On the Development of the Turtle Scute Pattern and the Origins of Its Variation

Home , Australochelys, Eunotosaurus, Kallokibotion, Pappochelys, Proterochersidae, Reptile scale

On The Development Of The Turtle Scute Pattern And The Origins Of Its Variation

February 20, 2019

Roland Zimm

Institute of Biotechnology

Helsinki Institute of Life Science (HiLife) and Department of Biosciences Faculty of Biological and Environmental Sciences Doctoral Programme in Integrative Life Science (ILS) University of Helsinki

Academic dissertation

To be presented for public examination with the permission of the faculty of Biological and Environmental Science of the University of Helsinki in Sali 1041 (Biokeskus 2, Viikinkaari 5, Helsinki) on 12. March 2019 at 12:00. Supervisor Isaac Salazar-Ciudad, University of Helsinki

Thesis advisory committee Jukka Jernvall, University of Helsinki Mikael Fortelius, University of Helsinki Ritva Rice, University of Helsinki

Pre-examiners Ann Campbell Burke, Wesleyan University, United States of America Shigeru Kondo, Osaka University, Japan

Opponent Marcelo S´anchez-Villagra, University of Z¨urich, Switzerland

Custos Liisa Holm, University of Helsinki

Copyright c 2019 Roland Zimm ISSN (print) : 2342-3161 ISSN (e-thesis) : 2342-317X ISBN (paperback) : 978-951-51-4953-4 ISBN (e-thesis) : 978-951-51-4954-1 http://ethesis.helsinki.ﬁ Helsinki 2019 Unigraﬁa Acknowledgements

Working on my thesis has been a process of many years, at times joyful, at times strenuous. All this would have hardly been possible without the support of many people and institutions who believed in me and many more people whose contributions are usually rarely mentioned. At first, I want to express my thanks to my supervisor Isaac Salazar-Ciudad who taught me new ways of thinking in the contexts of biology, science and society in general, and my thesis in particular, as well as new ways “cross-country” the wilderness of Finland and Catalonia. In particular, I am grateful for all the honest and detailed feedback I received from him, and it is safe to say that without all our discussions I would be scientifically and personally far from where I stand today. Another important person to whom I owe a lot in context of my thesis is Jacqueline Moustakas-Verho. She has been my closest collaborator on the turtle project and become a great friend to me. Also, I owe to Scott Gilbert, without whose committment the turtles would never have arrived to Finland. I want to thank Jukka Jernvall, who, despite being usually busy with more important matters, was always approachable and extremely helpful, whenever some help and major or minor feedback was needed. Together with him, I want to thank the other members of my thesis committee, Mikael Fortelius and Ritva Rice, who gave equally helpful advice by pointing out risks, finding solutions and offering different views upon my ongoing research, especially by taking all my concerns seriously and telling me when my objectives were too ambitions. My thesis committee meetings have always been very easy-going, motivating, supportive and greatly contributing to the advancement of my research. During my time at Helsinki University, I was financially supported by the Academy of Finland, the Cluster of Excellence in Computational and Developmental Biology, and a GPBM grant that I was lucky to be awarded with during my first years of my studies. I want to acknowlege the support and guidance I got from my grad school(s), GPBM and ILS, who helped making my time at the university enjoyable, as well as guiding me through the duties towards the end of my PhD. As my work was mainly based on heavy computation, I do not want to miss to acknowledge the help by the CSC. The CSC Grand Challenge Grant I received with my collaborators played a crucial role in enabling us to perform computational experiments efficiently. The transition time between the end of my thesis-related research and my defense was substantially facilitated by the opportunity to work at the Konrad Lorenz Institute for Evolution and Cognition Research (KLI) for half a year and essentially write my thesis there. It has also been a very inspiring place where people with very diverse, theoretical questions in Biology come together. In particular, I want to thank Isabella Sarto-Jackson, Eva Lackner and Gerd Müller,but also the former director Johannes Jägerwho convinced me to apply there. Since I meanwhile started working in the lab of Nicolas Goudemand, I want to thank him for his readiness to accept me as “pre-postdoc”. I also do not want to forget to acknowledge the lab in which I completed my Diploma of Biology, the group of theoretical biology at Dresden University of Technology, especially Andreas Deutsch, Lutz Brusch and my former supervisor, Walter DeBack, who had crucial impact in diverting my scientific path towards the mathematical and computational field. Working inside a group in which it is equally natural to collaborate in scientific terms and to share an after-work beer, while discussing about all possible aspects of life, is a great luck which I have been endowed with. So, I am grateful to have been with my colleagues and friends, Irepan, who shared a lot of time with me and introduced me into the vast Mexican “morphospace of spices”, an indispensable asset in resisting the Finnish winter, Miguel, who has been a great companion in exploring the morphospace of unknown places and theories, Miquel, with whom I survived the wild slopes of Barcelona’s hinterland, Pascal, who has become my “scientific brother” in continuing in the projects I have begun, Lisandro and Jhon. I am also grateful for the wider environment at the Helsinki EvoDevo community: Mona, who rescued me whenever I was unable to cope with the Finnish language, Susanna, who facilitated my arrival and helped me coping with all “practical contingencies” in uncountable ways, Richi, Outi, Teemu, Ian, Mia, Yoland, Tuomas A., Tuomas K., Maria, Solja, Alison, Ewelina, Otto, Vinod, Isabel, Ana-Maria, Martin, Julia, Lotta, Fabien, Blanca, Joni, Simone, Filipe, Umair, Irma Thesleff, Marja Mikkola and Nicolas DiPoi, and all the other people whom I may have forgotten to mention here. Although personally disliking the concept of “life-work balance”, my life in Finland would have been severely disbalanced without the different activities and distractions that diversified and en- riched my past six years, such as participating in the Aland˚ bird rallyes with Tuomas K, Kalle and Joni, joyful and creative weekly board game evenings with Maja, Billy, Julia, Barbie, Teemu and Juhani, evenings to break down philosophical frontiers with Matan, Jani and Aura, French cheese challenges with Fabien and Valerio, after-work “Stammtisch” with Julia and Wojciech, bike trips with Irepan, Helene and Ekaterina, international evenings at CaféLingua with Julian and Jimena, great social events with Pekko and Anna, and Birgitta Ahlberg, who offered me to stay with my frinds at her Mökkis,thereby possibly providing the most finnish times ever. Thank you Loic, Joan, Irepan and Erno for being almost perfect flatmates. You all have been great friends and contributed to the experience I have had at Helsinki and will remember. Finally, my biggest thank goes to the people who spent more time, effort, nerves and love on me than anybody else, during most of my scientific journey and much beyond, and, thereby, have shaped what I am: my parents and grandparents, my brother, and Vitória,the one person who has accompanied me during the last years through all distances in time and space, understanding me, providing support in so many ways, and bearing all the additional stresses and restrictions that a shared life in academia conveys. Contents

1 Literature review 1 1.1 Novelties ...... 2 1.2 The origins and diversity of the turtle shell ...... 4 1.3 Development of turtle scutes ...... 7 1.4 Development of periodic structures ...... 10 1.5 Reaction-diﬀusion systems ...... 14 1.6 Epidermal appendages and their development ...... 18 1.7 The origins of variation through development ...... 22 1.8 Origins of variation in turtle scute patterns ...... 26 1.9 Models of development ...... 31 1.10 Towards a uniﬁed model of development ...... 35

2 Aims 38

3 Methods 38

4 Results 42

5 Discussion 43 5.1 Caveats and Limitations ...... 43 5.1.1 Caveats due to model limitations ...... 43 5.1.2 Caveats due to limited data ...... 46 5.2 Classification of the turtle scute pattern development and implications for variation 48 5.3 The carapacial ridge as a developmental organizer ...... 50 5.4 Comparison between scutes and other ectodermal appendages from a developmental perspective ...... 52 5.5 Interspecific variation of turtle scutation ...... 58 5.6 A genetic origin of turtle scute anomalies? ...... 61 5.7 Origins of intraspecific variation: Hypotheses revised ...... 62 5.8 Possible mechanisms by which epigenetic factors may interfere with development . . 66 5.9 Turtle anomalies and climate change ...... 70 5.10 Has the role of epigenetic factors in phenotypic variation been underrated? . . . . . 70 5.11 Evo-devo and the value of non-model organisms ...... 71 5.12 A dialogue of experimental and mathematical approaches in evo-devo ...... 73

6 Concluding remarks 74

7 Bibliography 77

Contributed Publication I ...... 100 Contributed Publication II ...... 122 Contributed Publication III ...... 134 Abstract

The turtle carapace has been considered one of the classic examples of an evolutionary novelty. It is diagnostic of the turtle clade and probably largely responsible for its success in evolution. In more detail, it consists of ossified dermal plates overlaying laterally extended ribs, as well as a regular array of epidermal scutes. Interestingly, the unique and complex mosaic pattern formed by the carapacial scutes is highly conserved amongst almost all extant turtle taxa. Since the morphogenesis of this remarkable pattern has not been studied much before, I present, in this dissertation, a novel computational approach to elucidating carapacial scute development. In a mathematical model, a combination of two mechanistically different reaction-diffusion systems together with growth reproduces scute formation in the carapace. In order to disentangle the underlying gene regulatory network, we disturb carapace development experimentally in embryos of Trachemys scripta and use the model to understand the resulting phenotypes. We show how the carapacial ridge acts as a signaling center which is required for the onset of scute development. Furthermore, the model suggests how phenotypic differences between species might arise and explains, in line with statistical data from different turtle species, how environmental stress, rather than genetic mutations, leads to the very specific patterns of variation found within turtle populations. Thus, our results suggest that environmental stress might be an underrated source of phenotypic variation within species. Developmental insight helps us understand why this particular variation occurs and at what frequencies. In order to further understand how environmental factors might interfere with development and to be able to account for more realistic tissue growth a1nd biomechanics, I present a more general multiscale model of animal development that integrates biomechanics, GRN dynamics and cell behaviours in different, interacting cell types and tissues. Thus, this model does not only allow an advancement of our understanding of mechanisms by which phenotypic variation in the developing turtle scute might be generated, it is also a useful tool to understand more general features of development and variation. Finally, I discuss my work in the context of previous hypotheses about the origins of turtle scutes as well as their variation. By a comparison with the development of other ectodermal organs in different vertebrates, I try to integrate turtle scutes into a wider evolutionary background. List of Contributed Publications

The following publications, all submitted and accepted in peer-reviewed journals, are contributed to the dissertation:

I : Jacqueline E. Moustakas-Verho, Roland Zimm, Judith Cebra-Thomas, Netta K. Lem- piläinen,Aki Kallonen, Katherine L. Mitchell, Keijo Hämäläinen,Isaac Salazar-Ciudad, Jukka Jernvall and Scott F. Gilbert: The origin and loss of periodic patterning in the turtle shell. Development (2014) 141, 3033-3039 doi:10.1242/dev.109041

II: Roland Zimm, Blair P. Bentley, Jeanette Wynecken and Jacqueline E. Moustakas-Verho. Environmental Causation of Turtle Scute Anomalies in ovo and in silico. Integrative and Comparative Biology, Volume 57, Issue 6, 1 December 2017, Pages 1303–1311, https://doi.org/10.1093/icb/icx066

III : Miquel Mar´ın-Riera,Miguel Brun-Usan, Roland Zimm, Tommi V¨alikangas,Isaac Salazar- Ciudad. Computational modeling of development by epithelia, mesenchyme and their interactions: a uniﬁed model. Bioinformatics, Volume 32, Issue 2, 15 January 2016, Pages 219–225, https://doi.org/10.1093/bioinformatics/btv527

Author contributions In study I, J.E.M-V. and J.C-T. acquired specimens and performed culture experiments; L.E.M-V. and N.K.S. performed gene expression studies; A.K.,J.E.M-V. and K.H. performed the micro-CT experiments; R.Z. and I.S-C. designed the model and R.Z. implemented it; K.M. participated in designing the project; J.E.M-V., R.Z., I.S-C., S.F.G. and J.J. designed the project and wrote the paper. In study II, all authors contributed to designing the study; data from live turtles was collected by B.P.B. and J.W. R.Z. performed model simulations and data analysis. R.Z. and J.E.M-V. wrote the paper. In study III, M.M-R., M.B-U., R.Z. and I.S-C. designed the project, implemented the model, with particular focus of each on diﬀerent software modules, and wrote the paper; T.V. helped in implementing the graphical interface; R.Z. conceived and implemented the GRNMaker interface; all authors contributed in testing and applying the model to biological exemple problems. 1 Literature review

“The harmony of the world is made manifest in Form and Number, and the heart and soul and all the poetry of Natural Philosophy are embodied in the concept of mathematical beauty.” D’Arcy Wentworth Thompson

“Cell and tissue, shell and bone, leaf and ﬂower, are so many portions of matter, and it is in obedience to the laws of physics that their particles have been moved, moulded and conformed. [...] Their problems of form are in the ﬁrst instance mathematical problems, their problems of growth are essentially physical problems, and the morphologist is, ipso facto, a student of physical science”. D’Arcy Wentworth Thompson, On Growth And Form, 1917/1942 [1]

“[...] morphological evolution must be studied from an ontogenetic perspective. [...] We need to view the organism as an integrated whole, the product of a developmental program and constrained by developmental and functional interactions. In evolution, selection may decide the winner of a given game but development non-randomly deﬁnes the players.” Pere Alberch, 1980. Ontogenesis and Morphological Diversiﬁcation [2]

Since old times, humans have shown interest in peculiar morphological structures of animals and tried to build hypotheses about their origins, as evidenced by a plethora of ancient legends and tales trying to explain natural phenomena. Darwin, and a few others, proposed that a sequence of organismal descent with mutations should be sufficient to explain how organismal features originated, but struggled to accommodate into this theory features and structures, whose descendence could not be traced back to an evident sequence of intermediate homologues. Later evolutionary biologists tried to account for different magnitudes of morphological variation by distinguishing between mico- and macroevolution, attributing many prominent morphological structures with lacking or questionable homologues to the latter process. Such structures have commonly be termed novelties and received much attention, despite (or, arguably, because of) the difficulty to define novelties in the first place. For several reasons, novelties have been met with particular interest from evolutionary and developmental biologists. First, novelties have been often used as diagnostic features distinguishing clades of organisms and, thereby, used to characterize organismal biodiversity and suggest phylogenetic relationships between clades. Second, novelties have often been associated with ecological or physiological functions which allowed species to occupy ecological niches and eventually triggered further diversification. Hence, many novelties have been considered drivers of evolution. Third, the elucidation of their developmental origins has often been a challenge, making them an interesting study objective for scientists interested in development: In order to ultimately understand how novelties emerge and undergo further modification and diversification, we need to understand the mechanisms of their development and which range of phenotypes (or phenotypic variation) these mechanisms are capable of producing. This has been one of the main goals of evo-devo, a discipline aiming at explaining phenotypic change in evolution by understanding how the underlying development changes. In my thesis, I attempt to contribute to this endeavour. An evident example of a morphological novelty is the turtle shell. This structure has been used as a diagnostic feature of the taxon of Testudines (turtles), as it is the most blatant autapomorphy in this monophyletic group and has endowed these animals with an exclusive fitness advantage by providing mechanical protection, albeit at the expense of agility. Anatomically, the turtle shell is composed of a dorsal and a ventral shield, called carapace and plastron, and develops by ossification

1 of mesenchymal tissue overlying the ribs and vertebrae. On top of those shields, an array of heavily keratinized scutes forms in a characteristic pattern. Despite its prominence, no study has, to my knowledge, been dedicated to explaining the development of the turtle scute pattern. One of the main reasons for this might be that turtles are, as compared to classic model organisms, less amenable to experimentation. In this thesis, I am using a novel mathematical model to shed light on the developmental mechanisms that produce the scute pattern in the turtle carapace. I also use this model, in comparison with real turtle data, to understand which variation emerges in the scute pattern within and between turtle species and which processes cause anomalous scute patterns, as frequently observed in many species. Finally, I try to place turtle scutation in the wider context of ectodermal appendages in order to discuss how this novelty can be connected to other structures, if viewed from an evo-devo perspective. Furthermore, I present another, more general model of development which may allow to overcome limitations of the speciﬁc turtle scute model, but also to investigate the origins of phenotypic variations, and possibly of novelties, in a more general manner.

1.1 Novelties In order to understand evolution, it is crucial to know how phenotypic characters change through mutations and how their distribution in populations changes through drift and selection. The main focus in evolutionary biology has traditionally been placed on selection, but studies about how phenotypic traits evolve under specific selective pressures in a specific organism are only informative, if some knowledge - or at least a set of sophisticated assumptions - is given about the options selective forces can act upon, i.e. phenotypic variation arising from genetic or epigenetic mutations. Since phenotypic variation is ultimately produced by development, knowledge about the underlying developmental processes is required if we wish to predict the scope of phenotypic variation to which mutations can give rise by affecting developmental dynamics. However, not all phenotypic changes are equally likely or even possible at all. In simple traits, the potential directions of phenotypic change can typically be expressed as a vector of low dimensionality, often as an only one-dimensional measure. Examples of such traits are body size, mass, lengths of particular body parts, pigmentation etc. In such cases, any mutation will simply affect the single phenotypic variable and, usually, any value between physiological boundaries can be attained or “selected” for, although increases and decreases may not be equally easy to achieve. Complex phenotypes can change simultaneously in a multitude of directions. For example, the shape of a human face has many traits that can differ by different amounts, partly varying independently, partly covarying [3, 4, 5]. Yet, it is possible that, due to complex interactions of developmental factors (genetic and environmental), some trait combinations are developmentally impossible, at least without substantial changes to the architecture of the underlying developmental processes [6, 7]. More generally, it has been argued that phenotypic complexity is intrinsically correlated with complexity of the patterns of variation [8]. This means that, in complex traits, the direction of phenotypic change becomes more unpredictible and less gradual. Understanding development can therefore contribute to predict which phenotypic changes are expected to be more or less likely. Another important concept linked to variational properties is the one of novelty. The idea is that, in some taxa, phenotypes and traits that differ significantly from their ancestors’ have emerged. Often comparably easy to recognize, defining novelties has proven substantially difficult and, thus, numerous different definitions have been proposed [9, 10]. Some of those definitions are based on an evolutionary perspective. Ernst Mayr defined novelties (innovations) as any structure that allows the organism to perform a function that its ancestors were not capable of [11]. This definition has evident intrinsic limitations. First, it translates the problem of distinguishing between quantitatively and qualitatively different forms to the problem of distinguishing between quantitatively and

2 qualitatively different functions, which may be equally problematic. Second, qualitatively different functions can be often achieved by quantitative changes in structure. For example, the ability of flight may be achieved by a gradual increase of the surface area of some appendages or tissues. Also, the functional context of the original novelty, i.e. the origination of the respective appendage or tissue in one ancestor, may have had nothing to do with the function for which we characterize it as novelty. In my example, the original function of a wing structure whose further evolution has enabled flight may have been thermoregulation, communication between individuals, improved climbing or running abilities, etc. Thus, Mayr’s definition takes an a posteriori perspective which does not include any inferences on the actual origination of a novel structure. A different definition was suggested by Müllerand Wagner [12]. These authors proposed to call a structure a novelty if neither homology to structures in the ancestors nor serial homology to any structure within the same organism can be established. This definition is quite restrictive, since most paradigmatic examples of novelties (e.g. appendages in Arthropods) have been suggested to show some sort of homology (morphological, developmental, molecular) with previously existing morphological structures [13]. Even more problematic, the concept of homology is at least as troublesome as the concept of novelty itself. Introducing a more permissive definition, Gerd Müller[14] suggested to define novelties as qualitative changes of structures that represent a discontinuous departure of a character from its ancestor (i.e. a non-gradual character change). Although it is still unclear how a “discontinuous departure” can be unambiguously identified, particularly if the specific changes in the ancestors are unknown, this definition appears intuitive and less ambiguous than the previously introduced ones. As a response to criticisms calling the previous definition proposed by Müller and Wagner too restrictive and difficult to apply to real structures, G.Müllerpublished an update to it in 2010 [15], suggesting to distinguish three main categories of novelties: Type I novelties, comprising the emergence of major body plans, type II novelties, including additional structures without homologues and type III novelties, comprising changes in the variational patterns of existing structures, roughly corresponding to what has been often called a “breakdown” of developmental constraints Hallgrim- son et al. [16] tried to create a synthesis between Mayr’s evolutionary and Müller’s homology-based (i.e. ultimately developmental) approaches, by identifying novelties as transitions between peaks in the fitness landscape (i.e. changes in the evolutionary trajectory) coupled with overcoming “developmental constraints” [6, 17, 7], by allowing for the development of variation that could previously not be produced. While the first part of this definition is basically a more general and more quantitative version of Mayr’s, the second links novelties to changes in phenotypic variation, which is interesting. Yet, it requires knowledge about the specific patterns of variation that may be cumbersome or impossible to collect. This definition also includes changes that might appear very subtle. Peterson and Müller[18, 19] suggested distinguishing between process-based definitions, such as Hallgrimsson’s, that focus on how novelties emerge in evolution and become fixed in populations (the process of innovation [20]), and character-based definitions, such as Müller’s,which define novelties based on their structural features. While the former fall short on being helpful in the identification of novelties and tend to overrate their frequency, the latter require additional hypotheses about their accommodation within the evolutionary theory. There are further attempts to define what a novelty is [18, 9, 21, 22], but, as the reviewed and discussed definitions, all of them include limitations, mainly a distinction between qualitative and quantitative changes that is difficult to establish, or a reliance on other, similarly problematic, concepts. The turtle shell has been often regarded as a paradigmatic case of novelty - almost independently of the definition -, as it is a structure which has not been successfully homologized with any structure in related taxa and, thus, fulfills even the most restrictive definition of novelty proposed by Müller and Wagner. Functional novelty sensu Mayr, too, can be attributed to the turtle shell, since it clearly fulfills a protective function that is not realized by structures in extant sister clades. A few other functions of the turtle shell have been suggested [23], but an improved protection from predators

3 remains the most prominent one. However, it is impossible to tell with certainty whether protection of the trunk was indeed the initial function of the first phenotypic changes to the reptile body plan that paved the later evolutionary road towards the turtle shell. Comparing it to the original reptile body plan, the turtle shell represents a clear departure from the original body plan of the ancestors of turtles. Finally, the emergence of the turtle shell created new directions of phenotypic variation, while certainly affecting the direction in which selection was acting in the ancestors of the turtle lineage, fulfilling the respective criterion of Hallgrimson’s definition of novelties.

1.2 The origins and diversity of the turtle shell Like the integument of other reptilians, turtle skin contains mainly α-Keratins, making it rather flexible, while lacking the characteristic reptilian scales made from β-Keratins [24], as found in all Squamata and in Archosauria (although the different scale appendages in Archosauria may differ substantially from the ones in Squamata [25]). Instead, protection is achieved in the Testudinata by a large ossified shell which consists of a dorsal (carapace) and ventral (plastron) part. In most species, these structures are composed of two layers: First, bony plates that in the case of the carapace overlay flattened and laterally extended ribs, and, second, a integumental layer of horny scutes inserted between epidermis and dermis, which are made of a variety of β-Keratins [26]. Due to the apparent similarity between turtle scutes and reptilian scales, homology has been suggested for these two kinds of structures [27]. The arrangement of the carapacial scutes follows, with certain exceptions as detailed below, a stereotypic pattern 1.2: A central series of large scutes, called vertebral scutes, stretches anterior-posteriorly from neck to tail. This series is flanked on both sides by a series of costal scutes (also called pleural scutes). The centers of vertebral and costal scutes are positioned alternatingly, if seen from a lateral perspective. The outer margin of the carapace is covered circumferentially by a series of significantly smaller scutes, which are called marginal scutes. In addition, one nuchal scute, that may eventually be split into two, connects the neck region of the animal with the vertebral scutes. Carapacial structures covered by minuscule osteoderms have been identified in mesozoic non- turtle reptiles. One notable example is the shell integument of the Placodont Sinosaurosphargis yunguiensis, a Middle Triassic Placodont species which has been interpreted as a member of a sister clade of Testudinata [28], although the homology between turtle scutes and those osteoderms is questionable. Another Triassic Placodont, Henodus, even possessed larger scute elements (with both ossified and cornified elements) inside a carapacial and plastral shell, thus resembling turtles anatomically even more. Presumably, these fossil forms represent parallel attempts by independent, yet somewhat related, reptile lineages to evolve a protective shell based on a shared developmental innovation that might have occurred in a common ancestor. This innovation was suggested [29, 28] to be the developmental mechanism which allowed for rib entrapment in a specialized dermis and subsequent dermal ossification, possibly in the wake of the presence of large aquatic predators. Importantly, this hypothesis would imply that the origin of the carapace predates the origins of the turtle clade, making it an autapomorphy of the so-called Pantestudines subclade of Reptiles. Within the Pantestudines, turtles are simply the only group to have survived until today. However, it has to be emphasized that such an ancestral autapomorphy in the Pantestudines might have, in the beginning, only allowed for the development of a substantially less spectacular carapace than in extant turtles and the mentioned Placodont taxa. Size and shape of the turtle carapace can vary substantially. The smallest species Homopus signatus measures only 6-10 cm, while the carapace of Dermochelys coriacea can reach a length of more than 2 meters. Fossil forms ( Archelon sp., a close relative of Dermochelys ) did reach even larger dimensions of up to around 3 meters. Carapace outlines can be round or elongated and the 3D shape usually reflects the particular habitat a species lives in: While terrestrial species often

4 show saddle- to extensively dome-shaped carapaces, the shell of many aquatic species, as well as of terrestrial species that need to squeeze into narrow niches such as Malacochersus, is flat. In some clades, ridges extending in anterior-posterior direction form on the scutes, a feature that might be correlated with aquatic environment. Even closely related species may have very differently shaped carapaces [30] and studies in limnic turtles have found intra-species carapace shape variation to be correlated with habitat differences [31]. Other authors [32, 33] found consistent changes of carapacial shapes in the transition from terrestrial to aquatic habitat in two independent lineages, using morphometric methods. Further diversity is seen between the scutes of an individual: their shape can differ substantially from anterior to posterior. In some species, vertebrate scutes are larger than costal scutes, whereas the opposite is true in others. Scutes can be elongated anterior-posteriorly or laterally. They can be thin, thick, flat or domed, which is mainly a consequence of the specific way keratinization proceeds during growth periods of the animal, either by proliferation of keratinocytes or increase in β-keratin production [34]. Some species form additional ornaments, (e.g. Chryse- mys, Trachemys, Malaclemys, Rhinoclemmys, Terrapene ornata) or spines (e.g. Macrochelys). In addition, melanocytes and other kinds of chromatophores [35] can produce characteristic patterns which may serve protective (i.e. through aposematic or camouflage patterns) or communicative purposes. An extreme modification of the turtle shell occurs in softshell turtles, which show either partially or completely reduced ossification and no scutes. Such a phenotype has evolved at least twice independently (Trionychidae/Carettochelyidae, Dermochelyidae). However, it is not clear yet if the loss of the shell was the consequence of the same developmental modifications in all lineages in which it happened. Despite of all this variation in extant turtles, it is quite surprising that the scute number seems to be highly conserved within the turtle clade [36], especially amongst non-marine turtles. In all extant terrestrial, freshwater hard-shell, and most marine species, the carapace comprises 5 vertebral, 4 pairs of costal (or pleural) and 11-12 pairs of marginal scutes (the most posterior pair is also named supracaudal), as shown in 1.2. Macrochelys may show a few additional submarginal scutes between the marginal and the costal scutes in the central part of the carapace [37], the only terrestrial genus with a numerically divergent scute pattern (although not all individuals may show this particular pattern of variation). Sometimes, Notochelys platynota is considered to have a divergent scute formula, as supernumerary vertebral scutes are unusually common [37]. More diversity is seen in marine turtles in which the number of scutes is diagnostic of certain species. Caretta caretta and Lepidochelys kempii usually have 5 vertebral and 5 pairs of costal scutes, and 13 pairs of marginal scutes [38]. While in C.caretta observations of up to 6 vertebral or costal scute pairs are common [39], even much lower or higher scute numbers have been reported in L.kempii [40], although rather exceptionally. In Lepidochelys olivacea, the most frequent numbers of vertebral scutes are 6 or 7 and the number of vertebral scute pairs ranges between 6 and 9 [38, 37], while 13 pairs of marginal scutes are present, on average. While a “normal” scute formula can be identified for virtually all turtle species, allowing to clearly distinguish between “normal” and “abnormal” scute patterning, this seems not to be the case, according to Pritchard, for L.olivacea, in which variation is so common that a dominant scute formula can be hardly established [41]. The most different carapacial integument of all marine turtles is found in Dermochelys, the largest extant turtle. In this species, the carapace is covered by a dense skin and, instead of scutes, an array of tiny osteoderms is present (possibly in a way similar to the shell integument in Triassic Placodonts) [42]. These extraordinary departures from the stereotypical scute formula are one of the reasons why, in the past, sea turtles had been placed as a sister group to all terrestrial turtles. More recent studies group sea turtles within the Cryptodira [43], rendering terrestrial turtles paraphyletic.

5 1 N 2 1 3 1 4 2 5 2 6 3 7 3 4 8 4 9 10 5 11 12

Figure 1. Turtle scute patterning. One example of standard carapace scute patterning in hatchlings of Natator depressus, representing the most common scutation. This pattern consists of a series of ﬁve vertebral scutes along the midline of the carapace (yellow numbers), anteriorly anteceded by one or two nuchal scutelets (light blue), two series of four vertebral scutes (magenta numbers) and a circumferential band of usually 24 marginal scutelets (teal numbers). Besides, two hatchlings with anomalous scute pattern (arrowheads pointing to additional scutes) are depicted.

As a group, extant turtles show a nearly cosmopolitan distribution, although omitting extreme climate zones. Already in the Triassic, a nearly world-wide distribution of turtles can be assumed (e.g. Proganochelys) [37]. The earliest fossil that has been classified and commonly accepted as a turtle (although some doubts about this classification have been raised [44]) is Odontochelys [45], an aquatic species from the Norian period (late Triassic, 310 Ma ago), which differed notably from all extant turtles by the fact that it lacked the dorsal shell (carapace), while showing a ventral shell (plastron) similar to the one in extant species. While some authors regard Odontochelys as evidence of an independent evolutionary origin of plastron and carapace [45], or of plastron and carapace ossification [46, 47], others have argued that the lacking carapace might rather evidence a secondary reduction in an analogous way to the loss of scutes in softshell turtles following adaptation to an aquatic habitat [48]. Even earlier reptiles that are considered close relatives of the last common ancestor of all turtles, such as Pappochelys [49] and Eunotosaurus [50], already showed laterally elongated and flattened ribs, but most likely no shell. The first undisputed turtle which bore a complete shell was Proganochelys, living only 10 Ma after Odontochelys in different places of the Laurasian continent. This genus already showed both a plastron and a carapace, the latter composed of ventral, costal and marginal scutes. In Odontochelys, the number of vertebral scute pairs was 4 and two parallel series of marginal scutes (marinals and supramarginals) were present, a pattern not found in any extant turtle species. A similar pattern has been found in the North American Chinlechelys [51] as well as in the South American Australochelys [52]. Additional supramarginal scutes could still be seen in the late-Jurassic Platychelys oberndorferi [53], albeit only in the anterior part of the carapace and most fossils of related taxa do not show any supramarginals. Idiochelys, a species from the Upper Jurassic, still showed 4 vertebral scutes [37], whereas the majority of Jurassic turtles had already the same number of vertebral scutes as recent taxa. These data suggest that turtles may have undergone a gradual transition from four to three vertebral scutes during their evolution. A very different scute pattern was found in Sayka, reviewed by [54], which exhibited up to 10 vertebrals and 9 costal scutes. Since this was a cenozoic genus, its divergent scute pattern

6 cannot be considered ancestral. Unlike extant turtles, some ancestral turtle taxa exhibited very wide vertebral scutes, while their costal scutes were more similar in size to the marginals. Examples for this scute pattern were Idiochelys, Proterochersis [55], Kayentachelys [56] or the Upper Createous Kallokibotion [57]. Intriguingly, however, the vast majority of fossil turtles from Jurassic onwards show the very same scute patterns as extant terrestrial species, meaning that the turtle carapace pattern has been a highly conserved structure throughout most of the time of the existence of the turtle clade. Interestingly, there are frequent numerical departures from the main scute pattern. Throughout the turtle clade, individuals with aberrant scute formulae have been found, examples of which are displayed in 1.2 and 1.3 as well as in the literature (see 1.8 for a list). Rivera [31] counts 114 species in which scute anomalies have been reported and suggests that they occur predominantly in terrestrial (versus aquatic) species, although high rates of scute anomalies are as well and frequently reported in the marine turtle clade. While some individuals have additional scutes, i.e. supernumerary scutes either within or between scute rows, others show scute fusions. Some individuals present an alternating “zigzag” pattern within a part of or the entire vertebral scute series. While some anomalies occur symmetrically in both carapacial hemispheres, the majority of anomalies is asymmetric with respect to the left-right axis [58, 59]. Anomalies may affect marginal, costal, and vertebral scute series independently, or several series at the same time. Marginal anomalies are probably the most common, but may often go unnoticed. Mast et al. [40] found that both vertebral and marginal anomalies are significantly more common in sea turtles than costal anomalies (although the authors excluded non-meristic scute modifications). Female individuals seem, in some species, to be more often affected than males [60]. Although it has been proposed that anomalous turtles are less likely to reproduce, based on allegedly higher anomaly frequencies in hatchlings than in adults [61, 62, 63], there are, to my knowledge, no experiments provide solid evidence in favor or against this claim. Other authors suggest instead that only highly divergent anomalies are actually selected against [40] and that individuals with moderate scute modifications do not face any survival disadvantages compared with their conspecifics without scute anomalies.

