless parameters were used, in addition to nuclei radius and density, developmental timing and cell compartment volume at nc14. In our modified model, a total of 19 parameters can be manipulated independently (Table 3), and their effects on the Dl gradient shape can be directly analyzed. The original values of most of these 19 parameters could be estimated from the representative nondimensionalized parameter set chosen here, while others were determined by direct measurements and assumptions (Text S3 available online). Revisions of the parameter values from this set were performed by manually testing a combination of parameters able to reproduce the gradients from wild type, ssm, and finally gyn.

To further validate our revised parameter set, we also quantified Dl from D. melanogaster embryos derived from dl-/dl+ heterozygote mothers (referred to as dl-/dl+ embryos for simplicity) and tested the model ability to reproduce this mutant Dl gradient. These embryos have normal embryo size and Tl signaling, but only half of normal Dl protein amount.

The analysis of dl-/dl+ embryos provided valuable insights about the

model parameters. In agreement to a previous report (Louisa M Liberman,

Reeves, and Stathopoulos 2009), we verified that these mutants have a flattened

Dl gradient (Fig. 12), which suggests that near the ventral midline, all cytoplasmic

Dl is translocated into the nucleus. Therefore, it is reasonable to assume that in the

wild type, the Dl nuclear import rate (ki) is not the limiting factor for the

formation of the gradient peak. In other words, given enough Tl receptor

activation and cytoplasmic Dl, peak levels of nuclear Dl can be achieved in the

wild type. This result motivated us to increase the ki value (Table 3). We next

asked if decreased Dl levels could simulate the dl-/dl+ gradient shape. However,

our model showed that the shape of the Dl gradient is insensitive to the initial

concentrations of Dl, Cact and Dl-Cact, unless these initial concentrations are zero, in which case the Dl gradient is not formed. This finding suggests that the gradient shape observed in dl-/dl+ embryos is caused by additional parameter

changes besides initial Dl concentration.

Several studies report that Cact is stabilized in the presence of Dl and that

Cact levels are reduced if Dl levels are diminished (Belvin, Jin, and Anderson

1995; Drier, Govind, and Steward 2000; Whalen and Steward 1993). Based on

this information, we tested if changes in the rate of Cact degradation (kDeg) were

able to reproduce the mutant gradient. We found that doubling the wild type kDeg

value was not sufficient to completely reproduce the dl-/dl+ flattened gradient.

This finding suggests that the relationship between Dl amounts and Cact stabilization is not linear and probably involves cooperativity. Indeed, Dl is

reported to form dimers, such that the Dl-Cact complex is formed by one unit of

Cactus bound to two units of Dl (Drier et al. 2000; Whalen and Steward 1993).

By increasing kDeg four times, our model could correctly reproduce the Dl

gradient from dl-/dl+ embryos (Fig. 12, Table 3).

After implementing this adjusted parameter set, our simulations still failed

to reproduce the gyn gradient, unless three additional changes were made: (1) an

increased diffusion rate among compartments, (2) an increased Dl nuclear export

rates, and finally (3) an increased embryo radius (Fig. 13). Increasing Dl, Cact

and Dl-Cact transport rates between adjacent compartments (Γ) from 0.03 to 2

sharpened the gyn gradient simulation (Fig. 13C, simulation 1) and still kept a

good fit between the simulated and experimentally obtained gradients of wild

type, ssm, and dl-/dl+ embryos (Fig. 13A, B, D). Indeed, the fit was actually

improved for the wild type (Fig. 13B, black dots. Table S5 available online). The

most common value of transport rate within the parameter vectors in the Kanodia

model is 0.0064. We verified that the transport rate constant of 2 tested here falls

within the parameter vectors found in the Kanodia model, albeit at low frequency

(Kanodia et al. 2009). By also increasing the Dl nuclear export rate (ke) from 0.44

to 1, the fit for gyn improved significantly (Fig. 12C, simulation 2) without

compromising the wild type (Fig. 13B, black dots) and ssm simulations (Fig. 13D,

black dots), and only having a small increase of the Dl peak levels in the dl-/dl+

gradient simulation (Fig. 13A, black dots; Table S5 available online).

In sum, increasing diffusion across compartments and Dl export rates

greatly improved the gyn gradient simulation and did not impact significantly

87 other D. melanogaster mutants and wild type simulations. Finally, an almost perfect fit for the gyn gradient was obtained by increasing the embryo radius (Fig.

13C, simulation 3). Although the main motivation to use ssm and gyn mutants to test the influence of nuclei size and density was the fact that these mutants should a priori have wild type egg sizes and a normal DV signaling pathway, actual measurements indicate that gyn has a slightly larger radius of 117 µm in comparison to 100 µm in the wild type.

The lack of a perfect simulation of the steep gyn gradient may be due to simplifications in the model. For instance, while Dl-Cact dissociation is the main response to Tl activation, the removal of Dl and Cact interaction is insufficient to promote maximum peak levels of nuclear Dl (Drier et al. 2000). However, the model does not include alternative pathways for Dl nuclear translocation or possible interactions between Dl and other IkB related proteins. Also, the model does not represent other alternative DV polarizing sources involving components upstream of Toll, but the effect of this second polarizing signal is reported to be subtle and may not necessarily have measurable effects in a wild type background

(Zhang et al. 2009). Together, our simulations nonetheless clearly indicate that embryo morphology affects the Dl gradient shape, and is likely to play an important role in the modifications seen in the other Drosophila species.

3.3.4 Embryonic Morphology alone does not fully explain species-specific

Dl gradient shapes

Since embryo geometry, nuclear density and nucleus radius affect the

shape of the Dl gradient, we first addressed how this restricted set of parameters

act together to generate the species-specific Dl gradients we analyzed previously

(Chahda et al. 2013). Embryos from D. busckii, D. simulans and D. sechellia have

different sizes and geometries, as well as distinct nuclear density and size (Table

4). After adjusting these parameters to the values obtained experimentally, we

verified that the model fails to reproduce the species-specific Dl gradients (Fig.

14, Table 4, simulations 1). Furthermore, additional simulations also discarded other parameters relative to morphology with no significant impact to the gradient shape; namely embryonic AP length, width of cortical layer and the total number

of nuclei in the entire embryo. We conclude that the evolutionary morphological

modifications in these species alone are not sufficient to generate their final Dl

gradient shape.

3.3.5 Modulating a small subset of parameters affecting the Toll signaling

pathway can reproduce species-specific Dl gradients

We reasoned that the next logical step requiring minimal model

manipulations to achieve good gradient fits for the species should involve

adjusting parameters that regulate the Tl signaling pathway. This idea is supported

by the fact that the Tl pathway is a fast-evolving pathway in Drosophilids, which

is required for immune response in addition to DV patterning (Clark et al. 2007;

Jiggins and Kim 2007; Obbard et al. 2009; Schlenke and Begun 2003; Sousa-

Neves and Rosas 2010). Furthermore, we previously showed that this pathway is indeed modified in the species, as seen by their distinct ranges of peak Tl

activation levels measured as the percentage of arc-length occupied by the

mesodermal marker sna (Chahda et al. 2013). This variation goes from 21% in D.

melanogaster to 17% in D. busckii, 26% in D. sechellia and 27% in D. simulans.

We tested the effect of three parameters (R, S and ξ) that influence the amplitude

and shape of the space-dependent Dl-Cact dissociation rate constant (kD) and as

such control the range of Tl signaling strength extending dorsally from the

midline. By modifying either R or ξ, we could obtain simulations with good fit for

each species. For instance, D. sechellia gradient can be reproduced using an R value of 50,000, but D. simulans requires a much larger value of 114,000.

However, such large difference in R values is not supported by the experimental data showing these two species have nearly identical mesodermal percent arc-

length (Chahda et al. 2013).

Assuming a linear relationship between the percent arc-length of the mesoderm and R, we tested adjusted R values of 12,142 (busckii), 19,285

(simulans) and 18,571 (sechellia). These more modest changes in R slightly improve all simulations (Fig. 14, simulations 2; Table 4). Most importantly, the gradients are correctly reproduced by few additional changes in Tl pathway parameters, and these changes agree with the phylogenetic relationship of these species. For instance, in the two most closely related species D. simulans and D. sechellia, either increasing Cact degration rates (kDeg) or reducing Cact production

rates (PCact) can correctly simulate their gradients (Fig. 14C, D, simulations 3;

Table 4). In other words, the significantly different gradients observed in these

species, which vary in nuclear and embryo size, are generated by changes in the

same parameters and place them apart from D. melanogaster. In contrast, the

model predicts that D. busckii, a more distantly related species from the

melanogaster subgroup, requires an opposite change over Cact regulation, i.e., a

decrease in kDeg or increase in PCact in order to simulate its gradient (Fig. 14B,

simulation 3; Table 4).

To gain further insights about the more closely related species D. simulans

and D. sechellia, we tested additional parameters that regulate Dl and Cact functions. We found that decreasing binding of Cact to Dl (kb) also generates a

good fit for these two species (Fig. 13C, D, simulations 4; Table 4). Another

prediction made by the model was that decreasing Dl export rates in both D.

simulans and D. sechellia can also improve the simulation of their gradients.

Finally, by simultaneously modifying more than two Tl-related parameters at a time, we also obtained good fits for D. simulans and D. sechellia (Fig. 14C, D,

simulations 5; Table 4). We also observe the same overall model behavior when

using various randomly generated parameter sets within the range of the

parameter cloud identified in the original model (Kanodia et al. 2009) (Figure S7

and Table S6 available online). These results indicate that the very distinct Dl

gradient shapes found in these closely related species can be correctly simulated

by making similar modifications in selected parameters involved in Tl pathway.

3.3.6 Analyses of another pair of closely related sibling species suggest

evolutionarily shared mechanisms for Dl gradient formation

As seen above, our simulations indicate that making similar adjustments in

parameters that affect Cact regulation or Dl export rates generate good fits for D.

91 simulans and D. sechellia. We expected that the model could reveal if there were common evolutionary mechanisms for the formation of the Dl gradient in another pair of sibling species, D. santomea and D. yakuba, which would also set them apart from D. melanogaster. D. santomea emerged as recently as D. sechellia

(Fig. 9C) and is also reported to have enlarged egg size (Lott et al. 2007), but the speciation of these two species took place in geographically distinct regions

(David et al. 2007).