1.3 Development of turtle scutes The turtle shell is a structure that is diagnostic and unique to the turtle clade. It is composed of two parts, a ventral part, the plastron, and a dorsal part, the carapace. Surprisingly, the specimens of Odontochelys unearthened so far show exclusively the bony plates of the plastron and the ribs, but no parts of the dermal carapace, i.e. neither dorsal dermal plates nor scutes. It is therefore assumed that the evolution of plastron and carapace took place independently [45]. However, two alternative explanations are possible: It is hard to rule out, although unlikely, that the carapacial part of the shell was, by accident, not preserved in the fossil record. Also, it has been suggested that the turtle shell evolved several million years earlier in a yet-to-be-discovered ancestor of Odontochelys and that the latter already displayed a secondarily reduced shell [48, 64]. This secondary reduction would have been due to the aquatic habitat; however, slightly later Triassic stem turtles such as Proganochelys and Palaeochersis were identified as terrestrial species based on morphometric evidence [65]. Shell reduction has been observed independently in several lineages (see 1.2 ) as a result of adaptation to an aquatic habitat. A shell reduction that only affects the carapace would make sense in a marine environment in which predatory attacks from below are much more likely or dangerous than from above. However, in extant aquatic turtles, shell reduction typically affects both the ventral and dorsal part of the shell. Arguing against the hypothesis of secondary carapace reduction, Burke and Nagashima et al. [46, 47] highlight similarities between the morphology of Odontochelys and early stages of shell development in turtle embyos, suggesting that carapace development might

7 recapitulate early stages of carapace evolution and that the endochondral parts (the ribs) of the carapace might have evolved before its dermal parts. Thus, understanding the development of the carapace is crucial in elucidating the origins of this particular novelty. In the following, I focus on the development of the carapace rather than the plastron. Although many of the developmental processes by which both structures are built are likely to be homologous, there are clear differences between carapace and plastron development: the origin of the mesenchymal cells that make up the plastron as well as the nuchal scutes, is, unlike the carapace, of neural crest origin [66, 67], while subsequent developmental processes such as dermal ossification show differences as well as similarities between plastron and carapace [68, 69]. Future developmental studies on plastron development will give evidence in favour or against the hypothesis of an independent evolutionary origin of carapace and plastron. Turtle trunk development begins to diverge significantly from the development of sauropsid reptiles (including birds) with the formation of the so-called carapacial ridge, a swelling of the lateral axial mesenchyme consisting of condensed mesenchyme and overlaying columnar epithelium [70] that forms in the dorsal region. Unlike in developing sauropsids, where the rib mesenchyme precursors grow downwards, following the lateral body wall to form the later rib cage (“axial arrest”), rib mesenchyme precursors in turtle embryos extend instead perpendicularly to the body wall and towards the carapacial ridge [71, 28, 72, 70, 73], schematically shown in 1.3. Comparative gene expression studies [29] suggest that the formation of the carapacial ridge might constitute a cooption of developmental pathways that are commonly active in early amniote limb bud development and that one of the co-opted genetic pathways (one candidate being the canonical Wnt-pathway [74]) may be involved in the failure of the developing ribs to continue growing downwards. It has been shown that paracrine signaling synchronizes the growth of rib mensenchyme and carapacial ridge: The distal tip of the rib mesenchyme expresses Fgf8, while Fgf10 is expressed in the condensed mesenchyme of the carapacial ridge, but only in proximity to the distal rib tips [75, 76]. Both Fgfs mutually activate each other, establishing a positive feedback loop (see scheme in 1.3). In the absence of Fgf signaling during this stage, neither the formation of the carapacial ridge nor a lateral outrgrowth of the rib mesenchyme is observed. If the carapacial ridge is removed partially, ribs in the affected region will bend towards one another, rather than growing straight and separately [71, 72, 77]. Fgf10 also appears to have a chemotactic effect on the growing rib mesenchyme, as ectopically implanted beads that release Fgf10 will divert the rib growth towards them [75]. In addition, Fgfs have been reported to mediate chemotaxis also in other developing organ systems [78, 79]. These results suggest that it is mainly the chemotactic effect of Fgf10 on Fgf8-expressing rib mesenchyme that is responsible for the peculiar rib growth in turtles. Thus, the dynamical interaction of Fgfs expressed in rib precursor cells with the condensing mesenchyme in the carapacial ridge gives rise to the shape of the future carapace. In addition, the rib mesenchyme expresses growth factors, such as BMPs, along the carapacial ridge. These growth factors are crucial for the further development of the carapace and its scute pattern [77, 75]. BMPs, in particular, have been suggested to be responsible for intramembranous ossification in the carapacial dermis [75], as they are known to be crucially involved in various aspects of bone formation and mesenchymal condensations in vertebrates [80, 81, 82]. Other growth factors expressed in the carapacial ridge are Fgfs [76], Shh, Gremlin [83], Msx-1 [84], and Sostdc-1 [Moustakas-Verho, oral communication]. The fact that some of those growth factors are known to regulate each other in various other tissues (Shh, BMPs, Gremlin) indicates that they are probably part of a dynamic gene regulatory network in the carapacial ridge. Some of those growth factors (Shh,Msx1) are expressed in the ectoderm while others (Gremlin, BMPs, Fgfs) are expressed in the underlying mesenchyme. Thus, ectoderm and mesenchyme appear to regulate the patterning of the carapacial ridge through paracrine crosstalk, in a similar manner as in other ectodermal appendages (more detailed discussion in 1.6).

8 It has been proposed that the carapacial ridge acts as a lateral organizing center for the subsequent patterning of the entire carapace and that loss of that signaling has prevented further patterning in softshell turtles [77, 85]. In fact, patterning along the carapacial ridge, which will also ultimately form the marginal scutes, precedes patterning of the more central scute primordia (giving rise to vertebral and costal scutes), with costal scute primordia appearing prior to the vertebral ones (i.e. the sequence of appearance in development is lateral to medial) [59]. Costal and vertebral scute primordia become visible as geometrically arranged patterning centers whose placement already resembles the positions of the future scutes [59]. The vertebral scutes form as independent pairs of primordia which are, usually, going to fuse [86]. Gene expression in those primordia is distinct from gene expression in the surrounding tissue [87]. Those scute primordia are considered anatomical placodes, as they consist of a dermal and epidermal thickening (as histologically identified by elongated epidermal cells that, in later stages, become flatter and organize in multiple layers) [59, 54, 70, 69]. Concomitantly with the expanding rings of gene expression, tissue bulges, i.e. mesenchymal thickenings that are initially confined to the placodal areas extend rapidly in centripetal direction from the placodes, forming the future scutes [59, 54]. Histological sections reveal that the actual placodal center may in later stages even form a slight invagination into the scutal bulges, but this might be just a transient morphology [59]. In the places at which scutes collide, the epithelial cells undergo some cell-morphological modifications, invaginate partially, and will later form the interscutal sulci or furrows. Interestingly, the furrows between vertebral scutes seem to form before the furrows of the costal scutes [59], i.e. their temporal sequence is the contrary of the order in which scute primordia appear first. Also, the epithelial cells inside the furrows have been reported to growth downwards (i.e. they invaginate) and proliferate and, thus, contribute to a deepening of the inter-scutal areas [34]. At the same time, the entire carapace tissue grows, leading to a laterally widened carapace, although the rate of growth may differ greatly between different carapacial regions. Differences of growth have been suggested to be responsible for differences in scute shapes and proportions between different turtle species [86]. Concomitantly with the centrifugal expansion of the scutal placodes, underlying mesenchymal cells between ribs and integument start undergoing dermal ossification (a process which comences and completes at different times or stages for different scutes). Dermal ossification is generally regulated by expression of BMPs [82] and continues, in the turtle shell, throughout late embryogenesis and even after hatching [85, 88]. Neural crest cells have been shown to critically contribute to this process [66, 67]. It is debated whether ossification in the turtle carapace and plastron evolved as a derivation from the endoskeletal rib mesenchyme or if it has emerged as an independent dermal ossification in evolution [28]. In parallel, a thick layer of keratins is secreted between dermis and epidermis in order to consolidate the scutes [89]. Although the scutes of adult turtles are only made of β-keratins,α-keratin can be observed embryonally above the layer of β-keratines [69]. In softshell turtles, however, the carapace is covered mainly or exclusively by α-Keratins [69, 34]. While the characteristic flattened rib shape is found in all extant turtle species, dermal ossification and scute keratinization may be reduced or absent secondarily in soft-shell turtles. Thus, these turtles, especially Pelodiscus sinensis, have been chosen as model organisms to identify, by comparison with hard-shell turtle species, molecular pathways and developmental mechanisms that are essential for the development and ossification of the carapace. Interestingly, progressive de-ossifiction has been described in Malacochersus [90, 91, 92] as an adaptation to body flexibility, suggesting that the process of dermal ossification in the carapace is not too constrained to be modified in evolution. In those softshell turtle species in which developmental studies have been performed, key developmental events have been found to differ from those of the remainder of turtles. Although the formation of the carapacial ridge and the lateral outgrowth of the ribs are observed equally in softshell and hardshell turtles [29, 93, 94, 95], several downstream growth factors (Shh,Fgfs,BMPs) are not expressed in the carapacial ridge of the former. In consequence, neither scute primordia

9 formation nor dermal ossification take place in those species. It is worthwhile mentioning that there is, in most turtles, a correlation between integumental scutes and the underlying dermal bones, which form by dermal ossification around the ribs. The dermal bones are found underneath both the centers of vertebral scutes and the borders between these. A causal role for those ossifications in the process of scute formation has therefore been discussed. However, turtles hatch without complete carapacial ossification [88, 96], while the scute pattern is completely established. If signals from the rib are crucial for both the induction of the dermal ossifications [77, 75] and the pre-pattern in the carapacial ride (which will, as shown in , give rise to the scute primordia formation), the spatial overlap of both structures may be indirected correlated due to different inductive signals stemming from a common source.

other Amniotes

Tu r t l e s

CR induces Fgf10

Rib mesenchyme Fgf8 Chemotaxis

Figure 2. Turtle carapace development initiation. The schematics represent sections through the trunk at two diﬀerent stages of embryogenesis. Dark green represents the developing sclerotome-derived rib mesenchyme, red the muscle plate, grey the neural tube. On the right side of each trunk section, the lateral body wall is depicted as a downward extension. Dark blue marks the spot in the turtle embryo in which Fgf10 is expressed, the place of the future carapacial ridge (CR). While, in other amniotes, the developing rib mesoderm invades the lateral body wall, giving rise to the downgrowth of the rib cage, Fgf signaling in turtle embryos causes the rib mesenchyme to be diverted towords the carapacial ridge, giving rise to the laterally expanded carapace. The inset shows the positive feedback between Fgf8 in the tip of the developing rib mesenchyme and Fgf10 in the carapacial ridge that lead to directed rib mesenchyme expansion. After [71, 75, 97, 28].

1.4 Development of periodic structures Many structures in animals appear at periodic or more or less regularly spaced positions throughout the body or a particular part of it. Examples of repeated structures are plentiful: the body segments of Arthropods and Annelids (although secondarily reduced in some groups such as the Hirudinata

10 and some parasitic arthropod taxa), the segments of arthropod limbs, somites in vertebrates, fingers, different types of ectodermal appendages, lobular and branching structures in many organs. Such structures have been called serially homologous. Especially obvious are repetitions within animal integuments: most ectodermal appendages, as discussed in 1.6, develop as repeated units, and body pigmentation often exhibits periodic patterns, i.e. spots, stripes and more complex combinations of these. Once the positions of the repeated units within the animal body have been specified in development, later gene expression and morphogenesis may differ between them, for instance, along the anterior-posterior axis. Thereby, different structures that began developing as parts of a larger periodic pattern may end up showing morphological distinction or assume different functions, sometimes rendering the identification of different structures as serially homologous difficult.

How to form periodic structures in the ﬁrst place

1. Periodic structures by mechanical stresses Periodic structures can be the consequence of epithelial folding processes that result from mechanical stresses and tensions. If a surface consists of different cell layers (e.g.epithelium and mesenchyme), stresses can occur due to different growth rates between the layers which then relax by wrinkling and folding [98, 99, 100]. Theoretical models showed that this process does not only explain the regular arrangement of the intestinal crypt and villi structures of the murine intestine, it also accounts for the morphological differences that exist between the intestinal folding patterns in different vertebrate species [101]. A similar model explains the gyration differences between cerebral cortices of different mammalian species [102]: The observed patterning differences are the consequence of different growth rates or different mechanical properties between adjacent tissue compartments (e.g. elasticity differences of the cell layers involved due to a different content of muscular fibers or different local compositions of the basement membrane). Similar folding patterns can even result from mechanical stresses within a single epithelial layer composed of one cell type [103, 104], e.g. by differential tensions between the apical and basal side of an epithelium or local differences in adhesion to the baseline membrane, as revealed by simulations, or even by local differences in apoptosis within epithelia [105].

2. Periodic structures by short-range (juxtacrine) signaling Cell-surface proteins of neighbouring cells can interact with each other and, thereby, integrate information about their local cellular environment. Such local cell-cell interaction can lead to global patterning of the tissue. One of the most important cell contact-mediated signaling pathways which can give rise to tissue-scale patterning is the Delta-Notch pathway [106]. It mediates lateral inhibition in adjacent cells by competitive binding of Notch to its ligand Delta, both of which are transmembrane proteins. Notch bound to Delta expressed on a neighbour cell’s surface will undergo enzymatic cleavage and, consequently, represses Delta expression in the same cell. As a result, any given cell expresses either Delta or Notch at high levels in a mutually exclusive manner, and adjacent cells end up expressing Delta (or, respectively, Notch) alternatingly. On the tissue level, this produces a regular chequerboard pattern. Diﬀerent interaction strengths between Delta and Notch, the involvement of further proteins modulating this interaction or interaction with other signaling pathways have been demonstrated to produce other regular tissue patterns (mostly distributions of two cell types in which one is more frequent than the other) [107]. Several developing organ systems (e.g. neuronal precursors [108], Pancreas cell type precursors [109], Drosophila hindgut [110], Drosophila thoracic bristles [111]) have been shown to be patterned by the Delta-Notch

11 pathway. However, one limitation is that this kind of tissue patterning takes place on the scale of a few cell diameters, although the absolute size of such a regular pattern can be increased later in development, e.g. by subsequent tissue growth and other morphogenetic processes. Another mechanism based on juxtacrine signaling that has been suggested to be capable of producing periodic patterns is adhesion-protein-mediated cell sorting with autoactivation [112]: Cells can express adhesion proteins (e.g.ﬁbronectin) that cause other cells to adhere to them. In addition, this binding upregulates the expression of further adhesion proteins (e.g. N-Cad), which consolidates the cell-cell binding process and increases “stickiness” of the newly bound cells. This mechanism has been proposed to play a role in mesenchymal condensation processes.

3. Periodic structures by long-range (mostly paracrine) signaling Other types of tissue patterning mechanisms use paracrine signaling, i.e. diffusion of growth factors (or other types of diffusive molecules that bind to receptors or are capable of directly entering into cells), allowing for transmission of information across larger fields of cells without necessarily re- quiring direct cell contact, and, thus, pattern formation on larger spatial scales. Some animals, such as Drosophila, have early developmental stages during which nuclear divisions take place without the formation of separating cell membranes (blastoderm stage). In this scenario, long-range signaling between nuclei is even more straightforward, as no cell boundaries have to be crossed to transmit information. Yet, it has to be mentioned that, alternatively to diffusion, signals can be transmitted across larger distances using only juxtacrine signaling mechanisms, too: Signals arriving in a cell by juxtacrine ligand-receptor-binding may activate the expression of more juxtacrine ligands in that cell and, thus, lead to a propagating signalling wave. Signaling molecules can also be transmitted via cytonemes or filopodia (cellular protrusions) which allow cells to signal to other, distant cells without releasing diffusive molecules into the extracellular space [113]. Such types of juxtacrine signaling mechanisms behave, macroscopically, like long-range paracrine mechanisms, which is why I do not distinguish between them and actual diffusion here. There are two fundamentally different ways how long-range signals can interact to generate patterns at large tissue scales [114]:

a) Hierarchical In hierarchical patterning mechanisms, patterning occurs through a regulatory network of paracrine signaling factors and intracellular transcription factors (all of which I will refer to as “factors” in here) with a clear hierarchy between different levels of interacting factors: Factors on a higher level (“upstream”) either activate or repress, transcriptionally or post-transcriptionally, the activity of factors on lower levels (“downstream”), but not vice versa, thereby forming a gene regulation network (GRN) with only top-down interactions. In order for downstream factors to be expressed in a stripe- or spot-like manner, the concentration of the most upstream factor or factors has to be non-homogeneous (forming a gradient, a field of high and a field of lower concentration or a more complex initial pattern). The strength by which upstream factors affect the production of downstream factors is often different for each interaction in the gene regulation network. In addition, diffusing factors can have different diffusion rates, but this is not strictly necessary for hierarchical mechanisms to form patterns. Such a cascade of activation or repression of downstream factors, together with paracrine signaling, can subdivide a tissue into a set of spatial territories each of which ends up with a characteristic gene expression pattern different from the gene expression patterns of the adjacent territories. A paradigmatic case of hierarchical patterning is the segmentation of Drosophila in which different sets of hierarchically structured genes progressively subdivide the anterior-posterior body axis.

12 b) Emergent In emergent patterning mechanisms, there exists no distinction between hierarchical levels of signaling factors and even the initially expressed factors can be signaling targets of their own targets (directly or indirectly), forming positive and negative feedback loops. This means that the dynamics of emergent GRNs are often difficult to predict, and can be highly counterintuitive. Emergent patterning mechanisms can also give rise to oscillations (periodic changes of gene expression in time), waves (periodic changes in space and time) and chaotic gene expression patterns. However, some emergent GRNs give rise to spatially and temporally stable patterns. The most important group of emergent mechanisms is called reaction-diffusion mechanisms [115] and has been associated with integumental patterning [116]. In paradigmatic reaction-diffusion systems, one signaling factor promotes (usually, but not exclusively), by increasing transcription, its own production as well as the production of a second signaling factor which suppresses the production of the former. Both factors are diffusive, but the inhibitor diffuses faster. From any initial inhomogeneity in the concentration of the two factors, the dynamics of such a mechanism can cause the concentration of the activator or the inhibitor to form periodically distributed local minima or maxima. Since this mechanism is perhaps the most versatile and complex way of forming repeated patterns, and especially relevant for the patterning studied in this thesis, I am going to focus on it in more detail in the following chapter.

4. Periodic structures by paracrine signaling and cell-autonomous processes Another class of patterning mechanisms integrates cell-autonomous processes and paracrine signaling. Cell-autonomous processes are cellular processes that do not depend on interactions between diﬀerent cells. Most prominently, the clock-and-wavefront mechanism [117] that is responsible for somitogenesis in vertebrates [118] as well as segmentation in short-germband arthropods [119], be- longs in this group. In this mechanism, a molecular oscillator is activated locally and autonomously in each cell. In addition, there is a paracrine factor interacting with the oscillator, which is provided either in a spatially deﬁned growth zone of a tissue or by a travelling wavefront. As the cells exit the growth zone (e.g. by being pushed away by continuous cell divisions), or as the travelling wave propagates away, the cell-autonomous oscillations are irreversibly arrested, giving rise to evenly- spaced segments. In other words, long-range signaling translates temporal oscillations into a spatial readout, typically through cells moving relatively to a front of inductive signaling in a developing tissue. Interestingly, this mechanism has been found to recruit components of the Delta-Notch pathway (e.g. [118]), the key mediator of lateral inhibition between neighboring cells (see above).

Evolutionary implications Even though all the different classes of mechanisms are equally capable of producing periodic patterns, there are fundamental differences between them with respect to evolution. Salazar-Ciudad [120] simulated a huge number of random gene regulatory networks and found that pattern formation networks could be classified into a small number of different core topologies capable of complex pattern formation (namely: hierarchical, several classes of emergent mechanisms, and a few others) and discussed evolutionary implications of these. In a different approach [114], artificial evolution was applied to a similar model of GRNs with signaling in a 1D tissue, selecting for different numbers of stripes. When selecting for a small number of stripes, the GRNs that arose in evolution to produce these stripes were hierarchical networks, while emergent networks were found to be the networks producing the optimal patterns when selecting for a large number of stripes. Furthermore, developmental systems drift acts continuously on GRNs underlying development. Even if a given phenotype develops by a hierarchical mechanism, it is unlikely that it remains hierarchial throughout

13 evolutionary time, since random mutations will change its topology, creating feedback loops between lower (downstream) and higher (upstream) levels of signaling factors (and, thus, the mechanism will cease being hierarchical). These reasons suggest that, when complex periodic patterns emerge in evolution, they should rarely develop by hierarchical mechanisms. In a review comparing different studies of in silico stripe formation [121] tenTuscher et al. concluded that the clock-and-wavefront mechanism is the most likely segmentation mechanism to evolve, which is in line with its presumably independent origination in vertebrates and arthropods (and possibly in Annelids [122]). Yet, hierarchical mechanisms do occur in animal development and some of them produce regular and somewhat complex patterning, a paradigmatic case being the early segmentation in long-germband insects. One reason for the emergence of hierarchical patterning mechanisms in evolution was proposed by Isaac Salazar-Ciudad and co-workers: [114]. The authors found that the phenotypic fine-tuning capacities of hierarchical mechanisms are substantially higher than of emergent mechanisms. Together, these findings suggest that, while emergent, as well as clock-and-wavefront, mechanisms, are more likely to emerge in evolution than hierarchical mechanisms, the latter may be selected for under a fine-tunning and stabilizing selection regime. Such a hypothesis is in line with the fact that clock-and-wavefront mechanisms are the ancestral segmentation mode in insects that was eventually, in some clades, replaced by a hierarchical mechanism. Such a transition of mechanisms might be generally likely in stable environments, when organisms are already well adapted and only incremental changes in phenotype may improve fitness further. Interestingly, theoretical work by Salazar-Ciudad and Sharpe suggests that evolutionary transitions between different patterning mechanisms might actually be counterintuitively easy. Isaac Salazar-Ciudad [123] showed that enabling diffusion alone can be sufficient to transform a clock- and-wavefront patterning mechanism into a hierarchical patterning mechanism. Sharpe et al. [124] who used a different scheme to classify stripe-forming mechanisms with small GRNs found that, in general, different mechanisms can be transformed into one another by only a few mutations. Even a simple GRN circuit consisting of three interacting components can give rise to completely different behaviours, such as bistable switches and stable or instable oscillations, simply by modifying the initial activation of the network [125]. These studies suggest that, once some mechanism capable of producing stripes arose in evolution, replacement of that mechanism due to developmental systems drift may have occurred relatively easily and frequently.

1.5 Reaction-diffusion systems One class of emergent pattern formation mechanisms that has been associated with periodic integumental structures is called reaction-diffusion systems [126, 127, 116] and was first proposed by Alan Turing in 1952 [128]. He developed the idea that a system of two active substances can form a regular and stable spatial distribution pattern of their concentrations (regularly spaced spots of high or low concentrations), only by virtue of interacting with each other and diffusing in space: One substance acts as an activator for its own and the other substance’s production. The second substance, in turn, downregulates the expression of the former. Such a situation can lead to the establishment of a stable equilibrium between the two substances (which depends on their interaction strengths and degradation rates), or a global oscillation (if the degradation rates are very different). However, if the two substances diffuse at different rates, they can give rise to very different scenarios. If the activator diffuses faster than the inhibitor (cf. 1.5, right column), a third possible situation can emerge. Let us assume a spot of higher activator concentration than in its surroundings, due to either local activation by an upstream activator, or random local variation of concentration in an otherwise homogeneous activator distribution. Due to diffusion, an activator wave will begin spreading from this point. In the initial place of activation, however, the inhibitor, produced by the activator, will accumulate and lead to a local reduction of activator concentration,

14 which, in turn, will reduce the production of inhibitor as well. Only much later, the concentration of inhibitor will decrease enough to permit a new accumulation of the activator. As a consequence, a concentric activator wave will spread centripetally and be closely followed by a subsequent wave of inhibitor. Such a situation is commonly termed travelling wave. Where several travelling waves collide, they can merge and interference patterns may arise. A very different picture can emerge if the inhibitor diffuses faster than the activator (cf. 1.5, left column). In this scenario, a locally high activator concentration will produce a local increase in its own inhibitor. This excess of inhibitor will diffuse away into the surroundings of the spot in which it was produced, thus suppressing the activator in the surroundings. Depleted from its activator, inhibitor production will cease in the surroundings, too, allowing activator concentration to increase at larger distances from the initial spot of induction and to form secondary activator maxima there. The same way, tertiary maxima will form and so on. As a result, we arrive at a periodic pattern whose wavelength (the distance between neighboring maxima) mainly depends on the differences between the diffusion rates of activator and inhibitor. Nevertheless, this is a simplification, as the formation of periodic patterns also depends on interaction strengths between the substances and their degradation rates. The critical parameter ranges between which periodic patterns occur delimit the so-called Turing instability and can be calculated analytically in some models (for a detailed demonstration see e.g. [129]). Further steady states that can be attained by Turing instabilities comprise “salt-and-pepper patterns” (i.e. very small wavelengths that cause neighboring cells to assume alternating cell states, similar to Delta-Notch signaling (see above)) and oscillatory variants thereof. A concise review can be taken from [130]. Although Turing explicitly connected his theory of pattern formation to the problem of the development of patterns in biology, his paper received no attention by developmental biologists of the time. Therefore, Alfred Gierer and Hans Meinhardt developed about two decades later a very similar hypothesis independently from Turing [115] and applied it to a multitude of different biological phenomena [131, 132]. In biological systems, gene regulatory networks (GRNs) can produce the dynamics of a reaction-diffusion system, if they include some paracrine signaling protein (usually a growth factor) that directly or indirectly activates its own production as well as the production of another signaling protein which, in turn, directly or indirectly represses the first signaling protein and travels faster than that one. In other words, extracellularly diffusive and transcriptionally interacting proteins can function as Turing’s “active substances”. Periodic patterns in nature are, evidently, more complex than a regular array of spots as predicted by the Turing model. In many cases, regular and irregular stripes appear, and some reseachers (such as Hans Meinhardt and Shigeru Kondo) dedicated a lot of work to explain those, using Turing’s reaction-diffusion model and modifications of it. Mutagenesis experiments in the zebrafish revealed [133] that genetic changes are capable of transforming a stripe pattern into a spot pattern and vice versa. In fact, stripe patterns are a regular outcome of reaction-diffusion systems that differ from spotted patterns only by the precise parameters used and transitional patterns in which stripes locally break down into spots have been observed [133]. Even more complex patterns exist, in which alternating stripes of different widths, spots, stripes and rings form. Some of these patterns can indeed be explained by a simple reaction-diffusion system, while the majority of them can be achieved by more complex GRNs with reaction-diffusion dynamics or nested combinations of reaction diffusion systems that allow for combinations of pattern formation mechanisms [134]. Sometimes, the same reaction-diffusion system acts in different, subsequent developmental stages, thereby interacting with different developmental processes. For instance, tissue growth can increase the spacing between existing spots or stripes, allowing for new, second-order spots or stripes to form in between [135, 136]. Such a mechanism has been proposed for the stripe patterning of the palatal rugae [137], the integumental pattern of certain tropical fish species with different, alternating stripe types [138, 139], and the hair arrangement in mammalian fur, in which long hairs form in a first array of activator spots originated by a reaction-diffusion system, while smaller hair types [140, 141]

15 form in secondary activator spots that arise, after tissue growth between the primary hair placodes, by a second round of reaction-diﬀusion dynamics.

Reaction-Diusion systems: Turing patterns Reaction-Diusion systems: Travelling waves a b a b

c d c d

e e Activator Inhibitor

Activator concentration Activator Inhibitor Activator after simulation (2D) Inhibitor Concentration Concentration Concentration Concentration Concentration

y time time time time

0 0 0 0 x x x x x

Figure 3. Pattern generation by reaction-diffusion systems. Two of the most important ways by which reaction- diffusion systems generate spatial patterns are Turing systems and travelling waves. In Turing systems (left side), the inhibitor (green) diffuses substantially faster than the activator (orange). In all schematics a-e, the y-axis denotes the concentration of activator and inhibitor. In order to initiate pattern formation, there is a local inhomogeneity in point P of the activator concentration (a). The locally higher concentration of the activator induces higher inhibitor production (b), so that, after a while, a local equilibrium between activator and inhibitor concentrations may form. Since the inhibitor diffuses faster than the activator, it represses activator production in the surroundings of P (c), which, in turn, also represses inhibitor production there (d). Depleted of inhibitor, secondary activator peaks form (e) at a distance from P that depends on the diffusion rates of activator and inhibitor, but also on other interaction parameters of the reaction-diffusion system. The spatial outcome of Turing system is a regular and stable stripe or spot pattern. Other reaction-diffusion systems give rise to travelling waves, if the activator diffuses faster than the inhibitor. Such a case in shown on the right side. Following an initial inhomogenity of activator concentration in point P (a), more inhibitor is produced in point P, which represses the activator (b). Since the activator diffuses away faster than the inhibitor, the surroundings of P experience an excess of the activator (c), while the inhibitor accumulates in P and represses the activator there strongly. Since a lower activator concentration in this point leads to a reduced inhibitor production, P becomes depleted of both activator and inhibitor (d). With slowly diffusing inhibitor, the area of depletion enlarges centripetally (coloured arrowheads), resembling waves of activator and, subsequently, inhibitor concentrations travelling away from P (e). Only much later, the activator concentration in P will eventually recover (e, center) and give rise to another traveling wave. Below, exemplary one-dimensional space (x) - time plots of activator and inhibitor concentrations, as well as a 2D plot of the activator distribution at the end of the simulation of the Turing patterning mechanism, all taken from the turtle model of the first publication. Time starts at the bottom in the time-space plots; brighter pixels indicate higher concentrations. The red line in the 2D plot indicates the x-values at which the concentrations in the respective time/space plots were taken. Note also the discrete stripes forming by superposition of two travelling wavefronts.

16 Very complex patterns are found, for instance, in the integument of animals inhabiting tropical reefs, in bird plumage, in butterfly wings (although in the latter, many patterning elements are non-periodic and probably develop by different mechanisms [142, 143]), but also, to a lesser extent, in the fur patterns of mammals and scute patterns of squamates in which patterning mainly serves protective or aposematic purposes or play a role for sexual selection. As already mentioned, natural GRNs exhibiting reaction-diffusion systems are usually more complex than the classic Turing/Gierer-Meinhard models. Activation and inhibition can be effec- tuated by chains of transcription and growth factors that activate and repress one another. Often, an activation can result from the inhibition of an inhibitor. Parameters of a given reaction-diffusion model may also change over developmental time [144] and this might even cause transitions between different model dynamics, such as Turing patterns and traveling waves [145]. There can even be several nested reaction-diffusion systems that affect one another or act autonomously in parallel to one another [146, 147]. In addition, Hans Meinhardt suggested an alternative, very similar model, that shared most properties with the classic Turing/Gierer-Meinhardt models, namely the substrate-depletion model [115]. In this model, the activator is not repressed by an inhibitor, but by competitive depletion of an upstream factor or a substrate on whose presence the activator depends, but which regenerates slowly or not at all. Since this type of model can give rise to most of the patterns produced by the classic Turing reaction-diffusion model, it is often not easy to decide between the two alternatives without detailled knowledge about the specific molecular interactions. In several biological systems, different tissues are involved in the patterning process (such as in ectodermal appendages in which part of the involved signaling factors reside in either epithelium or mesenchyme or in both tissues [148, 149]), which might affect the dynamics of the patterning mechanism. Finally, activation of reaction-diffusion systems can depend on larger-scale gradients or more complex signaling backgrounds, generating patterning differences or even pattern transitions between different parts of the tissue [150, 135, 136]. The latter case occurs in many animals in which spots gradually merge into stripes. Several objections have been made against the hypothesis that reaction-diffusion systems be frequently responsible for patterning in biology. First, it has been shown that there is no or insufficient diffusion of crucial signaling factors [151] between cells in some classic examples of stripe or spot patterns, such as the zebrafish pigmentation pattern [152]. Yet, pigment cells in Zebrafish, i.e. mainly melanophores and xanthophores, interact in a unique and dynamical way: Xanthophores induce transient membrane depolarization in nearby melanophores, leading the latter to migrate away [153]. In vitro studies showed that repulsed melanophores are, in turn, chased by xanthophores, suggesting a contact-depending positive feedback from melanophores to xanthophores. Kondo and coworkers [154] have argued that such a scenario, even in the absence of diffusion, fulfills the criteria for a system with all the properties of the Turing/Gierer-Meinhardt model. In this case, pattern formation by differential diffusion of interacting signaling proteins is mimicked mechanistically by the differential induction of repulsive and adhesive cell behaviours in two different cell types. In other words, instead of being based on molecular interactions, the Turing mechanism would, here, be the result of cell type interactions. Other authors [155, 156] have suggested different mechanisms that can give rise to Turing dynamics. Transport of signaling molecules via cytonemes has been suggested to be a common, and efficient way of signaling across distances. Although molecule transport via cytonemes is usually directed and aided by motorproteins, the travelling speed of different molecules differs. Also, different cell types form cytonemes of different lengths and with different densities and frequencies [157]. From a mathematical perspective, there is little difference between molecules that diffuse at different rates and cell types that produce cytonemes at different lengths, as in both cases, information will spread at different velocities across the tissue. Dagmar Iber and coworkers suggested that cooperative ligand-receptor binding fulfills the requirements for a Turing

17 reaction-diffusion system under certain conditions and conclude that these are met rather often in real ligand-receptor pairs [146]. These cases exemplify how the requirements under which pattern formation by Turing instabilities is possible can be met by a large number of different molecular, cytological and histological mechanisms that are commonly encountered in developing biological tissues. Another objection against the frequent use of reaction-diffusion systems in biological pattern formation is that the Turing space (i.e. the part of the parameter space of a reaction-diffusion system that produces periodic patterns) tends to be rather small, making it unlikely for GRNs acting in an organism’s development to have the correct parameters (diffusion rates, degradation rates, interaction strengths between activator and inhibitor). However, Kurics et al. [146] analyzed in detail under which conditions a Turing space, as arising by cooperative ligand-receptor binding, could expand substantially. The authors found that very low diffusion rates of the activators (such as in cell surface receptors), additional negative feedback loops, coupling between different reaction- diffusion systems (e.g. different receptor-ligand pairs) and cooperative clustering of the activators, in case it is a receptor protein, will all increase the Turing space. This makes it more likely that sets of parameters that allow the formation of Turing patterns, can be found repeatedly and independently in evolution. The authors argue that most of these features are rather common in biological systems, so that it would be reasonable that Turing mechanisms, although sensu lato, will frequently emerge in evolution.

1.6 Epidermal appendages and their development 1. Epidermal Appendages The epidermis is the animal tissue which is in most permanent and immediate contact with the environment. As such, it has been a hotspot of phenotypic adaptation. In vertebrates, one major step in conquering terrestrial habitats was the evolution of an integument that prevented desiccation of the underlying dermis and thereby permitted habitat expansions beyond humid places. Such an integument also serves as an efficient barrier against pathogens and mechanical damage. This evolutionary step occurred at the evolutionary transition from the Anamnia to the early Amniotes [25, 158]. Although anamniotic Tetrapodes have survived until today, they are mostly restricted to moist habitats where water loss does not act as limiting factor to survival (although there are integumental adaptations to drier habitats in some adult amphibia [159]). On a histological level, that evolutionary novelty consisted in the appearance of a multilayered stratum corneum [160, 25, 158], the uppermost layer of the epidermis, which is built up by a layer of keratinocytes. The key structural proteins (for a more exhaustive review of molecular innovations in the amniote integument see [158]) that render this cell layer an efficient barrier are keratins which exist in two main classes [161, 162]. In all Amniotes, α-keratines can be found [161], whereas sauropsids also produce β-keratines [163], which have been named a novelty in the reptile lineage [164]. α- keratines are predominantly organized in α-helices, making them strong against mechanical stress while preserving high viscoelasticity, whereas β-keratines contain a higher percentage of β-sheets and are, therefore, more rigid thanα-keratines. Besides consolidating the protective integument, Amniotes evolved various structures formed by or inside the epidermis that serve a wide range of functions and are considered to be crucial, in many cases, for their clades’ adaptive radiations. Such structures are commonly termed epidermal appendages [25]. There is a number of different ectodermal appendages in different clades of vertebrates: In extant reptiles, we find several different [165] types of scales, the finger pads in Geckoidea, the shell of the Testudines and a particular scale type in the head region of Crocodilea. Birds show different types of scales (two types of leg-scales), a keratinized beak and different types of feathers, besides specialized lobular structures in the head region in some groups. Mammals have hairs, bristles, and different types of glands. Many species

18 develop horns in the head region. Teeth and claws occur in all groups of Amniotes, although they have been lost in extant birds. (however, stem clades of paleozoic birds had dentition and tooth development could be reinduced experimentally in recent birds [166, 167, 168]).

2. Diversity of epidermal appendages The following section characterizes the major types of epidermal appendages in Amniotes, albeit there are further types. In here, I decided to focus on structures that appear in periodically repeated units, rather than structures usually appearing as singular units, such as beaks and claws.