We obtained measurements of embryo size, nuclear size and density for these species (Table 5). Dl quantifications in both D. yakuba and D. santomea reveal an overall gradient shape similar to D. melanogaster and D. sechellia, except for slightly lower peak levels in D. yakuba. Interestingly, the percent arc- length of sna in D. yakuba and D. santomea (22.06%, SD=1.92, n=5; and 20.44%,

SD=1.54, n=5, respectively) is similar to D. melanogaster, suggesting that the broadening of Tl range is an innovation in the branch of D. simulans and D. sechellia.

After adjusting the model parameters with the D. yakuba and D. santomea measurements of embryo, nuclear size and density, the resulting gradients were sharper than the experimentally measured gradients (Fig. 15, simulations 1). We were able to correctly simulate their gradients by modifying parameters related to the Tl pathway, such as decreasing kb, or increasing kDeg to a same value in both species (Fig. 15, simulations 3 and 4; Table 5). Decreasing Dl export rates also improves the simulations, but a comparison of Dl protein sequence did not indicate modifications in the Nuclear Export Sequences (NES) from D.

92 melanogaster (see below). In sum, our model indicates that in the D. yakuba and

D. santomea lineages, the Dl gradient formation appears to depend on similar modifications in Cact regulation, setting these species apart from D. melanogaster as was the case for D. simulans and D. sechellia.

3.3.7 Dl and Cact protein sequence comparisons of melanogaster subgroup

species support predictions made by the model

To further investigate the biological relevance of Cact regulation and Dl export rates in the formation of the species-specific Dl gradients, we analyzed the amino acid sequences of these proteins from the melanogaster subgroup species.

We aligned D. melanogaster Dl with D. simulans and D. sechellia Dl sequences and found that all known functional domains of the protein are conserved, with the exception of the nuclear export sequence 3 (NES3), which contains 3 amino acid (aa) substitutions in D. simulans and D. sechellia (Fig. 17A). These changes could potentially decrease Dl export rates in these species (Isoda, Roth, and

Nusslein-Volhard 1992; Xylourgidis et al. 2006), as predicted by our model. In contrast, D. yakuba and D. santomea exhibit identical sequences of all NES domains to D. melanogaster.

Although the model equations do not capture the full complexity of the

Cact degradation pathways in vivo, the comparison of Cact sequences from these species also provided further support for possible changes in its regulation (Fig.

17B). The Cact C-terminal contains six ankyrin repeats (Gay and Ntwasa 1993) which are necessary for its binding to Dl. We found that D. simulans and D. sechellia contain an insertion of 15 aa within the beginning of ankyrin repeat 4.

Using Phyre2 software to predict protein structure (Kelley and Sternberg 2009), we verified that this insertion does not eliminate this ankyrin motif itself but it may create two α-helixes between ankyrin repeats 3 and 4, in contrast to only one long helix present in D. melanogaster Cact. Likewise, D. yakuba Cact is also predicted to have two α-helixes in the same region, due to some nearby aa substitutions. It is possible that the alteration nearby the ankyrin domains could modify the binding between Dl and Cact in these species, which would further support the model prediction that using a lower kb rate than D. melanogaster yield good fits of the other species simulations.

Another important regulatory region in Cact sequence is located in the N- terminal (Fig. 17B). This region is rich in serine residues that are phosphorylated in response to Tl activation, leading to Cact degradation. D. simulans contains only one serine substitution (S94R) in relation to D. melanogaster, but this site has never been tested for its function in vivo. D. yakuba contains more Cact modifications in relation to D. melanogaster, with a total of 18 aa substitutions, including 4 serine substitutions. In addition, D. yakuba Cact has a deletion of 9 aa at positions 124-132, nearby a domain previously implicated in Cact degradation in vivo (Fernandez et al. 2001). Together, these variations in Cact and Dl suggest that subtle and additive, but possibly biologically relevant changes in components of the Tl pathway are shared by the most closely related species and may contribute to their final Dl gradient shape, as suggested by our model simulations.

3.3.8 Model robustness and sensitivity analysis reveals non-linear

interaction between species morphology modifications and other

relevant parameters

We next carried out a sensitivity analysis to test how robust the model is to simultaneous changes in the relevant parameters identified above. Instead of an exhaustive test for all possible combinations of parameter values, we focused on the effects on the model output when changing only two concomitant parameters at a time. We observe that for most combinations tested, the simulations stay within robust regions of the model (Fig. 16A). Two simulations in particular tend to fall within slightly more unstable regions of the parameter range, namely gyn and D. busckii (Fig. 16B; also see Fig. S9E, L-K, S, U available online).

Our analysis confirmed that the model is indeed sensitive to changes in two key parameters for reproducing the species gradients, Cact degradation (kDeg)

and Dl-Cact binding rates (kb) (Fig. 16B, D). Furthermore, we observe non-linear interactions between kDeg and kb and parameters related to embryonic morphology

(Fig. 16B, D). For instance, in D. simulans and D. sechellia, changes in Dl-Cactus binding rates (kb) affect the Dl gradient distribution outcome caused by changes in

embryonic radius (Er; Fig. 15D) and nuclear radius (r). In addition, while the

model is mostly robust to changes in nuclear Dl export rates (ke; Fig. 16C; also

see Fig. S9 R, T available online), it does display more sensitivity to ke when

paired with the species-specific nuclear radius (Fig. 16C) and embryo radius

variations (Fig. S9Q available online). Together, these results support an overall

robustness of the model simulations and reveal an interaction between

morphological modifications and the few selected parameters of Tl pathway

regulation that improve the species-specific simulations.

3.4 Discussion

Variation in species size creates a challenge on how gene expression patterns can accommodate to new embryonic dimensions without compromising cell fates and viability. Our study of DV patterning response to physical and biochemical changes in mutants and Drosophila species provided new insights on

Dl gradient scaling. First, the model indicates that changes in parameters such as embryo size and nuclear size, which are commonly found in several Drosophila species, are not sufficient to recreate the Dl gradient shapes seen in these species.

However, these parameters interact with a small subset of parameters related to Tl pathway, which when modified, are sufficient to generate simulations with good fits with the experimental Dl gradients. Our results also suggest that those changes in Tl pathway are likely to have been shared within closely related lineage branches, which is further supported by the sequence comparisons of Dl and Cact proteins from these species. Thus, the mathematical modeling used here advances our understanding on how gradient shapes are acquired during evolution, which could not be explained by solely quantifying and comparing Dl levels across species.

3.4.1 Dorsal scaling within and across species

Garcia et al. (Garcia et al. 2013) recently investigated the Dl gradient scaling within the same species using D. melanogaster lines artificially selected to have small or large embryos (C. M. Miles et al. 2011). Their study indicates that the Dl gradient width is positively correlated with DV axis length and the number

96 of nuclei along the DV axis. Our experimental data from ploidy mutants and mathematical simulations support the claim that an increase in the number of DV nuclei causes a widening of the Dl gradient. Garcia et al. (Garcia et al. 2013) also suggest that changes in the range of Tl signaling could explain the observed scaling of the Dl gradient width within D. melanogaster species. We previously found variations in the range of peak Tl signaling across species (Chahda et al.

2013), and in this work we provide evidence for species-specific changes within the Tl signaling pathway as a means of influencing the Dl gradient shape.

We also show that increasing Dl nuclear export rate and diffusion between cellular compartments more accurately recreates D. melanogaster wild type and mutant Dl gradients (Fig. 13). With regard to diffusion rates, the majority of parameter sets found in the Kanodia model is in agreement with a cell autonomous steady state behavior, which is supported by live-imaging experiments showing that a GFP-tagged version of Dl has limited diffusion between neighboring compartments (DeLotto et al. 2007). We verified that our adjusted diffusion rates do not exclude the possibility that the embryo is fully compartmentalized, but we also observe that the final Dl gradient shape is influenced by a non-cell-autonomous process. Future work testing native Dl diffusion without GFP may resolve whether the Dl gradient formation is a non- cell-autonomous process with increased lateral diffusion that may be required for scaling the final gradient shapes observed in nature. The difference in embryo morphology across species is also expected to either increase or decrease the diffusion of Dl by itself, as it has been shown before in experiments that measured

diffusion constants of injected dextran in species with small and large embryos

(Gregor et al. 2005). These experiments revealed a trend of increased dextran diffusion in large embryos versus decreased diffusion in small embryos.

Consistent with this finding, we also note that our calculated diffusion coefficient of Dl, Cact and Dl-Cact slightly increases in larger embryos (e.g. D. sechellia and

D. santomea) and decreases in smaller embryos (e.g. D. busckii), after doing a

unit conversion of Г that inputs the measured values for embryo morphology.

Prior work showed a scaling of the antero-posterior gradient Bcd in the

inbred D. melanogaster lines mentioned above (Cheung et al. 2011, 2014) and proposed a mechanism in which more maternal bcd mRNA is loaded into larger embryos to compensate for their increased size. With respect to the Dl gradient, an increase in nuclear Dl concentration can occur with or without a corresponding increase in embryo size or altering the maternal contribution of Dl. For instance, we found that D. sechellia and D. santomea do have greater concentrations of Dl in ventral nuclei in relation to their smaller sibling species D. simulans, D. melanogaster and D. yakuba (Chahda et al. 2013). However, despite the fact that

D. simulans produces embryos of comparable size to D. melanogaster, the nuclear

Dl concentration levels in the former species are more elevated (Chahda et al.

2013). We show that changes in nuclear size and density, range of peak Tl activation and changes within the Tl signaling pathway provide additional strategies to altering nuclear Dl concentrations and distributions, which can work in conjunction with altering the maternal dosage of Dl.

3.4.2 Model sensitivity analysis support evolution of Dl gradient by small

additive changes in Tl regulation pathway

Two interesting properties of this system emerged from our robustness and

sensitivity analysis. First, it can be seen that lowering Dl nuclear export rates for

D. simulans and D. sechellia allows the model output to change from a flat to a

sharp gradient shape after correcting for the species-specific nuclear radius (Fig.

16C, white arrows). A similar non-linear interaction is observed between Dl-Cact

binding constant (kb) and nuclear radius. Second, we notice that for D. simulans

and D. sechellia, the simulations stay within robust regions when more than two parameters are modified at a time (e.g. ke, kDeg and kb, Table 4, simulations 5). In

contrast, simulations that sharply decrease only one parameter at a time in D.

simulans, such as decreasing Dl-Cact binding rates (kb) (e.g. Table 4, simulation

4) fall within more unstable regions (Fig. 16D, yellow arrow; also see Fig. S8 F-J,

yellow dots, available online). In the case of D. busckii, we also note that the

simulations fall within more unstable regions of the model upon changes in the

parameter of Cact degradation (kDeg) only. These results suggest that it is unlikely

that these species acquired their Dl gradient shapes by drastic regulatory changes

that affect only one component of the Tl pathway.

3.4.3 The Dl gradient model predicts changes in the Tl pathway in

Drosophila species that are consistent to their phylogenetic

relationships

The results obtained from the use of computational modeling revealed

important properties about the behavior of gradient formation and evolution of the

Tl signaling pathway in the Drosophila species tested. First, our analyses of ssm

and gyn mutants demonstrate that the rapid changes in embryo size, nuclear size

and density of these species can modify the Dl gradient shape, but those changes

alone are not sufficient for the final species-specific Dl gradient shapes. The

second significant prediction made by the modeling is that additional changes in

the Tl pathway regulation are required for obtaining good fits with the

experimental gradient shapes in these species.

At first, it was surprising to find that the gradients of the most distantly

related species (i.e. D. sechellia and D. melanogaster) have an identical

distribution, whereas the gradients of the closest related species D.

simulans and D. sechellia acquired completely different shapes. However,

predictions made by our model reconcile the fact that these quite different

gradient shapes can in fact be generated by a similar dynamics of Tl signaling in

D. simulans and D. sechellia, after adjusting for their divergent anatomy. First, as

suggested in our previous work, the range of peak Tl activation is broader in D.

simulans and D. sechellia, compared to D. melanogaster. Second, our present data

suggest that additional modifications in components of the Tl pathway affecting

Cact regulation and Dl export rates also diverged in the newest species. By

altering these parameter to similarly higher (e.g. increased kDeg) or lower values

(e.g. decreased ke or kb), good fits of the gradients are generated for both D.

simulans and D. sechellia (Fig. 14, Table 4; see also Table S5 available online). In support of these findings, we verified that D. simulans and D. sechellia share similar changes in amino acid sequences of Cact and Dl within or near domains

100

previously implicated in Dl-Cact binding and Dl nuclear export, respectively (Fig.

15).

The simulations of Dl gradients in another closely related pair, D. yakuba

and D. santomea, also suggested shared modifications in Cact regulation. Either

lowering Dl-Cact binding rates (kb) or increasing Cact degradation (kDeg) to same

values can generate good fits for the gradients of these species (Fig. 15, Table 5).

Genomic data available for D. yakuba confirmed the prediction that Cact sequences within domains involved in degradation and Dl binding are indeed modified in relation to D. melanogaster. In contrast, the Dl protein domains in D. yakuba are well conserved in relation to D. melanogaster. We partially sequenced

Dl from D. santomea and found that these domains are similar to D. yakuba.

In sum, despite the fact that melanogaster subgroup species have particular egg sizes, nuclear size and density, and their Dl gradient shapes appear at odds with their phylogenetic relationships, the use of mathematical modeling reveals that most closely related species share similarly modified regulation of the

Tl pathway inherited from their common ancestor.

3.5 Material and Methods

3.5.1 Fly stocks yw D. melanogaster was used as wild type. Haploid and triploid embryos were generated in our previous work (Chahda et al. 2013) using the mutations sesame

(ssm) (Loppin et al. 2000) and gynogenetic-2; gynogenetic-3 (gyn) (Fuyama

1986). The D. busckii, D. sechellia and D. simulans strains used in (Chahda et al.

101

2013) were obtained from the Drosophila Species Center at UCSD. The D.

yakuba (tai 6 line) and D. santomea (CAR 1495.5 line) stocks were obtained from

Daniel Matute (Univ. of Chicago).

3.5.2 Gradient quantification and measurements of nuclear size

Quantification of Dorsal gradient and normalization method are described in

detail in (Chahda et al. 2013). Briefly, embryos were stained for anti-Dorsal

antibody (Iowa Hybridoma Bank) and a Donkey anti-mouse Alexa 647, manually

sliced in cross-sections within trunk region and imaged using a LSM700 Zeiss

Confocal microscope. Fluorescent intensity from the 30-most ventral nuclei was

obtained using Axiovision software (Zeiss). Position of midline was estimated

with a double staining for snail RNA. For nuclei diameter measurement, early-

stage embryos stained with anti-Laminin (Iowa Hybridoma Bank) were mounted

longitudinally with glass beads (150-210 µm size, Polysciences), to prevent

flattening caused by the coverslip. Confocal slices were taken from the embryo

surface to its mid-section and nuclei diameter was determined using ImageJ

software. In the case of ssm and gyn mutations, some additional measurements

were taken from embryos stained with DAPI nuclear dye.

3.5.3 Reproduction and modification of the Kanodia model in Mathematica

The nondimensionalized model of nc10-14 was reproduced as described by

Kanodia et al. (Kanodia et al. 2009). Simulations of gyn and ssm gradients employed same equations, with the following genotype-specific changes in the parameter values. Nuclei radius and density along the DV axis at the last nuclear cycle were directly measured as described above and in (Chahda et al. 2013).

Total embryonic nuclei density at final cyles in ssm (nc15) and gyn (nc13) were

102

estimated at 1200 and 3000, respectively, based on the fact that wild type

embryos have an estimated 6000 nuclei at nc14 and on previous data for haploid

embryos (Edgar et al. 1986). The number of nuclei along the DV axis (n) at early cycles was obtained as in Kanodia et al. (Kanodia et al. 2009) by multiplying n

by 2 after each cycle. In our modified model, we adjusted the final number of

DV√ nuclei at nc14 for D. melanogaster wild type from 100 to 92, as

experimentally obtained in (Chahda et al. 2013). Adjustments for nuclei size in

early nuclear cycles and developmental timing in ssm and gyn embryos are

explained in Text S2 (see also Table S4 and Fig. S4), and were estimated based

on (Edgar et al. 1986; Gregor et al. 2007; Lu et al. 2009). Parameter changes for

other Drosophila species were done according to data measured here, and in

previous work (Chahda et al. 2013; Fowlkes et al. 2008; Kanodia et al. 2009) as

described in the main text.

3.5.4 Dimensionalized model of the last nuclear cycle

Dimensionalized equations were written in Mathematica using original mass-

balance equations from the Kanodia model. Additionally, to better represent the changes in embryo volume between species, instead of linearizing the cellular compartments as in the original model, those were represented as circular trapezoids organized in a circle. The whole cross-section was modeled, with no need for no-flux boundary conditions.

3.5.5 Model validation

The original Kanodia model was validated here against three mutant conditions

within the same species D. melanogaster (dl-/dl+, ssm, gyn). Manual adjustments

in the parameter ki was made for dl-/dl+, and adjustments of Г, ke were made for

103

gyn (See main text). Those same values were maintained for a second round of

simulations for wild type, ssm and dl-/dl+, which served as internal validation

controls. Fit between experimental and simulation graphs remained roughly

similar for ssm and dl-/dl+, and it was improved for wild type.

3.5.6 Fit calculation and confidence intervals

For fitness comparison, the square root of the square differences between the

simulated gradients and respective experimental data was calculated. Standard

deviation of the mean (SD) are indicated by error bars (Fig. 10) or shadowed

areas (Fig. 12-15). Pink shadowed area in Fig. 4 (dl-/dl+ mutants) indicates SD,

gray shadow indicates the 99% confidence interval for the experimental mean, as

explained in the figure legend. Simulations “1” for D. busckii, D. simulans and D.

sechellia (Fig. 14) lie outside of the 99% confidence interval (not shown), and are

statistically different from best fit simulations (black dots). Simulations “1” for D.

yakuba and D. santomea (Fig. 15) lie within the 99% confidence interval.

However, even though there is no statistical significance, simulations “3” and “4”

have an improved fit according to our fit calculations.

3.5.7 Sequence comparison of the Dl and Cact in melanogaster subgroup species

Available coding sequences of Dl and Cact for D. simulans, D. sechellia and D. melanogaster were obtained from FlyBase and aligned using tblastn (NCBI). For

D. santomea, genomic DNA was amplified, sequenced and analyzed as described above. Fig. 16 summarizes the comparison for the sequences obtained. Protein structure analysis was done using Phyre2 software

(http://www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index). Location of Cact

104

and Dl conserved domains was based on previous work (Fernandez et al. 2001; Ip

et al. 1991; Isoda et al. 1992; Kidd 1992; Xylourgidis et al. 2006).

3.6 Acknowledgements

We are grateful to Stas Shvartsman (Princeton University) and Jitendra Kanodia

for sharing the MATLAB code of the Dl model, and Daniel Matute (University of

Chicago) for sending us the D. yakuba and D. santomea strains. C.M.M. is indebted to R. Sousa-Neves for helpful discussions. The anti-Dl antibody

developed by Ruth Steward was obtained from the Developmental Studies

Hybridoma Bank, created by the NICHD of the NIH and maintained at The

University of Iowa, Department of Biology, Iowa City, IA 52242.

105

3.7 Tables and Figures

Table 3. Parameter values used in model simulations for D. melanogaster wild type and mutant conditions shown in Figures 10 and 11. Revised parameter values discussed in the text are indicated in bold.

106

Table 4. Selected parameter sets used in model simulations of Dl gradient for three different Drosophila species, D. busckii, D. simulans and D. sechellia. Bold numbers indicate modified values in relation to D. melanogaster.

107

Table 5. Selected parameter sets used in model simulations of Dl gradient for Drosophila yakuba and Drosophila santomea. Bold numbers indicate modified values in relation to D. melanogaster. Simulations 3 and 4 (shown in Fig. 13) provide the best fit.