Commonalities in the development of ectodermal appendages The different types of epidermal appendages share key features of development. First, the development of all epidermal appendages involves several stages of crosstalk between the epithelium and the underlying mesenchyme [169, 170] and any failure in this crosstalk will result in a prevention of subsequent developmental events. This crosstalk then induces morphological changes in epithelial and mesenchymal cells [171]. While mesenchymal cells are often condensating, the part of the epithelium which will later undergo further changes and specializations, will form a histologically distinct primordium, the so-called placode. Second, skin appendage development involves the same set of molecular pathways (Wnt, Fgf, BMP, Shh, Edar), that often seem to play analogous or similar roles in different kinds of appendages [172, 169, 25]. Particularly Wnts [173], but also Fgfs, act as upstream inducers or activators of placode formation and dermal condensation, while BMPs antagonize them [173]. Shh and the EDAR pathway further regulate the interactions between epidermis and dermis as well as the specific tissue modifications and transformations they induce [174, 173, 175, 176, 177, 178]. Together, these signaling pathways first determine the places at which appendage placodes will form (and often in a regular spatial pattern, as discussed in 1.4). Typically, dermal tissue will condensate. In the hair placode, this condensation is the consequence of changes in cell adhesive properties leading to mesenchymal cell immigration [179]. Changes in the expression of matrix components such as tenascins and syndecans have been reported to induce such changes in cell adhesion [180, 181, 182]. The placodal epithelium, on the other hand, undergoes proliferation rate changes, invaginates or changes cell shapes and sizes (or a combination of these tissue transformations) [173, 183]. It had been formerly believed that the appearance of placodes in the development of ectodermal organs were a novelty of birds and mammals (e.g. by exaptation of signaling pathways involved in other placodal organs) [184, 185], but DiPoi and Milinkovich [186] identified placodes as appearing in the development of reptilian scales as well, although more transiently than in the other types of appendages. As discussed, interactions between epithelium and mesenchyme are crucial for the development of ectodermal appendages. Classic xenotransplantation experiments [187, 188, 189, 190, 191] in which mesenchymal and epithelial tissue of different animals, such as lizards, chicken and mice or mice and rats, were recombined, proved that the mesenchyme determines the places at which appendages will form, while the epithelium will transform specifically according to the respective appendage of the part of the organism it was excised from. For all these processes, however, the presence of both tissues is required (i.e. isolated dermal and epidermal tissue will not produce any appendages). In addition to the common molecular pathways, the different epidermal appendages express also specific gene products or secrete specialized extracellular matrices, many of which are essential for the specific structural and material properties that differ between the epidermal appendage types. Yet, the fact that xenografts undergo “normal” development visualizes the high

19 degree of conservation between the interacting signal molecules and points strongly to a common evolutionary origin.

Sauropsid scales, i.e. keratinized, more or less flattened small platelets with periodic arrangement within parts of the sauropsid integument, are primarily made fromβ-keratines and are present in all extant reptiles (except for some softshell turtles and the giant leatherback turtle Dermochelys [69]). They are typically organized in different areas throughout the integument of the reptilian body, between which their appearance, size and direction of alignment can differ. Studies by Alibardi [192] show that these structures form by localized increase of epidermal cell size (hypertrophy), while the underlying basal cell layer undergoes proliferation. In most scale types, there is also a pronounced AP-polarity, which may be a novelty as ancestral amniotes and some early amphibia were mainly equipped with simple round scales [193]. Round scales also occur in some extant squamates (Geckoidea, although the lack of polarity in their scales is most likely a derived feature). DiPoi et al. [186] discovered that the reptilian scale development involves a transient placodal stage, which is a hallmark of ectodermal appendage development [171, 173]. Besides, different types of scales can be found on avian hindlimbs [194].

Feathers are considered the most complex epidermal structures found in vertebrates, made mainly (but not exclusively) fromβ-keratins [25]. Although many different feather precursors have been identified in fossils [195], it is still not clear how feathers originated in the first place, although recent studies have identified molecular pathways whose modification allows to transform scales into feather-like structures and feathers into scales, as well as into different intermediate appendages [196]. Comparisons between such intermediate structures and appendages from fossilized therapods have instigated hypotheses about the putative scale-feather transition in evolution, both from a developmental and evolutionary perspective [25, 197, 198, 199, 200, 201, 202, 194, 203]. In development, activators like Shh, Fgf4 are antagonized by BMPs and are involved in the formation of a regular array of feather primordia [149, 204]. The fact that feather primordia first appear around the dorsal midline and later spread laterally and ventrally, points to an upstream activator emanating from the central dorsal region of the embryo. Some factors such as β-catenin and Fgfs have been demonstrated to be capable of inducing feather bud formation [205, 206]. Dermal-epidermal crosstalk leads to dermal condensation [207, 208] and epidermal differentiation and proliferation. Each feather bud shows a clear antero-posterior symmetry indicated by exclusive anterior expression of Shh and posterior expression of BMP, due to antagonism of these two signaling molecules [209], and expression of Wnts [210, 211] which is lost when Shh is blocked [208]. Interestingly, the same pattern can be found in avian scale primordia and some mutants indicate that both types of ectodermal appendage can be converted into one another [212, 213] relatively easily. Feather primordia continue growing in a polarized manner, thereby forming a tubular structure with radi- ally arranged lobular structures. These structures eventually grow out and form the future feather branches. Again, Shh-BMP antagonism defines the positions of the these branches by periodically differentiating proliferating and non-proliferating areas along the rachis [209, 214, 215]. Later feather morphogenesis involves changes in adhesion protein expression [216] and apoptosis [208]. There are different types of feathers whose structural differences are largly due to Wnt7a-dependent symme- tries or asymmetries [211] and the differential activation of a number of subsequent developmental events [198, 209].

Hair is the main epidermal appendage of mammals. Morphologically, hairs are much simpler than feathers, consisting of single, non-branching filaments made mainly from α-keratines [217], although some modifications exist. Like feathers, hair primordia, or placodes, become visible first as a field of scattered dermal thickenings [141]. Although the distances between single primordia

20 are highly regular [218, 141], the overall distribution is, with the exceptions of some specialized hair (such as sensory whiskers [219]) less regular than feather primordia. The places of those dermal thickenings are determined by antagonistic interaction between the growth factors Wnt and Dkk [141]. Whereas Dkk localizes in the mesenchyme [141] and Wnt10b is expressed more strongly in the epithelium [220], its target β-Catenin acts in both the epithelium and the mesenchyme compartments and causes in the latter the formation of dermal condensates [221], together with Fgf signaling [222, 223]. The dermal condensate induces differentiation and proliferation in the epithelium which starts folding inwards, giving rise to the dermal papilla. In mammals, hair placodes arise in several developmental “waves” [141, 140]. A first wave of hair follicle morphogenesis gives rise to the main body hairs, that soon differentiate and become refractory to further growth factors. A second, later, wave induces secondary hair placodes in between the existing ones [140, 141]. Hairs whose development is induced by this second wave are typically smaller and have a different shape.

One of the most stereotyical examples of the importance of epidermal appendages for adaptation and evolution are teeth. Since there is a direct link between the shape of teeth and their function [224], their morphogenesis has to be controlled with precision. Tooth development begins with local thickenings in the dental lamina (part of the oral epithelium) [225, 226] that provide inductive signals (BMP4) for the odontogenic mesenchyme [227], which contains cell populations of cranial neural crest origin [228]. The epithelium eventually grows under the control of the Ectodysplasin pathway to become dental placodes [229, 175]. The growth of the placode is repressed by signals from the oral mesenchyme [173], while ectodermal signals are causing the mesenchyme to proliferate and condense [230]. This interaction leads to the formation of the enamel knot [231, 232, 233], a signaling center that causes cells at the ectodermal-dermal interface to differentiate and later produce enamel and dentine [234, 235]. While the cells that are part of the enamel knot do not proliferate any more, they induce proliferation in the surrounding mesenchyme and epithelium [235], causing the morphogenesis of the cervical loops, a specific structure at which the outer and inner enamel epithelia meet. At a later stage, by interaction of growth factors (Fgfs, Shh, Wnt) and their inhibitors (BMPs) [234, 229, 236, 235], secondary enamel knots may appear that define the final shape of the tooth [237, 238] by defining areas of growth (lateral cervical loops in the surroundings) and areas of increased mineralization. The positions of those secondary signaling centers are the sites of the later cusps. A number of studies has revealed molecular processes involved in each of those developmental stages [229]. In despite of minor differences, tooth development appears to be relatively conserved across vertebrate classes [239].

Origins of ectodermal appendage diversity This diversity of epidermal appendages in different groups stresses the importance they had in driving morphological adaptation. It has been suggested that most kinds of integumental appendages are actually derived from some ancestral reptilian skin appendages [25, 240]. Certain epidermal structures, like denticles (i.e. extraoral tooth-like structures) and teeth, are indeed evolutionary old [241, 242, 243, 244, 245]. Some studies show that tooth development is very conserved between mammals and osteichthyes [246, 239] and that chondrichthyian teeth (and denticles) develop in similar, albeit not entirely identical ways [239]. Furthermore, denticles and dermal ossifications were common in early vertebrates, especially in Placoderms [247, 248] and even Conodonts [249], although the interpretation of mineralized conodont elements as teeth or denticles is contested [250], and it is not sure whether dermal modifications and ossifications in extinct taxa were homologous to mesenchymal tissue modifications occurring during the formation of epidermal appendages in extant vertebrates [248, 251]. Taking an even wider evolutionary perspective, inductive tissue interactions, particularly epithelial-mesenchymal interactions, are not only crucial for the development of ecto-

21 dermal appendages, but all sorts of placodal organs in vertebrates [252, 253, 254, 255, 170, 256, 171]. Overall, paleontological and developmental studies in evolutionarily old vertebrate taxa suggest that key mechanisms acting in the development of different epidermal appendages are likely to be highly conserved and evolutionarily old. Oster and Alberch [170] show how the developmental program of all ectodermal appendages starts with a common placodal stage, but might bifurcate at later stages by undergoing organ-specific tissue modifications (epithelial invaginations versus evaginations as first developmental bifurcation) and recruitment of different genetic toolkits acting downstream. While it appears clear that integumental appendages of extant amniotes (and probably in all extant vertebrates) are, at some level, homologous, there has been significant debate about the particular evolutionary connections of derived integuments in birds and mammals (hairs, feathers) with ancestral scale appendages in reptiles. Since intermediates that would convincingly show how scales could have transformed into hairs and feathers have not yet been identified (although a variety of feather precursors have been recently unearthened [25, 257], there is substantial discussion about whether feathers and hairs did directly emerge from a common scale ancestor. Some authors [186, 258] strongly suggest that both avian feathers and mammalian hair are derived from highly modified reptilian scales. They hypothesize that scales began forming tubular shapes and undergoing distal elongation [25] and, in the case of feathers, further modifications. Arguments supporting this hypothesis would be the developmental, including molecular, commonalities between different types of appendages, namely the shared existence of a placodal stage, mesenchymal modifications that can be linked to an ancestral ability to form osteoderms (i.e. specialized mesenchymal condensations [248, 186] and differentiation) throughout vertebrates [246] (although the presence of mesodermal condensations in Squamata had been questioned previously [185]) and the finding that disrupting the same pathways [186] induces a loss of appendages in reptiles, birds and mammals alike. Other authors suggest an independent origin of scales, feathers and hair [185, 259, 184]. These authors suggest either that scales of extant reptiles were an independent novelty within the ancient tetrapod integument and not truly or directly homologous to the appendages of stem amniotes [259, 184] or that feathers might stem from scales, while hairs would be a modification of mechanosensory bristles in the interscale areas or sebaceous glands that can be found in some reptile species [185]. Additional confusion came from evident differences between reptile and avian scales [203, 260], suggesting structural similarity, but no close homology. Further similarities in the morphogenetic molecules involved would then be explained by cooption of signaling pathways due to an analogy of function. It is noteworthy that the pathways specifically regulating differentiation of the appendages, i.e. genes downstream of the pathways initializing placode growth, act as hierarchically individualized network modules. This means that, once evolved, they can be easily recruited in different places of the skin, and be combined with one another, thus facilitating fast phenotypic change and a high degree of variability [171, 170]. In other words, there is almost certainly a strong developmental homology between ectodemal appendages, not only in the particular pathways involved, but also in the particular way these are interacting and regulate tissue interactions. This may also be the reason why appendages which were lost in evolution (such as teeth in birds) could be reinduced experimentally [166, 167, 168]. In comparison to feathers and hair, relatively little attention has been dedicated to the origin of the turtle shell. Yet, similarities of the formation of placodes [83] as well as histological studies [86, 54, 59] suggest a similar evolutionary origin, presumably by modification of reptile scales.

1.7 The origins of variation through development Phenotypic variation exists at different cladistic levels: between individuals, populations, species or larger clades. Understanding what in particular, from a mechanistic perspective, causes phenotypic variation in different taxa and at different levels, has intrigued scores of geneticists, developmental

22 and evolutionary biologists. Traditionally, a lot of research has been dedicated to elucidating genetic differences between organisms, and many studies tried to link those to phenotypic differences, in a qualitative and quantitative manner. It is clear that genetic changes are causally linked to a large part of phenotypic variation. Population genetics explains why the frequency of phenotypic anomalies is usually higher in small populations than in larger ones, especially after bottleneck events [261, 262, 263, 264, 265, 266]. However, genetic differences alone usually fail to fully explain or predict the particular type or direction of observed variation, or its magnitude. Briefly summarized, the main reason is that any phenotype, together with its variation, is generated by development [267, 268, 269]. During development, gene products interact with each other leading to complex, and usually non-linear, spatio-temporal gene expression dynamics. Furthermore, gene products interact with and cause changes in cell and tissue biomechanics. In return, changes in biomechanics will affect gene expression dynamics. In addition, epigenetic, i.e. non-genetic factors, are crucial for animal development in many ways and, consequently, for the generation of variation. Thus, in order to understand the origins of variation in a given lineage or between lineages, we must not only consider the genetic variation in the population, but also how development integrates epigenetic, (i.e. mostly environmental) and genetic information [270, 271], in a dynamic way, to produce the phenotype. How exactly genetic mutations translate into differences in the structure and concentration of proteins has been analyzed and described, in detail, in genetics and cell biology textbooks. It has to be emphasized, however, that the magnitudes of the phenotypic effects caused by point mutations differ substantially and that the majority of mutations in complex organisms is likely to be neutral or effectively neutral (as the majority of genetic variants in populations of complex animals does not appear to produce phenotypes differing significanty in fitness) [272, 273]. As suggested above, the dynamics of gene regulatory networks have substantial effect on how mutations in a specific gene are going to affect development and, thus, the phenotype. When gene products interact with each other or with the DNA, the molecular interactions usually show second-order reaction kinetics and sigmoidal or saturating binding or interaction curves [274, 275, 276, 277, 278, 279, 280, 281]. This means that linear changes in transcription factor concentration or their binding affinity will typically lead to non-linear changes [276, 282, 283] in the transcription levels of their target genes and, thus, potentially on the phenotype. Also, GRNs form a number of positive or negative feedback loops around transcription factors that will lead to changes in protein levels in often counter- or non-intuitive ways. Such feedback loops can also amplify or reduce (buffer) initial differences in gene expression caused by genetic mutations [284, 285]. Even when phenotypes persist without much change in evolution, the underlying developments producing them might change over time, a phenomenon often called developmental system drift [286, 287]. Such mutations that change developmental GRNs have no visible effect on the phenotype [288], but on the variation that is produced if further, small-scale, mutations occur [289] or if the epigenetic background changes (e.g., different types of mechanisms that are able to produce periodic patterns have been associated with different kinds and amounts of variation and, thus, with a different potential to give rise to novel phenotypes in evolution [290, 114]). For instance, animal lineages that remain phenotypicly stable throughout long evolutionary time spans might show less variation nowadays than in previous periods, due to conservative selection during which ancestral GRNs were replaced by GRNs that produce the same phenotype under “normal” conditions, but less variation [291] under mutational or environmental stress. Interestingly and in line with this, it has been claimed that the magnitude of phenotypic variation in wildtype populations is smaller than in mutant populations [292, 293]. Finally, there has been a debate to what extent phenotypic variation is gradual or not. Al- though much of evolutionary theory relies on adaptation by small, gradual phenotypic changes, such variation alone is hardly able to account for phenotypic novelties. Thus, it has been argued

23 that mutations might lead to either no change or large changes in phenotype, depending on the non- linearities of the mechanisms acting in particular developmental events. Such scenarios of sudden large-scale changes in phenotypes followed by periods of stasis have been called “punctuated equilibria”, proposed as a hypothesis of how macroevolutionary changes might take place by Gould and Eldredge [294]. According to these and other authors [295, 296], morphological evolution tends to proceed quite often in steps of larger phenotypic changes, visible as episodes of increased speciation, or periods without phenotypic changes. While such large phenotypic changes might usually not be viable or be selected strongly against, some of them may actually produce novelties that future evolution can act upon, a scenario that Goldschmidt dubbed the theory of “hopeful monsters” [296]. Development does not only produce morphological changes in some embryonic cells or tissues through GRN dynamics that affect cell behaviours and tissue biomechanics, it also integrates various epigenetic factors. It is crucial to point out that there is no one-way relationship between gene expression dynamics and morphogenesis. While gene products cause cells to undergo changes (in position, behaviour or any cell property), cell biomechanics, shapes and distributions will also affect the distribution of gene products in space and time, e.g. by shaping the relative spatio-temporal distribution of the cells sending and receiving paracrine factors [297, 298]. Recently, several studies have even discovered that mechanotransduction acts as a trigger for gene expression, constituting a pathway by which mechanical changes directly cause changes in gene expression [299, 300, 301]. Mechanotransduction may also be required for more permanent changes in cell state, e.g. by inducing cell differentiation [302]. More common are passive effects of tissue-level morphogenesis on gene expression. For instance, epithelial cells may form diffusion barriers, thus canalizing the directions in which diffusive signals can propagate [303]. In general, the intricate interactions between morphogenesis and GRN dynamics can either increase variation by effectively forming a positive feedback or, in other cases, reduce variation by forming negative feedback mechanisms [304] or by spatially directing morphogenesis [305]. In the development of many placodal organs, mutual across-tissue signaling (between epithelial and underlying mesenchymal tissue) may be a classic example of the latter scenario. Also, by making further morphogenesis dependent on the successful completion of previous morphogenetic processes in adjacent tissue compartments, development is effectively parsed into well-concerted subsequent stages, reducing amplification of variation through spatio- tempoarally interacting developmental processes. Isaac Salazar-Ciudad suggested that the relative timing of inductive cell-cell signaling and morphogenesis has general implications on the amount of variation produced in development [297]. According to this hypothesis, morphostatic developments (in which cell-cell signaling precedes cell and tissue rearrangements) show less and rather gradual variation than morphogenetic developments (in which signaling and morphogenesis take place at the same time) 1.10. A paradigmatic example of morphostatic development is the AP segmentation of Drosophila, in which the body axis becomes compartamentalized prior to significant cell movements (in fact, this patterning precedes cellularisation), while the clock-and-wavefront segmentation mode used by Tribolium (and vertebrates) can be categorized as highly morphodynamic. It needs to be noted that no development is completely morphostatic and different developmental stages can be more or less morphodynamic in comparison to one another.

Epigenetic causes of variation have received proportionally less attention than genetic causes of variation. In this thesis, I am going to use this term in the original or wider sense, comprising all factors that aﬀect development, but are not hard-wired in the genome. Important epigenetic factors are:

1.) Developmental information provided by the mother at the very beginning of development. The initial symmetry break that usually deﬁnes the main body axis cannot come from the genetic information of the zygote alone (unless the symmetry break occurs in a random direction). Instead,

24 it is often defined by external gradients of maternal factors [306], mechanical stimuli [307] or gradients of internal factors resulting from the entrance point of the sperm cell [308]. Also, maternal gene products, which are necessary for the onset of developmental GRN dynamics, are deposited into the oocyte. The crucial contribution of the maternal organism is not the specific nature of those gene products (although in the particular organism, they are specific and indispensable), but that they are distributed unevenly within the oocyte, usually due to non-homogeneity of the cellular environment inside the gonad, thus allowing for spatial gradients inside the oocyte. Even more generally, the mechanical and chemical properties of the oocyte can be considered as epigenetic factors, since they are not due to the genetic information of the zygote itself, but provide the initial cellular “material” that will subsequently undergo transformation in development.

2.) The immediate environment in which the embryo develops. In oviparous animals, this is characterized by the chemical substances contained in as well as the mechanical and osmotic constraints imposed by the egg. In viviparous animals, the embryo develops in continuous and mutual contact with the maternal organism, providing an environment of extraembryonic cells and matrix products as well as diffusive factors (e.g.growth factors). Furthermore, and very obviously, nutri- ents are crucial for the development. While nutrition is provided directly by the maternal organism in viviparous and ovi-viviparous animals, yolk fulfils this task in oviparous species. Embryos developing under conditions of food deprivation, often form smaller or anomalous adult phenoypes whose further life-history traits are often critically compromised or modified, too. In extreme cases, the development or growth of certain organs can be severely impaired, resulting in phenotypic deformations. [309, 310, 311, 312]

3.) The external enviroment. Differences in temperature, moisture, pH, oxygen levels, salinity, as well as diverse chemical and mechanical conditions, such as pressure inside the nesting site or a high concentration of toxic substances in the surrounding media that can diffuse into the egg or damage its protective layers, have a substantial effect on embryonal development. In many cases, interactions with different organisms [313, 314, 315] will critically affect development. While some of these interactions are important for development to take place at all, others are a potential source for variation in development and, ultimately, the adult phenotype.

4.) Epigenetic factors in the narrow sense, i.e. factors that affect the efficiency by which specific gene products are produced, typically, but not exclusively, by specific, reversible chemical modifications of the DNA or to the histones. Some of these modifications can be passed on by one or a few generations. Epigenetic factors from the environment can generally affect the trajectory of development, even in late developmental stages, by changing gene expression in different ways: physico-chemically (e.g. by mechanotransduction pathways or changing the chemical milieu in which proteins interact, as explained above), endocrinologically, by modifications of the chromosomal methylation patterns and, especially in very late stages, neuro-endocrinologically and by changes in the signaling between bacterial symbionts and host [316].

Speciﬁc epigenetic variation is often a necessary element of normal development [270, 317, 318] and, therefore, should not be treated as merely some addition that causes “abnormal” development. On the contrary, without epigenetic factors, especially the initial asymmetries and maternal factors inside the egg, development would simply not take place. In many animals, it serves as an important input of information about the environment that can serve as a switch between diﬀerent alternative developmental trajectories [270], in order to help adapt to the presumptive environmental conditions the organism is likely to face. For instance, nutrition deprivation during development may lead to

25 changes that will enable the adult organism to cope better with starvation. In insects, different pheromones may be critical in choosing between alternative developmental trajectories that will lead them to occupy different niches [319]. Epigenetic factors can also be crucial for sex or cast determination, which is for example the case in turtles [320]. Since environment usually changes on the timescale of one or a few generations, epigenetic factors are likely to play a more important role in the generation of variation between related individuals, especially siblings and littermates, than between higher taxonomic levels. Importantly, the intricate and mutual relationship between genotype, GRN dynamics, morphogenesis and epigenetic factors means that, even if we can point out a causal link between a genetic or epigenetic factor and a given instance of variation in phenotype, we may neither understand how exactly that genetic or epigenetic change translates into the phenotypic variation, nor be able to predict how a weaker or stronger change or the same change at a different stage would affect the phenotype. Sometimes the phenotypic effect of epigenetic differences may be highly unintuitive. Therefore, understanding the particular processes of development is often indispensable for understanding how, and which, phenotypic variation is produced in a given developing system.

1.8 Origins of variation in turtle scute patterns A surprising result from the comparative study of extant and fossile carapacial scute patterns is the overall low amount of variation between species. Basically, nearly all terrestrial and most aquatic species have shown the very same scute formula since most of the Jurassic. One of the few exceptions to the dominant scute formula in turtles are the scute patterns in the oldest known turtle taxa, which showed an enlargement of vertebral scutes at the expense of costal scutes, a vertebral scute number of four, and an additional, supramarginal scute row between the marginal and costal scutes. Combinations of these features have been reported mainly from turtles from the late Triassic and early Jurassic (see references below). The second exception are extant sea turtles in which higher scute numbers occur, albeit not in all species. Finally, the carapacial scute patterning has been entirely lost in soft-shell turtles and Dermatochelys. Thus, it seems that, in non-marine turtles, variation of the carapacial scute pattern between species tends to mainly exist in an “everything- or-nothing” manner; in other words, either the classic scute pattern exists or no scutes form at all. On the other hand, variation within species can be substantial. While the majority of individuals follows the stereotypical scute pattern, anomalously patterned individuals are widespread and seem to occur amongst many, if not all, species [31] and there is a large body of publications reporting such anomalies in extant [321, 322, 323, 58, 324, 325, 326, 327, 328, 329, 330, 331, 332, 333, 334, 60, 335, 336, 337, 338, 40, 339, 42, 340, 341, 342, 343, 344, 345, 31, 346, 61, 347, 348, 39, 349, 350, 36] and fossil [351] turtles. See 1.8 for examples of different anomaly types. Interestingly, there appears to be no correlation of any specific anomaly type with particular species, meaning that the same set of anomalies do occur independently from the specific genetic and morphological differences between turtle taxa, although, for most species, reports of anomalies are only qualitative. Some authors [31] have suggested that anomalies occur more often in terrestrial than aquatic species, but some marine species (especially L.olivacea) show even more variation than any terrestrial species [41].

26 additional vertebral costal+vertebral split additional costal

zigzag global zigzag local nuchal/anterior deformity additional marginals

partial scute splittings

Figure 4. Compliation of examples of different types of scute anomalies in Caretta caretta, Natator depressus, Chelonia mydas and Trachemys picta, hatchlings and adult individuals, that were used in the study in the Contributed publication II. Yellow arrowheads were placed to easily localize anomalous scutes. Classes of scute anomalies were defined based on which scute series were affected, although various intermediates occur, and show that anomaly types are largely independent of the species. Other and more severe deformities have been described in several studies (e.g. [58, 40] ).

27 As turtle scale anomalies are a highly apparent phenomenon, potential causes of the observed anomalies have been discussed for over a century. One of the first hypotheses was brought up by Gadow [329] (discussed in [352]) who had observed a higher amount of variation in hatchlings than adult sea turtles. He speculated that split scutes in freshly hatched turtles were still on the way to fusing together, in order to re-establish the stereotypical adult scute configuration, a process he termed “orthogenesis”. The fact that anomalies are still visible in some adults could be explained by assuming that orthogenesis, in some cases, would fail to terminate, and similar ideas of undetermined or arrested development were issued by other authors [350, 332]. While New- man [42] critizised Gadow on methodological grounds, namely for presumably including hatchlings from different turtle species with different phenotypic variability into his studies, leading to wrong conclusions, it was Coker who most outspokenly argued against the existence of orthogenesis as a developmental mechanism [353, 354]. He argued that the positions of scute pairs which, according to Gadow’s hypothesis, ought to merge during subsequent development, do often not correspond with the actual scute positions of the normal adult scute pattern. Gadow replied [355] that the mechanism of “orthogenesis” would only affect local scute splittings, but not apply to larger-scale “malformations” such as a global zigzag patterns across the carapace, although he did not define how to clearly tell these two cases apart. Since in the past, marine turtles [356, 357], or only Der- mochelys [358], used to be classified as sister clade of all other extant turtles, Dermochelys (which has a high number of small bony plates instead of a carapacial shell) was considered the one extant species retaining a scute pattern most similar to the one in the last common ancestor of all extant turtles. This misinterpretation, as well as the generally high scute numbers in other reptiles, led Newman [42] to conclude a trend towards scute number reduction in turtle evolution and to hypothesize an ancestral condition in which each somite/rib pair would correspond to one set of scutes. Gadow, too, [329] had previously expressed similar thoughts. Thus, in these authors’ view, split scutes were interpreted as accidentally occurring atavistic phenotypes. Yet, this hypothesis shared some of the problems of Gadow’s orthogenesis: a poor alignment of additional scutes in anomalous individuals with their “ancestral” or “orthogenetic” positions as well as a difficulty to deal with instances of scutal fusions or zigzag offset patterns in vertebral scutes, in which there is no obvious correlation between rib and scute elements [343]. Therefore, some authors assumed independence between bony and scutal elements of the carapace [359]. Berry [360] went in the opposite direction of Newman and speculated that scute deformations were instead potential novelties with uncertain evolutionary success. Together with Zangerl [36] and Ewert [62], Coker [58] opposed the idea that single scute malformations had any phylogenetic relevance and proposed that scutes be interpreted rather as a developmental entity than as developmentally individualized, separated organs. In this line, Pritchard [37] saw in misshapen scute patterns instances of compensatory growth ensuing developmental malformations. A common interpretation of scute anomalies has been that they may represent an ontogenetic, i.e. developmental defect which could stem from epigenetic, rather than from genetic factors. Sev- eral authors tried, therefore, to focus more on the mechanisms causing scute anomalies and how developing turtle embryos are affected by different sorts of environmental stress. Already in 1910, Coker suggested that anomalies might be caused by the pressures embryos experience in ovo inside crowded nests [58], and Parker [343] reported a reduced length-to-width ratio of anomalous carapaces, consistent with the assumption of mechanical compression during development. Unlike avian eggs, turtle eggs (as reptile eggs in general) are covered by an elastic shell that allows for egg deformations which, in turn, may result in embryonic malformations, too. In a similar vein, Mast [40] suspected that extrinsic mechanical forces could leave imprints on development and correlated the occurrence of scute anomalies with the “roughness” of turtle egg handling in different

28 controlled breeding sites, from which he concluded a significant correlation. Hill [63] arrived at a similar conclusion. Other extrinsic factors were suggested to cause anomalies, too. Hildebrand [332] suggested that hypoxia during development may lead to tissue deformations and Lynn and Ulrich [336] performed detailed desiccation experiments on turtle eggs, concluding that drying severely affects turtle scute development (some of their deformations were substantially stronger than the variation more commonly observed in nature). Furthermore, temperature has been suggested as another extrinsic factor interfering with carapacial development [342]. This suggestion has been particularly driven by the fact that, in many turtle species as well in other reptiles, temperature determines the animals’ sex [361], albeit how exactly differs from species to species [362, 363]. Par- ticularly, the period of carapacial formation was reported to coincide with the developmental period of sex determination [320]. Some authors reported that females, which develop under higher temperatures during stages of sex determination, show significntly more deformations [347, 60, 364]. Telemeco and coworkers [348] reported a correlation between the duration of heat stress experienced during development and anomaly frequencies. Another line of evidence supporting a role of high temperatures in the production of scute anomalies comes from studies of methylation patterns in turtles with and without carapacial anomalies. Caracappa et al. [323] reported a significantly higher percentage of methylated DNA in anomalous C.caretta. Other studies [365, 366] had shown that high methylation patterns are correlated with increased temperatures in other reptiles and that the temperature-dependency of sex determination in T.scripta is due to methylation differences of transcription factor binding sites [367], suggesting a correlation of anomalies and methylation patterns via temperature. O’Steen [368] and Hewavisenthi [331] argued that high temperatures would lead to an increased growth rate. Thus, it could be argued that increased temperatures might cause and amplify tissue deformation, as well as changes in water exchange inside the tissue, whereas other authors [369] stated the opposite, connecting anomalies also to low temperatures. Interesting, variation in the scute pattern has also been associated with temperature in different species of Squamata [370, 371, 372]. Low temperatures may give rise to growth failures, delays and even tissue damage that could translate into deformed carapaces or altered scute patterns. In a similar way, chemical substances could affect development by interfering with development and this has been suggested in several studies [373, 374]. Kazmaier [334] measured site-specific variation in the frequency of anomalies and hypothesized that salinity might affect osmotic pressure in tissues (However, he also suggested that, alternatively, high salinity might increase anomaly frequencies by increased mutation rates). Lynn and Ulrich [336] drew attention to the idea that developmental stages exist which are more susceptible to physiological or mechanical stress, an idea that had been called ’critical developmental period’ by Stockard [375]. In mammals, certain chemicals (e.g. Thallidomide [376] or Cyclopamine [377]) have been demonstrated to cause severe deformations or even death, if applied during a certain developmental period, whereas not having any phenotypic effect at other stages, which has proven a challenge to medical research. The reason for such a manifest heterogeneity of effect is, ultimately, the complexity of development, during which different mechanisms and different GRNs are active at different times and in different parts of the organism. In addition, later developmental stages depend on earlier stages, although variation originating during the latter may be compensated, maintained or amplified during subsequent development. Which of these cases takes place depends on the particular developmental mechanisms involved. As introduced in the previous chapter, Salazar- Ciudad [305] proposed that developmental events in which patterning (change in cell states within a tissue, usually by gene regulation dynamics and cell-cell signaling) takes place at the same time as morphogenesis (change of cell positions, e.g. by growth) tend to show more disparate variation than developments in which patterning and morphogenesis do not coincide 1.10. This is due to the mutual interactions and dependencies between morphogenetic events and cell-cell signaling, which suggests that then, small-scale environmental stress might also be amplified during development. In general,

29 understanding the specific development of the turtle carapace will be necessary to understand if such a “susceptibility stage” during development exists and to which extent later developmental stages will amplify pre-existing variation even further. Ewert [62] observed that the severity of anomalies tend to increase in the posterior part of the carapace, a phenomenon which he called “dovetail syndrome”. Given that somitogenesis progresses from anterior to posterior direction in all vertebrates by a clock-and-wavefront patterning mechanism [378], it has been hypothesized that more posterior parts of the carapace develop later than more anterior parts and, therefore, have had more time to accumulate and amplify developmental imprecisions [59]. However, this hypothesis has been challenged both on grounds of Ewert’s data, as other authors have questioned whether the number of anomalies in the posterior carapace used in Ewert’s argument is really higher, and developmentally, as there is hardly any anterior-posterior bias in the onset of scute formation. More recently, several authors have suggested that there might be a genetic origin of misshapen turtle carapaces. Mast and Carr [40] observed that there is a strong correlation between scute anomalies and other organ deformations as well as albinism, suggesting a link between scute deformations and genes that are also crucial for the development of different structures. Some authors [60, 334, 347] observed that adult females with scute anomalies give rise to offspring with anomalies more frequently than phenotypicly normal females. Using morphometrics, others concluded that plastron shape [379, 380] is hereditary. On the other hand, others [40] described that they found no significant clutch-depending accumulation of carapace deformities and, thus, no significant genetic effect. Throughout the debate of whether, and inhowfar, there is a hereditary component in scute malformations, the existing data has been interpreted in different ways: Some authors [59] propose that only certain variation is genetic, while other variation is exclusively due to environmental factors, thereby reconciling contradictory findings. In detail, Cherepanov suggests that symmetric variation may have a strong hereditary component, while asymmetric variation is unlikely to do so. Other authors suggested population effects on scute malformations. Velo-Anton et al. [349] found a negative correlation between genetic and phenotypic variation in different terrestrial turtle populations, leading him to explain his data by a bottleneck effect (i.e. inbreeding). This is in line with many studies that have found an increase of phenotypical deformations in animals to be due to small population sizes and, thus, low allelic variation [261, 262, 263, 264, 265, 266]. How- ever, an association between genetic mutations and scute anomalies does not necessarily imply that genetic variation explains which phenotypic variation arises, it may instead determine how much phenotypic variation to expect under certain conditions. Genetic mutations may affect the stability of development in a more general way, in other words, they may effect how well development is able to compensate enviromental variation during embryogenesis. If environmental factors cause development to produce different phenotypes, such a development can be called plastic, and this developmental plasticity may either produce a few discrete phenotypes in a switch-like manner or different and hardly predictible amounts of variation [270]. In the case of turtle scute development, Mast and Carr proposed that anomalies might represent instances of non-discrete, plastic developmental responses towards environmental stress during critical developmental stages [40], although it is unclear weather and to which extent such developmental plasticity is increased by genetic mutations. Overall, there is good evidence that environmental stress causes turtle scute variation by interfering with development and some evidence that genetic variation may play a role for some sort of variation. However, no study has concretely linked those factors to specific changes in scutal development dynamics, which could help evaluating the different hypotheses that have been coined so far.