108

Figure 9. Dl gradient model rationale. A) The embryo is modeled as a single string of n cuboid cellular compartments. B) Chemical reactions and transport processes considered by the Kanodia model (Kanodia et al., 2009). The DV Toll receptor activation gradient (red) is represented by the space-dependent Dl-Cactus dissociation constant (kD) and results in higher nuclear concentrations of Dl (gray) at the ventral side of the embryo (V) and higher cytoplasmic concentrations at the dorsal side (D). (Parameters shown in blue are explained in Table S2 available online). C) Phylogenetic tree of melanogaster subgroup species.

109

Figure 10. The Dl gradient is modulated by changes in nuclear size and density. A–C) Increasing nuclear size and decreasing nuclei density from haploids (ssm, B) to diploids (A) to triploids (gyn, C) stained with anti-laminin (magenta). D–F) Normalized graphs showing distinct Dl gradient shapes from D. melanogaster (D), ssm (E) and gyn(F, mean±SD). G–J) Cross-section schemes for wild type (G), ssm (H) and gyn (I), and aD. melanogaster embryo (J) representing the Toll signaling gradient. K) Simulated Toll signaling gradient based on the equation for kD, the space-dependent Dl-Cactus dissociation constant. As illustrated in (J), nuclei density affects the angle subtended by 30 cells in a cross- section, resulting in a larger rate of Toll signal decay for gyn and a smaller rate for ssm. (L) Normalized Toll signaling gradients, emphasizing the relationship between the simulated Toll signaling gradient and experimental Dl gradients. Figures (A–F) were modified from (Chahda et al. 2013); V, ventral midline; color-coded arrowheads in D–J delimit the 30 ventral most cells.

110

Figure 11. Comparison between experimental data and model output. A) Simulations of nuclear Dl levels at the last nuclear cycle of each genotype for the entire cross-section. Note that the cross-section has the same size in all genotypes, but the number of nuclei changes, due to extra or fewer nuclear cycles. B) Non-normalized and (C) normalized simulated gradients considering the 30 most ventral cells only. D–F) Direct comparison between experimental data and the model normalized output. Experimental data indicated by solid lines was reproduced from (Chahda et al, 2013). Shaded areas represent average±SD. G–H) Individual effects of changing nuclei radius and density on the Dl gradient at nuclear cycle 13. G) Increasing nuclei radius and (H) density flattens the gradient, as indicated by arrows. V indicates ventral midline, y axis indicate absolute (A, B) and normalized Dl levels (C–H), x axis indicate nuclei.

111

Figure 12. Changes in kDeg allow the reproduction of the Dl gradient in embryos derived from dl−/dl+ mothers. Solid line and pink shadow, experimental quantification of Dl nuclear levels in the 30 most ventral nuclei (mean±SD, n = 5). Dotted lines indicate simulations with original values (light pink dots) for Dl nuclear import rates (ki) and Cact degradation (kDeg), and adjusted values (dark pink and black dots). The other parameters values used in the simulations are shown in the third column of Table 3. Gray shadow is a larger confidence interval (99%) due to the small sample size, and indicates that the slight deepening in the experimental gradient peak is not significant. V indicates ventral midline, y axis indicate normalized Dl levels, x axis indicate nuclei. See Table S5 available online for fit calculations.

112

Figure 13. Simulations of wild type and mutant gradients suggest increased diffusion and Dl export rates. A–D) Comparison between experimental data (solid lines) and simulations (dotted lines) using parameter sets shown in Table 3. Simulations gyn 1–3 and ssm 1–2 from table are indicated by the number in parenthesis in (C) and (D), respectively. V indicates ventral midline, y axis indicate normalized Dl levels, x axis indicate nuclei. See Table S5 for fit calculations. Shaded areas represent average±SD.

113

Figure 14. The model predicts similar adjustment in parameters consistent to the phylogenetic relationship of species. Experimental quantification of the Dl gradient (solid line) and model simulations (dotted lines). Shadowed area indicates mean±SD A–D) D. melanogaster (A, adjusted parameters shown in Table 3); and species gradient simulations according to Table 2 (B, busckii; C, simulans; D, sechellia). Experimental data was obtained from (Chahda et al. 2013). V indicates ventral midline, y axis indicate normalized Dl levels, x axis indicate nuclei. See Table S5 for fit calculations.

114

Figure 15. Simulations of Dl gradient in an additional pair of closely related species with varying egg size suggest a shared Dl-Cact binding rates or Cact degradation rates. Experimental quantifications (solid lines) and simulations (dotted lines) in D. santomea (A) and D. yakuba (B) based on parameters indicated in Table 5. Shadowed area indicates average±SD. Best fitting curves are obtained with the same lowered kb or the same increased kDeg value for both species (black dots, simulation 3 and 4 respectively). V indicates ventral midline, y axis indicate normalized Dl levels, x axis indicate nuclei. See Table S5 for fit calculations.

115

Figure 16. Sensitivity analysis for main parameters tested. A–D) Contours represent a drop of 0.01 in fit (square root of the sum of square differences between the gradient produced with the parameter values showed in the y and x axis and the gradient produced with wild type melanogaster parameters), with exception to (A), in which the drop in fit is 0.0094. A) Example of all simulations within highly robust regions of the model

(ki x R). B) Example of gyn (“i”) and D. busckii simulations (“b”) that fall within less stable regions of the model compared to other samples. C) Lowering Dl export rates (ke) in D. sechellia (“d, arrow”) and D. simulans (“c, arrow”) allows good fits for these species after correcting for nucleus radius (r). D) Simulations of D. simulans that rely on drastic changes in only one parameter, in this case, kb, fall within unstable regions of the model (compare “c” white and yellow dots indicated by arrows.) a, D. melanogaster; b, D. busckii; c, D. simulans, d, D. sechellia; e,D. yakuba; f, D. santomea; g, dl−/dl+; h, ssm; i, gyn.

116

Figure 17. Comparison of the Dorsal and Cactus proteins across species supports model findings. Amino acid sequence comparison between D. melanogaster (mel), D. simulans (sim), D. sechellia (sec), D. yakuba (yak) and D. santomea (san) for relevant domains of Dl (A) and Cact (B). The genome sequence of D. santomea is not available, thus we partially sequenced the D. santomea Dl and Cact. Location for the following domains are shown: rel homology domain (Ip et al. 1991; Isoda et al. 1992); nuclear localization signal (NLS) and nuclear export signals (NES1-4) (Isoda et al. 1992; Xylourgidis et al. 2006); validated Cact serine phosphorylation sites and functional domain (Fernandez et al. 2001); ankyrin repeats (red boxes)(Kidd 1992).

117

3.8 Conclusions and future directions

Our previous work implicated alterations to the range of Toll signaling and

morphological traits, such as nuclear size and density, as the causal factors behind

Dl gradient modification (Chahda et al. 2013). Here, we used mathematical

modeling of the Dl gradient to show that divergence in morphological traits alone

is not sufficient to recreate the gradients seen in the Drosophila species. However,

morphological changes along with changes in components of the Toll signaling

pathway have a stronger effect over the Dl gradient formation, than if these

parameters are modified in isolation. Our model also predicted that specific

functional changes to Toll signaling components have evolved in separate

lineages to reshape the Dl gradient.

3.8.1 The Dl import, export and diffusion rates are greater than previously

thought

The original parameter set obtained from Kanodia et al. (2009) was not

validated in mutant strains or other species. Moreover, their parameter values for

Dl import, export and diffusion rates were inferred from GFP-tagged Dl, which

may not mimic the biophysical properties of endogenous Dl (Kanodia et al. 2009).

GFP is a large protein that most likely alters the dynamics of the Dl protein,

which is supported by the fact that the Dl-GFP transgene used in live-imaging fails to complement mothers homozygous for Dl null alleles (L M Liberman et al.

2009). Furthermore, the Dl-GFP transgene is lacking the Dl C-terminus that

contains a nuclear export signal (L M Liberman et al. 2009; Xylourgidis et al.

2006).

118

Different combinations of parameter values for measurements that could not

be directly measured, such as the nuclear import rate of Dl or the Cactus

degradation rate, were fit against WT and mutant Dl gradients to improve the

parameter set in a way to ensure biological relevance. Notable updates to the

original parameter set included doubling the Dl nuclear import and export rate,

and greatly increasing the diffusion between compartments. Our modified

parameter set more accurately fit WT data than the original parameter set from

Kanodia et al. (2009), and was able to correctly fit the Dl gradients from ssm, gyn,

and dl heterozygote mutants, as well (Fig. 13).

3.8.2 Evolution within the Toll signaling pathway can account for species-

specific Dl gradient distributions

Our original hypothesis was that differences in the size and density of nuclei along with the range of peak Toll signaling could account for the species-

specific Dl gradient shapes. After adjusting for these parameters, the Dl gradient

shapes for D. busckii, D. simulans and D. sechellia could not be recreated by the

mathematical model without additional parameter changes to Toll signaling

components (Fig. 14). Importantly, necessary changes to Toll signaling pathway components were consistent with phylogenetic relationships, validating our

approach. For example, parameter sets with changes in the same direction to Toll

signaling components recreate the Dl gradients of sister-species D. simulans and

D. sechellia, while changes in the opposite direction are necessary to simulate the

gradient of the distantly diverged D. busckii (Table 4; Fig. 14).

119

Sensitivity analysis further revealed that evolution has most likely resulted

in a combination of small changes to Toll signaling components, as opposed to a large change to single Toll component that would result in an unstable Dl gradient

(Fig 16). Furthermore, a non-linear interaction between morphological traits and

Toll signaling components was observed, demonstrating how changes in nuclear

size (r) and embryo size (Er) can have large effects on Dl gradient shape when

coupled with changes in Dl nuclear export rate (ke) and Dl-Cactus binding affinity

(kb), respectively (Fig. 16C,D).

Finally, we found changes to the amino acid sequence of Dl and Cactus

across species that confirms predictions made by the mathematical model that

components within the Toll signaling pathway have been modified. For instance,

in D. simulans and D. sechellia, there are conserved amino acid substitutions in

the third nuclear export signal (NES3) of the Dl protein (Fig. 16) that differ from

the D. melanogaster sequence. Our modeling predicts that the Dl nuclear export

rate is reduced in D. simulans and D. sechellia compared to D. melanogaster.

To confirm that Dl nuclear export is reduced in D. simulans and D.

sechellia, the CRISPR/Cas9 system could be used to replace the endogenous dl

allele in D. melanogaster with that of D. simulans and D. sechellia. The Dl

gradient should be flatter in the allele-replaced D. melanogaster background. This

same strategy could be used to replace the D. melanogaster cactus allele with that

of D. yakuba and D. santomea, as our modeling predicts that rate of Cactus

degradation and the Dl-Cactus binding affinity is reduced in the D. yakuba/D.

santomea lineage.

120

Chapter 4: Discussion and Future Direction

4.1 Dorsal gradient amplitude and distribution can explain species-specific

tissue allocation along the dorsal-ventral axis

In this work we first confirmed previous observations that demonstrated the lack of scaling of the embryonic mesoderm and neuroectoderm in different

Drosophila species (Belu and Mizutani 2011). This result shows that there is sharp contrast between the mechanisms that generate periodic patterns along the

AP axis and those that subdivide the embryo in three distinct layers across the DV axis. To better understand why the mesoderm and neuroectoderm do not scale to size, we first asked whether there were differences in the shape of the Dorsal gradient that defines gene expression patterns within these domains.

By directly measuring the Dorsal gradient in different species and normalizing the data, it became evident that the each species has a gradient with an unique shape

(Chahda et al. 2013). Next we asked whether these different shapes were associated to different thresholds of gene activation in different species. The test of whether genes from different species respond to the same or different levels of

Dorsal was particularly challenging because we needed an experimental setting that allowed us to measure the transcription of alleles from each species in response to a single known Dorsal gradient. We created this experimental setting and solved the problem by analyzing transcription dots of two mesodermal genes

(snail and twist) in hybrid embryos subjected to either a D. melanogaster maternal

Dorsal gradient, or either of its sibling species, D. sechellia and D. simulans, maternal Dorsal gradients. From these experiments in hybrid embryos, we found

121 that sna and twi from different species respond to the same levels of Dorsal. Since each gene responds to the same level of Dorsal and they define the limits of the mesoderm, it follows that the position of the mesoderm and the ventral neuroectoderm boundary is dictated by the local levels of Dorsal that vary across the DV axis in each species, not by differential sensitivity of these genes to the gradient (Chahda et al. 2013). Such a mechanism allows the specification of the mesoderm and the ventral boundary of the neuroectoderm at different positions across the DV axis. Since at the mesoderm/neuroectoderm boundary the levels of

Dorsal are too low to induce the expression of mesodermal genes and likely to fade dorsally at a similar rate in different species, the dorsal and ventral boundaries of the neuroectoderm remain constant. Thus, this constitutes a remarkable self-organizing system that easily adjusts tissues in response to embryo size by increasing or decreasing the mesoderm while maintaining a fixed number of cells allocated to the neuroectoderm. This mechanism is of great functional significance to the larval stage since it the assures that no matter the size of the embryo, the neuroectoderm field will be generating the same number of neuroblasts and later motor neurons required to coordinate the locomotion of animals with same numbers of muscle fibers with more or less muscle mass (Belu and Mizutani 2011).

122

Figure 18. Summary of Chapter 2 findings: differences in morphological traits, relative range of peak Toll signaling and DV tissue specification across Drosophila species. Arrows indicate the phylogenetic emergence of the traits indicated. Box in the right side are findings from haploid ssm and triploid gyn mutants that alter nuclear size and density and mesoderm specification. Note the difference in the number, size of and density of red balls that represent mesoderm nuclei.

123

4.2 Morphological traits and fast evolution of the Toll signaling pathway

reshape the Dorsal gradient

From the previous experiments we uncovered a self-organizing system involving variations in the gradient of Dorsal that expands or retracts the mesoderm but maintains a neuroectoderm with a fixed size. However, these experiments did not inform how these variations in the Dorsal distribution are generated. From the analyses of embryos of different species, it became evident that these embryos also vary in nuclear size and density, which must modify the distribution of the Dorsal gradient. Also, we noted that range of Toll was altered by analyzing the range of the Dorsal target gene snail.

To test the extent to which nuclear size and density affect the Dorsal gradient we created embryos with more or less nuclei and directly measured the distribution of Dorsal in this embryos. This was achieved by using mutations in sesame (ssm), and a combination of gynogenetic2 and gynogenetic3 (collectivily called gyn). ssm produces haploid embryos with one less cell division than the wild type and sparsely distributed small nuclei. In constrast, gyn embryos undergo one more division than the wild type and generate embryos with large and densely packed nuclei. These experiments demonstrated that the Dl gradient shape is indeed affected by variations in these two parameters (Fig. 7). However, these experiments could not separate the effects of nuclei density and nuclei size and whether either parameter was a major or minor contributor to the shape of these gradients.

124

To address this issue, we turned to a mathematical modeling that tests the individual and combined contribution of nuclear size, density and range of Toll on

Dl gradient formation (Chapter 3). This modeling revealed that the combined changes to the number, size and density of nuclei and range of peak Toll signaling are insufficient to recreate the experimentally determined Dl gradient distributions across species (Ambrosi et al. 2014). The model also suggested that specific changes in components of the Toll signaling pathway, such as the rate of Cactus degradation or the Dl nuclear export rate, were necessary to recreate the species- specific Dl gradients. Furthermore, the model resolved the seemingly paradoxical gradients observed in D. melanogaster, D. simulans and D. sechellia. For instance, D. sechellia is more closely related to D. simulans than D. melanogaster but its Dorsal gradient shape resembles more the D. melanogaster gradient than

D. simulans. What the model shows is that differences observed in the gradients

of the large embryo of D. sechellia or a small embryo of D. simulans are

generated by similar changes in Toll signaling components that set these two

species apart from D. melanogaster. In other words, predicted evolutionary

changes within the Toll signaling pathway are in agreement with phylogenetic

relationship of these species (Fig. 26).To test whether this was also true for other

species with different egg size, similar divergence time and that were formed after

a separate speciation event, we performed the same analysis in two other species,

D. santomea and D. yakuba. Here again we noted that similar changes in

components of Toll signaling were capable of reproducing the different Dorsal

gradients of these species, which provides evidence that this has been a recurrent

125 evolutionary mechanism in species that differ in embryo size. Finally, we asked whether the predictions of the model were actually supported by the existence of mutations the components of the Toll signal pathway. These analyses revealed that the model was not only accurately predicted the components involved, but the mutations in these components are in domains known to modify Cactus degradation and nuclear export of Dorsal, as well as the binding between Cactus and Dorsal.

Together, the results that manipulated the size and density of nuclei, and modelled the formation of these gradients reveal that the evolution of body scaling is subjected to some expected and some unexpected influences. It was somewhat expected that the geometry and density of nuclei might impact the gradient formation. However, it was somewhat surprising to verify that genetic changes in the Toll pathway, which evolves fast due to its shared role in immunity, would greatly exceed the impact of these physical changes.

Ultimately, the fast evolution of Toll signaling components coupled with changes in the size and density of nuclei, plus a versatile self-organizing system capable of positioning the mesoderm and neuroectoderm at different positions allow embryos to go through dramatic changes in the way they allocate cells across the

D/V axis within a very short evolutionary time.

126

Figure 19. Diverse Dorsal gradient shapes are generated by similar changes in parameters related to Toll signaling pathway in agreement with phylogenetic relationships between Drosophilids. Phylogenetic tree (left) of Drosophila species used for mathematical modeling of the Dl gradient. Phylogenetic tree ends with representative blastoderm cross sections stained against Dl and plotted graphs for Dl gradient levels from experimental and simulation data. Solid line represents experimentally obtained Dl gradient and the dotted line represents the simulated Dl gradient. The color of the dotted line and species name indicate necessary changes to the Toll signaling parameters relative to D. melanogaster (color spectrum on right) to recreate species-specific Dl gradients. Note that most closely related pairs of sister-species are in the same color.

127

Summary of Conclusions

• While the mesoderm is species-specific in size, the absolute size and sub-type

composition of the neuroectoderm is conserved in Drosophila sibling species

and D. busckii.

• Variation in the Dl gradient explains differences in mesoderm size across

Drosophila species that produce embryos of different sizes.

• Modification to the Dl gradient repositions the borders of the neuroectoderm.

• Dl levels are similar across sibling species in lateral regions of the embryo

where the neuroectoderm is specified.

• Changes in nuclear size and density, the range of peak Toll signaling, and

components within the Toll signaling pathway interact to shape the Dl

gradient.

128

BIBLIOGRAPHY

Ambrosi, Priscilla, Juan Sebastian Chahda, Hannah R. Koslen, Hillel J. Chiel, and Claudia Mieko Mizutani. 2014. “Modeling of the Dorsal Gradient across Species Reveals Interaction between Embryo Morphology and Toll Signaling Pathway during Evolution.” PLoS computational biology 10(8):e1003807.

Anderson, K. V, L. Bokla, and C. Nusslein-Volhard. 1985. “Establishment of Dorsal-Ventral Polarity in the Drosophila Embryo: The Induction of Polarity by the Toll Gene Product.” Cell 42(3):791–98.

Ashe, Hilary L. and James Briscoe. 2006. “The Interpretation of Morphogen Gradients.” Development (Cambridge, England) 133(3):385–94.

Baker, J., W. E. Theurkauf, and G. Schubiger. 1993. “Dynamic Changes in Microtubule Configuration Correlate with Nuclear Migration in the Preblastoderm Drosophila Embryo.” J Cell Biol 122(1):113–21.

Barth, K. a et al. 1999. “Bmp Activity Establishes a Gradient of Positional Information throughout the Entire Neural Plate.” Development (Cambridge, England) 126(22):4977–87.

Barth, K. a and S. W. Wilson. 1995. “Expression of Zebrafish nk2.2 Is Influenced by Sonic Hedgehog/vertebrate Hedgehog-1 and Demarcates a Zone of Neuronal Differentiation in the Embryonic Forebrain.” Development (Cambridge, England) 121(6):1755–68.