30 1.9 Models of development There are many different notions of what a model in biology is. Here, I want to introduce different types of mathematical modelling approaches that have been used to understand aspects of developmental biology. Key to building a model of biological processes is a hypothesis about the process that one wishes to elucidate. For instance, a hypothesis might be that a particular combination of GRN dynamics and morphogenetic processes explains the transformation of some embryonic state X of a part of a specific organism into its adult state Y. In a model, we would use some mathematical representation of key features of X as initial conditions and formalize the developmental processes of our hypothesis in mathematical terms. Provided that the mathematical tools are adequate, we would expect the simulation to lead to a phenotype that corresponds to Y, if the hypothesis is correct. Such a procedure is, for instance, very useful to decide between different, equally plausible, hypotheses. Nevertheless, even in the case that the simulation confirms our hypothesis, it cannot be ruled out that other hypotheses might prove equally capable of explaining the phenomenon under consideration. In the opposite case, however, if a model does not reproduce the phenomenon in any way, it clearly means that the hypothesis is either wrong or insufficient to explain the phenomenon. Thus, models are an even more powerful tool to winnow out and discard wrong hypotheses. Models can also prove very useful in understanding the development of organisms that are only poorly accessible to experimentation. Turtles are a good example for this, given that most of their life-history traits as well as development are unusually slow, many species are only seasonal breeders and require difficult keeping conditions. Other organisms might not be accessible to experimentation at all, because they are prehistorical, very rare, living in hardly accessible habitats, or cannot be bred in captivity. Modeling, however, still allows assessment of the consistency or plausibility of different developmental hypotheses (without being able to decide which one is the “right” one) about those taxa [381, 382, 383]. Furthermore, mathematical models permit exploration of organismal variation by changing genetic or developmental parameters. In a systematic manner, such experiments are only possible in a small number of species in the lab and under large time and financial efforts. Exploring possible variation systematically, mathematical models can also be helpful tools in the prediction of evolutionary trajectories [384], which is hardly possible in real populations. There are, of course, plenty of caveats. First, the representation of cells and tissues has to be adequate. This does not mean that every biological detail has to be faithfully reproduced. If the processes included in the hypothesis take place on macroscopic (tissue, organs and entire organisms) levels, it might not be necessary to represent single cells at all, while, if juxtacrine signaling or cell shape changes are important, single cells and even subcellular elements must be represented. Also, biological processes taking place on different spatial scales involve different time scales, which should be taken into consideration. Finally, there is an intrinsic danger that crucial systems components are left out and that a model “works” for the wrong reasons. John von Neumann is credited with the famous saying “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk” [385]. In a more generalized sense, models can fit almost anything if a sufficiently high number of parameters or variables are adjusted. However, this statement refers to statistical models, in which a given distribution of measurements is fitted by a function, and the aim of such models is prediction without necessarily understanding causal factors. In our case, we use models to understand how interactions of physico-chemical mechanisms explain dynamic phenomena, i.e. a mechanistic model. Such phenomena cannot simply be approximated by mathematical functions with a fixed number of parameters. In order for a model to qualify as correct with high probability, it should be parsimonious. This means, statistical models should include as few parameters and variables as possible, and mechanistic models should be able to explain as many different, independent phenomena, as possible. For instance, a model that reproduces normal development can be considered more likely to be correct,

31 if it also reproduces laboratory experiments that interfere with development or extant mutants. Yet, ﬁnally, what is easy to evolve in nature might not be parsimonious in the model and vice versa. Therefore the previously stated heuristic only holds if alternative hypotheses are equally plausible to arise in nature.

Types of models Generally, models can be classified as analytical or numerical models. In analytical models, processes are described by mathematical functions that are often continuous in time and space and can be solved in a rigorous manner. Thus, we can learn under which precise parameter settings a change in behaviour of the process under consideration can be expected. However, this type of model can only be used if the number of variables is low or the function of their interaction simple. Such limitations do not exist for numerical models in which time and space are parsed into small discrete steps and equations are typically differential equations that describe the change in variables values per time step. The main drawback is that numerical models do not allow to determination of precise conditions under which the behaviour of the system changes, but only approximate them by tight sampling of the parameter space. Numerical models are also only able to sample a limited part of the parameter space, making it likely that the “right” or critical parameter value set is missed, especially if the interactions of the model components are highly non-linear, while analytical models can encompass the behaviour of a given system entirely. Another basic model distinction is deterministic versus stochastic models. In deterministic models, the outcome of any simulation is completely reproducible, while in stochastic models, changes in the values depend on some probability distribution. i) Representation of cells and tissues As indicated, cells and tissues can be represented in different ways and with different details. This is an overview of some of the most widely used approaches to modelling cells in tissues. Their basic differences are the aspects of cells and tissues that are represented. All approaches have advantages and disadvantages, depending on the particular research question; for comparative reviews of several classes see [386, 387, 388]

1. Cellular automata (CA) [389, 390]: Cells are arranged as positions on a regular grid, implying that the possible neighbourhood conﬁgurations of cells are discrete and limited. Grid positions can remain empty, as well. Each cell contains a number of properties, such as gene product concentrations, cell cycle stage, mechanical properties etc. that take the form of a vector associated with each grid positions. There are rules by which the values of those vectors are updated, which may depend on the values of neighbouring cells, some global variables or gradients or be completely cell-autonomous.

2. Cellular Potts models (CPM) [391, 392]: This type of models uses a fixed grid as well, but represents cells as clouds of filled, coherent grid positions that share the same cell identity index. Unlike the cellular automata, different cellular substructures (e.g.nucleus, cytoplasm, cell membrane or cell wall in plants) can be distinguished. In order to be able to change the shape of cells dynamically, CPMs incorporate the Metropolis-Hastings algorithm which, at each time step, quantifies the total systems energy of the current state to a stochastically changed, alternative, cell configuration. Alternative cell configurations with a lower energy, relatively to the old configuration,

32 are always accepted, while cell configurations with a higher energy are accepted with a probability that depends on the difference of energy and a systems temperature parameter. “Energy” may incorporate a number of biological features [393, 394], such as relative cell perimeter (very digitated cell shapes are to be avoided), average position of all cell parts with respect to a chemoattractant source (to emulate cell movement), cell volume (to emulate cell growth), cell anisotropy (to emulate polarization), the kinds and concentrations of surface proteins of the surrounding cells, etc., depending on the hypothesis one wants to model. This places a considerable onus on the choice and parameterization of the factors of the energy function, since this will ensure both a natural tissue behaviour as well as the right tissue response to changes of developmental processes. Classic applications of this model class are adhesion-driven tissue behaviours, such as Steinberg’s differential adhesion hypothesis [395, 396, 391], but also grouping dynamics of unicellular organisms [397].

3. Immersed boundaries models [398, 399]: This model type is similar to the CPM, but models explicity the cell boundaries that interact and might change conﬁguration. It also allows modelling of more explicit cellular surface and interaction angles. Thus, it increases resolution where it is needed, which can make it less computationally heavy than the classic CPMs.

4. Cell-centre models [400]: This modelling approach comes without a ﬁxed grid. Cells are represented as nodes that are free to move in a continuous space. As in the previous model classes, update rules that depend on both cell-autonomous variables as well as on global variables and cell- cell interactions ensure that the cells cohere, form a tissue, and undergo tissue transformations. Often, a Metropolis-Hastings algorithm is used.

5. Subcellular element models [401]: This approach is similar to the previous one, except for the fact that cells are represented as a cluster of subcellular nodes that cohere more strongly to one another than to other cells. In a way, it relates to the cell-centered models as the CPM relates to the CA.

6. Vertex models: In this model class, cells are deﬁned wholly by their boundaries [402, 403]. Springs between cell boundaries and cell centroids, as well as mathematical functions quantifying surface tensions, ensure that a basic, realistic cell shape is maintained, although vertex models allow for a lot of freedom of cell shape variation. These models are often used to study changes in epithelia as well as tissue biomechanics, as the forces between cells are modelled explicitly. Gene expression can aﬀect mechanical constants and, thereby, lead to tissue transformations. ii) Representation of GRN dynamics In biological systems, genes are transcribed into gene products that can localize intracellularly, intramembranously or extracellularly and that can interact with other genes by interfering with their transcription or other gene products by binding to them, or mediate cell and tissue changes. Transcription of genes is restricted to the cell nuclei, which, depending on the model class used, may be represented explicitly or not. Thus, gene concentrations and GRN dynamics need to be incorporated in most models of development.

33 1. Boolean networks [404, 405, 406] Many models consider gene expression changes in a binary way: Transcription of a given gene is either active or not, depending on the activation state of activators and repressors (which, as an input, are binary as well). Such networks are called Boolean networks and provide a substantial simplification of gene expression dynamics that is easier to analyse and quantify, because the number of possible gene expression profiles, per network, is finite.

2. Differential equations and Partial differential equations [406, 407, 408, 409] Continuous changes of gene product concentrations can be approximated using differential equations. These predict the amount of change in concentration of any given gene product discretely for any given time step. In GRNs, the amount of change usually depends of the concentrations of other gene products, i.e. transcriptional activators (note that there can be autoactivation and autorepression, too). Decay terms are usually incorporated, accounting for the specific half-lives of molecules. If diffusion is considered, partial differential equations are used, which include a diffusion term. In this case, gene expression changes would depend as well on the concentration differences between neighboring cells and the diffusion rate.

a b c

d e f

Figure 5. Different types of model representations of cells and tissues. Cells can be either represented by positions on a fixed grid (a-c) or as entities that can move freely in 2D or 3D (d-e). Entities or grid positions may represent entire cells (a and d), cell parts (b and e) or cell boundary portions (c and f), depending on the level of detail considered by the model. The different model types: (a) Cellular Automata (CA): Cells are positions on a grid. (b) Cellular Potts Models (CPM): Cells are coherent clusters of grid positions which may, by displacement of cell parts, move over time. (c) Immersed boundary models: Similar to CPM, only cell boundaries are modelled. (d) Cell center models: Cells are positions or nodes in space that can move and interact freely. (e) Subcellular elements models: Cells are clusters of nodes that can move and interact. Nodes belonging to the same cell have stronger cohesion between each other than nodes belonging to different cells. (f) Vertex models: Only edges between cells are modelled. Cells interact and change shape by changes of forces between edges and edges and cell centers. Realistic cell shapes are maintained by modelling mechanical springs within cells. The choice of the model type depends on the particular research question one wishes to address. Different intensities of grey are used for easier distinction of adjacent cells.

34 iii) Uniﬁed (or multi-scale) models

While most models focus on a particular problem, some authors have tried to unite GRN dynamics, paracrine signalling, biomechanics and different cell behaviours in unified modelling frameworks. While some models include several or all of those processes in order to simulate morphogenesis of a particular organ or biological structure [410, 411, 412], others have created general model frameworks that can be applied to a broader range of problems [413, 414, 415, 416]. Such frameworks can be very useful to implement a range of complex problems, since they have the generality to include very different biological processes and allow for their interaction. In most developmental processes, both GRN dynamics and biomechanical changes interact in development and, thus, integrating them is crucial for representing and understanding their development. Multi-scale models also implement different organisational levels, such as subcellular, cellular, tissue and organ levels, permitting to explore how the consequences of modifications in the lower levels propagate across intermediate and higher levels. While for some biological problems, phenomena can be traced down to a particular cause, others require the integration of many different processes that cannot be disentangled easily. Particularly for such problems, in which causes and effects cannot be defined easily and thus, dynamics of the biological processes may be highly unintuitive, modelling can be a valuable tool to advance understanding. The drawback of such unified models is that they are usually computationally heavy. iv) Generality in mathematical models There has been some debate on the meaning of “general” in the context of models, usually referring to the possibility of drawing general conclusions from mathematical models of biological processes rather than only understanding a very specific problem. It has been argued that general models must be very simplistic or abstract, since only if a very abstract model reproduces some features of biology, such a result is supposed to be very robust and likely to apply to many different biological problems. On the other hand, it can be argued that the opposite should be true. If some mechanism that has been shown to play a role in development is deliberately left out from a model in the process of simplification, the model loses biological realism and may only be suitable to draw abstract or vague conclusions. Results might as well be misattributed to the wrong causes. Very “abstract” or simplified models usually add implicit assumptions about the importance or “negligibility” of some biological factors or processes which, although not explicit, may limit the generality of the model or the scope of biological aspects they are capable of explaining. Thus, I argue that models which are considered general should at least be allowed to contain all relevant biological mechanisms and therefore be applicable to any biological problem without dismissing crucial causative factors.

1.10 Towards a unified model of development As stated and detailed before, gene expression dynamics and cell behaviours as well as changes in biomechanics are usually not independent of one another. Changes in the production rate of structural and surface proteins, due to changes of gene expression, will affect shape and biomechanical properties (e.g. stiffness, cell layer thickness, compressibility, plasticity, surface tension, etc.) of the developing tissue. In a similar manner, changes in the expression of genes that affect the cell cycle will lead to changes in tissue size and shape. However, there are also ways in which shape, size and mechanical properties of the tissue will affect gene expression: First, GRNs often involve signaling molecules diffusing between cells in a tissue. The amount of signaling molecules that reach their target cells will depend on the size and shape of the tissue in which they diffuse [298, 297].

35 In reality, developing tissues with signaling molecules diffusing within them are usually not static, but continuously changing their shape, thus affecting the local concentrations of signaling molecules in a non-linear manner. If changes in tissue shape and size depend directly or indirectly on those molecular signals, a dynamic feedback between GRN dynamics and tissue shape emerges. Isaac Salazar-Ciudad even argued that the distinction between developmental events in which cell-cell signaling and changes in the shape of the tissue temporally coincide (morphodynamic) and other processes of development in which this is not the case (morphostatic) is a very fundamental one and has implications for evolution [297, 305]: Morphodynamic developments will produce more disparate variation, while morphostatic developments produce less and rather gradual variation, as schematically depicted in 1.10. In addition, the mechanism of mechanotransduction represents a very explicit way in which mechanical forces in biological tissue induce gene expression changes. This process has been found crucial for the dorsal closure in Drosophila [417], differentiation of mammalian epithelium [302] and others [418], although the full scope to which it acts in morphogenesis has only begun to be appreciated and future experiments will demonstrate how important this mechanism really is.

Morphostatic Morphodynamic

cell movement cell movement signalling signalling developmental time developmental time Trait 2 Trait Trait 2 Trait

Trait 1 Tr ait 1

Figure 6. Morphostatic versus Morphodynamic developments. In morphostatic developments, cell movement and cell-cell signaling does not take place simultaneously, while this is the case in morphodynamic developments. These two types of development diﬀer in the variation they produce (After [305]). In the schematics, grey areas represent combinations of two phenotypic traits that can be produced by genetic mutations of an individual whose phenotype is depicted as a dot in the center; darkness of grey is proportional to the likelihood of any trait combination to arise. While variation generated in morphostatic development is likely to be rather gradual, allowing for phenotypical ﬁne-tunning, variation in morphodynamic developments is potentially larger and less gradual, allowing for rapid macroevolutionary changes.

Thus, in order to fully understand development of structures that involve a substantial amount of morphological change, we need to unite GRN dynamics, cell behaviours and cell biomechanics. This is a particular challenge for modelling approaches. Even more importantly, such a unified, multiscale model is crucial, if we want to incorporate how environment affects or may affect development. As described in 5.8, environment (and other epigenetic factors) can affect tissues mechanically as well as GRN dynamics during development, in a permanent or transient way. However, only by means of a united model of development, allowing for both biomechanics and gene expression to change, can we understand how exactly environmental perturbations translate into phenotypic variation [270, 271]. Depending on the particular mechanisms acting during development, perturbations may be either compensated for, or amplified or will translate into unintuitive changes of development,

36 due to the mutual interaction of cell-cell signaling and biomechanics.

Such an attempt may be very useful in further elucidating the effect of environmental stress on turtle development, as will be discussed in detail in 5.8. Since various phenotypic anomalies in the turtle shell have been linked to environmental causes (see 1.8), studies of carapacial development would benefit substantially from such a unified model of development. This could, in future, also help to decide between competing hypotheses about the origination of particular anomalous scute patterns. Furthermore, a unified model connects cellular dynamics, such as changes in cell shape, cell biomechanics and cell divisions, to morphological changes at the tissue or organ level. Yet, the model that I used to simulate turtle scute development does not include much detail at the cellular level and only quite simplistic cell division mechanisms. While I could successfully model the general development of the turtle scute pattern as well as major patterns of its variation by means of a simpler model, incorporating cell shape and cell biomechanics in a dynamical manner might be necessary to explain phenotypic variation in turtles, between species and within populations, in more detail and to understand how exactly environment affects morphogenesis. Implementing cell divisions and tissue growth patterns in more biologically realistic ways will essentially improve the model of turtle scute development. Morphogenesis of the turtle carapace involves the expansion of the carapace from an elongated to a shield-shaped structure that can be, depending on the species, flat, domed, saddle-shaped or anything in between, and the outline of the carapace can have different widths and various shapes, suggesting substantially different tissue growth patterns. As discussed in detail in 1.2, despite the general conservation of their number and distribution over the carapace, the shapes of scutes vary substantially between species. From model simulations with different growth rates, we learned that the dynamics of the model are relatively susceptible to variation in growth, suggesting that more realistic tissue growth might make a substantial difference for the simulations. To sum up, tissue growth, which coincides with the formation of the scute pattern in Trachemys, has a substantial impact on carapacial development. Therefore, any advanced study of scute formation in turtles that considers phenotypic variation between turtles will benefit from a model that is able to implement different modes of growth, such as isotropic versus polarized and regionalized versus homogeneous cell divisions as well as tissue growth as a consequence of changes in cell sizes and shapes. Taken together, in order to understand realistic scute pattern variation, such as differences in scute shapes, and how environmental stress affects the developing tissue, we need a model that incorporates 3D tissue biomechanics, more explict modelling of cells and, in particular, realistic cell division. A simpler model, such as the one I used in the first two publications contributing to this thesis, can account for some major aspects of variation, while facing important limitations, particularly a very simplified representation of cells and tissue growth. In 5.1.1, I will discuss model limitations and how they can be overcome by a general model of development, as the one presented in paper III.

37 2 Aims

1. Advancing the understanding of the development of the characteristic carapacial scute pattern of turtles using a speciﬁc computational model of turtle scute development.

2. Understanding the origins of interspeciﬁc and intraspeciﬁc variation in turtle scutation

3. Embedding turtle scute development into the wider evolutionary context of ectodermal appendages

4. Presenting a general, mutilevel computational model of animal development.

3 Methods

The model of turtle scute development The model aims to help understanding the scute pattern development between embryonic stage Y14 and Y20 [419]. At stage Y14, the carapace is a mainly longitudinal structure, barely wider than the trunk of the embryo, i.e. the lateral outgrowth of the carapace has not started at that stage. In situ staining for Shh, BMP or Gremlin mRNA reveals two lines of 12 approximately equidistant spots along the carapacial ridges. Between stage Y14 and Y16, spots of gene expression appear on the carapace at the positions of the future scutes, while the carapace is undergoing lateral growth. At stage Y18-Y19, the shape of the carapace is, in its proportions, already resembling closely the shape of the carapace in adult turtles. Shh and BMP are expressed as thin lines, defining the scute outlines. I proposed that a set of two reaction-diffusion systems with different patterning dynamics and lateral tissue outgrowth might explain turtle scute development and built a cellular-automaton model to test this hypothesis.

1. Representation of the tissue I used a grid of 100x140 positions (which was extended for simulations with high growth rates) arranged in rows and columns on which I defined a central rectangle of 80x40 as “cells” at the beginning of the simulations. Cells could contain gene products and interact mechanically. For all cells, I defined a 4-neighborhood, i.e. each cell would interact with all adjacent cells in vertical and horizontal directions, but not diagonally. Note that a larger grid would have been possible, but make simulations take substantially more time. Since the grid size used was sufficient to address our research questions, I decided not to enlarge the grid size of the model.

2. Implementation of growth Since there is no data from turtle embryos about when and where in the carapacial tissue cell divisions take place, I decided to model growth in a very simple manner. Each cell row contained

38 a cell cycle clock that caused, according to the specific division rate, the addition of a new cell at the anterior-posterior midline, pushing the already existing cells outwards. Division rates were determined according to observations of lateral outgrowth in embryos: Since the lateral outgrowth in the center of the embryonic trunk is stronger than in the anterior or posterior end of the carapace and also stronger in the anterior than in the posterior half, different division rates were used for different cell rows, so that at the end of the simulation, a realistic carapace shape would emerge. Also, based on our observations of the speed of carapacial growth between embryonic stages Y16 and Y18, I implemented an acceleration of the cell division cycles during mid-development in the model.

3. Implementing the GRNs and diffusion Two reaction-diffusion systems were implemented as partial differential equations [406, 407, 408, 409]: . . 2 ∂A1 RA1A1 2 = + A10 − MA1A1 + DA1∇ A1 ∂t 1+RI1A1I1

∂I1 2 2 ∂t = RA1I1A1 − MI1I1 + DI1∇ I1

2 2 ∂A2 RA1A2A1 +RA2A2 2 = − MA2A2 + DA2∇ A1 ∂t 1+RI2A2I2

∂I2 2 2 ∂t = RA2I2A2 − MI2I2 + DI2∇ I2

This is one of the most common ways to implement reaction-diffusion systems in a numerical model, although other methods exist [420, 146]. A1, A2, I1 and I2 represented the concentrations of all signaling molecules in a given cell (A being an activator, I an inhibitor). Rij are the transcriptional strength of gene product i on gene j, A10 is the initial baseline activity of the first activator. Mi are degradation rates and Di diffusion rates for each gene product i. Diffusion occurred at all times between neighboring cells. The second activator is only activated at a advanced developmental time T.

4. The initial conditions According to the embryonic phenotype observed at stage Y14, I defined two vertical rows with 12 equidistantly placed inductive spots in the outer margins of the area of the grid filled with carapacial cells. In those spots, concentration of the activator was set invariably as 1. These initial conditions are a simplification, given that the signaling processes in and between these spots are presumably more complex, but sufficient to reflect their indispensability for the induction of the reaction diffusion dynamics.

5. Further experiments: bead implants and mechanical stresses The model permits to place beads as cells with a high concentrations of speciﬁc gene products. In order to be able to simulate the full scope of experimental bead experiments we conducted in Contributing Publication I, I introduced further activators and repressors Si of the four signaling molecules (only one Si per simulation):

39 . . 2 2 ∂A1 RA1A1 +S1 2 = 2 + A10 − MA1A1 + DA1∇ A1 ∂t 1+RI1A1I1+S2

2 2 ∂I1 RA1I1A1 +S3 2 ∂t = S42 − MI1I1 + DI1∇ I1

2 2 2 ∂A2 RA1A2A1 +RA2A2 +S5 2 = 2 − MA2A2 + DA2∇ A1 ∂t 1+RI2A2I2+S6

2 2 ∂I2 RA2I2A2 +S7 2 ∂t = S82 − MI2I2 + DI2∇ I2

∂Si 2 ∂t = Si0 − MSiSi + DSi∇ Si

Furthermore, mechanical stress was modelled in several ways: I introduced changes to the cell cycles (temporarily) and division rates (permanently) in either one of the carapacial hemispheres or in a smaller area of the carapace, in order to cause asymmetric or regionally compromised tissue growth. I also modelled mechanical offsets in one of the carapacial hemispheres, as presumably caused by external pressures embryos may suffer in ovo [336]. This was done by introducing a shift upwards or downwards to all cells (by n grid positions, usually a small number) of one hemisphere at a developmental time t1 and for a duration t2. Upon reaching t1+t2, i.e the time of stress release, tissue relaxation was modelled to occur at a certain speed, by cells shifting towards their original vertical position on the grid that they had at t1 (while maintaining their vertical positions) at a velocity v. Thereby, I accounted for different tissue plasticities, although modelling them in a simplified manner. Thus, I systematically explored the anomalies produced by different combinations of t1, t2, n and v.

40 a c y

d e

t1 t2

1 f

v 0.5 * * cell cycle

mid t1 mid t2

Figure 7. The model of turtle carapacial scute development. (a) In situ staining for BMP4 at stage Y15 in an embryo of Trachemys scripta. Note mainly the alternate strong expression along the Carapacial ridge (CR). (b) Overlay of different in situ data reveals that this is true for a number of different morphogenes (from Contributed Publication I). (c) The turtle model is a Cellular Automaton in which cells are positions on a rectangular grid. In the beginning, the activator is expressed in regular spots arranged in two nearly parallel marginal lines along the CR, based on the observations seen in (a) and (b). In this schematic, cells with high activator concentration are painted blue. (d) Diffusion of molecules takes place between neighbouring cells. (e) A qualitative representation of a reaction-diffusion system in which the activator A induces the expression of itself and a more diffusive inhibitor I. (f) Each cell row has an autonomous cell cycle and cell divisions are implemented as additions of new cells at the carapacial midline, causing an immediate shift of the entire cell row towards the CR. Cell cycle vaues are denoted as shades of red; cell rows with a cell cycle value of 1 will undergo mitosis. (g) In order to model mechanical pressures that were hypothesized to be experienced by embryos in ovo ([58, 40, 336]), shifts of an entire lateral carapace hemisphere were introduced in the model at different times, magnitudes and different hypotheses of tissue elasticity. Other kinds of tissue deformations and local changes in the cell cycle were implemented as well (not shown here).

41 4 Results

Contributed Publication I

In this paper, we elucidate some of the key mechanisms underlying the development of turtle carapace scute patterning. Using in situ hybridization in Trachemys scripta, we show that normal development of the carapacial scute pattern starts from a line of alternating signaling centers along the carapacial ridge, whose positions coincide with the underlying tips of the outgrowing ribs. Also evidenced by their absence in softshell turtles, these signaling centers are required for the subsequent emergence of secondary signaling centers that appear at certain positions on the carapacial tissue. Subsequently, waves of gene expression extend centripetally from all signaling centers. The collision sites of those waves mark the outlines of the future scutes. Simultaneously, the carapace grows from a narrow tissue stripe, covering the dorsal part of the embryo, to its characteristic, laterally widened shape in the adult. I built a mathematical model that implements a GRN of four signaling factors, tissue growth and, as initial conditions, two lines of punctuate signaling centers along the lateral edges, similar to the pattern of marginal signalling centers. Using this model, we show that two coupled reaction-diffusion systems, together with growth, are sufficient to explain the development of the carapacial scute pattern. By interfering with early scute patterning, both in vitro and in silico, we corroborate our model further and identify some of the molecular factors involved in the reaction-diffusion systems. Furthermore, we suggest mechanisms explaining observable variation between and within turtle species. While most turtles have the same carapacial scute formula, a few species show numerical differences. Performing a parameter screen in the model, we show that slight changes in the size of the developing carapace can account for these differences. We also find that the most common scute formula covers a larger area of the parameter space than rarer scute formulae. Exploring further ways to produce phenotypic variation in the model, we conclude that common intraspecific variation cannot be produced by changes in the genetic parameters, but by introducing mechanical stresses or biases into the growing carapace, suggesting an environmental causation of scute pattern anomalies. Overall, this paper shows how a close dialogue between experimental and modelling approaches can advance understanding of development and the causation of variation in a non-classic model organism.

Contributed Publication II

This paper explores further the causation of carapacial anomalies that are found in different turtle species, using quantitative data from different turtle species and parameter screens of the in silico turtle model. In detail, we counted incidences of anomalies in turtle hatchlings bred in nature as well as in controlled laboratory conditions and correlated them with temperature and humidity. Independently of the setup and the turtle species, we find a strong correlation between temperature, but not humidity, and the production of anomalies, in line with our more general hypothesis of an environmental causation of scutal variation. A closer analysis suggests that particularly mid- development appears to be susceptible to environmental stress. We also compare the frequency of specific types of anomalies in nature and in the turtle model and find that, both in vivo and in silico, vertebral scute anomalies occur more frequently than costal scute anomalies. Interestingly, this bias is independent of the particular turtle species considered. We explain this finding by the developmental mechanisms underlying turtle scute development. This paper further corroborates our hypothesis about the mechanisms underlying intraspecific variation by statistically analyzing scute anomalies in extant turtles and comparing our finding quantitatively with our model of scute development.

42 Contributed Publication III

Here, we introduce EmbryoMaker, a model framework that unites GRN dynamics, cell-cell signaling, realistic biomechanics of epithelial and mesenchymal cells and ECM, and all cell behaviours that have been described to play a role in animal development. Therefore, it is general enough to simulate most developmental processes in animals and can help to approach a vast array of diﬀerent problems and hypotheses. It allows implementation of cells and tissues at a customizable degree of detail so that, depending on the question, subcellular, cellular and tissue dynamics can be modelled. Thus, unlike many other more speciﬁc models, it allows study of how exactly GRN-dynamics and morphogenesis interact to produce morphological variation, which makes it a valuable tool for developmental biologists. I suggest that this model framework is capable of providing the biological realism required to address further, and more detailled questions about variation in turtle scutation.

5 Discussion

5.1 Caveats and Limitations As all mathematical or conceptual models, and particularly models of biological processes, the model of turtle scute development comes with simplifications of the real developmental processes. In different words, any model of development is based on assumptions about which processes and mechanisms to include, which to substantially simplify, and which to omit altogether. Since mathematical models in development are used as a tool to elucidate which mechanisms explain specific phenomena, it is necessary to simplify and omit some of those mechanisms. Once such a model has proven to reproduce the biological phenomenon under consideration, it allows distinguishing mechanisms that are indispensable to explain the phenomenon from mechanisms that are less important. While it can be very difficult to prove that certain mechanisms are sufficient to explain a phenomenon by using mathematical models, modelling is a very efficient method to detect which mechanisms do not explain it. On the other hand, however, simplifying always comes with the risk of neglecting important factors and mechanisms or failing to explain the phenomenon completely. In the turtle model central to this thesis, I simplified the developing carapace in several ways, thus limiting how much the model is able to explain.

5.1.1 Caveats due to model limitations Instead of explicitly modelling cells with their particular shapes and arrangements, the turtle model is spatially parsed into a rectangular grid of up to 140x100 positions (referred to as “cells” within the model) that can contain concentrations of gene products. Thus, “cells” in the model represent tissue parcels rather than actual cells. Since computing a high number of individual cells is computationally costly, most mathematical models reduce tissue resolution, i.e. cell number. The question is to what extent such a simplification affects the biomechanical realism of the processes we attempt to understand. There are further important differences between cells in the model and in real tissues. Apart from gene product diffusion and simplified growth as well as growth-induced cell-cell pushing, there are no further cell interactions in the model. The cells in the model also lack most of the biomechanical properties real cells or tissues possess. Indeed, all these aspects that distinguish real cells and tissue portions from the simplified “cells” in the model seem not to be required to explain large-scale tissue patterning. The dynamics of reaction-diffusion systems work both on coarser and more fine-grained grid resolutions. Enabling cells to move in all directions or permitting adaptive cell shape changes would, presumably, not change the way anomalies are produced: by compromising the

43 diffusion space during carapacial growth. Independent of how naturally or realistically cells respond to mechanical stresses [421, 422], shifting or compressing the carapacial tissue laterally will, overall, lead to a restriction of the diffusion space and, thus, reproduce anomalies at least coarsely. However, it is probable that more natural tissue biomechanics might let scute anomalies appear even more realistic. This means that these model limitations do not prevent us from reproducing many major aspects of turtle scute development, but exclude reproduction of smaller-scale variation such as antero-posterior shape differences between scutes within the carapace. Since cells in the model are not able to undergo morphological deformations and can only be displaced in four directions, there are aspects of phenotypic variation, such as elongated or unusually shaped scutes, that cannot be modelled by the turtle scute model. Cells capable of actively or passively moving into all directions would also allow for more realistic tissue deformations, which might be necessary to explain better the interactions between environmental stresses and development. In real cell divisions, daughter cells can be placed in various directions with respect to their mother cell, although these directions may be statistically biased by cell polarization. In the model, cell division is modelled as the addition of a new “cell” at the carapacial midline, while its sister cell as well as all cells between this one and the outer fringe of the carapace are shifted laterally, towards the outer carapacial margins (i.e. to the left or right). In natural tissues, cell divisions will usually introduce isotropic mechanical pressure in the surrounding cells that, by relaxation, may ultimately lead to a displacement of these as well as more distant cells, but in a different way than in the model: The way pressurized tissue relaxes depends on all the biomechanical properties (cell shapes, cell densities, epithelial cell junctions, stiffness, surface tension, cell polarizations, intracellular distributions of structural proteins, etc. [423, 424, 425, 422]) of the developing tissue as well as their changes during development, and cells surrounding a dividing cell are more affected than more distant cells. In addition, cells adapt actively to sustained mechanical stresses, for instance by modifying their structural proteins to become stiffer towards the direction of stress, by changing cell polarizations, by local repositioning of the cells [425, 426, 427] or by reducing or increasing the number of cells, which can occur by contact-mediated cell cycle inhibition or apoptosis (also called aniokis) [428, 429]. Density-dependent decrease or increase of cells and cell divisions might be a common feature in epithelia [430] and mesenchymal tissues [431] in order to maintain tissue homeostasis. However, since we observe a fast lateral outgrowth during the carapacial development, it makes sense to assume that cell divisions will, at least in sum, result in a lateral increase of tissue. In addition, we assume that growth takes place mainly in the central region of the carapace (the midline in our model). This assumption was introduced ad hoc, because there is no data about where within the developing carapace cell divisions occur. Although it is more likely that growth takes place all over the carapace, shifting entire rows in the model from the midline has one important advantage: Any cell division affects all cells within a given cell row and leads therefore to the same amount of movement per time in all places (cells) of the cell row. If cell divisions were implemented randomly throughout the entire model carapace, cells close to the carapacial ridge would, statistically, be shifted more often than cells in the center of the carapace, and tissue movements thereby be biased to the more peripheral regions of the carapace. Thus, I believe that implementing cell divisions the way I decided to do is the most parsimonious manner to implement it in the model, although alternative scenarios are not unlikely to be more realistic. Also, as mentioned before, “cells” in the model are much larger than actual cells, which means that the biomechanic effects of a single cell division in the model and in reality are hardly comparable. Therefore, introducing more realistic cell divisions does not necessarily improve the realism of the model, unless the cellular resolution of the carapacial tissue is increased, too. Nevertheless, it could be argued that more realistic cell divisions, together with a higher tissue resolution, might be necessary to account for all kinds of scute variation seen in turtles, as well as more realistic anomalies. It would also allow study of how exactly different environmental stresses (such as mechanical stresses

44 and high temperatures) affect scute development, by being able to affect single cells and different aspects of cell division in more biological detail. There are other important shortcomings. While the model reproduces qualitatively the precise number and placement of scutes, it is not, without including additional assumptions, capable of reproducing the elongated scute shapes seen in some species (substantially elongated vertebral scutes can be seen in marine turtle species and elongated costal scutes occurred most blatantly in several Triassic and Jurassic turtles, as reviewed in 1.2), although it faithfully reproduces the shapes of more isotropic scute shapes as common in Trachemys. Note that it is not likely that any parameter combination of the current scute model would be, even in theory, able to account for this sort of variation, since the first reaction-diffusion system would form additional placodal signaling centers, rather than forming elongated scutes, if additional space for diffusion was given. However, there are additional mechanisms that could produce more diversity of scute shape, and, thus, might account for more detailled scute shape variation: secondary lateral outgrowth of the carapace after defining the scute outlines, individual elongation of the already formed scutes, changes of cell shape, diffusion, or biomechanics (biasing the direction of signaling and growth) during development, fusion of several placodes (unlikely), more complex GRN dynamics, limitation of the time window during which new placodes are allowed to form despite continuing lateral growth, etc.. Some of these hypotheses could be implemented easily in the model, whereas others would require substantial changes to it, particularly the implementation of biomechanics and more explicit simulation of cells. Testing them experimentally would have been very difficult and not possible within the temporal scope of this thesis. Future studies, however, may continue addressing the shortcomings of the current model. Limitations also apply to the study of environmental causation of anomalies, which I have proposed to underlie much of the intraspecific variation observed in turtles. First of all, we conclude both from correlations between environmental stress and anomaly frequencies, and from the phenotypic consequences of mechanical interfering with development in the model, that mechanical stress causes the majority of anomalies in turtle scutes, but we do not know yet how exactly this happens. In more explicit words, we have no detailed mechanistic model of how mechanical stress translates into phenotypic deformations by affecting cells and tissues. Since we did not have, as described above, an explicit implementation of cells and tissues in the turtle shell model, mechanical stress had to be modelled in a rather coarse manner: by hemispheric changes of growth (magnitude and direction) or explicit deformation of the grid representing the carapacial tissue. However, it is easy to generate various hypotheses of how specific environmental factors could affect or damage cells or induce them to undergo specific changes in their biomechanical properties. In order to test them thoroughly, one would need to produce a predicted anomaly by experimentally interfering with development, for instance, by compressing or partly desiccating the developing embryo, in ovo. While still difficult in experiment, a model that takes into account tissue biomechanics might already help winnowing out some wrong hypotheses. As discussed previously, one way to overcome a number of the mentioned limitations and to advance mechanistic knowledge about scute development and its variation would be the implementation of cell and tissue biomechanics into an extended model of development that allows to connect more levels of organismal organization, namely cellular, tissue and organ development dynamics. Thus, we built a more general model of development, the EmbryoMaker, which is able to link gene regulatory networks, realistic biomechanics and all cell behaviours described to act in animal development. Apart from enabling us to overcome limitations like the ones described above, such a model framework can be used to address questions about morphogenesis in systems different from the turtle carapace development. In the following, I briefly discuss in which ways the EmbryoMaker can advance the study of carapacial development. 1) Whereas the limited way in which the current turtle model implements biomechanics and

45 cell behaviours does not permit to explain the diversity of scute shapes, such as laterally elongated scutes, EmbryoMaker could implement more realistic biomechanics of interacting cells without a fixed grid and in 3D and implement some of the hypotheses listed above. It can represent both epithelial and mesenchymal cells as a number of connected nodes in 3D, thus accounting for a large number of different cell shapes and tissues with different biomechanical properties. We chose to build this model based on Newman’s subcellular element representation of cells [401], because this permits to include cellular details ad libitum and because replacing the static grid of a cellular automaton by nodes interacting freely in 3D is necessary to simulate realistic tissue shapes and shape transformations. Mechanical forces are calculated between interacting cell portions (nodes) and can result both from endogenous (e.g. expression changes of structural proteins or cell divisions) or exogenous (e.g. mechanical pressures exerted onto the tissue) causes. Differences in biomechanical properties, such as surface tension or cell compressibility are implemented as well. Thus, EmbryoMaker allows cells and tissues to respond to all sorts of stresses that occur during development in all ways that natural tissues do. In a similar way, EmbryoMaker accounts for polarized and non-polarized cell divisions that occur in natural tissues. Mechanical tissue feedback (such as density-dependent cell cycle inhibitions) can be implemented as well. Thus, EmbryoMaker can implement very different scenarios of realistic tissue growth, which might be a particularly important limitation of the current model. Together with realistic cell shapes and biomechanics, EmbryoMaker permits to address all of the previously mentioned hypotheses that may explain the differences between scute shapes. 2) With EmbryoMaker, the development of the carapacial scute patterns could be embedded easily into the wider context of carapacial development. For instance, we could model earlier developmental stages [75, 71] during which lateral outgrowth of ribs is coupled to the development of the carapacial ridge. This could be important in order to study deforming effects of environmental effects on carapacial development. 3) EmbryoMaker can give mechanistic insight in how specific environmental stress (mechanical compression, desiccation, high temperatures, local necroses) impact tissue morphogenesis. As previously discussed, the interaction of dynamic changes of tissue shape and developmental signaling can produce highly unintuitive phenotypic outcomes. During development, environmental noise might be buffered, amplified, or direct development into a different direction [271]. Using a model explicitly combining cell-cell signaling and tissue morphogenesis based on realistic cell and tissue biomechanics can help answering those questions and, thus, bridging the gap in our understanding of how environmental stress will affect tissue morphogenesis and thereby produce phenotypic anomalies.