Bate, CM. 1976. “Embryogenesis of an Insect Nervous System I. A Map of the Thoracic and Abdominal Neuroblasts in Locusta Migratoria.” Journal of embryology and experimental morphology 35:107–23.

Belu, Mirela et al. 2010. “Upright Imaging of Drosophila Embryos.” J. Vis. Exp. (43), e2175, DOI: 10.3791/2175 (2010). (43).

Belu, Mirela and Claudia M. Mizutani. 2011. “Variation in Mesoderm Specification across Drosophilids Is Compensated by Different Rates of Myoblast Fusion during Body Wall Musculature Development.” PloS one 6(12):e28970.

Belvin, M. P., Y. Jin, and K. V Anderson. 1995. “Cactus Protein Degradation Mediates Drosophila Dorsal-Ventral Signaling.” Genes & Development 9(7):783–93.

129

Ben-Zvi, Danny, Ben-Zion Shilo, Abraham Fainsod, and Naama Barkai. 2008. “Scaling of the BMP Activation Gradient in Xenopus Embryos.” Nature 453(7199):1205–11.

Biehs, B., V. Francois, and E. Bier. 1996. “The Drosophila Short Gastrulation Gene Prevents Dpp from Autoactivating and Suppressing Neurogenesis in the Neuroectoderm.” Genes & Development 10(22):2922–34.

Campos-Ortega, JA. 1995. “Genetic Mechanisms of Early Neurogenesis inDrosophila Melanogaster.” Molecular neurobiology 111–22.

Chahda, Juan Sebastian, Rui Sousa-Neves, and Claudia Mieko Mizutani. 2013. “Variation in the Dorsal Gradient Distribution Is a Source for Modified Scaling of Germ Layers in Drosophila.” Current biology : CB 1–7.

Cheung, David, Cecelia Miles, Martin Kreitman, and Jun Ma. 2011. “Scaling of the Bicoid Morphogen Gradient by a Volume-Dependent Production Rate.” Development (Cambridge, England) 138(13):2741–49.

Cheung, David, Cecelia Miles, Martin Kreitman, and Jun Ma. 2014. “Adaptation of the Length Scale and Amplitude of the Bicoid Gradient Profile to Achieve Robust Patterning in Abnormally Large Drosophila Melanogaster Embryos.” Development (Cambridge, England) 141(1):124–35.

Chu, Hsin, Carlos Parras, Kalpana White, and Fernando Jiménez. 1998. “Formation and Specification of Ventral Neuroblasts Is Controlled by Vnd in Drosophila Neurogenesis.” Genes and Development 12(22):3613–24.

Clark, Andrew G. et al. 2007. “Evolution of Genes and Genomes on the Drosophila Phylogeny.” Nature 450(7167):203–18.

Coppey, M., A. N. Boettiger, A. M. Berezhkovskii, and S. Y. Shvartsman. 2008. “Nuclear Trapping Shapes the Terminal Gradient in the Drosophila Embryo.” Curr Biol 18(12):915–19.

Cornell, Robert a. and Tonia Von Ohlen. 2000. “Vnd/nkx, Ind/gsh, and Msh/msx: Conserved Regulators of Dorsoventral Neural Patterning?” Current Opinion in Neurobiology 10(1):63–71.

Cowden, John and Michael Levine. 2002. “The Snail Repressor Positions Notch Signaling in the Drosophila Embryo.” Development (Cambridge, England) 129:1785–93.

Cowden, John and Michael Levine. 2003. “Ventral Dominance Governs Sequential Patterns of Gene Expression across the Dorsal–ventral Axis of the

130

Neuroectoderm in the Drosophila Embryo.” Developmental Biology 262(2):335–49.

Crocker, J., Y. Tamori, and A. Erives. 2008. “Evolution Acts on Enhancer Organization to Fine-Tune Gradient Threshold Readouts.” PLoS Biol 6(11):e263.

Crocker, Justin, Nathan Potter, and Albert Erives. 2010. “Dynamic Evolution of Precise Regulatory Encodings Creates the Clustered Site Signature of Enhancers.” Nature communications 1:99.

D’Alessio, Marina and Manfred Frasch. 1996. “Msh May Play a Conserved Role in Dorsoventral Patterning of the Neuroectoderm and Mesoderm.” Mechanisms of Development 58(1-2):217–31.

David, Jean R., Françoise Lemeunier, Leonidas Tsacas, and Amir Yassin. 2007. “The Historical Discovery of the Nine Species in the Drosophila Melanogaster Species Subgroup.” Genetics 177(4):1969–73.

DeLotto, R., Y. DeLotto, R. Steward, and J. Lippincott-Schwartz. 2007. “Nucleocytoplasmic Shuttling Mediates the Dynamic Maintenance of Nuclear Dorsal Levels during Drosophila Embryogenesis.” Development 134(23):4233–41.

Dessaud, Eric, Andrew P. McMahon, and James Briscoe. 2008. “Pattern Formation in the Vertebrate Neural Tube: A Sonic Hedgehog Morphogen- Regulated Transcriptional Network.” Development (Cambridge, England) 135(15):2489–2503.

Doe, C. Q. 1992. “Molecular Markers for Identified Neuroblasts and Ganglion Mother Cells in the Drosophila Central Nervous System.” Development 116(4):855–63.

Doe, Chris Q. 1992. “Molecular Markers for Identified Neuroblasts and Ganglion Mother Cells in the Drosophila Central Nervous System.” 863:855–63.

Drier, EA, S. Govind, and R. Steward. 2000. “Cactus-Independent Regulation of Dorsal Nuclear Import by the Ventral Signal.” Current Biology 10(1):23–26.

Driever, W. and C. Nüsslein-Volhard. 1989. “The Bicoid Protein Is a Positive Regulator of Hunchback Transcription in the Early Drosophila Embryo.” Nature 337(6203):138–43.

Driever, Wolfgang, Gudrun Thoma, and Christiane Nusslein-volhard. 1989. “Determination of Spatial Domains of Zygotic Gene Expression in the

131

Drosophila Embryo by the Affinity Fo Binding Sites for the Bicoid Morphogen.” Nature 340.

Edgar, B. A., C. P. Kiehle, and G. Schubiger. 1986. “Cell Cycle Control by the Nucleo-Cytoplasmic Ratio in Early Drosophila Development.” Cell 44(2):365–72.

Ekker, M. et al. 1997. “Relationships among Msx Gene Structure and Function in Zebrafish and Other Vertebrates.” Molecular biology and evolution 14(10):1008–22.

Erickson, J. W. and J. J. Quintero. 2007. “Indirect Effects of Ploidy Suggest X Chromosome Dose, Not the X:A Ratio, Signals Sex in Drosophila.” PLoS Biol 5(12):e332.

Ferguson, E. L. and K. V Anderson. 1992a. “Decapentaplegic Acts as a Morphogen to Organize Dorsal-Ventral Pattern in the Drosophila Embryo.” Cell 71(3):451–61.

Ferguson, E. L. and K. V Anderson. 1992b. “Localized Enhancement and Repression of the Activity of the TGF-Beta Family Member, Decapentaplegic, Is Necessary for Dorsal-Ventral Pattern Formation in the Drosophila Embryo.” Development (Cambridge, England) 114(3):583–97.

Fernandez, N. Q., J. Grosshans, J. S. Goltz, and D. Stein. 2001. “Separable and Redundant Regulatory Determinants in Cactus Mediate Its Dorsal Group Dependent Degradation.” Development (Cambridge, England) 128(15):2963–74.

Foe, VE and BM Alberts. 1983. “Studies of Nuclear and Cytoplasmic Behaviour during the Five Mitotic Cycles That Precede Gastrulation in Dorosphila Embryogenesis.” J. Cell Sci. 61:31–70.

Fowlkes, Charless C. et al. 2008. “A Quantitative Spatiotemporal Atlas of Gene Expression in the Drosophila Blastoderm.” Cell 133(2):364–74.

Fowlkes, Charless C. et al. 2011. “A Conserved Developmental Patterning Network Produces Quantitatively Different Output in Multiple Species of Drosophila” edited by Kopp. PLoS Genetics 7(10):e1002346.

François, Vincent, Mark Solloway, Jason W. O’Neill, John Emery, and Ethan Bier. 1994. “Dorsal-Ventral Patterning of the Drosophila Embryo Depends on a Putative Negative Growth Factor Encoded by the Short Gastrulation Gene.” Genes and Development 8(21):2602–16.

132

Frohnhofer, H. G. and Christiane Nusslein-volhard. 1986. “Organization of Anterior Pattern in the Drosopila Embryo by the Maternal Gene Bicoid.” Nature 321(13).

Fuyama, Y. 1986. “Genetics of Parthenogenesis in DROSOPHILA MELANOGASTER. II. Characterization of a Gynogenetically Reproducing Strain.” Genetics 114(2):495–509.

Garcia, Mayra, Marcos Nahmad, Gregory T. Reeves, and Angelike Stathopoulos. 2013. “Size-Dependent Regulation of Dorsal-Ventral Patterning in the Early Drosophila Embryo.” Developmental biology 381(1):286–99.

Gay, Nicholas J. and Monde Ntwasa. 1993. “The Drosophila Ankyrin Repeat Protein Cactus Has a Predominantly Α-Helical Secondary Structure.” FEBS Letters 335(2):155–60.

Gómez-Skarmeta, José Luis, Sonsoles Campuzano, and Juan Modolell. 2003. “Half a Century of Neural Prepatterning: The Story of a Few Bristles and Many Genes.” Nature reviews. Neuroscience 4(7):587–98.

Gregor, T., W. Bialek, R. R. de Ruyter van Steveninck, D. W. Tank, and E. F. Wieschaus. 2005. “Diffusion and Scaling during Early Embryonic Pattern Formation.” Proc Natl Acad Sci U S A 102(51):18403–7.

Gregor, T., E. F. Wieschaus, A. P. McGregor, W. Bialek, and D. W. Tank. 2007. “Stability and Nuclear Dynamics of the Bicoid Morphogen Gradient.” Cell 130(1):141–52.