5.1.2 Caveats due to limited data There are aspects of the biological processes involved in turtle scute development and the developmental origins of variation that we cannot understand because the data as well as the scope of biological insight we have about them is limited. The distinction between epithelium and mesenchyme has not been implemented into the model, because we do not know much yet about the distinct contributions of both tissue compartments in scute formation, apart from the fact that some of the signaling proteins which are likely involved in the reaction-diﬀusion dynamics localize only in the epidermis, mesenchyme or in both of them. In the development of other ectodermal appendages (such as tooth, hair and feather morphogenesis [229, 432, 433], the distinctive crosstalk between the diﬀerent tissue compartments has been shown to play important roles. It appears to be indispensable to explaining the emergence of inductive prepatterns (e.g. the morphogenesis of the carapacial ridge through guidance of the growing ribs) and later morphogenetic events (the particular events that give rise to the shape of the tooth cusp or epithelial folding and mesenchymal condensations in feather buds, hair follicles and scales).

46 The distinctive placement of the ectodermal organs, however, (hair, feather bud primordia, tooth cusps) can be modelled without explicit tissue compartment distinction. This does not mean that epithelial-mesencymal are not involved, but that our particular hypothesis of pattern formation, namely the placement of primordia, is not strictly dependent on their distinction. In fact, it does not matter for the dynamics of the model if the reaction-diffusion system involves signaling in different, but adjacent and interacting tissue compartments. Thus, I decided not to include the dermal-epidermal distinction in the turtle scute development model either. During animal development, different cell behaviours, such as cell divisions, apoptosis, changes in cell adhesion and cytoskeleton that lead to cell shape changes and cell migration, cell polarization, apoptosis and secretion of extracellular matrix, interact in various manners [297]. In this model, we only include growth. Yet, some experimental data suggest that some other cell behaviours might be involved in scute development. Immigration of neural crest cells has mainly be documented for the plastron and only to a very limited extent to carapacial development (namely, the nuchal scute(s)) [66, 434]. There is evidence that both scute epithelium and mesenchyme undergo morphological changes (shape changes in epithelial cells that lead, e.g. to invaginations around the scute borders, mesenchymal condensation [435, 69]), but these changes affect either very late stages of scute development or, in the case of dermal condensations, do not determine the positions or the approximate shapes of the scutes. Taken together, there is, to my knowledge, no sufficient experimental evidence that cell behaviours other than growth play a crucial role in those stages of carapacial scute development that I focused on. In addition, the fact that the model does reproduce scute development sufficiently well, suggests that growth, but not further cell behaviours, play a dominant role in scute development, at least in Trachemys scripta. As detailled in 1.7, gene product interactions can take place as transcriptional, receptor-ligand, catalytic or various kinds of posttranslational interactions. Since we model all genetic interactions using the same kind of partial differential equations (i.e. we only change coefficient values in the equations, rather than the equations themselves or the exponent of the term calculating concentration change by genetic interaction), we might not encompass sufficiently well the different interaction kinetics that different gene product interactions show. In addition, there is no distinction between nucleus, cytosol, cell membranes and extracellular matrix. However, our knowledge about the particular dynamics of molecular signals in the developing carapace is very limited. Also, the use of partial differential equations is a standard method to represent the interaction dynamics of gene products which has been implemented in a multitude of models (see 1.9). There is also very limited knowledge about the different time scales of signaling, gene product interactions and tissue growth. In addition, and as stated before, the size of “cells” in the model is different from cell sizes in real tissues. We determined the model parameters which indirectly define those time scales (gene product interaction strengths, diffusion rates, growth rates) by parameter screening and qualitative comparison of the sequence of patterning stages in the in silico development with the in situ patterns observed in developmental stages in real turtle embryos. This means that even if the particular time and space scales substantially differ from the ones in real development, we identified parameters that ensured overall similarity between turtle scute development in silico and in vivo. One important question is whether, and to what extent, very different mechanisms could give rise to the same pattern. Evidence points quite clearly to reaction-diffusion systems as the mechanism inducing and shaping the turtle scute pattern. The fact that new scute primordial spots appear on the developing carapace (evidenced by in-situ hybridization experiments at different developmental stages, see Contributed publication I) as the tissue expands or changes shape due to mechanical stresses is a typical feature of reaction-diffusion systems, because the distance between maxima of signaling molecule concentration is constant for each given reaction-diffusion system and, thus, adding space between existing activator maxima will create space for additional maxima in between

47 [136, 138, 135]. Also, the travelling waves that, in subsequent stages, initiate from the placodes to delimit the outlines of the future scutes are a typical phenomenon that can be produced by reaction- diffusion mechanisms, although it might arise by different mechanisms as well. Another argument in favour of the proposed role of reaction-diffusion systems in turtle scute development is the strong evidence for the action of reaction-diffusion systems in virtually all periodic ectodermal appendages in which developmental studies have been conducted (see 1.6). However, it is possible that the actual gene regulatory network underlying the reaction-diffusion dynamics that act in turtle scute development is substantially different from the one used in the model. Substrate-depletion models, a subclass of reaction-diffusion systems, can produce patterns very similar to activator-inhibitor systems ([131, 115], even in the particular case of turtle scute outlines pattern, H. Meinhardt, oral communication), and they are equally parsimonious. We could identify Shh as behaving like an activator, and Fgf4 as behaving like an inhibitor, but these signaling molecules’ roles could probably be reformulated within the framework of a substrate-depletion model. Ultimately, we do not have enough experimental data to reconstruct how exactly these signaling molecules interact and therefore cannot rule out either of the two options discussed. Yet, we decided to focus on the more classic activator-inhibitor system, since this model has been successfully used to model hair and feather follicle placement as well as tooth development. One of the most important limitations during my studies has been the scarcity of data about most of the aspects of development listed above. Turtles are a rather cumbersome model organism and gathering detailed knowledge about turtle development takes time (turtles are slowly developing organisms and seasonal breeders) and, compared with classic model organisms of developmental biology, relatively few studies have been conducted. Even more difficult, or barely possible, are developmental experiments in the few marine species with divergent scute patterns, as most sea turtle species are endangered and, therefore, strictly protected. Many conclusions of our studies would have benefited from experimental testing, but this was often not possible for the reasons explained. Despite the clear limitations imposed by the biology and protective status of the animals themselves, however, the model results are, according to our judgement, intriguingly close in line with real turtle development. This is not only the case for regular turtle development, the model also reproduced experiments in which we interfered with the paracrine signaling (by placement of beads soaked with those molecules or their inhibitors), as well as inter- and intraspecific variation qualitatively. The number of different experiments reproduced in silico suggests that we indeed included the crucial processes into the model, while omitting the less important ones for the development of scutal patterning. It also renders it less likely that our assumptions simply reproduce turtle scute development by accident, because in that case, we would expect it to have failed in at least some of those validation experiments.

5.2 Classification of the turtle scute pattern development and implications for variation Several periodic ectodermal appendages have been shown to be patterned by reaction-diffusion systems 1.6. Therefore, I suggested that this should also be the case in the development of the carapacial scute pattern of the turtles. More concretely, I hypothesized that the scute pattern could develop by two coupled reaction-diffusion systems in a coupled two-step process. In the first step, the inhibitor would diffuse much faster than its activator, producing an array of regularly spaced spots (i.e. a Turing pattern), while in the second step, the activator would diffuse faster than its inhibitor, giving rise to travelling waves. These waves collide and thereby determine, in a semi-stable manner, the future scute boundaries. In addition, outward growth of the carapacial tissue, taking place simultaneously with the signaling, is crucial for this specific development. In Contributed Publication I we demonstrate by an intertwined series of in vitro and in silico experi-

48 ments that a model based on our hypothesis can reproduce observed development and even propose candidate genes whose products might act, directly or indirectly, as activator or inhibitor. Such a patterning mechanism can be classified as an emergent, but hierarchically structured patterning process [120, 114] (i.e. different emergent subprocesses have a clear temporal sequence) which is, due to the temporal coincidence with carapacial morphogenesis, slightly morphodynamic [297]. This classification comes with a set of expectations with respect to phenotypic variation, since authors like Isaac Salazar-Ciudad [298] predict different types of development to be associated with different variational patterns. First, emergent and morphodynamic mechanisms are expected to produce more, rather complex and non-gradual variation than hierarchical and morphostatic mechanisms. Indeed, some of the most manifest variation observed in turtle carapaces tends to be non-gradual: the occurrence of additional scutes or the fusion of existing scutes. Usually, there are no intermediates: either do scutes split or they do not (nevertheless, a few instances of half-split scutes have been reported). Other sorts of variation, however, are perfectly gradual: Additional scutes can occur in all possible sizes. Once the vertebral scute series has split up in more than one single scute, zigzag patterns in the vertebral scute series occur in all degrees of severity (slight offset up to 50% offset between vertebral scutes) and the severity of this zigzag pattern may vary antero-posteriorly throughout the vertebral scute series. Also, there is a lot of variation of scute size and shape between numerically stereotypical scute patterns. Since I mainly focused on numerical variation, non-numerical, gradual, scute shape variation has not been studied much within the scope of this thesis. However, it is not straightforward to determine the actual degree of non-graduality or discontinuity of scute variation and to what extent it results from the emergent patterning mechanism. To assess this, we would need to be able to compare the extent of turtle scute variation with the variation produced by a hypothetical alternative, fully hierarchical pattern mechanism, which we do not have. While stripes can be easily produced by hierarchical mechanisms, the same is not true for regular spot patterns, which restricts the range of potential mechanisms that could possibly produce the turtle scute patterning. The hierarchical arrangement of the two reaction-diffusion systems also ensures that variation generated by the first RD system does not become amplified by the second RD system, which would be likely to happen if (a) there was any regulatory feedback between the two RD systems or if (b) they were acting simultaneously. In the latter case (b), travelling waves from the developmentally earlier arising costal scute placodes might interfer with the later arising vertebral scute placodes, leading to very different scute sizes. In the former case (a), positive feedback between the two reaction-diffusion systems might easily lead to rounds of noise amplification [436]. Unlike many hierarchical patterning mechanisms, RD systems do not depend much on the precise initial conditions [437, 438]. Indeed, modifications of the inductive pattern typically consisting of 12 regularly arranged spots along the carapacial ridges in the model do not have any substantial effect on the resulting scute primordia pattern, as long as they are not too extreme. The reason for this is that, in patterns produced by RD systems, the initial conditions mainly serve to break the symmetry, while the RD dynamics produce the periodic pattern in a self-organizing manner. This is most likely also the reason why experimental ablations of parts of the carapacial ridge does not lead to even partial absence of vertebral and costal scutes [77]. Second, scute pattern development is a morphodynamic process, as patterning and growth co- occur simultaneously. As discussed in previous chapters, it has been argued that [8] morphodynamic processes often amplify developmental noise and are likely to produce non-gradual and large-scale variation (see 1.10). In the model, non-linear variation (such as the emergence of additional placodal spots) arises by compromising the diffusion space in the developing carapace, as it happens by growth defects or mechanical tissue deformations. Although this would also be true if mechanical stresses did affect the patterning without concomitant growth (i.e. in a hypothetical alternative morphostatic turtle scute development), it is likely that there is some amplification of the develop-

49 mental noise by growth. Particularly, the fact that anomalies tend to occur more often within the vertebral scute series than anywhere else in the carapace is linked to the morphodynamic patterning process: The correct placement of the vertebral scute primordia depends on the synchronized arrival of signaling molecules from both sides of the carapacial ridge. If biases, either temporal or spatial, occur in either side of the carapace costal scute primordia might still form in the correct places, but the vertebral scutes might not develop symmetrically or fail to fuse later. A selection of such scenarios is schematically depicted in 5.3. Since the space where vertebral scutes are going to form opens up during carapacial growth, any global or local growth anomaly (delay, acceleration, directional bias) will affect the vertebral scute series and become amplified, but will hardly affect the costal series. Although this bias exists even in the hypothetical morphostatic case, it is amplified in morphodynamic developments, because then, misfits between the vertebral scute primordia can result from both spatial and temporal signaling asymmetries. If development was morphostatic, temporal noise would not matter much. On the other hand, growth occurs rather slowly (as compared to the GRN dynamics) and, thus, its impact on noise amplification might be moderate. Also, there is no feedback between signaling and growth, which explains why we do not see any extreme variation that might be expected in highly morphodynamic developments. Taken together, the hierarchical arrangement of the two RD systems as well as the rather low rate of concomitant growth contribute to an overall reduction of high variation, which would otherwise be expected in a morphodynamic development involving emergent patterning mechanisms. This is in line with the overall high robustness of the turtle carapace scute pattern. Yet, the concomitant growth might contribute to a higher frequency of anomalies and explains the higher frequency of anomalies in the vertebral than in the costal scutes.

5.3 The carapacial ridge as a developmental organizer One of the oldest key concepts in developmental biology is the concept of the organizer, coined by the German embryologist Hans Spemann [439, 440, 441]. In a series of transplantation experiments, Spemann and his student Hilde Mangold identified specific fields of cells as capable of inducing subsequent development of an organ or body part in the surrounding embryonic tissue, using newt embryos as a model organism [442]. Transplantation of, e.g. presumptive head organizer tissue into a different part of the embryo would induce the development of an ectopic head. The organizer tissue releases signaling molecules into the surrounding tissue, where they are bound by receptors and induce morphogenetic changes. This also implies that, it such cases, the development of some organs and body parts depend on the presence of inductive signals, but not on other developmental processes in the remainder of the embryo. In this sense, such organ development can be considered independent “modules” [443, 444].

50 scute placodes possible (schematic) scute pattern

normal development

wider CR

oset CR

skewed CR

Variation (max.=1)

local growth anomalies

unequal CR lengths

Figure 8.: Possible generation of scute anomalies by anomalous positioning of the Carapacial Ridge (CR) or mechanical stresses. In the schematics, only a part of the developing carapace is shown; purple spots denote the positions of scute primordia after applying the change to the positioning of the CR or after applying mechanical stresses, black lines in the plots on the right side the possible outlines of the scutes. Since the placement of the vertebral scute primordia depends on signaling coming from both the left and right CR, vertebral scutes show anomalies more frequently, while costal scute primordia only depend on induction from one CR side, which reduces the amount of anomalies in the costal scutes. In real developing turtle scutes, the final scute pattern also depends on how different parts of the developing carapace change over developmental time, e.g. due to growth asymmetries or transient environmental factors. On the right side, variation of scute placode placements is quantified in a number of Trachemys scripta embryos (n=14). Above, an example embryo at stage Y15 is shown, with white circles around scute primordia. Below, filled circles show the average positions of vertebral and costal scute primordia, whose colors represent the measured variation (1=highest variation). These circles are connected to smaller circles by grey whiskers, which represent the individual scute primordia positions of all embryos used in the study. The data shows that higher variation in verebral than costal scutes is already manifest in variation of the scute primordia positions during development.

51 In the development of the turtle carapace, inductive signals have been proposed to emanate from signaling centers along the carapacial ridge [75]. This is part of the hypothesis upon which I constructed the model of turtle carapacial development and in line with the observation that the costal scute primordia appear in development before the more central vertebral scute primordia, suggesting a wave of inductive signals coming from the carapacial margins. In situ studies of scute- less softshell turtles (Pelodiscus) reveal that the carapacial ridge of this species does not show the regular punctuate gene expression pattern that is seen in Trachemys (see Contributed Publication I). Also, if the gene expression pattern in the carapacial ridge is prevented from forming in Tra- chemys through blocking of important signaling pathways, no scute primordia spots appear. These experiments suggest that the expression of key signaling molecules in the carapacial ridge is required for the development of scutes. Although it is not clear yet which particular signals originating from the carapacial ridge are responsible for the induction of scute development, bead experiments (Con- tributed Publication I) revealed that additional Shh will cause the formation of additional scute primordia, while additional Fgf4 will prevent scute primordia formation. Thus, the carapacial ridge appears to act as an organizer of carapacial scute formation in the turtle. Yet, the absence of signaling from the carapacial ridge does not prevent the formation and outgrowth of the carapace, as evidenced by the softshell turtles; it prevents the formation of scutes and possibly interferes with dermal ossification [435], since the carapace is present, but lacks the characteristic scute and ossification patterns. Consequently, carapacial outgrowth is not part of the developmental module induced by the carapacial ridge, but a developmental process largely independent from it. However, unlike in the classic organizer studies, no grafting experiments have been conducted in order to test if tissue from the carapacial ridge is also sufficient to induce subsequent scute development. It is likely that the carapacial tissue flanked by the carapacial ridge, either the future carapacial tissue itself or other underlying tissues such as the somitic mesenchyme, plays some active role in scute development [28] and that the carapacial ridge can only act as an organizer in this very specific tissue environment, but so far this is mere speculation. Thus, depending on the strictness of the definition of organizer used, the carapacial ridge can be classified as organizer or potential organizer (if we demand a proof of sufficiency for induction, i.e. by grafting experiments) of scute development. Even in the classic case of induction by an organizer, as described by Spemann, induction of development could only happen, because the organizer was transplanted into tissue that expressed the correct proteins to be repressed (most importantly BMP4 [445]) by the signals emanating from the organizer tissue. This suggests that induction usually involves a commitment of both inducer and inducible tissue and that development usually depends on gene expression changes in all interacting tissues. Elucidating the molecular signals by which the carapacial ridge induces reaction-diffusion dynamics and subsequent steps of scute formation will be an aim of future studies.

5.4 Comparison between scutes and other ectodermal appendages from a developmental perspective Turtle scutes count amongst ectodermal appendages, such as scales, feathers, hairs and teeth. Now that our studies have elucidated some crucial mechanisms involved in turtle scute development, we can compare it to the development of other ectodermal appendages. Since there are obvious morphological differences between the different types of appendages, which most palpably come from the secretion and morphogenesis of different keratins or other specialized ECM molecules, my comparison will mainly focus on early developmental stages in which differences and commonalities may be less clear. One common feature shared between all ectodermal appendages is the involvement of both epidermal and mesenchymal tissues in signaling which, by mutual crosstalk, leads to morphological changes in both tissues. The signaling pathways involved in ectodermal appendage are largely

52 conserved between hair, tooth and feather development: Shh, Fgf, BMP, Wnt, and Edar pathways, with dynamical roles and localizations throughout different developmental stages. Although the knowledge about turtle scute development is, in comparison to hair, tooth and feather development, scarce, our study (and a few others [83, 75, 84, 74]) have shown that BMPs, Fgfs, Shh and EDAR (Moustakas-Verho, unpublished) are dynamically expressed during scute development and that some of them are critically important for it. In the beginning of the turtle carapace development, Fgf signals induce the carapacial ridge in a manner similar to the induction of limb buds [75, 76], involving epithelial-mesenchymal crosstalk and mesenchymal condensation. The carapacial ridge will subsequently serve as an organizer of turtle scute development. In the developing avian skin, Fgfs have been shown to be required for feather bud induction [206], while in the developing odontogenic mesenchyme, they are responsible for the specification of the part of oral tissue in which teeth will subsequently form, by inducing the expression of specific transcription factors (Pax9,Lhx) [227]. If induced in avian jaw tissue, Fgf8 is even capable of reinducing the development of an odontode (the tooth-forming tissue unit) leading to the development of simple teeth [168, 166, 167]. Only in the turtle carapace, there is evidence that the first source of the Fgf signal that will induce changes in the integument emanates from an external mesenchymal structure (the ribs), while in other ectodermal appendages, its source lies within the developing tissue itself, mainly in the epithelium. Fgf signaling is in all instances required to induce the specification of the tissue that is going to give rise to ectodermal organs [446, 227, 206]. Unlike other organizers of integumental patterning, the carapacial ridge also induces changes to the axial skeleton. Experiments have demonstrated that preventing Fgf signaling in the carapacial ridge leads to an arrest of the rib outgrowth required for the further development of the carapace [75]. Thus, the carapacial ridge is responsable for both the formation of the carapace (which is the case in all turtle species) and its integumental patterning (which is lost in softshell turtles and Dermochelys). Subsequent roles of Fgfs consist in the induction of changes in the mesenchyme, usually by activating or regulating BMPs, in order to form dermal condensations [204, 447, 222, 223]. This is the case in hair, tooth and feather follicle formation. In the developing odontodes, Fgf signaling causes, in addition, proliferation in both mesenchyme and epithelium [235]. So far, there is no experimental proof for a comparable role of Fgfs in the turtle scute development, but the formation of dermal condensation in the carapacial ridge, colocalizing with the expression of Fgf10, is at least suggestive of this function [75, 85]. During later development, dermal condensation does play an important role in the 3D morphogenesis of carapacial scutes [70, 69]. The carapacial ridge is a structure characteristic of and confined to turtle development. However, development of the feather placodes on the dorsal trunk in chicken embryos appears to begin with two lines parallel and adjacent to the dorsal midline laterally to which, in subsequent stages, further feather primordia form, while the tissue grows, thus apparently resembling the induction of scute primordia [127]. Like in the turtle embryo, induction of feather buds seems to proceed in a wave- like manner, albeit in the contrary direction (inside-out in avian feathers versus outside-in in the turtle, in a similar manner as the appearance of chondrichthyian skin denticles [448]). However, while the inductive signaling centers for integumental patterning along the carapacial ridge appear to be patterned by the underlying ribs, nothing comparable has been described, to my knowledge, for the incipient feather bud patterning. Nevertheless, it has been shown that signals from the dermomyotome, such as Wnts, are crucial for the induction of appendage development in avian and mammalian skin [449], suggesting a common inductive mechanism for skin appendage development through signals from competent underlying somitic mesenchyme. As described before, we detected a characteristic alternating expression pattern of BMP2, its inhibitor Gremlin, and Shh within the carapacial ridge. Although the specific mutual interactions between those factors have not been revealed yet, mutual repression is very likely to be involved. Antagonistic interactions between Shh and BMP (although BMPs might also activate Shh expression

53 [204, 450]) have been reported in teeth, hair and feather follicle development [191, 204, 451]. Shh has been usually restricted to the epidermis, while both BMPs and Gremlin have been found expressed in both dermis and epidermis, consistently in all the ectodermal organs considered in this comparison. Interestingly, repression of Edar (Moustakas-Verho, unpublished) as well as Fgf4 leads to a randomization of the regular pattern along the carapacial ridge as well as an overall reduction of spots. Although there might be no structure homologue to the carapacial ridge in the development of other ectodermal appendages, this pattern is at least strikingly reminiscent of the Tabby mutant (a mutation in the Eda gene [176, 452]), which leads to a reduced and irregular dentition, a sparser fur, plumage, or scale coverage [176, 186, 453]. In tooth development, increased Fgf signaling has been reported to partially rescue Tabby phenotypes in mouse [454] and genes interacting with Fgf signaling are involved in the formation of the rodent diastema [455], suggesting a link between Edar and Fgf signaling. Mutagenesis experiments in zebrafish, in which several phenotypic malformations, most prominently a defect in regular body scalation, reveal that the developmental roles of the Edar pathway are indeed very conserved in vertebrates [456]. All ectodermal organs in our comparison have been suggested to involve reaction-diffusion dynamics [126, 127, 116, 450, 141, 448]. The signaling molecules involved are mainly BMPs and Shh, but also Fgfs and Gremlin. Shh is usually expressed at the center of the periodic structures which undergo morphogenetic change, i.e. the center of the feather buds [209], the place of hairbud ingrowth [457], the enamel knot in the odontode [235], and possibly changes of the epithelial cells in or in the folds between developing turtle scutes where slightly invaginating epithelial cells form sulci (i.e. in the places in which the traveling waves in the model collide), suggesting that the general morphogenetic role of Shh might be twofold: defining the places of ectodermal appendage formation by reaction-diffusion dynamics and regulating local morphogenesis (e.g. by inducing morphological changes and growth in the epithelium). High concentrations of BMPs, in general, inhibit proliferation [458, 233, 459]. In the turtle carapace development, Shh and BMPs are expressed together in the spots in which scute primordia form (although not completely overlapping, suggesting mutually antagonistic behaviour), while Gremlin is expressed circumferentially around these spots. This suggests that Shh and BMPs are indeed crucially involved in the reaction-diffusion dynamics and that Shh expression might induce tissue proliferation and scute morphogenesis. Since placing a bead coated in Fgf4 into the developing carapace prevents the formation of scute primordia locally, which is in line with increased inhibitor production in a reaction-diffusion model, Fgf4 might play an important role inside the reaction-diffusion system. On the other hand, adding Shh protein to the developing carapacial tissue leads to the formation of additional scute primordia, suggesting an activator role for Shh. In the developing bird plumage, adding BMP4 by bead implantation gives rise to an inhibitory zone around the bead [204], while Fgf4 increases the number of feather primordia and causes them to fuse. Shh appears to act as an activator, like in the turtle. This suggests that the role of Fgf4 in turtle scute and feather bud development might be contrary, while BMP4 in the feather bud appears to have an analogue role to Fgf4 in the turtle scute. In the developing tooth, BMPs have been suggested to act as an activator in the reaction-diffusion system, whereas Shh and Fgfs to act as inhibitors [450]. In the hair follicle, different signaling molecules, Dkk and Wnt have been named as being mainly responsible for the reaction-diffusion dynamics [126] which lead to a regular spacing of hair follicles. Taken together, although the molecules that create reaction-diffusion dynamics by interacting are the same in feather bud, tooth and turtle scute primordia formation, their particular roles within the reaction-diffusion systems may be remarkably different. Interestingly, scute morphogenesis involves a second round of activator-inhibitor dynamics, this time not producing stable patterns, but a transient travelling wave. Again, Shh and BMPs are expressed dynamically in this wave-like manner, until these waves collide and form the future scute outlines, suggesting that at least some of the molecules involved in the first reaction-diffusion

54 system are part of this second reaction-diffusion system as well. Several rounds of reaction-diffusion dynamics are also involved in feather, and, to some extent, in tooth and hair development. In feather development, the Shh expression centers in the epithelium of the feather buds form longitudinal stripes while the feather buds elongate. These longitudinal stripes induce the formation of rugae and valleys that deepen in a progressive manner, and, ultimately, give rise to the delicate feather branches. Harris et al. [214] demonstrated that this process can be explained by a reaction-diffusion system involving Shh and BMP. During the development of complex teeth (such as molars), primary enamel knots may give rise to secondary enamel knots, while the tissue proliferates, to ultimately define the position of the tooth cusps [235]. However, this apparent two-step process is not based on two different reaction-diffusion systems, but a splitting of spots due to an increasing diffusion space. Similarly, the placement of hair placodes proceeds, as previously explained, in two rounds of activation [141, 140]. Also in this case, both “rounds” are probably due to the very same reaction- diffusion system acting over a longer time in a growing tissue. Interestingly, a recent study found evidence for a transition between travelling waves and Turing pattern during the establishment of tooth prepatterns [145], but in this case, the travelling waves phase does not appear to give rise to any spatially distinct pattern. Thus, the sequence of two reaction-diffusion systems acting in scute development has mainly similarities with the development of feathers, in which two distinct reaction-diffusion systems act sequentially. Yet, the morphogenetic events that are patterned by them are completely different: While the second reaction-diffusion system leads in avian feathers to a longitudinal fine subdivision of the growing buds, involving Turing patterning, it leads in turtle scute development to the formation of the 3D scute shape by a travelling wave, potentially inducing morphogenetic changes in the cells within and between scutes. Finally, the patterning of the tissues that give rise to the different ectodermal organs also involves the expression of structural and extracellular matrix proteins that will define the mechanical properties and the specific functions of the different ectodermal organs. In the case of the turtle carapace, this is a firm keratinization (β-keratins) which increases the protective function of the carapace. Keratins are also expressed in the feather and hair placodes (although the keratins involved in hair formation are different, as explained in 1.6), while the odontode gives rise to cells secreting dentin and enamel. Thus, one of the reasons why ectodermal organs might have become so successful in evolution is the versatility of keratins as well as the option of secreting different hard biomaterials, assuming very different mechanical properties and, thus, allowing for a multitude of different functions. Taken together, the development of carapacial scutes has many commonalities with the development of different ectodermal organs. These commonalities lie in the signaling pathways involved and their particular roles, the basic structure of development, with inductive signaling and one or two stages of reaction-diffusion dynamics, some specific morphogenetic events in epithelium and mesenchyme, including dermal condensations, ectodermal proliferations and foldings, and multiple rounds of epidermal-mesenchymal crosstalk. Differences consist in the particular structure of the reaction-diffusion systems: Although the same molecular pathways are involved, the role of each of them appears to differ. In some ectodermal appendages, such as turtle scutes and feathers, two different reaction-diffusion systems are involved. While the first one leads to a regular distribution of placodes, similar in most ectodermal organ developments, the second orchestrates a very different morphogenesis linked to the future function of the appendage (which is sometimes called “secondary induction”). Thus, while the core inductive events and patterning processes are conserved between different ectodermal organs, it is the subsequent morphogenetic processes that are mainly responsible for the crucial differences between them.

55 Table 1: Comparison between diﬀerent ectodermal appendages.

Feature Mammalian Hairs Teeth Feathers Scutes Inductive Wnt [460] BMP4, b-Cat[205], Fgfs [75], Wnt signal Fgfs[227, 461], Fgfs[446, 213] [29] Wnt[462, 463] Prepattern none, tooth odontogenic signal from dorsal carapacial primordia form mesenchyme, midline[464], ridge, induced more or less regular spacing of induced by somitic from ribs (i.e. simultaneously in wider area, future odontodes mesenchyme [449] somitic induced by somitic mesenchyme) mesenchyme [75] [449] Mesenchymal dermal papilla mesenchymal by BMP underneath the Condensation formation by Shh condensation interactions scute epidermis from Epidermis induced by Msx1 [464, 467] [77, 70] [465, 457]. [466] Epithelial proliferation and proliferation proliferation leads ingrowth at proliferation ingrowth into outside Enamel to evagination, the scute dermal Knot, folds give elongation and borders mesenchyme rise to cusps subsequent feather [468, 54] morphogenesis, induced by Shh[209] Reaction- Dkk –| Wnt [141] Sostdc1, Shh –| BMP –| Fgf, Shh proposed: Fgf4 diffusion Wnt, Act [204] –| Shh, BMPs system [410],[469] might be involved Primordial salt-and-pepper diverse cusp regularly regularly pattern pattern, spacing formulae, species hexagonal hexagonal, regular. Addition dependent influenced by of secondary growth placodes that may be morphologically different [140] Developmental primary, secondary odontode placode formation, prepattern Stages hair follicle patterning, feather bud formation in formation, hair primary enamel outgrowth, feather the CR, morphogenesis knot, secondary morphogenesis primordia enamel knots, placement, enamel/dentin scute growth secretion and final morphogensis Bilateral PCP in nascent AP polarity often AP polarity by AP polarity Symmetry/ hair follicles compromised, Shh-BMP presumably by AP polarity [470] functional antagonism, Wnt-7 Shh-BMP constraints [471] dependence[209, antagonism 210] specialized α-Keratins Enamel, Dentin β-Keratins β-Keratins ECM Wnts required for induction of Induction of Induction of induction odontogenic tissue feather follicle CR [29] , [221, 472, 460], [462], tooth fate[212], feather interference and spacing [141] renewal [475], bud AP leads to similar of placodal maintenance of asymmetry [210] phenotypes as pattern, secondary enamel EDAR maintenance of knot pathway:sparse inductive signals morphogenesis or absent CR from the dermal [476] prepattern. papilla [473]. Sustained activity disrupts hair shaft morphogenesis [474]

56 Feature Mammalian Hairs Teeth Feathers Scutes Epithelial- several inductive several inductive several inductive proposed by events: Induction mesenchymal events: events: of mesenchymal comparison with interac- Mesenchymal odontogenic condensation by other ectodermal tions activation by mesenchyme and epithelial Shh; appendages and Epidermal Wnts; BMP expression placode formation for development of induction and induced by signals in epithelium CR[70]; proposed proliferation of from epithelium; induced by in this thesis: Fgf placode in odontogenic mesenchymal signaling in epidermis by mesenchyme signals; placode mesenchyme mesnechymal induces placode in signals back and induces prepattern signals that are epithelium; induces gene in CR mesenchyme expression in induced by placode signals to mesenchyme; Shh and epithelium, epidermal Shh; induce in the epidermis which leads to epidermal signals mesenchymal induces patterning of the required for condensation; proliferation in carapace. dermal enamel knot mesenchyme and Signaling between proliferation and induced by epithelium mesenchymal condensation mesenchymal [216, 433, 208] condensate and [477, 432] BMP; enamle knot scute placodes has affects to be elucidated. proliferation in Reduction of epithelium and dermal ossification mesenchyme in soft-shell turtles [478, 229] suggests epidermal-dermal signal is required for dermal ossification. Shh activated by β-Cat Marker of Induction AP polarity of [472], Hair follicle presumptive dental mesenchymal scutes, in morphogenesis: epithelium [463], condensation [208], opposition with epithelial tooth cusp cell death [208], BMP, possible role ingrowth[465], patterning feather bud (with BMP) in the formation of [410, 462, 479], elongation [481], induction of the dermal papilla tooth size patterning of barb scute margine [465, 457] regulation ridge structures ingrowth (possibly by [481] AP polarity restricting the area of feather buds of the enamel [209] knot) and control of epithelial invagination [479], Regulation of Epithelial cell survival [480] Fgfs control of cell induction of Feather bud Expressed in the cycle[482] odontode and induction CR [76], Induction definition of its [206, 446] of CR [75], area[227, 455], patterning of scute proliferative signal primordia from enamel knot [235, 231] BMPs induced by b-Cat BMP4: restriction feather bud fate CR patterning [472, 474], with area of inhibition [488], together with Shh, noggin: induction presumptive feather bud ossification of the of secondary hair odontode [486], placement [204] mesenchyme [84] follicle[483, 484], secondary and barb ridge hair follicle mesenchymal- structure suppressor epithelial patterning with [485, 458], induction[487] Shh [209] orchestrates tissue interactions [477] Edar interfering: sparse interfering: interfering: sparser interfering: pathway hair [176]. irregularly spaced, plumage [178, 453], irregular, sparse Increase: hair sparser teeth, increase: placode prepattern on CR clusters without simpler cusp fusions [453, 178] interplacodal patterns[490] space[489]. Refinement of pattern established by Wnt [221]

57 5.5 Interspecific variation of turtle scutation Most variation between turtle species is subtle and lies merely in differences in shape and relative sizes of the different carapacial scutes. While in some species, vertebral scutes tend to be larger than costal ones (e.g. many Graptemys species), the opposite is true for others (most marine turtles) and there is substantial diversity even within species (including Trachemys scripta used in my studies). However, as already discussed before, such variation cannot be simulated with the current model of scute pattern development. The reason is that reaction-diffusion systems, which I propose as the main mechanism in scute patterning, will position the activator maxima that will give rise to scute primordia equidistantly. This, in the case of turtle scutes, leads necessarily to equally sized scutes. Thus, the spacing of periodic ectodermal appendages, such as hair, plumage and scales in Squamata, is usually quite homogenous. Yet, there are mechanisms that can give rise to differences in size and spacing of periodic structures within the same organism. Such differences are often due to heterogeneity in tissue growth between different parts of the tissue area to be patterned by reaction- diffusion dynamics. Very importantly, the timing of growth matters. If the tissue area grows while tissue patterning by reaction-diffusion dynamics is still ongoing, either do existing activator maxima split, become displaced, or new maxima will form in between existing ones [491]. Which of these scenarios will take place mainly depends on the particular reaction-diffusion system, but also on the particular growth dynamics. For instance, Kondo [138] reported that in some species of tropical fish, an alternating integumental pattern of thick and thin stripes is formed in two steps: First, the thick stripes appear by reaction-diffusion dynamics. Subsequently, the tissue grows, leaving enough space in between two thick stripes for another, slimmer stripe to form in between. Splitting stripes have been described in the development of palatal rugae [137] and suggested for the development of distal limb bones [144, 492], and lung branching [493, 494]. If tissue growth mainly occurs after the patterning by reaction-diffusion dynamics has completed, individual appendages in different places that originated as part of the same regular pattern can assume substantially different sizes. The fact that the scute primordial pattern is already established in mid-development (around Y18 [419]), before the carapace has undergone most of its outgrowth, implies that different growth schemes in later turtle development are likely to cause phenotypic differences, both within and between individuals. Distally elongated scutes could be produced by increased or prolonged lateral growth within the area of those scutes, or by changes in the cell size or shape. In general, shape differences between scutes within the carapace might imply different growth rates or growth timing in different parts of the carapace. Alternatively, different scute shapes could also result from biases in the diffusion field by locally impaired diffusion (e.g. by local differences in the expression of structural proteins that may hamper diffusion), by local expression of proteins that interfere with the reaction diffusion dynamics (e.g. [144]) or by local differences in tissue bending. The latter could happen if the developing carapacial tissue locally expands away from the plane (forming a buckle or a depression perpendicularly to the carapace plane), thus locally increasing the space diffusive signals have to travel across, if signaling takes place inside the epithelium or relatively superficial layers of mesenchyme. Some of these hypotheses about the origins of different scute shapes could be implemented in the existing model, e.g. local growth biases, while most of them require explicit modelling of cell shapes and mechanical properties, as well as more realistic cell divisions. This could be achieved with the EmbryoMaker software. However, in order to address these hypotheses, we would also need to gather experimental data about cell division patterns in the developing carapace at different stages, which could be done e.g. by quantitative analysis of staged BRdU-stainings, as well as histological data about cell shape changes during development. While some studies about the latter exist [54, 59, 86, 495], including descriptions of mesenchymal condensations and epithelial cell shape changes, the former still needs to be conducted. Furthermore, we need embryological data about the

58 carapacial development of species with higher scute shape heterogeneity within the carapace, since in the species mostly used in developmental studies, Trachemys scripta, vertebral and costal scutes are of rather similar size (although substantially different scute size proportions have been observed in individuals). Once those data have been collected, EmbryoMaker will prove a helpful tool to decide about which hypothesis is best suited to explain the development of phenotypic differences between scutes. Interestingly, a large heterogeneity of scute sizes can be seen in marine turtles whose costal scutes are expanded laterally at the expense of the vertebral scutes. One hypothesis which could explain this pattern proposes a temporal shift between the timing of scute pattern formation and carapacial outgrowth. If the patterning of scute primordia takes place almost entirely prior to the lateral expansion of the carapace, pairs of vertebral primordia might fuse early due to insufficient space, leading to very narrow vertebral scutes. If growth then mainly affects the area of the already formed costal scute series, the typical phenotype of marine turtle carapaces may result readily. However, since we do not have any embryological data, this hypothesis is, at best, highly speculative at the moment, although heterochronic changes in development have already been proposed to be a major mechanism in developmental evolution[496]. Interspecific variation that has been more amenable to the model is symmetrical numerical variation in scutes. This kind of variation only applies when comparing the normal turtle scute pattern to certain marine turtle species, namely C.caretta and Leptochelys spp.. In the model, such variation can be produced in several ways, most easily by simple differences in length and size of the tissue that will give rise to the carapace. By slightly (<20%) increasing or decreasing the carapacial tissue area anteriorly and posteriorly, I was able to reproduce the numerical patterns found in the aforementioned turtle species. It has to be disclaimed that the scute patterns generated this way would look indeed closely reminiscent to the scute patterns of those species. Nevertheless, the scute numbers and appoximate positions are very similar between real and in silico carapaces and further differences are arguably due to scute shapes that are very different from T.scripta, which could only be implemented in the model by adapting different growth schemes or even more realistic ways of growth implementation. The finding that carapacial elongation can easily produce, in a symmetric way, additional vertebral or costal scutes is in line with the visibly elongated carapace of Notochelys platynota, a limnic turtle species with a high frequency of supernumerary vertebral scutes, as well as with observations by Parker [343], who measured an altered length/width ratio in anomalous turtle carapaces. Interestingly, all scute patterns that we considered in our study appear to occupy different, partly disparate areas of the morphospace spanned by combinations of different amounts of anterior or posterior carapace extensions or compressions. This is interesting for two reasons: First, the particularly patchy and discoherent distribution of the scute formulae of marine turtle species might explain why the divergent scute patterns of these species tend to show much more numerical phenotypic variation than most other species [41, 38], since individuals of C.caretta and Leptochelys spp. show both symmetrical and asymmetrical numerical scute aberrations much more frequently than most terrestrial, limnic and marine species with the standard scute pattern. If the area a specific phenotype occupies in the parameter space is small or disparate, it is likely that subtle developmental variation will cause a phenotypic pattern transition. Second, the distribution of different phenotypes in the morphospace is non-linear. This means that we do not see a simple correlation between scute number and total length of the carapace, which one might have expected by intuition: Increases in carapace length do not always lead to an increase in scute number, it is sometimes even the opposite. Such a non-linearity might be a common feature of both emergent patterning mechanisms and morphodynamic mechanisms [8]: The continuous impact of morphogenesis on signaling (and vice versa, although this does not appear to be the case in turtle scute pattern development) in morphodynamic developments causes that small differences in early development (the initial conditions in the model) will become amplified or reduced. In addition, emergent mech-

59 anisms are more likely to produce non-intuitive patterns (such as spots and stripes and transitions between them) as they, unlike hierarchical mechanisms, feature feedback loops between the signaling factors, which, again, lead to non-linear phenotypic responses to parameter variation [436]. In the case of turtle scute patterning, the placement of scutes, particularly the vertebral ones, depends on where the waves of inductive signals, coming from both lateral sides of the carapacial ridge, collide and how much space there is at the particular time of their confluence, as determined by growth. It has to be emphasized that growth occurring simultaneously with the signaling creates a time dependency for the patterning process which introduces another dimension in which the collision of the travelling waves at the “right” spot might fail and, thus, makes it more likely that small variations in earlier stages can give rise to large phenotypic variation. In other words, if growth does not happen concomitantly with signaling, patterning anomalies will mainly be caused by spatial variation, whereas, if growth and signaling co-occur simultaneously, anomalies can be caused by spatial or temporal variation, making anomalies more likely to happen overall. Small differences in carapacial size or timing can therefore cause scutes to divide or merge. In order to give rise to the particular shape of the carapace in Trachemys, growth rates must be larger in the center of the carapace than in the front and rear, and largest slightly anterior to the center. Thus, a growth anomaly in the anterior part of the carapace might not produce the same number of scutes as the same growth anomaly in the posterior part. In situ staining for genes that mark scute primordia reveals that there is, indeed, a lot of variation in size, shape and proportions of the developing carapace between individuals (although it cannot ruled out that a part of this variation is due to the handling of the embryos in vitro) with the highest amount of variation in the anterior mid-center of the carapace, exactly where tissue growth is supposed to be largest 5.3, in line with our hypothesis and in silico observations 5.7. However, since the largest portion of the morphospace of carapacial length appears to be occupied by the standard scute pattern, it can be assumed that most of this carapacial size variation observed in real embryos will not give rise to anomalous scute formulae in the adults. Furthermore, several of the earliest turtle taxa (from the late Triassic and early Jurassic) displayed substantially different scute patterns. Specimens often show larger vertebral scutes than any extant turtle species, while having costal scutes of very reduced size, and sometimes an additional series of supramarginal scutes with a size rather in the range of marginal than of costal scutes (i.e the opposite deviation from the standard scute pattern than in the marine turtles). Such a pattern might be explained by a strong growth in the center of the carapace during an embryonic stage at which the scute margins are already defined, while virtually inhibiting growth at the outer parts of the carapace. Since the regular scute pattern is defined by a reaction-diffusion system, adding another scute series between the costal and marginal scutes is easily accomplished by either increasing the diffusion space locally (i.e by growth in later developmental stages), or reducing the distances at which scute primordia form, e.g. by differences in diffusion rates. Less parsimoniously, there might have been a local expression of proteins interfering with the reaction-diffusion system in a way that reduces the distance between activator maxima or reduces the diffusion of the travelling waves. Another possibility is the splitting of marginal scutes that form from the signaling centers on the carapacial ridge. Even in extant turtles, supramarginal scutes can be found as a rare anomaly, probably due to an excess of diffusion space, in the same way as tiny additional scutes can sometimes be found between more central scutes if there is space for them. Which of the aforementioned possibilities is really most likely to be responsible for the production of such additional scutelets has to be tested in a more comprehensive model of scute formation, although a definitive answer, as far the unique pattern of early turtles is concerned, will not be possible.

60 5.6 A genetic origin of turtle scute anomalies? We were not able to reproduce typical turtle scutes anomalies by systematically introducing genetic mutations (i.e. changes in GRN topology, interaction strengths of activator and inhibitor, degradation and diffusion rates) in the present model. Since most anomalies observed in turtle carapaces are asymmetric, based on my own and other authors’ observations [59, 86], and genetic changes would be likely to affect development globally (i.e in the entire carapace) and almost always symmetrically, a genetic explanation of asymmetric scute anomalies is highly unlikely to be true (also because asymmetrical anomalies occur equally often in the left and right carapacial hemisphere: mutations that lead to left-right asymmetries tend to produce asymmetries consistently directed towards one side [497]). However, genetic mutations could decrease “developmental stability”, a term that describes how much noise affects development and the resulting phenotype. Developmental stability would be considered low if a given turtle’s developmental trajectory would only require a small change in order to induce a phenotypic shift. Note that this property can change as a result of genetic changes. For instance, mutations might subtly increase the diffusion space in the developing carapace, e.g. by increasing cell division rates, so that, while unperturbed development still produces the normal phenotype, very small perturbations (such as slight growth asymmetries) would create just enough space for a scute placode to split into two. In such a case, the anomaly itself would still be produced by environmental noise, but genetic changes would permit this to happen much more frequently and by much smaller environmental perturbations. As a result, anomalies would appear hereditary, although the hereditary property would actually be the susceptibility to noise or “developmental instability”. This might be the reason why some authors have suggested that anomalies in turtles were hereditary and, therefore, at least partly genetic [321]. Thus, although we cannot completely rule out that some turtle anomalies might be actually hereditary, our simulations suggest that, even if some degree of heredity of scute anomalies was clearly proven, this would rather show heredity of developmental stability, while the causation of anomalies would still be strongly environmental. If anomalies can be linked to mutations that affect developmental susceptibility to noise in a given population, those mutations might have accumulated due to bottleneck effects [261, 262, 263, 264, 265, 266], as such events increase the likelihood of deleterious mutations being passed on or to become homozygotic (i.e. giving rise to an inbreeding depression). Such bottleneck effects, however, might also have detrimental effects on other traits, increasing overall morbidity of the individual. Alternatively, if the development of one particular organ is strongly affected by environment, this could hint at other organs being affected by it as well. More concretely, if scute anomalies are visible in some individuals, the development of other organs or body parts might have been compromised too, as most genetic pathways and developmental mechanisms are pleiotropic and disturbing them will rarely only affect one organ. In fact, many pathways important for scute development are also involved in the development of many other organs and body parts. Note that this might be equally true for both genetic and environmental causation of anomalies. Defects in other organs are also likely to have more substantial relevance for the individual’s viability and survival probability. Some authors found that hatchlings with scute anomalies tend to be smaller or slimmer than individuals with the standard scutation [61]. In such a scenario, a reduced body size might on one hand cause scutation anomalies by compromising the diffusion space of the signaling molecules producing the scute pattern, and, on the other hand, reduce survival probability due to an overall weaker habitus. Thus, it is very likely that selective disadvantages and carapacial anomalies might be correlated, but not causally linked. Furthermore, it is unclear if there are selective disadvantages for turtles with anomalous scutation. Clearly, many anomalous turtles survive into adult age, and, since the main function of the carapace is protection and not integrity of the scute pattern, it is not evident why selection should strongly disfavour those individuals. Extreme asymmetries in the carapace might compromise loco-

61 motion, but as high agility is not the main survival strategy of turtles anyway, such a disadvantage might arguably be of minor selective importance [345, 61]. Yet, the fact that the basic carapacial pattern is so highly conserved amongst turtles may point to selective advantages of symmetrical scute patterns, although this is not the only possible explanation. One possibility is that symmetrical carapaces are favoured in sexual selection. This is not completely implausible, given that scutes are often decorated by ornamentations. So far, however, there is no experimental evidence for this.

5.7 Origins of intraspecific variation: Hypotheses revised As discussed in 1.8, hypotheses about the processes causing the formation of anomalies have been proposed for more than a century. One of the first hypotheses was Gadow’s orthogenesis [329], according to which morphologies would tend to develop towards a defined bauplan and deformations or perturbations commonly occurring during development would be compensated until the target bauplan has been achieved. Anomalies would, then, represent incomplete compensations of such transient developmental perturbations. Unfortunately, Gadow didn’t explain which processes might be responsible for guiding aberrantly patterned embryos back towards their pre-defined “stereotypical” developmental trajectories and how development would recognize that it has been deviating from its stereotypical trajectory. He also did not address the question of how anomalies form in the first place, instead he assumes them to naturally occur during development. Although this hypothesis lacks important elements and does not answer the question of where anomalies come from (instead he seems to speculate about why anomalies are not even more common), developmental processes which make that phenotypical deformations or anomalies do not persist in subsequent stages, are common in tissue repair and regeneration, often using mechanisms different from the ones in primary embryonal development [498, 499]. Also, depending on which mechanisms act in a specific development and when (cf. the discussion about stage-dependent susceptibility to environmental stress 1.7), environmental stress which an organism experiences during early developmental stages may either become largely amplified and lead to major deformations, or, on the opposite, fail to give rise to phenotypic anomalies in the adult organisms at all [271, 436]. In case of our turtle scute pattern model, the first reaction-diffusion system induces the scute primordia pattern reliably, even if there are substantial changes to the inductive pattern in the carapacial ridge, as evidenced by both in-situ stainings and model simulations. This is in line with experiments in which relatively normal scute development was observed after partial ablation of the carapacial ridge [77]. The reason is that the dynamics of reaction-diffusion systems only require a symmetry break in an initially homogeneous field of activator concentration [128], while moderate differences in the inductive signals have often no important impact on the resulting pattern. In contrast, large differences in the inductive signals may have dramatic effects, often in an all-or-nothing manner. However, such large changes to the inductive signaling might be less frequent than moderate changes or simply not occur in the anomalous patterning cases discussed here. Arguably, the complete scute absence of scutes in softshell turtles might be considered the result of such a large-scale perturbation of “stereotypical” turtle shell development. Yet, the place where the initial symmetry break occurs has nevertheless an impact on where spots will appear first: the fact that feather bud induction proceeds in a directed manner beginning from the midline [204] might be responsible for the more orderly placement of feather buds (as compared to e.g. hair placodes, whose symmetry breaks tend to happen in different places throughout the tissue). In summary, it can be argued that the reaction-diffusion dynamics are capable of balancing out some particular sort of variation originating in previous embryonal stages. Thus, although we can roughly compare Gadow’s idea with the fact that reaction-diffusion systems, as acting in early scute development, prevent certain variation from translating into phenotypic differences, it is not true that any developmental process would reduce variation progressively over developmental time, a central tenet of Gadow’s orthogenesis hypothesis.

62 Another hypothesis was brought up by Newman [42]: that there might be an evolutionary trend within turtles towards lower scute numbers and that the occurrence of additional scutes would therefore be considered an atavism. However, there is no clear indication that such a trend actually exists. Newman invokes the higher scute numbers in some species of marine turtles to support his hypothesis, assuming that marine turtles, or only Dermochelys, would be more similar to the last common ancestor (LCA) of all extant turtles [357, 358, 356]. While marine turtles had been traditionally placed as a sister clade to all other extant turtles species, more recent phylogenetic studies nest the marine turtles within one of two main turtle clades. Even if Newman’s phylogenetic assumption had proven correct, there would be no reason to conclude that the marine turtle bauplan was more similar to the turtle LCA than the bauplan of their terrestrial (and limnic) sister clade, because both clades would have diverged for exactly the same time since their LCA. What is more, only certain marine turtles display divergent scute formulae [38], whereas the other follow entirely the “standard” turtle scute pattern. Also, there is no indication of higher scute numbers in turtle taxa from the early Mesozoic, apart from additional supramarginal scutelets and a few extraordinarily divergent taxa from rather late geological periods (cf. 1.2), casting further doubt on Newman’s hypothesis. Newman’s idea was built upon the basic hypothesis that in the first turtle species, each rib would be overlaid by a corresponding scute. This implies a hierarchical hypothesis about scute formation, in which signals from the ribs would directly induce scute formation, rather than only the prepattern at the carapacial ridge. Yet, this hypothesis is completely at odds with the hypothesis of scute formation by reaction-diffusion systems that is supported by our studies. It could nevertheless be argued that the developmental processes that give rise to turtle scutation have changed during evolution. There are examples of closely-related and phenotypicly similar species in which major developmental processes have been replaced, such as [500, 501, 502, 503], by developmental systems drift. Such a scenario is, however, unlikely to have occurred in the turtle scute development for two reasons: First, reaction-diffusion systems play a crucial role in most other periodic ectodermal organs and are, compared to hierarchical patterning mechanisms 1.4, relatively easy to evolve. It is therefore probable, or at least parsimonious to assume, that reaction-diffusion systems are the ancestral patterning mechanism of ectodermal organ development. Second, Newman uses his hypothesis to explain recent scute anomalies. But even if turtle scute development had substantially changed between ancestral and recent turtles, it is highly unlikely that anomalous individuals might suddenly (i.e. by a relatively subtle change in their genes) deploy an ancestral patterning mechanism in a part of their carapaces that is fundamentally different from the one in their siblings. Contrary to Newman, Berry [360] hypothesized scute anomalies to be morphological novelties. Unlike Newman’s, this hypothesis does not make any assumptions about the developmental events that are responsible for scute anomalies. However, in order to be classified as novelties (sensu Müller[14] or Mülleret Wagner [12]), scute anomalies would have to be, to some extent, heritable (even if this does not necessarily imply a non-environmental, i.e. purely genetic, causation of anomalies). Unfortunately, we do not have transgenerational data about the heritability of certain anomalies, but the high diversity in position and severity of different scute anomalies (as described in [40]) even within the same nesting site, while similar anomalies occur in phylogenetically distant turtle clades, rather suggest a high degree of phenotypic unpredictability that is hardly compatible with most concepts of novelty (that at least require a defined novel structure). Coker came up with a different perspective on scutes [58], considering them as part of a whole pattern rather than regarding each scute as a separate character. Given the space-dependent manner in which scute primordia are placed and the fact that the travelling waves of the second reaction- diffusion system fill the space around the placodes, Coker’s intuition might have anticipated important features of scute development, without knowing about reaction-diffusion systems. Pritchard [345] assumed that, then, anomalous scutes would simply result from compensatory filling of empty

63 space between scutes. This is certainly in line with actual scute development: whenever there is too much space between two scute placodes, another placode or placode primordium is going to form in their middle. Also, it cannot be ruled out that some scutelets might develop without placode formation, simply by “filling the gaps”, but I would predict clear morphological differences between this case and regularly developing scutes. Interestingly, experimental partial removal of the carapacial ridge before scute formation does not prevent costal and vertebral scutes from forming [77]. In the light of the scute development hypothesis discussed in this thesis, this result is not surprising, as induction of scute development from adjacent intact areas of the carapacial ridge is sufficient for the formation of scute primordia centrally to the site of CR excision. One of the core findings of the papers contributed in this thesis is the environmental induction of scute anomalies. Various hypotheses suggesting that environmental stress might be responsible for scute anomalies have been made in the past, starting with Coker in 1910 [58]. One line of authors [347, 342, 341] attributes scute anomalies to high or low temperatures during embryogenesis. This suggestion is indeed in agreement with our findings, as we observed a clear correlation between high temperatures and the relative frequency of hatchlings with scute anomalies. However, no conclusive data have been provided to explain why this correlation exists, although some authors suggest that high temperatures induce deformities by increasing tissue growth rates [341, 368, 331]. This could be a realistic scenario, since many anomalies result from a splitting of scutes that would have not split, had the carapacial tissue been growing at normal growth rate, i.e. slowlier. Yet, high temperatures might as well affect protein diffusion and degradation rates, which might counteract the described effects on growth. Future studies, involving both laboratory experiments and modelling including realistic growth implementations, are needed to elucidate the mechanism by which higher temperatures affect turtle development. On the other hand, very low temperatures have been connected with scute anomalies as well, which could be explained in an analogous manner: due to reduced growth, some scutes that would otherwise split, might remain fused. Therefore, it would be interesting to compare which anomalies arise in high temperatures and which in cold temperatures. If the hypothesis that high temperatures mainly affect development by accelerating tissue growth is correct, additional scutes should occur more often at high temperatures, while anomalous scute fusions should predominate if temperatures during scute development are too low. Hypoxia, as suggested by Hildebrand [332], would possibly cause the opposite, reducing growth and causing scutal fusions and, in extreme cases, tissue necroses. Tissue necroses may lead to impredictible anomalies. Other authors propose that chemicals might be a source of anomalies [373, 374, 321]. Like hypoxia, some chemicals might impair tissue growth and even cause local necroses, which would definitively make the formation of anomalous scute patterns likely. Chemicals might also interfere with the signaling involved in scute development, preventing scute primordia from forming or delaying scute formation. Since there are more ways to completely disrupt the scute patterning process by interfering with signaling (most changes in the model parameters will prevent it from forming any Turing patterning) than causing it to form additional scute primordia, it is more likely that chemicals would cause the loss of scutes or scute fusions or even severe carapacial deformations than the emergence of additional scutes. Finally, since we were able to increase the frequency of anomalies simply by increasing incubation temperatures during turtle embryogenesis in the lab, thereby ruling out other factors, I conclude that temperature differences have certainly an effect on turtle scute development which might be, in the absence of any mechanical stresses, the dominant environmental cause of anomalies. The specific effect of different chemicals, however, still needs to be tested. Finally, another group of authors argues that, given the deformability of the turtle egg shell, mechanical stress could translate into tissue deformation and, thus, into anomalous scute patterns [336, 40]. Some of those researchers suggest increased pressure onto the embryo by nest crowding or rough handling to cause tissue deformations and anomalous scute patterns, while others point

64 to indirect effects of osmotic changes or desiccation. Although these scenarios have not been tested explicitly in experiment by us, more or less strong deformations and asymmetries appear to be common in embryonic trunk cultures, in which the decapitated embryo is kept in vitro solution, in order to be able to perform in situ stainings and other experiments. Such asymmetries could stem from mechanical stresses during sample transport and the experimental procedure, but also from desiccation or tissue damages issuing from the cultivation in vitro. Despite the rigidity of the adult carapace, embryonic carapaces are flexible and adapt readily to deformations of the underlying tissues [88]. The shell bones which are mostly responsible of maintaining the mechanical rigidity develop late, well after the keratinization of the scutes [504, 88]. Furthermore, the simulation of mechanical stresses is the easiest way to reproduce most of the asymmetrical anomalies in the turtle model. Some particular anomalies, such as zigzag patterns could only be produced this way in the model. In detail, I biased growth inside the carapace (growth speed and direction) and introduced hemispherical offsets by shifting the tissue of one of the carapacial halves up- or downwards, at different developmental time points, for different durations and with different assumptions about tissue plasticity. Since both permanent and transient tissue deformations would give rise to anomalous phenotypes, it is possible that even transient pressures, as produced by rough egg handling in turtle farms, can produce anomalous scute patterns. Timing, therefore, appears to matter as well, since tissue deformations have more severe effects on the phenotype, if they coincide with the stage in which the scute primordia are placed. Also, our studies suggest strongly the existence of a developmental stage during which the embryo is most susceptible to environmental stress (both in vivo and in silico, as shown in Contributed publication II 5.7). Yet, we have not been able to elucidate if this developmental time period is really identical with the stage of scute placode formation, due to the complications of using turtles in the lab, as discussed in 5.1.2). As scute placodes are placed where there is enough space, tissue deformations will alter the diffusion space and, thus, the future scute positions (cf. different scenarios 5.3). Earlier (during the consolidation of the inductive prepattern in the carapacial ridge) or later (during the action of the second reaction- diffusion system) mechanical stresses have much less effect on the phenotype. Desiccation might also play a role in the development of anomalies, as indicated by our climate-anomaly frequency association study, but temperature appears to be the more important factor by far. Interestingly, mechanical stresses have been found to explain variation in other ectodermal appendages as well. In teeth, the developing jaw bone exerts a lateral bias on the developing odontode and has been shown to be an important factor in tooth development. This generates a buccal-lingual bias which does not produce anomalous variation, but is functional and reproducible, defining position and number of cusps and, arguably, playing a role in the development of occlusion between corresponding upper and lower molars [471]. This suggests that mechanical induction of biases in organs patterned by reaction-diffusion dynamics and growth might be a common source of variation.

65 Phenotypical instability low

high

Figure 9.: Phenotypic instability and developmental stages. In the model, mechanical stresses of different directions and strengths (transient lateral offsets) were introduced into developing in silico turtle carapaces. The resulting variation in the adult scute phenotypes was quantified (the total difference of scute outlines positions between individuals exposed to different stresses at the same stage, lower graph), suggesting that early mid-developmental stages might be more susceptible to developmental or environmental stresses, which is in line with the observation that high temperatures correlated most with the emergence of carapacial anomalies if the elevated temperatures occurred during mid-developmental stages, as published in Contributed publication II The upper row shows where in the in silico carapaces variation tends to occur, namely the anterior vertebral scutes. Brighter colours indicate more rarely encountered phenotypes, meaning that such phenotypes only occurred for rather restricted combinations of developmental parameters.

5.8 Possible mechanisms by which epigenetic factors may interfere with development Although the articles presented in this thesis strongly point to an environmental causation of typical turtle scute anomalies, we have not discussed in much depth potential mechanisms by which mechanical, osmotic and temperature stresses could affect cells and developing tissues. Most studies about the effect of environmental stresses on tissues focus on adult tissues, not embryonic ones. This is an important difference, since there are a number of adult cell types and tissues (chiefly specialized mesenchymal cell types, such as muscle cells, tendons, bone and cartilage), whose main function is maintaining or changing the mechanical properties of differentiated organs, e.g. by alignment of structural proteins or by secreting specialized types of ECM. In the developing embryo, the diversity of cell types with distinctive biomechanical properties (including secretion of specialized ECMs) is much lower, as cell differentiation has not been concluded yet. Nevertheless, mechanics are mediated in the same ways as in adult tissues, although the absence of functional muscle cells and most cell types that secrete specialized ECM, suggests that tissue surface tension created by cell-cell adhesion proteins and the cytoskeleton, as well as by the osmotic pressure built up by the cells’ cytosol might play a more dominant role in the overall tissue mechanics of the developing early embryo than in the adult [396, 395, 505, 506]). These considerations are important if we want

66 to try to understand tissue deformations during embryonic development. At the early stages of carapacial development Y15-16 [419], the tissue is differentiated in a layer of epithelium that has already begun stratification [69] and precursors of the different mesenchymal cell types. It is only in later stages of scute development that the production of keratins commences (stages Y22-23 [69]) and the beginning of dermal ossification becomes manifest in the underlying mesenchyme only around stage Y22 [84] due to cell differentiation. Thus, during those stages at which most of the scute patterning takes place, cell-cell adhesion, cellular turgor as well as cytoskeletal rearrangements, as introduced before, are likely to be the most important factors in defining tissue biomechanics. Temperature is an obvious candidate for an environmental factor interfering with development [342, 507]. It has a direct and well-studied impact on the kinetics of chemical reactions. Since most processes in developmental biology ultimately rely on a plethora of intertwined chemical reactions (namely the chemical processes involved in gene expression and interactions between proteins, nuclear acids and lipids), temperature should affect development in multiple, partially even counteracting, ways. First, tissue growth rates should depend on temperature. For physiological temperature ranges, the higher the temperature, the faster the cell cycle is going to proceed [508]. Mammalian leukemia cells cultivated at different incubation temperatures [509] revealed that high temperatures caused cell cultures to reach the population plateau phase faster and overall higher population plateaus, independently of the total culture period, than cells cultivated at lower temperatures. Thus, higher temperatures do not only lead to shorter cell cycles, they will also lead to overall higher numbers of cell divisions (seen as a higher plateau phase in the diagram plotting time against cell number). Too high temperatures, however, will have the opposite effect. Although it is not clear to which extent these results attained in cell cultures in vitro can be applied to living tissues in developing embryos, it is reasonable to use them as a proxy. Experiments by Yntema [369, 510] in Chelydra demonstrate that higher temperatures will accelaterate gestation times, suggesting that cell cycle duration depends on temperature in developing turtle embryos. In the turtle, a direct influence of external temperatures on embryonic cells can be assumed, since turtle eggs are deposited in the ground, thereby exposing them to the surrounding environmental temperatures throughout embryogenesis. In the case of high temperatures, both more and faster cell divisions in the turtle carapace would lead to an increase of the carapacial diffusion space for the patterning of the scute primordia, thus increasing the likelihood for additional scutes to form. Too high temperatures, however, might have the opposite effect: They can compromise the structure and stability of proteins and, therefore, decrease the efficiency of the intracellular processes that underlie cell divisions. Even higher temperatures may cause tissue necroses, reducing overall tissue mass and integrity and giving potentially rise to extreme carapacial malformations. As mentioned above, protein turnover will be faster at higher temperatures, and so will diffusion. Both processes could, in the case of the turtle, substantially affect scute patterning. Higher diffusion rates can result from increased Brownian motion of proteins or from faster interactions with proteins that intercept or temporally bind diffusive molecules. Thus, signaling molecules might travel faster and farther at increased temperatures, possibly affecting scute patterning threefold: First, the spacing between scute placodes might increase (or decrease, depending on which of the specific signaling molecules is affected more strongly by higher temperatures). Second, the scute patterning might be completed earlier during development. As cell division rate is accelerated concomitantly, these two effects might balance one another out to some extent. Yet, an increased division rate might cause scute deformations by itself or might amplify existing ones at later stages. Third, fast waves of diffusive signals that originate from opposing signaling centers will produce collision fronts that are off-center more frequently. In other words, if diffusion from one signaling center starts with a delay t with respect to the onset of diffusion from the opposing signalling center, both