Gregor, Thomas, Alistair P. McGregor, and Eric F. Wieschaus. 2008. “Shape and Function of the Bicoid Morphogen Gradient in Dipteran Species with Different Sized Embryos.” Developmental biology 316(2):350–58.

Grosshans, J., H. A. Muller, and E. Wieschaus. 2003. “Control of Cleavage Cycles in Drosophila Embryos by Fruhstart.” Dev Cell 5(2):285–94.

Grosshans, J. and E. Wieschaus. 2000. “A Genetic Link between Morphogenesis and Cell Division during Formation of the Ventral Furrow in Drosophila.” Cell 101(5):523–31.

Hamaratoglu, Fisun, Aitana Morton de Lachapelle, George Pyrowolakis, Sven Bergmann, and Markus Affolter. 2011. “Dpp Signaling Activity Requires Pentagone to Scale with Tissue Size in the Growing Drosophila Wing Imaginal Disc.” PLoS biology 9(10):e1001182.

133

Hamburger, V. 1988. “Ontogeny of Neuroembryology.” The Journal of neuroscience : the official journal of the Society for Neuroscience 8(10):3535–40.

Hanes, S. D. and R. Brent. 1989. “DNA Specificity of the Bicoid Activator Protein Is Determined by Homeodomain Recognition Helix Residue 9.” Cell 57(7):1275–83.

Hashimoto, Carl, Kathy L. Hudson, and Kathryn V. Anderson. 1988. “The Toll Gene of Drosophila, Required for Dorsal-Ventral Embryonic Polarity, Appears to Encode a Transmembrane Protein.” Cell 52(2):269–79.

Hemmati-Brivanlou, Ali, Olivia G. Kelly, and Douglas A. Melton. 1994. “Follistatin, an Antagonist of Activin, Is Expressed in the Spemann Organizer and Displays Direct Neuralizing Activity.” Cell 77(2):283–95.

Hoch, M., E. Seifert, and H. Jäckle. 1991. “Gene Expression Mediated by Cis- Acting Sequences of the Krüppel Gene in Response to the Drosophila Morphogens Bicoid and Hunchback.” The EMBO journal 10(8):2267–78.

Holley, S. a et al. 1995. “A Conserved System for Dorsal-Ventral Patterning in Insects and Vertebrates Involving Sog and Chordin.” Nature 376(6537):249– 53.

Hong, Joung-Woo, David a Hendrix, Dmitri Papatsenko, and Michael S. Levine. 2008. “How the Dorsal Gradient Works: Insights from Postgenome Technologies.” Proceedings of the National Academy of Sciences of the United States of America 105(51):20072–76.

Hsieh-Li, HM et al. 1995. “Gsh-2, a Murine Homeobox Gene Expressed in the Developing Brain.” Mech Dev 50:177–86.

Ip, Y. T., R. Kraut, M. Levine, and C. a Rushlow. 1991. “The Dorsal Morphogen Is a Sequence-Specific DNA-Binding Protein That Interacts with a Long- Range Repression Element in Drosophila.” Cell 64(2):439–46.

Ip, Y. T., R. E. Park, D. Kosman, K. Yazdanbakhsh, and M. Levine. 1992. “Dorsal-Twist Interactions Establish Snail Expression in the Presumptive Mesoderm of the Drosophila Embryo.” Genes Dev 6(8):1518–30.

Isoda, K., S. Roth, and C. Nusslein-Volhard. 1992. “The Functional Domains of the Drosophila Morphogen Dorsal: Evidence from the Analysis of Mutants.” Genes & Development 6(4):619–30.

134

Isshiki, T., M. Takeichi, and A. Nose. 1997. “The Role of the Msh Homeobox Gene during Drosophila Neurogenesis: Implication for the Dorsoventral Specification of the Neuroectoderm.” Development 124(16):3099–3109.

Jazwinska, A., C. Rushlow, and S. Roth. 1999. “The Role of Brinker in Mediating the Graded Response to Dpp in Early Drosophila Embryos.” Development 3334:3323–34.

Jiang, J. and M. Levine. 1993. “Binding Affinities and Cooperative Interactions with bHLH Activators Delimit Threshold Responses to the Dorsal Gradient Morphogen.” Cell 72(5):741–52.

Jiggins, F. M. and K. W. Kim. 2007. “A Screen for Immunity Genes Evolving under Positive Selection in Drosophila.” J Evol Biol 20(3):965–70.

Jiménez, F. et al. 1995. “Vnd, a Gene Required for Early Neurogenesis of Drosophila, Encodes a Homeodomain Protein.” The EMBO journal 14(14):3487–95.

Johnston, R. Daniel St and William M. Gelbart. 1987. “Decapentaplegic Transcripts Are Localized along the Dorsal - Ventral Axis of the Drosophila Embryo.” 6(9):2785–91.

Kanodia, J. S. et al. 2009. “Dynamics of the Dorsal Morphogen Gradient.” Proc Natl Acad Sci U S A 106(51):21707–12.

Kelley, Lawrence A. and Michael J. E. Sternberg. 2009. “Protein Structure Prediction on the Web: A Case Study Using the Phyre Server.” Nature protocols 4(3):363–71.

Keranen, S. V et al. 2006. “Three-Dimensional Morphology and Gene Expression in the Drosophila Blastoderm at Cellular Resolution II: Dynamics.” Genome Biol 7(12):R124.

Kerszberg, Michel and Lewis Wolpert. 2007. “Specifying Positional Information in the Embryo: Looking beyond Morphogens.” Cell 130(2):205–9.

Kidd, S. 1992. “Characterization of the Drosophila Cactus Locus and Analysis of Interactions between Cactus and Dorsal Proteins.” Cell 71(4):623–35.

Kimelman, David and Benjamin L. Martin. 2012. “Anterior-Posterior Patterning in Early Development: Three Strategies.” Wiley Interdisciplinary Reviews: Developmental Biology 1(April):253–66.

Knipple, Dc, E. Seifert, and Ub Rosenberg. 1985. “Spatial and Temporal Patterns of Kruppel Gene Expression in Early Drosophila Embryos.” Nature.

135

Kohwi, Minoree and Chris Q. Doe. 2013. “Temporal Fate Specification and Neural Progenitor Competence during Development.” Nature reviews. Neuroscience 14(12):823–38.

Kosman, D. et al. 2004. “Multiplex Detection of RNA Expression in Drosophila Embryos.” Science 305(5685):846.

Kosman, D., Y. T. Ip, M. Levine, and K. Arora. 1991. “Establishment of the Mesoderm-Neuroectoderm Boundary in the Drosophila Embryo.” Science 254(5028):118–22.

Kraut, Rachel and Michael Levine. 1991. “Spatial Regulation of the Gap Gene Giant during Drosophila Development.” Development 609.

Kunisch, Marianne. 1994. “Lateral Inhibition Mediated by the Drosophila Neurogenic Gene Delta Is Enhanced by Proneural Proteins.” Proceedings of the … 91(October):10139–43.

Lachaise, D. et al. 2000. “Evolutionary Novelties in Islands: Drosophila Santomea, a New Melanogaster Sister Species from São Tomé.” Proceedings. Biological sciences / The Royal Society 267(1452):1487–95.

Lachaise, Daniel, Jean R. David, Francoise Lemeunier, Leonidas Tsacas, and Michael Ashburner. 1986. “The Reproductive Relationships of Drosophila Sechellia with D. Mauritiana, D. Simulans, and D. Melanogaster from the Afrotropical Region.” Evolution 40(2):262–71.

Lamb, TM et al. 1993. “Neural Induction by the Secreted Polypeptide Noggin.” Science 262(5134):713–18.

Lee, K. J. and T. M. Jessell. 1999. “The Specification of Dorsal Cell Fates in the Vertebrate Central Nervous System.” Annual review of neuroscience 22:261–94.

Lehmann, R. and C. Nüsslein-Volhard. 1987. “Hunchback, a Gene Required for Segmentation of an Anterior and Posterior Region of the Drosophila Embryo.” Developmental biology 119(2):402–17.

Liberman, L. M., G. T. Reeves, and A. Stathopoulos. 2009. “Quantitative Imaging of the Dorsal Nuclear Gradient Reveals Limitations to Threshold-Dependent Patterning in Drosophila.” Proc Natl Acad Sci U S A.

Liberman, Louisa M., Gregory T. Reeves, and Angelike Stathopoulos. 2009. “Quantitative Imaging of the Dorsal Nuclear Gradient Reveals Limitations to Threshold-Dependent Patterning in Drosophila.” Proceedings of the National Academy of Sciences of the United States of America 106(52):22317–22.

136

Loppin, B., M. Docquier, F. Bonneton, and P. Couble. 2000. “The Maternal Effect Mutation Sesame Affects the Formation of the Male Pronucleus in Drosophila Melanogaster.” Dev Biol 222(2):392–404.

Lott, S. E., M. Kreitman, A. Palsson, E. Alekseeva, and M. Z. Ludwig. 2007. “Canalization of Segmentation and Its Evolution in Drosophila.” Proc Natl Acad Sci U S A 104(26):10926–31.

Lu, Xuemin, Jennifer M. Li, Olivier Elemento, Saeed Tavazoie, and Eric F. Wieschaus. 2009. “Coupling of Zygotic Transcription to Mitotic Control at the Drosophila Mid-Blastula Transition.” Development (Cambridge, England) 136(12):2101–10.

Lynch, Jeremy a and Siegfried Roth. 2011. “The Evolution of Dorsal-Ventral Patterning Mechanisms in Insects.” Genes & development 25(2):107–18.

Markow, T. A., S. Beall, and L. M. Matzkin. 2009. “Egg Size, Embryonic Development Time and Ovoviviparity in Drosophila Species.” J Evol Biol 22(2):430–34.

Markstein, Michele, Peter Markstein, Vicky Markstein, and Michael S. Levine. 2002. “Genome-Wide Analysis of Clustered Dorsal Binding Sites Identifies Putative Target Genes in the Drosophila Embryo.” Proceedings of the National Academy of Sciences of the United States of America 99(2):763–68.

McDonald, J. A. et al. 1998. “Dorsoventral Patterning in the Drosophila Central Nervous System: The Vnd Homeobox Gene Specifies Ventral Column Identity.” Genes Dev 12(22):3603–12.

McHale, Peter et al. 2011. “Gene Length May Contribute to Graded Transcriptional Responses in the Drosophila Embryo.” Developmental biology 360(1):230–40.

Mellerick, D. M. and M. Nirenberg. 1995. “Dorsal-Ventral Patterning Genes Restrict NK-2 Homeobox Gene Expression to the Ventral Half of the Central Nervous System of Drosophila Embryos.” Dev Biol 171(2):306–16.

Miles, Cecelia M. et al. 2011. “Artificial Selection on Egg Size Perturbs Early Pattern Formation in Drosophila Melanogaster.” Evolution 65(1):33–42.

Miles, CM, SE Lott, and CL Luengo Hendriks. 2011. “Artificial Selection on Egg Size Perturbs Early Pattern Formation in Drosophila Melanogaster.” Evolution 65(1):33–42.

Mizutani, C. M. and E. Bier. 2008. “EvoD/Vo: The Origins of BMP Signalling in the Neuroectoderm.” Nat Rev Genet.

137

Mizutani, Claudia Mieko, Néva Meyer, Henk Roelink, and Ethan Bier. 2006. “Threshold-Dependent BMP-Mediated Repression: A Model for a Conserved Mechanism That Patterns the Neuroectoderm.” edited by Scott. PLoS biology 4(10):e313.

Morisalo, D. and KV Anderson. 1995. “Signaling Pathways That Establish the Dorsal-Ventral Pattern of the Drosophila Embryo.” Annual review of genetics.

Morisato, D. and KV Anderson. 1994. “The Spätzle Gene Encodes a Component of the Extracellular Signaling Pathway Establishing the Dorsal-Ventral Pattern of the Drosophila Embryo.” Cell 76.

Muñoz-Sanjuán, Ignacio and Ali H. Brivanlou. 2002. “Neural Induction, the Default Model and Embryonic Stem Cells.” Nature reviews. Neuroscience 3(4):271–80.

Nirenberg, M. et al. 1995. “The NK-2 Homeobox Gene and the Early Development of the Central Nervous System of Drosophila.” Ann N Y Acad Sci 758:224–42.

Nusslein-Volhard, C. and E. Wieschaus. 1980. “Mutations Affecting Segment Number and Polarity in Drosophila.” Nature 287(5785):795–801.

Obbard, Darren J., John J. Welch, Kang-Wook Kim, and Francis M. Jiggins. 2009. “Quantifying Adaptive Evolution in the Drosophila Immune System.” PLoS genetics 5(10):e1000698.

Von Ohlen, T. and C. Q. Doe. 2000. “Convergence of Dorsal, Dpp, and Egfr Signaling Pathways Subdivides the Drosophila Neuroectoderm into Three Dorsal-Ventral Columns.” Dev Biol 224(2):362–72.

Patel, N. H. 2000. “It’s a Bug's Life.” Proceedings of the National Academy of Sciences 97(9):4442–44.

Ray, R. P., K. Arora, C. Nüsslein-Volhard, and W. M. Gelbart. 1991. “The Control of Cell Fate along the Dorsal-Ventral Axis of the Drosophila Embryo.” Development (Cambridge, England) 113(1):35–54.

Reeves, G. T. and A. Stathopoulos. 2009. “Graded Dorsal and Differential Gene Regulation in the Drosophila Embryo.” Cold Spring Harb Perspect Biol 1(4):a000836.

De Robertis, Edward M. 2006. “Spemann’s Organizer and Self-Regulation in Amphibian Embryos.” Nature reviews. Molecular cell biology 7(4):296–302.

138

Rogers, Katherine W. and Alexander F. Schier. 2011. “Morphogen Gradients: From Generation to Interpretation.” Annu. Rev. Cell Dev. Biol. 27:377–407.

Roth, S., D. Stein, and C. Nusslein-Volhard. 1989. “A Gradient of Nuclear Localization of the Dorsal Protein Determines Dorsoventral Pattern in the Drosophila Embryo.” Cell 59(6):1189–1202.

Rushlow, C. A., K. Han, J. L. Manley, and M. Levine. 1989. “The Graded Distribution of the Dorsal Morphogen Is Initiated by Selective Nuclear Transport in Drosophila.” Cell 59(6):1165–77.

Sasai, Yoshiki et al. 1994. “Xenopus Chordin: A Novel Dorsalizing Factor Activated by Organizer-Specific Homeobox Genes.” Cell 79(5):779–90.

Schlenke, TA and DJ Begun. 2003. “Natural Selection Drives Drosophila Immune System Evolution.” Genetics 164:1471–80.

Schmid, Aloisia, Akira Chiba, and Chris Q. Doe. 1999. “Clonal Analysis of Drosophila Embryonic Neuroblasts : Neural Cell Types , Axon Projections and Muscle Targets.” 4689:4653–89.

Small, S., a Blair, and M. Levine. 1992. “Regulation of Even-Skipped Stripe 2 in the Drosophila Embryo.” The EMBO journal 11(11):4047–57.

Small, S., R. Kraut, T. Hoey, R. Warrior, and M. Levine. 1991. “Transcriptional Regulation of a Pair-Rule Stripe in Drosophila.” Genes & Development 5(5):827–39.

Sousa-Neves, Rui and Alexandre Rosas. 2010. “An Analysis of Genetic Changes during the Divergence of Drosophila Species.” PloS one 5(5):e10485.

Spemann, H. and H. Mangold. 1924. “Über Induktion von Embryonanlagen Durch Implantation Artfremder Organisatoren.” Wilhelm Roux Arch. Entw. Mech. Organ. 100:599–638.

Stathopoulos, A. and M. Levine. 2002. “Dorsal Gradient Networks in the Drosophila Embryo.” Dev Biol 246(1):57–67.

Stathopoulos, Angelike and Michael Levine. 2005. “Genomic Regulatory Networks and Animal Development.” Developmental Cell 9(4):449–62.

Steward, R., S. B. Zusman, L. H. Huang, and P. Schedl. 1988. “The Dorsal Protein Is Distributed in a Gradient in Early Drosophila Embryos.” Cell 55:487–95.

139

Tamura, Koichiro, Sankar Subramanian, and Sudhir Kumar. 2004. “Temporal Patterns of Fruit Fly (Drosophila) Evolution Revealed by Mutation Clocks.” Molecular biology and evolution 21(1):36–44.

Tautz, Diethard. 1988. “Regulation of the Drosophila Segmentation Gene Hunchback by Two Matern Morphogenetic Centres.” Nature 332(17).

Thomas, J. B., M. J. Bastiani, M. Bate, and C. S. Goodman. 1984. “From Grasshopper to Drosophila: A Common Plan for Neuronal Development.” Nature 310(5974):203–7.

Timmer, John R., Charlotte Wang, and Lee Niswander. 2002. “BMP Signaling Patterns the Dorsal and Intermediate Neural Tube via Regulation of Homeobox and Helix-Loop-Helix Transcription Factors.” Development (Cambridge, England) 129(10):2459–72.

Turing, Alan M. 1952. “The Chemical Basis of Morphogenesis.” Philos Trans R Soc Lond B Biol Sci 237(641).

Umulis, David M. and Hans G. Othmer. 2013. “Mechanisms of Scaling in Pattern Formation.” Development 140(24):4830–43.

Umulis, David M., Osamu Shimmi, Michael B. O’Connor, and Hans G. Othmer. 2010. “Organism-Scale Modeling of Early Drosophila Patterning via Bone Morphogenetic Proteins.” Developmental cell 18(2):260–74.

Ungerer, Petra and Gerhard Scholtz. 2008. “Filling the Gap between Identified Neuroblasts and Neurons in Crustaceans Adds New Support for Tetraconata.” Proceedings. Biological sciences / The Royal Society 275(1633):369–76.

Valerius, M. T. et al. 1995. “Gsh-1: A Novel Murine Homeobox Gene Expressed in the Central Nervous System.” Developmental dynamics : an official publication of the American Association of Anatomists 203(3):337–51.

Weiss, J. B. et al. 1998. “Dorsoventral Patterning in the Drosophila Central Nervous System: The Intermediate Neuroblasts Defective Homeobox Gene Specifies Intermediate Column Identity.” Genes Dev 12(22):3591–3602.

Whalen, a M. and R. Steward. 1993. “Dissociation of the Dorsal-Cactus Complex and Phosphorylation of the Dorsal Protein Correlate with the Nuclear Localization of Dorsal.” The Journal of cell biology 123(3):523–34.

Wharton, K. a, R. P. Ray, and W. M. Gelbart. 1993. “An Activity Gradient of Decapentaplegic Is Necessary for the Specification of Dorsal Pattern

140

Elements in the Drosophila Embryo.” Development (Cambridge, England) 117(2):807–22.

Whitington, P. 1996. “Evolution of Neural Development in the Arthropods.” Seminars in Cell & Developmental Biology 7(4):605–14.

Wolpert, L. 1969. “Positional Information and the Spatial Pattern of Cellular Differentiation.” Journal of theoretical biology 1968:1–47.

Xylourgidis, Nikos, Peggy Roth, Nafiseh Sabri, Vasilios Tsarouhas, and Christos Samakovlis. 2006. “The Nucleoporin Nup214 Sequesters CRM1 at the Nuclear Rim and Modulates NFkappaB Activation in Drosophila.” Journal of cell science 119(Pt 21):4409–19.

Zeller, Rolf, Javier López-Ríos, and Aimée Zuniga. 2009. “Vertebrate Limb Bud Development: Moving towards Integrative Analysis of Organogenesis.” Nat Rev Genet 10:845–58.

Zhang, Zhenyu, Xianjun Zhu, Leslie M. Stevens, and David Stein. 2009. “Distinct Functional Specificities Are Associated with Protein Isoforms Encoded by the Drosophila Dorsal-Ventral Patterning Gene Pipe.” Development (Cambridge, England) 136(16):2779–89.

Zimmerman, Lyle B., José M. De Jesús-Escobar, and Richard M. Harland. 1996. “The Spemann Organizer Signal Noggin Binds and Inactivates Bone Morphogenetic Protein 4.” Cell 86(4):599–606.

141