67 signaling waves will first meet at a point P that is closer to one of the former signaling centers. At higher temperatures, the distances of the two signaling centers from P will differ more from each other than at cooler temperatures, due to increased travelling speed of the signaling wavefronts. Since the precision of signaling wave collisions is important for scute development twice, first during the activity of the first reaction-diffusion system (meeting of the signaling waves at the midline to induce vertebral scute primordia) and the second reaction-diffusion system (definition of the scute boundaries by colliding travelling waves), this is a plausible way in which high temperatures can lead to more, and more extreme, scute pattern anomalies. This hypothesis is also in line with the prevalence of strongly asymmetrical carapacial patterns amongst anomalous individuals. In addition, intracellular diffusion rates increase at high temperatures, accelerating the dynamics of GRN interactions. Higher temperatures might thereby interfere with protein-protein, protein-DNA or RNA-DNA binding efficiencies, which might affect gene expression in hardly predictable manners. Furthermore, high temperatures are known to increase molecular noise in development, since they increase molecular motion, resulting in less stable protein folding (e.g. [511, 512]) and bindings. Frequent misfoldings of proteins result in their biological inactivity and degradation [513]. Overall, protein activity will become increasingly unpredictable with increasing temperatures, which can be measured as noise [284, 318]. Similarly, high temperatures increase the rate of transcription mistakes, leading to dysfunctional mRNAs, as well as translation mistakes, giving rise to wrongly processed proteins. Since both transcription and translation rely on the coordination of a number of specialized RNAs and proteins, a higher variation of their activity (and usually an overall reduced activity) might easily cause of amplify noise in the activity of downstream proteins. According to studies from van Oudenaarden and co-workers in bacteria [318], high temperatures will increase molecular noise mainly by increasing temporal variation in translational activity. In turtle embryos, such molecular noise might ultimately lead to higher scute variation within an individual. However, since the process of scute formation is a rather macroscopic event, involving gene expression in many cells, it could be expected that the effects of elevated gene expression noise might equal out throughout the carapace. In addition as discussed before, Turing patterns tend to be relatively stable against various kinds of noise, since the positions of activation maxima and minima depend mainly on the diffusion rates of activator and inhibitor, which ought to prevent major anomalies. However, they are not stable against cell division noise, suggesting that high temperatures increase scute pattern anomalies rather by affecting diffusion space and timing than by increasing gene expression noise. Changes in osmosis and hydrostatic pressure affect cell size and shape and, therefore, tissue biomechanics, but also indirectly gene expression [514, 515, 516, 517, 518]. As discussed above, such cellular changes should have a more pronounced effect in early embryos than in adults, since in early developmental stages, biomechanical tissue properties depend less on specialized ECM secretion, but intra- and intercellular biomechanics. If there are osmotic or hydrostatic differences within the carapace, they could lead to local alterations in cell biomechanics and cell sizes (although physiological processes will actively counteract them) [518], but also to active cell rearrangements and proliferation [519, 520, 518]. In consequence, local hydrostatic pressures may translate into asymmetries in the size of the tissue fields to be patterned by paracrine signaling. The diffusion rate itself might also be compromised locally [303]. Such a scenario might give rise to differences in scute numbers between the carapacial hemispheres. Osmotic differences might also have an unpredictible effect on the molecular interactions, since gene expression and protein activities might be affected (influences of osmolarity on membrane surface proteins: [521], impacts on gene expression profiles: [522, 523], impact on protein synthesis: [524]). Excessive changes in osmolarity will, like extreme temperatures, lead to tissue necroses and, thus, extreme carapace deformations or strong growth defects. According to our studies, however, the correlation between scute anomalies and temperature is much higher than the correlation between anomalies and drought, suggesting a rather minor role

68 of osmolarity in the generation of scutal variations. Mechanical stress, too, can lead to local tissue necroses, with similar effects as described for temperature and osmolarity. Milder pressures exert forces that can be withstood by cells and tissues. Inside cells, elements of the cytoskeleton and their distribution (which can be polarized) will bias and canalize the changes in cell size and shape generated by an extrinsically imposed force, whereas the adhesive complexes between cells transmit forces across the wider cell neighborhood to create tissue-wide responses. Pressurized tissues may, as a response, either reduce total size, undergo folding, or lead to compensatory stretching or extensions [422, 525], resulting in locally deformed tissues which potentially give rise to anomalous patterning. For instances, compressing a layer of epithelial cells may either lead to wrinkling of the entire cell layer [103], cause the epithelial cells to become shorter, but taller or wider, urge some cells to be extruded from the tissue, or simply reduce the effective cell size (in case of epithelial cells, the last option is rather unlikely). Pressurized mesenchymal cells may become deformed or condensed. Besides, cell divisions are likely to polarize under pressure, as it has been shown that mechanical tension can cause changes in orientation of the mitotic spindles [526]. All these scenarios would, in the case of the turtle, affect diffusion space and, thus, eventually lead to changes in the scute pattern. In addition, cells respond and adapt actively to deformations. While mechanical stretching on shorter time scales will typically induce a transient change in tissue shape, tissues will respond to persistent mechanical stretching by active cell rearrangements [525, 527], in epithelia mostly by large scale intercalations of neighboring cells, leading to a more permanent tissue deformation. Especially in epithelial tissues, the way how tissues respond to stress depends largely on cell adhesions proteins (mostly E-Cadherins) that form junctions between cells and how those are remodelled under the impact of mechanical stress. Mechanically induced remodelling of adhesion protein complexes usually leads to changes in cell neighborhoods and higher cell motility, but might also generate tissue invaginations [528, 529] by reorganisation of the apical/basal distribution of cytoskeletal proteins in epithelial cells. Another adaptation to tissue stress on larger time scales concerns the control of cell number and density. Cell packing will have an effect on cell division rates, with high cell densities inhibiting and low densities enhancing cell division [530, 531, 532]. Even transient tissue compression during a few embryonic stages might, therefore, cause a reduction of tissue growth altogether, which may give rise to size asymmetries. Very strong mechanical stresses can induce apoptosis or the extrusion of cells. Some authors [530] suggest a competition of neighbouring cells in growing and pressurized tissues, in which the smallest cells are driven into apoptosis. While some of those changes may be transient and reversible, others are irreversible and will easily lead to phenotypic anomalies. Other possible cellular responses to stress are mechanically induced changes in expression of genes, which are important for cell differentiation [302], or the secretion of ECM [533], thereby inducing alterations in tissue and organ development. In turn, such changes must also interfere with diffusion and, thus, development. Swartz [534] reports that stressed fibroblasts do not only express more ECM, they also induce ECM secretion in less stressed cells in order to create a synchronized stress response throughout the tissue. Other authors [535] associated de- creased ECM production with mechanical compression. Interestingly, not only the amount of stress or strain appears to be important for the modulation of ECM secretion, but also the direction [536]. Thus, rather than responding to stress by cell rearrangements, as in early embryonic stages, tissues in late stages can be expected to respond to stress by modulating tissue rigidity and differentially expressing ECM. In turtle carapaces, compression might lead to changes in the local ossification and keratinization processes, thereby affecting scute shape and size. Overall, it is likely that mechanical stresses in developing carapaces will lead to different tissue responses, depending on the stages during which they act, as well as on the duration of the stress. In the current model of scute development, none of the discussed cell and tissue responses is implemented, due to a strong simplification of cells and tissues.

69 Compressed tissue in the model either returns to its original state through relaxation, or remains deformed permanently, and the diﬀerence between short and long exposures to compression mainly lies in the stages of signaling that are aﬀected. In order to understand how mechanical stresses actually interfere with development, we need to implement more realistic biomechanics and growth into the model. My guess would be that mainly the impact of tissue reorganization by intercalation and growth polarization, together with increasing or decreasing growth rates, might explain how persistent mechanical stresses to the developing carapacial tissue will give rise to common scutal anomalies.

5.9 Turtle anomalies and climate change As extreme environmental conditions have been found to produce more anomalies in sea turtles, we would expect that climate change will have a considerable impact on the frequency of scute anomalies in the future [348]. In fact, we should expect an increase. Climate has seen a number of dramatic changes during the geological period in which turtles have existed and our study would suggest that mid- to late mesozoic and early cenozoic eras, during which average temperatures were substantially above the recent ones might have produced more scutal variation than the Triassic or in the late Cenozoic with lower average temperatures. However, in order to test this hypothesis, more and more complete turtle fossils have to be unearthened. Another reason why the frequency of anomalies might, in some cases, increase, are bottleneck effects in small populations. Since many turtle species are nowadays endangered or live in more fragmented habitats, effectively reducing population size, bottleneck effects might confound climatic effects, by increasing the frequency of anomalies. Despite this potential caveat, scute anomaly frequencies in widespread species could serve as an indicator or proxy for climatic changes, since scute anomalies are a “countable” type of variation and, therefore, easy to quantify.

5.10 Has the role of epigenetic factors in phenotypic variation been underrated? Much of the discussion about the origins of variation has been centered around genetic differences and how these affect the phenotype through development. Undoubtedly, only variation which has a genetic base will be passed on throughout many generations and will, thus, contribute to evolutionary change in a given lineage. Yet, not all of the visible phenotypic variation has a genetic base. In case of the turtle scute patterning, I found that the vast majority of phenotypic anomalies is not caused by genetic mutations. Epigenetic factors, such as temperature, drought, chemicals and the effect of mechanical stresses, were thought and partly shown to cause phenotypic variation. Although I hypothesized that genetic variation may contribute to making development more susceptible to epigenetic factors, the proximate explanation for the observed phenotypic variation is epigenetic. Phenotypic differences between turtle species can actually be linked to genetic variation, but the standard scute pattern seems to be so stable that only a few species actually deviate from it. Thus, taking all sorts of variation observed in turtle scutes together, it can be concluded that genetic changes only play a minor role in explaining phenotypic variation in this particular organ system. It can be asked if a high epigenetic contribution to the causation of anomalies is a feature of certain specific systems, such as the turtle scute pattern, or something more general that applies to the majority of organ systems and biological patterns. In many developing systems, epigenetic effects (other than maternal factors many of which are indispensable for development to initiate) have roles in actively switching between different developmental trajectories, which is often referred to as phenotypic plasticity, as discussed before 1.7. For instance, the amount or quality of food that is ingested during the larval stages of an insect may change size, morphology, coloration or diverse life-history traits of the adult animals

70 (the imago) [500, 501, 502, 503]. In many cases, even changes in only one environmental parameter (e.g. temperature, see 5.8) can lead to very different and unintuitive phenotypic consequences, by affecting several developmental processes at the same time, or by affecting processes, such as growth or tissue rearrangements, which affect cell-cell signaling and morphogenesis simultaneously. This is particularly true for heterothermic and oviparous animals, while the homothermic and protective conditions in the mammalian uterus shield the developing embryo from many external environmental factors. Thus, environmental, or more generally, epigenetic factors, can be considered as a source of information that is incorporated into development and provides a proxy about the current environmental conditions. This incorporation into development may render the adult animal better adapted to the particular environmental factors it is most likely to face during its life (the selective advantage of such a scenario being obvious). However, in the turtle, there is, to my knowledge and reasoning, no adaptive benefit from exhibiting scutal deformations. On the other hand, many genetic mutations do not produce beneficial variations, leading either to non-viable or very subtle variation [273]. This is particularly true for complex organisms, in whose development mutations often affect many different and intertwined developmental processes, so that beneficial changes in one aspect of the phenotype are often linked with negative changes in other aspects [537]. In the turtle carapace model, many genetic mutations in the signaling pathways involved in scute formation would prevent scute formation entirely. In addition, mutations are very likely to affect other developmental processes as well, given that many of the pathways involved in scute patterning are also involved in many other developmental processes. Thus, it could be suggested that a substantial proportion of viable phenotypical variation in phenotypes with complex development might actually be caused by environmental variation during development. Why is it then that studies of environmental causation of phenotypic variation seem to be so under-represented in biology? One answer might lie in a bias towards certain research questions. Outside human genetics (and some animal breeding studies), most studies of the genetic bases of phenotypes have usually focused on large-scale mutations, attempting to uncover the major processes underlying the development of wildtype phenotypes, or even novelties, rather than on small-scale variation within populations or between closely related species. An additional complication applies to the evo-devo research program: since our knowledge about the development of most species is marginal, it makes some sense to elucidate “normal” development as well as major differences to the development of well-studied model species first before trying to understand smaller-scale variation. On the other hand, many experimental setups explicitly remove or reduce environmental “confounding” factors, so that a lot of natural variation is usually invisible in the lab. In many cases of phenotypes found in nature, it is impossible to distinguish between genetic and epigenetic causation of phenotypic variation without further experimentation or extensive collection of data. Although it is difficult to estimate how high the relative contribution of genetic versus epigenetic factors in the production of phenotypic variation actually and in general is, current biases in research questions and experimental practices as well as theoretical considerations suggest that the role of environment in development is, at least, underrated.

5.11 Evo-devo and the value of non-model organisms Although turtles are, as I discussed before, a rather cumbersome model organism, I argue that it is beneﬁcial and even necessary to study such non-model organisms. The term “model organism” has been coined to express the assumption that general biological features can be learned by studying one, or a few, representative species. Since most of the major well-studied signaling pathways acting in animal development appear to have been conserved substantially throughout the Metazoa, and can be partly found even in unicellular organisms [538, 539, 540, 541], it has been assumed that large-scale commonalities be-

71 tween the developments of coarsely related taxa must exist. However, this is not necessarily true. Studies in non-model organisms have revealed that major features of development can differ substantially even between closely related taxa. The fact that development can change in evolution without considerable concomitant changes in the phenotype (or the general bauplan) of a given organism has been named “developmental systems drift”, see 1.7. Different developmental processes may produce very similar phenotypes, although their patterns of variation might largely differ. For instance, two species, closely related or not, might look very similar, but randomly occurring mutations might, in subsequent generations, lead to very divergent phenotypes in one species, while the other species only displays small, gradual variation if exposed to the same amount of mutations. The causes for such divergent variational properties must reside in differences in the mechanisms of their developments. Thus, understanding the development of a given model species does not always allow safe inferences about the evolution of development in closely related species. Research trying to understand developmental differences between related taxa is at the core of evo-devo (examples:[542, 543, 544, 545]). Adding even more problems to our understanding of nature, the criteria according to which model species have been selected in the past may have introduced substantial biases in the current and past views about the processes involved in development. For instance, classic model organisms were chosen to be prolific, small and fast developers, which implies that their development must be able to produce phenotypes with these specific life-history traits, imposing considerable constrains on its structure. [546, 547] This means that lessons learned about development from classic model organisms might not always be representative of a larger set of taxa. On the contrary, their selected life-history traits may have imposed a strong selective bias on the sort of mechanisms acting in their developments, suggesting that they are more likely to be “freaks” than a randomly picked species that can be somewhat representative of a larger clade. One example is Drosophila, whose development shows many features (patterning starts before cellularization, hierarchical segmentation etc.) that can be interpreted as facilitating a highly accelerated ontogenesis. In fact, developmental studies in different insect taxa showed that Drosophila’s mechanism of segmentation is not representative of the majority of arthropod taxa at all [542]. Since the currently popular model organisms are mostly chosen to be fast developers, comparing their developments to the development of animals with the opposite features, such as the consistently slow turtles, can be very interesting. Such a comparative strategy also permits us to assess how representative or general our results actually are. In addition, a substantial amount of data that can help us understand morphological transitions in evolution, comes from the paleontological record [548, 549, 382]. Although we cannot directly perform developmental experiments on extinct taxa, a comparison with developmental processes involved in the closest extant relatives as well as studies of their variation and mathematical models, may lead to a reconstruction of developmental changes in the evolutionary past [196, 382]. How well such an endeavour can succeed depends crucially on the knowledge about development in extant relative species. The value of departing from classic model organisms also depends on the particular research questions. While model organisms can indeed help answering many different questions about cell biology or biochemistry, as the basic toolkit of cellular processes is indeed relatively conserved amongst metazoa, many morphological structures are often restricted to specific lineages. One example is the turtle carapace which is restricted to its clade (and a few other paleozoic taxa). But even more common morphological features, such as segments, may be produced in various different manners, so that insight from one model species does not imply more than a sophisticated guess about the development of an analogous structure in another. Thus, in order to study the developmental - and evolutionary - origins of morphological structures, and especially of novelties, one cannot help studying non-model organisms. Also, if we want to understand how exactly devel-

72 opmental processes change over evolutionary time, being able to compare developments of species with diﬀerent degrees of phylogenetic relationship, and, if possible, very diﬀerent life-history traits, is crucial. Thus, studying non-model organisms enriches our understanding of the large-scale evolutionary transitions and endows us with a more complete picture of the strategies nature uses in order to build and transform complex morphologies.

5.12 A dialogue of experimental and mathematical approaches in evo-devo Although the importance of mathematical approaches in biology was already emphasized by D’Arcy Thompson one century ago [1], it is only by means of the current advances in computational technology that modelling approaches have become a common tool in biology. This thesis presents an example of what a successful dialogue between “wet lab” experiment and mathematical model can achieve. Both approaches together are important to advance understanding of a given biological process. A model requires real data to assess if the hypothesis implemented in it is correct or not. On the other hand, models can direct what sort of experiment to conduct or where to search for possible answers to a biological question, which is especially pivotal if the organism is not easily amenable to experimentation, such as the turtle. Models can make sense of unintuitive phenomena, such as the establishment of a regular pattern from inductive signals that do not convey any absolute spatial information. In order to judge if the modelled hypothesis is correct, it is, however, not sufficient to state correspondence between a real and a simulated phenotype; the model ought to be able to account for further, independent experiments, such as mutations or the reproduction of phenotypic variation as well. This has been the case for the turtle model presented in here. In turn, the model could narrow down which mechanisms are capable of producing observed patterns, thereby winnowing out a number of intuitive, but wrong hypotheses. For instance, by means of the model, we were able to rule out a genetic explanation of the observed anomalies, thus catalysing a research program which focuses more on the interaction between epigenetic factors and development. Models that implement in more detail different organizational levels such as intracellular, cellular, tissue and organ levels, such as EmbryoMaker, can further contribute to causally linking different processes, both top-down and bottom-up (which in reality are always intrinsically connected, but whose connection might give rise to complex and non-intuitive phenomena). Thus, such models will strengthen a holistic systems perspective on biological phenomena, rather than applying a purely reductionist paradigm. Another way in which mathematical models facilitate research is that they allow for systematic parameter screens (e.g. testing of all possible mutations of all possible magnitudes), which would not be possible in nature or in the lab. In case of the turtle, the model helped to systematically assess the effect of carapace length variation. Since the length – as the shape - of the carapace is determined by development itself, it would not be possible to restrict its size experimentally without affecting other developmental processes, even if we had enough turtle embryos to work on. Thus, models are a valuable tool especially for evo-devo, in which understanding of development in non-model organisms with intrinsic limitations to the scope of experiments that can be conducted in them, is key. Furthermore, mathematical models allow exploration of hypotheses on palaeontological data, e.g. addressing the question of which changes in development might explain phenotypic differences between extant and mesozoic turtle species. Of course, in this case, we could never validate model predictions with experimental results, but at least get some evidence of which hypotheses are more and which are less likely. Therefore, it is important to emphasize the role of mathematical models as a complementary and essential tool for evo-devo research - rather than a distinct and abstract discipline -, that can facilitate a widen- ing of the questions that can be approached in evo-devo, as well as other biological disciplines, and particularly for the understanding of both unintuitive and long-term processes as evolutionary change.

73 6 Concluding remarks

Although there are various attempts to define what a novelty is, most biologists will agree that the turtle shell and its peculiar scutation pattern can be considered a classic example of it, independently of the definition used. Understanding how novelties arise in evolution, however, always requires an understanding of the developmental processes and what particular evolutionary changes in those processes have or may have led to the production of the novel phenotype. In my thesis, I have contributed to this understanding, in the particular case of the turtle shell scutation, in two different ways: 1. I made a model of turtle scute pattern development that is likely to represent the main developmental mechanisms acting in real turtle scute development. 2. I elucidated probable mechanisms that can account for phenotypic variation in the scute pattern. The latter point is of evolutionary importance, since knowledge about the origins of variation ultimately permits to make predictions about possible directions of phenotypic changes in evolution. Intriguingly, the developmental processes acting in scute pattern formation share many commonalities with the developmental processes acting in the development of other ectodermal organs, such as hair, dentition, and, in particular, scales and plumage: Multiple mutual interactions of ectodermal and mesenchymal tissue compartments, orchestrated by a conserved toolkit of signaling pathways, and a nested sequence of two reaction-diffusion systems. Of these, the first is induced by signals from the underlying somitic mesenchyme and produces a regularly spaced prepattern of placodal primordia, the induction of mesenchymal condensations, epithelial modifications (involving growth, changes in cellular geometries and foldings), and finally the secretion of specific types of keratines (and eventually several types of specialized ECM). Yet, there are also important differences, particularly concerning the second reaction-diffusion system which orchestrates morphogenesis of the individual appendage units, rather than their relative positions to one another. This suggests that in organs or structures which develop by a clear temporal sequence of developmental events (i.e. with strongly hierarchically structured developments), evolutionary changes might mostly affect developmentally late processes or the properties of the prepattern that initiates the particular organ development, while maintaining intermediate developmental processes. This is reminiscent of the controversial hourglass of development hypothesis [550, 551, 552, 553, 554, 555, 556, 298], according to which intermediate developmental stages tend to be more conserved between different sister clades than early or late developmental stages. Another intriguing finding from the comparison of different ectodermal appendage developments is that, while the recruitment of the interacting molecular pathways seems to be highly conserved, the roles of the particular signaling molecules in the patterning process appear not to be so. In other words, both the logic of the patterning process and the underlying molecular pathways are conserved, but not the specific way the latter interact in order to give rise to the former. Viewed from an evo-devo perspective, this indicates that developmental drift might change development by rearranging interacting components of a specific molecular and morphogenetic toolkit which generate a specific morphogenetic process, while maintaining the developmental function of the entire process. Alternatively, certain morphogenetic processes, such as interacting morphogenes giving rise to Turing patterns, might have evolved in different ectodermal organs in parallel, as periodic pattern formation by reaction-diffusion dynamics may actually be relatively easy to evolve (at least compared to alternative patterning mechanisms). Whether these particular patterns of developmental change are common in evolution and, in particular, in the emergence of novelties, requires comparisons of development within other morphologically diverse organ systems (e.g. vertebrate and arthropod limbs, sensory organs etc.). Second, I have evaluated existing hypotheses about the origination of scute anomalies in the light of current knowledge of turtle scute development as well as of the results presented in this thesis. The study of variation in scute patterns suggests that epigenetic factors sensu lato, not genetic mutations, can explain anomalous scute variation within populations. This might be surprising,

74 given a perceived focus on genetic causes of trait variation in biological studies. However, since epigenetic factors interact with development in many different, partly indispensable, ways, and since high phenotypic plasticity in the face of environmental perturbations can sometimes be beneficial for the individual and selective in evolution, I have argued that environment is likely to be a rather underestimated source of variation in extant populations of complex organisms. Importantly, the particular developmental processes acting in scute pattern development were able to explain why certain anomalies are observed more frequently than others. Understanding developmental mechanisms underlying particular morphologies can, thus, help identifying causes of phenotypic variation and predicting their specific patterns of variation. Interestingly, while the predominantly environmentally caused scute pattern variation within species can be substantial, turtle scute variation between species is very low. The main exceptions are softshell turtles, which have lost scutes altogether, and some marine turtle species, which also display higher degrees of variation between individuals than turtle species with the standard scute pattern. This means that the development of the standard carapacial scute pattern in turtles is overall highly stable against mutations, possibly reflecting either long evolutionary periods of stabilizing selection or a high developmental stability for the particular combination of developmental mechanisms (i.e. there are different phenotypic stabilities in different areas of the generic morphospace of carapacial patterns, suggesting that the standard scute pattern might be a major developmental attractor [293, 404, 557, 558] and most developmental changes would not change this phenotype, while other scute patterns are more prone to suffer phenotypic changes when development changes). The latter interpretation is in line with observations in the morphospace of in silico scute patterns, in which the characteristic scute patterns of sea turtles only emerged as rather disparate morphological “islands”, although further experiments and more realistic models are needed to consolidate this hypothesis. Why complex phenotypes are usually stable enough to be faithfully produced by development despite mutations and developmental drift is a general question that has been traditionally answered in terms of conservative selection forces. However, theoretical work has cast doubts on the ability of selection to act with sufficient strength on all parts or traits in a complex phenotype [537]. Work on “developmental constraints” in specific morphological structures [7, 2, 6] has accumulated evidence that development itself defines the stability of the phenotype (in other words, the amount of phenotypic variation) it produces, both against genetic mutations and environmental perturbations. In the case of the turtle scute pattern, development “constrains” the possible phenotypes in several ways: The onset of turtle scute development is relatively independent of the precise inductive pattern, because reaction-diffusion dynamics unfold as soon as a symmetry break occurs. The first reaction-diffusion system defines how many scutes will form and only substantial changes in the diffusion space will change this number. The second reaction-diffusion system will only act after the places of the scutes have been defined. These “constraints” ensure that most developmental changes (due to genetic mutations or epigenetic changes) will only produce a limited set of phenotypes, unless the developmental modifications are massive or other developmental mechanisms are added. Many genetic and environmental perturbations will, thus, not be able to disturb development enough to produce a substantially deviant phenotype. Yet, very strong disturbances may severely affect the scute pattern or prevent it from forming at all, creating an everything-or-nothing response to developmental perturbances. This might be a way of looking at softshell turtles, in which some developmental modification prevents scute formation altogether. Since many different modifications would lead to the same consequence, modelling approaches are not sufficient to elucidate the mechanistic cause of the loss of scutes in soft-shell turtles, unless supplemented by experimental data. Reaction-diffusion dynamics are common in animal development and there are other developmental processes (also discussed in 5.7) that are resilient against non-extreme perturbations, suggesting that everything-or-nothing stress responses might be more common in the

75 development of complex phenotypic traits. This is in line with the finding that most mutations are likely to be either neutral or detrimental [273]. Yet, there are further aspects of variation between scute shapes within the carapace as well as between different species that cannot be explained by the turtle model presented in this dissertation, partly due to limitations of knowledge about the development of different turtle species, partly due to the limitations that are intrinsic to every mathematical model of biological processes. Therefore, I have proposed a general model of animal development that unifies dynamics of gene regulatory networks, cell-cell signaling, cell and tissue biomechanics and cell behaviours. Applying this model framework to the development of turtle carapaces, it would be possible to analyze the interplay of realistic tissue growth and biomechanics in the generation of scute shapes. Furthermore, it would allow testing different hypotheses of how environmental stresses might, through their particular effects on cells and tissues during development, translate into phenotypic deformities. Although my studies have provided evidence that both high temperatures and mechanical pressures might be among the most common causes of scute anomalies, we are still lacking the precise mechanistic understanding of those. Advancing this understanding, in the case of the turtle shell as well as in general, will be key to a richer and more inclusive model of development that would integrate environment and other epigenetic factors. Thus, we could use the different types of scute anomalies as a model system to understand the interaction mechanisms between ecology and development. Since our unified model of development was built not to serve only the purpose of simulating the development of one specific system in detail, results gained from its simulations aim at general conclusions that can be applied to most animal systems. Thus, the studies included in this dissertation aim at exemplifying how an integration of theoretical and experimental approaches does not only help elucidating development in a non-model organism, I also wish to propose how this knowledge can contribute to a more general understanding of the evolution of novelties, the generation of variation through development, and the interaction of development and environment. Future studies in other non-model systems, as well as more general modelling approaches to development, will contribute further insight that may allow assessment of the generality of the conclusions drawn from the “hopeful monsters” of Testudines, on the path towards a more integrated and powerful theory of eco-evo-devo.

76 7 Bibliography References [1] Thompson, D. W. On Growth And Form (Cambridge University Press, 1942).

[2] Alberch, P. Ontogenesis and morphological diversiﬁcation. Amer.Zool. (1980).

[3] Ackermann, R. R. Ontogenetic integration of the hominoid face. Journal of Human Evolution (2005).

[4] Mart´ınez-Abad´ıas,N. et al. The developmental basis of quantitative craniofacial variation in humans and mice. Evol. Dev. (2012).

[5] Shaﬀer, J. R. et al. Genom-wide association study reverals multiple loci inﬂuencing normal human facial morphology. PLOS Genetics (2016).

[6] Alberch, P. Developmental constraints in evolutionary processes. Evolution and Development (1982).

[7] Smith, J. M. et al. Developmental constraints and evolution: A perspective from the mountain lake conference on development and evolution. The Quarterly Review of Biology (1985).

[8] Salazar-Ciudad, I. & Jernvall, J. Graduality and innovation in the evolution of complex phenotypes: insights from development. Journal of Experimental Zoology Part B: Molecular and Developmental Evolution 304B, 619–631 (2005).

[9] Pigliucci, M. What, if anything, is an evolutionary novelty? Philosophy of Science 75, 887–898 (2008).

[10] Moczek, A. P. On the origins of novelty in development and evolution. BioEssays (2008).

[11] Mayr, E. Animal Species and Evolution (Harvard University Press, Cambridge, MA, 1962).

[12] Mueller, G. & Wagner, G. Novelty in evolution: Restructuring the concept. Annual Reviews of Ecology and Systematics (1991).

[13] Shubin, N., Tabin, C. & Carroll, S. Deep homology and the origin of evolutionary novelty. Nature (2009).

[14] Mueller, G. Developmental mechanisms at the origin of morphological novelty: A side-eﬀect hypothesis. (Univ. Chicago Press, Chicago, 1990).

[15] M¨uller, G. B. Epigenetic Innovation. In: Evolution - The Extended Synthesis (MIT Press, 2010).

[16] Hallgr´ımsson,B. et al. The generation of variation and the developmental basis for evolutionary novelty. Journal of Experimental Zoology (2012).

[17] Alberch, P. The logic of monsters: Evidence for internal constraint in development and evolution. Geobios (1989).

[18] Peterson, T. & M¨uller,G. B. What is evolutionary novelty? process versus character based deﬁnitions. Journal of Experimental Zoology (2013).

[19] Peterson, T. & M¨uller,G. B. Phenotypic novelty in evodevo: The distinction between continuous and discontinuous variation and its importance in evolutionary theory. Evol.Biol. (2016).

[20] M¨uller, G. B. & Newman, S. A. The innovation triad: An evodevo agenda. Journal of Experimental Zoology (2005).

[21] Brigandt, I. & Love, A. C. Conceptualizing evolutionary novelty: Moving beyond deﬁnitional debates. Journal of Experimental Zoology Part B: Molecular and Developmental Evolution (2012).

[22] Love, A. C. Explaining evolutionary innovation and novelty: A historical and philosophical study of biological concepts. Dissertation (2005).

[23] Jackson, D. C. How a turtle’s shell helps it survive prolonged anoxic acidosis. News Physiol. Sci. (2000).

[24] Alexander, N. J. Comparison of a and b keratin in reptiles. Zeitschrift f¨urZellforschung und Mikroskopische Anatomie (1970).

[25] Wu, P. et al. Evo-devo of amniote integuments and appendages. International Journal of Developmental Biology (2004).

77 [26] Li, Y. I., Kong, L., Ponting, C. P. & Hearty, W. Rapid evolution of beta-keratin genes contribute to phenotypic diﬀerences that distinguish turtles and birds from other reptiles. Genome Biol. Evol. (2013).

[27] Zangerl, R. The homology of the shell elements in turtles. Journal of Morphology (1939).

[28] Hirasawa, T., Nagashima, H. & Kuratani, S. The endoskeletal origin of the turtle carapace. Nature Communi- cations (2013).

[29] Kuraku, S., Usuda, R. & Kuratani, S. Comprehensive survey of carapacial ridge-speciﬁc genes in turtle implies co-option of some regulatory genes in carapace evolution. Evolution and Development (2005).

[30] Chiari, Y., van der Meijden, A., Caccone, A., Claude, J. & Gilles, B. Self-righting potential and the evolution of shell shape in galapagos tortoises. Scientiﬁc Reports (2017).

[31] Rivera, G. Ecomorphological variation in shell shape of the freshwater turtle pseudemys concinna inhabiting diﬀerent aquatic ﬂow regimes. Integrative and Comparative Biology (2008).

[32] Claude, J., Paradis, E., Tong, H. & Auﬀray, J.-C. A geometric morphometric assessment of the eﬀects of environment and cladogenesis on the evolution of the turtle shell. Biological Journal of the Linnean Society (2003).

[33] Benson, R. B., Domokos, G., V´arkonyi, P. L. & Reisz, R. R. Shell geometry and habitat determination in extinct and extant turtles (reptilia: Testudinata). Paleobiology (2011).

[34] Alibardi, T. M. B., Lorenzo. Ultrastructural and immunohistochemical observations on the process of horny growth in chelonian shell. Acta Histochemica (2006).

[35] Alibardi, L. Observations on the ultrastructure and distribution of chromatophores in the skin of chelonians. Acta Zoologica (2013).

[36] Zangerl, R. & Johnson, R. The nature of shield abnormalities in the turtle shell. Fieldiana (1957).

[37] Pritchard, P. C. in: Biology of Turtles: From Structures to Strategies of Life 3. Evolution and structure of the turtle shell (CRC Press, 2008).

[38] Wynecken, J. & Witherington, D. The anatomy of sea turtles (U.S. Department of Commerce, National Marine Fisheries Service, 2001).

[39] T¨urkozan, O., Ilgaz, C. & Sak, S. Carapacial scute variation in loggerhead turtles, caretta caretta. Zoology in the Middle East (2001).

[40] Mast, R. B. & Carr, J. L. Carapacial scute variation in kemp’s ridley sea turtle (lepidochelys kempi) hatchlings and juveniles. Conservation and Management (1989).

[41] Pritchard, P. Studies of the systematics and reproductive cycle of the genus lepidochelys. Ph.D. Dissertation, University of Florida, Gainesville (1969).

[42] Newman, H. The signiﬁcance of scute and plate ”abnormalities” in chelonia. a contribution to the evolutionary history of the chelonian carapace and plastron. The Biological Bulletin (1906).

[43] Thomson, R. C. & Shaﬀer, B. H. Sparse supermatrices for phylogenetic inference: Taxonomy, alignment, rogue taxa, and the phylogeny of living turtles. Systematic Biology (2010).

[44] Peters, D. H. Odontochelys: not a turtle. not even a turtle ancestor. close, but no cigar. beforeitsnews,com (2013).

[45] Li, C., Wu, X.-C., Rieppel, O., Wang, L.-T. & Zhao, L.-J. An ancestral turtle from the late triassic of southwestern china. Nature (2008).

[46] Burke, A. Turtles...again. Evolution and Development (2009).

[47] Nagashima, H. et al. Evolution of the turtle body plan by the folding and creation of new muscle connections. Science (2009).

[48] Reisz, R. R. & Head, J. J. Turtle origins out to sea. Nature (2008).

[49] Schoch, R. R. & Sues, H.-D. A middle triassic stem-turtle and the evolution of the turtle body plan. Nature (2015).

78 [50] Lyson, T. R., Bever, G. S., Bhullar, B.-A. S., Joyce, W. G. & Gauthier, J. A. Transitional fossils and the origin of turtles. Biol. Lett. (2010).

[51] Joyce, W. G., Lucas, S. G., Scheyer, T. M., Heckert, A. B. & Hunt, A. P. A thin-shelled reptile from the late triassic of north america and the origin of the turtle shell. Proceedings B of the Royal Society (2009).

[52] Rougier, G. W., de la Fuente, M. S. & Arcucci, A. B. Late triassic turtles from south america. Science (1995).

[53] Wagner, A. Beschreibung einer fossilen schildkroete und etlicher anderer reptilien-ueberreste aus den lithographischen schiefern und dem gruenen sandsteine von kehlheim. Abhandlungen der Bayerischen Akademie der Wissenschaften, Mathematisch-physikalische Klasse (1853).

[54] Cherepanov, G. O. Ontogenesis and evolution of horny parts of the turtle shell. Fossil Turtle Research (2006).

[55] Szcsygielski, T. & Sulej, T. Revision of the triassic european turtles proterochersis and murrhardtia (reptilia, testudinata, proterochersidae), with the description of new taxa from poland and germany. Zoological Journal (2016).

[56] Gaﬀney, E. S., Hutchinson, J., Jenkins, F. A. & Meeker, L. J. Modern turtle origins: The oldest known cryptodire. Science (1987).

[57] Gaﬀney, E. S. & Meyen, P. A. The transylvanian turtle, kallokibotion, a primitive cryptodire of cretaceous age. American Museum Novitates (1992).

[58] Coker, R. E. Diversity in the scutes of chelonia. Journal of morphology (1910).

[59] Cherepanov, G. Patterns of scute development in turtle shell: Symmetry and asymmetry. Paleontological Journal (2014).

[60] Kobayashi, S. et al. Sex diﬀerences and the heritability of scute pattern abnormalities in the green sea turtle from the ogasawara archipelago, japan. Zoological Science (2017).

[61] Sim, E. L., Booth, D. T. & Limpus, C. J. Non-modal scute patterns, morphology, and locomotor performance of loggerhead (caretta caretta) and ﬂatback (natator depressus) turtle hatchlings. Copeia (2014).

[62] Ewert, M. The embryo and its egg: development and natural history. In: Turtles: Perspective and Research. (John Wiley and Sons, New York, 1979).

[63] Hill, R. Surinam turtle notes - 1.polymorphism of costal and vertebral laminae in the sea turtle lepidochelys olivacea. Dtichting Natuurbehoud Suriname (1971).

[64] Rieppel, O. & Reisz, R. R. The origin and early evolution of turtles. Annual Review of Ecology and Systematics (1999).

[65] Joyce, W. G. & Gauthier, J. A. Palaeoecology of triassic stem turtles sheds new light on turtle origins. Proc. R. Soc. Lond. B (2004).

[66] Gilbert, S. F., Bender, G., Betters, E., Yin, M. & Cebra-Thomas, J. The contribution of neural crest cells to the nuchal bone and plastron of the turtle shell. Integrative and Comparative Biology (2007).

[67] Cebra-Thomas, J. et al. Evidence that a late-emerging population of trunk neural crest cells forms the plastron bones in the turtle trachemys scripta. Evolution and Development (2007).

[68] Rice, R., Kallonen, A., Cebra-Thomas, J. & Gilbert, S. F. Development of the turtle plastron, the order-deﬁning skeletal structure. Proceedings of the National Academy of Science (2016).

[69] Alibardi, L. & Thompson, M. B. Epidermal diﬀerentiation during carapace and plastron formation in the embryonic turtle emydura macquarii. Journal of Anatomy (1999).

[70] Burke, A. C. Development of the turtle carapace: Implications for the evolution of a novel bauplan. Journal of Morphology (1989).

[71] Kuratani, S., Kuraku, S. & Nagashima, H. Evolutionary developmental perspective for the origin of turtles: the folding theory for the shell based on the developmental nature of the carapacial ridge. Evolution and Development (2011).

[72] Nagashima, H. et al. On the carapacial ridge in turtle embryos: its developmental origin, function and the chelonian body plan. Development (2007).

79 [73] Ruckes, H. The morphological relationships between the girdels, ribs and carapace. Ann. N.Y. Acad. Sci. (1929).

[74] Pascual-Anaya, J., Hirasawa, T., Sato, I., Kuraku, S. & Kuratani, S. Comparative analysis of pleurodiran and cryptodiran turtle embryos depicts the molecular ground pattern of the turtle carapacial ridge. International Journal of Developmental Biology (2014).

[75] Cebra-Thomas, J. et al. How the turtle forms its shell: A paracrine hypothesis of carapace formation. Journal of experimental zoology (2005).

[76] LoredoKassai, Y. et al. Development of an evolutionary novel structure: Fibroblast growth factor expression in the carapacial ridge of turtle embryos. Journal of Experimental Zoology (2001).

[77] Burke, A. The development and evolution of the turtle body plan: inferring intrinsic aspects of the evolutionary process from experimental embryology. American Zoologist (1991).

[78] Park, W. Y., Miranda, B., Lebeche, D., Hashimoto, G. & Cardoso, W. V. Fgf-10 is a chemotactic factor for distal epithelial buds during lung development. Developmental Biology (1998).

[79] Li, S. & Muneoka, K. Cell migration and chick limb development: Chemotactic action of fgf-4 and the aer. Developmental Biology (1999).

[80] Salazar, V. S., Gamer, L. W. & Rosen, V. Bmp signalling in skeletal development, disease and repair. Nature Reviews (2016).

[81] Rosen, V. & Thies, R. The bmp proteins in bone formation and repair. Trends in Genetics (1992).

[82] Franz-Odendaal, T. A. Induction and patterning of intramembraneous bone. Frontiers in Bioscience (2011).

[83] Moustakas, J. E. Development of the carapacial ridge: implications for the evolution of genetic networks in turtle shell development. Evolution and Development (2008).

[84] Vincent, C., Bontoux, M., Le Douarin, N. M., Pieau, C. & Monsoro-Burq, A.-H. Msx genes are expressed in the carapacial ridge of turtle shell: a study of the european pond turtle, emys orbicularis. Dev. Genes Evol. (2003).

[85] Gilbert, S. F., Loredo, G. A., Brukman, A. & Burke, A. C. Morphogenesis of the turtle shell: the development of a novel structure in tetrapod evolution. Evolution and Development (2001).

[86] Cherepanov, G. O. Scute’s polymorphism as a source of evolutionary development of the turtle shell. Paleon- tological Journal (2015).

[87] Moustakas-Verho, J., Cebra-Thomas, J. & Gilbert, S. F. Patterning of the turtle shell. Current Opinion in Genetics & Development (2016).

[88] Werneburg, I., Hugi, J., Müller,J. & Sánchez-Villagra, M. R. Embryogenesis and ossification of emydura subglobosa (testudines, pleurodira, chelidae) and patterns of turtle development. Developmental Dynamics (2009).

[89] Alibardi, L. & L., D. Diﬀerentiation of epidermis of scutes in embryos and juveniles of the tortoise testudo hermanni with emphasis on beta-keratinization. Acta Zoologica (2005).

[90] Procter, J. B. A study of the remarkable tortoise, testudo loveridgii blgr., and the morphogeny of the chelonian carapace. Proceedings of the General Meetings for Scientiﬁc Business of the Zoological Society of London (1922).

[91] Mautner, A.-K., Latimer, A. E., Fritz, U. & Scheyer, T. M. An updated description of the osteology of the pancake tortoise malacochersus tornieri (testudines: Testudinidae) with special focus on intraspeciﬁc variation. Journal of Morphology (2017).

[92] Siebenrock, F. Ueber zwei seltene und eine neue schildkroete des berliner museums. Sitzungsber. Akad. Wis- sensch. Wien, Math-Naturw Cl 113 (1903).

[93] Nagashima, H. et al. Comparative study of the shell development of hard and softshelled turtles. Journal of Anatomy (2014).

[94] Sanchez-Villagra, M. R. et al. Skeletal development in the chinese soft-shelled turtle pelodiscus sinensis (testudines: Trionychidae). Journal of Morphology (2009).

[95] Tokita, M. & Kuratani, S. Normal embryonic stages of the chinese softshell turtle pelodiscus sinensis (trionychidae). Zoological Science (2001).

80 [96] Suzuki, H. K. Studies on the osseous system of the slider turtle. Annals of the New York Academy of Sciences (1963).

[97] Gilbert, S. F., Cebra-Thomas, J. A. & Burke, A. C. How the turtle gets its shell.

[98] Wang, Q. & Zhao, X. A three-dimensional phase diagram of growth-induced surface instabilities. Plos One (2014).

[99] Marin-Riera, M., Moustakas-Verho, J., Savriama, Y., Jernvall, J. & Salazar-Ciudad, I. Diﬀerential tissue growth and cell adhesion alone drive early tooth morphogenesis: An ex vivo and in silico study. PLOS Computational Biology (2018).

[100] Li, B., Cao, Y.-P., Feng, X.-Q. & Gao, H. Mechanics of morphological instabilities and surface wrinkling in soft materials: a review. Soft Matter (2012).

[101] Shyer, A. E. et al. Villiﬁcation: How the gut gets its villi. Science (2013).

[102] Tallinen, T., Chung, J. Y., Biggins, J. S. & Mahadevan, L. Gyriﬁcation from constrained cortical expansion. Proceedings of the National Academy of Sciences (2014).

[103] Odell, G., Oster, G., Alberch, P. & Burnside, B. The mechanical basis of morphogenesis: I. epithelial folding and invagination. Developental Biology (1981).

[104] Brezavscek, A. H., Rauzi, M., Leptin, M. & Ziherl, P. A model of epithelial invagination driven by collective mechanics of identical cells. Biophysical Journal (2012).

[105] Monier, B. et al. Apico-basal forces exerted by apoptotic cells drive epithelium folding. Nature (2016).

[106] Collier, J. R., Monk, N. A., Maini, P. K. & Lewis, J. H. Pattern formation by lateral inhibition with feedback: a mathematical model of delta-notch intercellular signalling. Journal of Theoretical Biology (1996).

[107] de Back, W., Zimm, R. & Brusch, L. Transdiﬀerentiation of pancreatic cells by loss of contact-mediated signaling. BMC Systems Biology (2013).

[108] Cabrera, C. V. Lateral inhibition and cell fate during neurogenesis in drosophila: the interaction between scute, notch and delta. Development (1990).

[109] Jensen, J. et al. Control of endodermal endocrine development by hes-1. Nature (2000).

[110] Iwaki, D. & Lengyel, J. A. A delta-notch signaling border regulated by engrailed/invected repression speciﬁes boundary cells in the drosophila hindgut. Mechanisms of Development (2002).

[111] Parks, A. L. & Muskavitch, M. A. Delta function is required for bristle organ determination and morphogenesis in drosophila. Developmental Biology (1993).

[112] Zeng, W., Thomas, G. L. & Glazier, J. A. Non-turing stripes and spots: a novel mechanism for biological cell clustering. Physica A (2004).

[113] Cohen, S. M. Developmental biology: Long-range signalling by touch. Nature (2003).

[114] Salazar-Ciudad, I., Newamn, S. & Sol´e,R. Phenotypic and dynamical transitions in model genetic networks i. emergence of patterns and genotype-phenotype relationships. Evolution and Development (2001).

[115] Gierer, A. & Meinhardt, A. A theory of biological pattern formation. Kybernetik (1972).

[116] Kondo, S. The reaction-diﬀusion system: a mechanism for autonomous pattern formation in the animal skin. Genes to Cells (2002).

[117] Cooke, J. & Zeeman, E. A clock and wavefront model for control of the number of repeated structures during animal morphogenesis. Journal of theoretical Biology (1976).

[118] Palmeirim, I., Henrique, D., Ish-Horowicz, D. & Pourqui´e,O. Avian hairy gene expression identiﬁes a molecular clock linked to vertebrate segmentation and somitogenesis. Cell (1997).

[119] Peel, A. D., Chipman, A. D. & Akam, M. Arthropod segmentation: beyond the drosophila paradigm. Nature Review Genetics (2005).

[120] Salazar-Ciudad, I., Garcia-Fernández, J. & Solé,R. Gene networks capable of pattern formation: from induction to reaction-diffusion. Journal of theoretical Biology (2000).

81 [121] ten Tuscher, K. Mechanisms and constraints shaping the evolution of body plan segmentation. Eur. Phys. J. E. (2013).

[122] Rivera, A. S. & Weisblat, D. A. And lophotrochozoa makes three: Notch/hes signaling in annelid segmentation. Dev. Genes Evol. (2009).

[123] Salazar-Ciudad, I., Sol´e,R. & Newman, S. Phenotypic and dynamical transitions in model genetic networks ii. application to the evolution of segmentation mechanisms. Evolution and Development (2001).

[124] Cotterell, J. & Sharpe, J. An atlas of gene regulatory networks reaveals multiple three-gene mechanisms for interpreting morphogen gradients. Molecular systems Biology (2010).

[125] Perez-Carrasco, R. et al. The power of the ac-dc circuit. arXiv (2017).

[126] Headon, D. J. & Painter, K. Stippling the skin: Generation of anatomical periodicity by reaction-diﬀusion mechanisms. Math.Model.Nat.Phenom. (2009).

[127] Painter, K., Hunt, G., Wells, K., Johansson, A. & Headon, D. Towards an integrated experimental-theoretical approach for assessing the mechanistic basis of hair and feather morphogenesis. Interface Focus (2016).

[128] Turing, A. The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences (1952).

[129] Edelstein-Keshet, L. Models in Biology (Society for Industrial and Applied Mathematics, Philadelphia., 2005).

[130] Kondo, S. & Miura, T. Reaction-diﬀusion model as a framework for understanding biological pattern formation. Science (2010).

[131] Meinhardt, H. Model of Biological Pattern Formation (Academic Press, London, 1982).

[132] Koch, A. & Meinhardt, H. Biological pattern formation: from basic mechanisms to complex structures. Reviews of Modern Physics (1994).

[133] Asai, R., Taguchi, E., Kume, Y., Saity, M. & Kondo, S. Zebraﬁsh leopard gene as a component of the putative reaction-diﬀusion system. Mechanisms of Development (1999).

[134] Yang, L., Dolnik, M., Zhabotinski, A. M. & Epstein, I. R. Pattern formation arising from interactions between turing and wave instabilities. Journal of Chemical Physics (2002).

[135] Maini, P. K. Bones, feathers, teeth and coat markings: a uniﬁed model. Science Progress (1997).

[136] Page, K. M., Maini, P. K. & Monk, N. A. Complex pattern formation in reaction-diﬀusion systems with spatically varying parameters. Physica D (2005).

[137] Economou, A. D. et al. Periodic stripe formation by a turing-mechanism operating at growth zones in the mammalian palate. Nature Genetics (2012).

[138] Kondo, S. & Asai, R. A reaction-diﬀusion wave on the skin of the marine angelﬁsh pomacanthus. Nature (1995).

[139] Varea, C., Arag´on,J. & Barrio, R. Conﬁned turing patterns in growing systems. Physical Review E (1997).

[140] Duverger, O. & Morasso, M. I. Epidermal patterning and induction of diﬀerent hair types during mouse embryonic development. Birth Defects Res. C Embryo Today (2009).

[141] Sick, S., Reinker, S., Timmer, J. & Schlake, T. Wnt and dkk determine hair follicle spacing through a reaction- diﬀusion mechanism. Science (2006).

[142] Nihout, H. Symmetry systems and compartments in lepidopteran wings: the evolution of a patterning mechanism. Development (1994).

[143] Martin, A. et al. Diversiﬁcation of complex butterﬂy wing patterns by repeated regulatory evolution of a wnt ligand. Proceedings of the National Academy of Science (2012).

[144] Sheth, R. et al. Hox genes regulate digit patterning by controlling the wavelength of a turing-type mechanism. Science (2012).

[145] Sadier, A. et al. Modeling edar expression reveals the hidden dynamics of tooth signaling center patterning. PLOS Biology (2019).

82 [146] Kurics, T., Menshykau, D. & Iber, D. Feedbacks, receptor clustering, and receptor restriction to single cells yield large turing spaces for ligand-receptor based turing models. Physical Review E (2014).

[147] Iber, D. & Menshykau, D. The control of branching morphogenesis. Open Biol. (2013).

[148] Ishida, K. et al. The regulation of tooth morphogenesis is associated with epithelial cell proliferation and the expression of sonic hedgehog through epithelial–mesenchymal interactions. Biochemical and Biophysical Research Communications (2011).

[149] Jiang, T.-X., Jung, H.-S., Widelitz, R. B. & Chuong, C.-M. Self-organization of periodic patterns by dissociated feather mesenchymal cells and the regulation of size, number and spacing of primordia. Development (1999).

[150] Green, J. B. & Sharpe, J. Positional information and reaction-diﬀusion: two big ideas in developmental biology combine. Development (2015).

[151] Kornberg, T. B. Cytonemes and the dispersion of morphogenes. Wiley Interdisciplinary Reviews: Developmental Biology (2014).

[152] Maderspacher, F. & N¨usslein-Volhard, C. Formation of the adult pigment pattern in zebraﬁsh requires leopard and obelix dependent cell interactions. Development (2003).

[153] Inaba, M., Yamanaka, H. & Kondo, S. Pigment pattern formation by contact-dependent depolarization. Science (2012).

[154] Nakamasu, A., Takahashi, G., Kanbe, A. & Kondo, S. Interactions between zebraﬁsh pigment cells responsible for the generation of turing patterns. Proceedings of the National Academy of Science (2009).

[155] Moreira, J. & Deutsch, A. Pigment pattern formation in zebraﬁsh during late larval stages: A model based on local interactions. Developmental Dynamics (2005).

[156] Bullara, D. & De Decker, Y. Pigment cell movement is not required for generation of turing patterning in zebraﬁsh skin. Nature Communications (2015).

[157] Stanganello, E. et al. Filopodia-based wnt transport during vertebrate tissue patterning. Nature Communica- tions (2014).

[158] Alibardi, L. Adaptation to the land: The skin of reptiles in comparison to that of amphibians and endotherm amniotes. Journal of Experimental Zoology Part B: Molecular and Developmental Evolution (2003).

[159] Toledo, R. & Jared, C. Cutaneous adaptation to water balance in amphibians. Comp. Biochem. Physiol. (1993).

[160] Sawyer, R. H. et al. The expression of beta keratins in the epidermal appendages of reptiles and birds. Amer. Zool. (2000).

[161] Vandebergh, W. & Bossuyt, F. Radiation and functional diversiﬁcation of alpha keratins during early vertebrate evolution. Mol. Biol. Evol. (2012).

[162] Toni, M., Dalla Valle, L. & Alibardi, L. Hard (beta-) keratins in the epidermis of reptiles: Composition, sequence and molecular organization. Journal of Proteome Research (2007).

[163] Alibardi, L. & Sawyer, R. H. Immunocytochemical analysis of beta keratins in the epidermis of chelonians, lepidosaurians and archosaurians. Journal of Experimental Zoology (2002).

[164] Gregg, K. & Rogers, G. Feather keratin: composition, structure and biogenesis. In: Biology of the Integument. Vol 2 Vertebrates (Springer Verlag, New York, 1986).

[165] Chang, C. et al. Reptile scale paradigm: Evo-devo, pattern formation and regeneration. Int. J. Dev. Biol. (2009).

[166] Kollar, E. & Fisher, C. Tooth induction in chick epithelium: Expression of quiescent genes for enamel synthesis. Science (1980).

[167] Mitsiadis, T., Cheraud, Y., Sharpe, P. & Fontaine-Perus, J. Development of teeth in chick embryos after mouse neural crest transplantations. Proceedings of the National Academy of Science (2003).

[168] Harris, M. P., Hasso, S. M., Ferguson, M. W. & Fallon, J. F. The development of archosaurian ﬁrst-generation teeth in a chicken mutant. Current Biology (2006).

[169] Biggs, L. C. & Mikkola, M. L. Early inductive events in ectodermal appendage morphogenesis. Seminars in Cell and Developmental Biology (2014).

83 [170] Oster, G. & Alberch, P. Evolution and bifurcation of developmental programs. Evolution (1981).

[171] Widelitz, R. B. & Chuong, C.-M. Early events in skin appendge formation: Induction of epithelial placodes and condensation of dermal mesenchyme. The Journal of Investigative Dermatology (1999).

[172] Mikkola, M. L. Genetic basis of skin appendage development. Seminars in Cell and Developmental Biology (2007).

[173] Pispa, J. & Thesleﬀ, I. Mechanisms of ectodermal organogenesis. Developmental Biology (2003).

[174] Chuong, C. M., Patel, N., Lin, J., Jung, H.-S. & Widelitz, R. Sonic hedgehog signaling pathway in vertebrate epithelial appendage morphogenesis: perspective in development and evolution. Cellular and Molecular Life Sciences (2000).

[175] Mustonen, T. et al. Ectodysplasin a1 promotes placodal cell fate during early morphogenesis of ectodermal appendages. Development (2004).

[176] Mikkola, M. M. & Thesleﬀ, I. Ectodysplasin signaling in development. Cytokine and Growth Factor Reviews (2003).

[177] Fliniaux, I., Mikkola, M. L., Lefevbre, S. & Thesleﬀ, I. Identiﬁcation of dkk4 as a trarget of eda-a1/edar pathway reveals an unexpected role of ectodysplasin as inhibitor of wnt signalling in ectodermal placodes. Developmental Biology (2008).

[178] Drew, C. F. et al. The edar subfamily in feather placode formation. Developmental Biology (2007).

[179] Ahtiainen, L. et al. Directional cell migration, but not proliferation, drives hair placode morphogenesis. Cell (2014).

[180] Vainio, S., Jalkanen, M. & Thesleﬀ, I. Syndecan and tenascin expression is induced by epithelial-mesenchymal interactions in embryonic tooth mesenchyme. J. Cell. Biol. (1989).

[181] Tucker, R. The sequential expression of tenascin mrna in epithelium and mesenchyme during feather morphogenesis. Roux Arch. Dev. Biol. (1991).

[182] Chiquet-Ehrismann, R., Mackie, E., Pearson, C. & Sakakura, T. Tenascin: an extracellular matrix protein involved in tissue interactions during fetal development and oncogenesis. Cell (1986).

[183] Schlosser, G. Induction and speciﬁcation of cranial placodes. Developental Biology (2006).

[184] Maderson, P. F. When? why? and how? some speculations on the evolution of the vertebrate integument. Am. Zoologist (1972).

[185] Dhouailly, D. A new scenario for the evolutionary origin of hair, feather and avian scales. Journal of Anatomy (2009).

[186] DiPoi, N. & Milinkovitch, M. C. The anatomical placode in reptile scale morphogenesis indicates shared ancestry among skin appendages in amniotes. Science Advances (2016).

[187] Dhouailly, D. Dermo-epithelial interactions between birds and mammals: diﬀerentiation of cutaneous appendages. Journal of Embryology and Experimental Morphology (1973).

[188] Dhouailly, D. Formation of cutaneous appendages in dermo-epidermal recombinations between reptiles, birds and mammals. Wilhelm Roux’ Arch. Entwicklungsmech. (1975).

[189] Dhouailly, D. Dermo-epidermal interactions during morphogenesis of cutaneous appendages in amniotes. Front.Matrix Biol. (1977).

[190] Rawles, M. E. Tissue interactions in scale and feather development as studied in dermal-epidermal recombinations. J. Embryol. exp. Morph. (1963).

[191] Cai, J. et al. Patterning the size and number of tooth and its cusps. Developmental Biology (2007).

[192] Alibardi, L. Scale morphogenesis during embryonic development in the lizard anolis lineatopus. J Anat (1996).

[193] Witzmann, F. The evolution of the scalation pattern in temnospondyl amphibians. Zoological Journal of the Linnean Society (2007).

[194] Sawyer, R. H. & Knapp, L. W. Avian skin development and the evolutionary origin of feathers. Journal of Experimental Zoology (2003).

84 [195] Chen, C.-F. et al. Development, regeneration and evolution of feathers. Annu. Rev, Anim, Biosci. (2015).

[196] Wu, P. et al. Multiple regulatory modules are required for scale-to-feather conversion. Molecular Biology and Evolution (2017).

[197] Maderson, P. On how an archosaurian scale might have given rise to an avian feather. American Naturalist (1972).

[198] Prum, R. O. & Dyck, J. A hierarchical model of plumage: Morphology, development and evolution. Journal of Experimental Zoology (2003).

[199] Prum, R. O. Development and evolutionary origin of feathers. Journal of Experimental Zoology (1999).

[200] Prum, R. O. & Brush, A. H. The evolutionary origin and diversiﬁcation of feathers. The Quarterly Review of Biology (2002).

[201] Regal, P. The evolutionary origin of feathers. The Quarterly Review of Biology (1975).

[202] Chuong, C.-M., Chodankar, R., Widelitz, R. B. & Jiang, T.-X. Evo-devo of feathers and scales: building complex epithelial appendages. Current Opinion in Genetics & Development (2000).

[203] Brush, A. H. & Wyld, J. A. Molecular correlates of morphological diﬀerentiation: Avian scutes and scales. Journal of Experimental Zoology (1980).

[204] Jung, H.-S. et al. Local inhibitory action of bmps and their relationships with activators in feather formation: Implications for periodic patterning. Developmental Biology (1998).

[205] Noramly, S., Freeman, A. & Morgan, B. A. Beta-catenin signaling can initiate feather bud development. Development (1999).

[206] Widelitz, R. B., Jiang, T.-X., Noveen, A., Chen, C.-W. J. & Chuong, C.-M. Fgf induces new feather buds from developing avian skin. The Journal of Investigative Dermatology (1996).

[207] Song, H.-K., Lee, S.-H. & Goetinck, P. F. Fgf-2 signaling is suﬃcient to induce dermal condensations during feather development. Developmental Dynamics (2004).

[208] Ting-Bereth, S. A. & Chuong, C.-M. Sonic hedgehog in feather morphogenesis: Induction of mesenchymal condensation and association with cell death. Developmental Dynamics (1996).

[209] Harris, M. P., Fallon, J. F. & Prum, R. O. Shh-bmp2 signaling module and the evolutionary origin and diversiﬁcation of feathers. Journal of Experimental Zoology (2002).

[210] Widelitz, R. B. et al. Wnt-7a in feather morphogenesis: involvement of anterior-posterior asymmetry and proximal-distal elongation demonstrated with an in vitro reconstitution model. Development (1999).

[211] Yue, Z., Jiang, T.-X., Widelitz, R. B. & Chuong, C. M. Wnt3a gradient converts radial to bilateral feather symmetry via topological arrangement of epithelia. Proceedings of the National Academy of Science (2006).

[212] Widelitz, R. B., Jiang, T.-X., Lu, J. & Chuong, C.-M. beta-catenin in epithelial morphogenesis: Conversion of part of avian foot scales into feather buds with a mutated beta-catenin. Developmental Biology (1999).

[213] Widelitz, R., Jiang, T., Lu, J. & Chuong, C. beta-catenin in epithelial morphogenesis: conversion of part of avian foot scales into feather buds with a mutated beta-catenin. Dev. Biol. (2000).

[214] Harris, M. P., Williamson, S., Fallon, J. F., Meinhardt, H. & Prum, R. O. Molecular evidence for an activator- inhibitor mechanism in development of embryonic feather branching. Proceedings of the National Academy of Science (2005).

[215] Yu, M., Wu, P., Widelitz, R. B. & Chuong, C.-M. The morphogenesis of feathers. Nature (2002).

[216] Chuong, C.-M. & Edelman, G. M. Expression if cell-adhesion molecules in embryonic induction. ii. morphogenesis of adult feathers. The Journal of Cell Biology (1985).

[217] Langbein, L. et al. The catalog of human hair keratins: I. expression of the nine type i members in the hair follicle. The Journal of Biological Chemistry (1999).

[218] Starck, J., Andl, T. & Millar, S. E. Hairy math: Insights into hair-follicle spacing and orientation. Cell (2007).

[219] Ohsaki, K., Osumi, N. & Nakamura, S. Altered whisker patterns induced by ectopic expression of shh are topographically represented by barrels. Developmental Brain Research (2002).

85 [220] Reddy, S. et al. Characterization of wnt gene expression in developing and postnatal hair follicles and iden- tiﬁcation of wnt5a as a target of sonic hedgehog in hair follicle morphogenesis. Mechanisms of Development (2001).

[221] Zhang, Y. et al. Reciprocal requirements for eda/edar/nf-kb and wnt/beta-catenin induction pathways in hair follicle induction. Dev. Cell (2009).

[222] Huh, S.-H. et al. Fgf20 governs formation of primary and secondary dermal condensations in developing hair follicles. Genes and Development (2013).

[223] Biggs, L. C. et al. Hair follicle dermal condensations forms via fgf20 primed cell cycle exit, cell motility, and aggregation. ELife (2018).

[224] Evans, A. R. & Sanson, G. D. The tooth of perfection: function and spatial constraints on mammalian tooth shape. Biological Journal of the Linnean Society (2003).

[225] Oo´e,T. On the early development of human dental lamina. Okajamas folia anatomica Japonica (1957).

[226] Mina, M. & Kollar, E. The induction of odontogenesis in non-dental mesenchyme combines with early murine mandibular arch epithelium. Archives of Oral Biology (1987).

[227] Bei, M. & Maas, R. Fgfs and bmp4 induce both msx1-independent and msx1-dependent signaling pathways in early tooth development. Development (1998).

[228] Lumsden, A. Spatial organization of the epithelium and the role of neural crest cells in the initiation of the mammalian tooth germ. Development (1988).

[229] Thesleﬀ, I. Epithelial-mesenchymal signalling regulating tooth morphogenesis. Journal of Cell Science (2007).

[230] Mammoto, T. et al. Mechanochemical control of mesenchymal condensation and embryonic tooth organ formation. Developmental Cell (2011).

[231] Vahtokaari, A., Aberg,˚ T., Jernvall, J., Ker¨anen,S. & Thesleﬀ, I. The enamel knot as a signaling center in the developing mouse tooth. Mechanisms of Development (1996).

[232] Tucker, A. et al. Edar/eda interactions regulate enamel knot formation in tooth morphogenesis. Development (2000).

[233] Jernvall, J., Aberg,˚ T., Kettunen, P., Ker¨anen,S. & Thesleﬀ, I. The life history of an embryonic signaling center: Bmp-4 induces p21 and is associated with apoptosis in the mouse tooth enamel knot. Development (1998).

[234] Thesleff, I., Keränen,S. & Jernvall, J. Enamel knots as signaling centers linking tooth morphogensis and odontoblast differentiation. Advances in Dental Research (2001).

[235] Jernvall, J., Kettunen, P., Karavanova, I., Martin, L. B. & Thesleﬀ, I. Evidence for the role of the enamel knot as a control center in mammalian tooth cusp formation: non-dividing cells express growth stimulating fgf-4 gene. Int. J. Dev. Biol. (2002).

[236] Jernvall, J. & Thesleﬀ, I. Reiterative signaling and patterning during mammalian tooth morphogenesis. Mech- anisms of Development (2000).

[237] Jernvall, J., Keranen, S. V. & Thesleﬀ, I. Evolutionary modiﬁcation of development in mammalian teeth: Quantifying gene expression patterns and topography. Proceedings of the National Academy of Science (2000).

[238] Kassai, Y. et al. Regulation of mammalian tooth cusp patterning by ectodin. Science (2005).

[239] Debiais-Thibaud, M. et al. Tooth and scale morphogenesis in shark: an alternative process to the mammalian enamel knot system. BMC Evolutionary Biology (2015).

[240] Sharpe, P. Fish scale development: hair today, teeth and scales yesterday? Current Biology (2001).

[241] Smith, M. M. & Hall, B. K. Development and evolutionary origins of vertebrate skeletogenic and odontogenic tissues. Biol. Rev. (1990).

[242] Smith, M. M. & Coates, M. I. Evolutionary origins of the vertebrate dentition: phylogenetic patterns and developmental evolution. European Journal of Oral Sciences (2014).

[243] Bryant, W. L. A study of the oldest known vertebrates, astraspis and eriptychius. Proceedings of the American Philosophical Society (1936).

86 [244] Ulrich, E. & Bassler, R. S. A classiﬁcation of the toothlike fossils, conodonts, with descriptions of american, devonian and mississippian species. Proceedings U. S. National Museum (1926).

[245] Ruecklin, M. et al. Development of teeth and jaws in the earliest jawed vertebrates. Nature (2012).

[246] Fraser, G. J., Graham, A. & Smith, M. M. Conserved deployment of genes during odontogenesis across oste- ichthyans. Proceedings of the Royal Society of London B (2004).

[247] Giles, S., Ruecklin, M. & Donoghe, P. C. Histology of ”placoderm” dermal skeletons: Implications for the nature of the ancestral gnathostome. Journal of Morphology (2013).

[248] Vickarious, M. & Sire, J.-Y. The integumentary skeleton of tetrapods: Origin, evolution and development. J. Anat. (2009).

[249] Krejsa, R. J., Bringas, P. J. & Slavkin, H. C. A neontological interpretation of conodont elements based on agnathan cyclostome tooth structure, function and development. Lethaia (1990).

[250] Donoghue, P. C. & Ruecklin, M. The ins and outs of the evolutionary origin of teeth. Evolution and Development (2016).

[251] Dhouailly, D. et al. Getting to the root of scales, feathers and hair: As deep as odontodes? Experimental Dermatology (2017).

[252] Alberch, P. & Kollar, E. Strategies of head development: workshop report. Development (1988).

[253] Baker, C. V. & Bronner-Fraser, M. Vertebrate cranial placodes i.embryonic induction. Developmental Biology (2001).

[254] McCabe, K. L. & Bronner-Fraser, M. Molecular and tissue interactions govgovern induction of cranial ectodermal placodes. Dev.Biol. (2009).

[255] Northcutt, R. & Gans, C. The genesis of neural crest and epithelial placodes: A reinterpretation of vertebrate origins. The Quarterly Review of Biology (1983).

[256] Webb, J. F. & Noden, D. M. Ectodermal placodes: Contributions to the development of the vertebrate head. Amer. Zool. (1993).

[257] Xu, X., Zheng, X. & You, H. Exceptional dinosaur fossils show ontogenetic development of early feathers. Nature (2010).

[258] Alibardi, L. Dermo-epidermal interactions in reptilian scales: Speculations on the evolution of scales, feathers and hairs. Journal of Experimental Zoology Part B: Molecular and Developmental Evolution (2004).

[259] Maderson, P. Mammalian skin evolution: A reevaluation. Exp. Dermatol. (2003).

[260] Sawyer, R. & Creaig, K. Avian scale development. absence of an ”epidermal placode” in reticulate scale morphogenesis. Journal of Morphology (1977).

[261] Gautschi, B., Widmer, A., Joshi, J. & Koella, J. C. Increased frequency of scale anomalies and loss of genetic variation in serially bottlenecked populations of the dice snake, natrix tessellata. Conservation Genetics (2002).

[262] Lande, R. Risk of population extinction from ﬁxation of new deletorious mutations. Evolution (1994).

[263] Lacy, R. C. Importance of genetic variation to the viability of mammalian populations. Journal of Mammalogy (1997).

[264] Räikkönen,J., Bignert, A., Mortensen, P. & Fernholm, B. Congenital defects in a highly inbred wild wolf population (canis lupus). Mammalian Biology - Zeitschrift fürSäugetierkunde (2005).

[265] Madsen, T., Stille, B. & Shine, R. Inbreeding depression in an isolated population of adders vipera berus. Biological Conservation (1996).

[266] Keller, L. F. & Waller, D. M. Inbreeding eﬀects in wild populations. Trends in Ecology and Evolution (2002).

[267] Carlborg, O. & Haley, C. S. Epistasis: too often neglected in complex trait studies? Nature (2004).

[268] Wagner, G. & Zhang, J. The pleiotropic structure of the genotype-phenotype map: the evolvability of complex organisms. Nature Review Genetics (2011).

87 [269] Kroymann, J. & Mitchell-Olds, T. Epistasis and balanced polymorphism inﬂuencing complex trait variation. Nature (2005).

[270] Smith-Gill, S. J. Developmental plasticity: Developmental conversion versus phenotypic modulation. Amer. Zool. (1983).

[271] Houle, D., Govindaraju & Omholt, S. Phenomics: the next challenge. Nature Reviews (2010).

[272] Kimura, M. The Neutral Theory of Molecular Evolution (Cambridge University Press, 1983).

[273] Eyre-Walker, A. & Keightley, P. D. The distribution of ﬁtness eﬀects of new mutations. Nature (2007).

[274] McEntee, K., Weinstock, G. M. & Lehman, I. Binding of the reca protein of escherichia coli to single- and double-stranded dna. The Journal of Biological Chemistry (1981).

[275] Phair, R., Gorski, S. & Misteli, T. Measurement of dynamic protein binding to chromatin in vivo, using photobleaching microscopy. Methods in Enzymology (2003).

[276] Tyson, J. J., Chen, K. C. & Novak, B. Sniﬀers, buzzers, toggles and blinkers: dynamics of regulatory and signalling pathways in the cell. Current Opinion in Cell Biology (2003).

[277] Riggs, A. D., Bourgeois, S. & Cohn, M. The lac repressor-operator interaction. iii. kinetic studies. J Mol. Biol. (1970).

[278] Parkhurst, K. M., Brenowitz, M. & Parkhurst, L. J. Simultaneous binding and bending of promoter dna by the tata binding protein: Real time kinetic measurements. Biochemistry (1996).

[279] Kalir, S. et al. Ordering genes in a ﬂagella pathway by analysis of expression kinetics from living bacteria. Science (2001).

[280] Elf, J., Li, G.-W. & Xie, X. S. Probing transcription factor dynamics at the single-molecule level in a living cell. Science (2007).

[281] Fried, M. G. & Crothers, D. M. Kinetics and mechanism in the reaction of gene regulatory proteins with dna. Journal of Molecular Biology (1984).

[282] Carey, M. The enhanceasome and transcription synergy. Cell (1998).

[283] Veitia, R. A sigmoidal transcriptional response: cooperativity, synergy and dosage eﬀects. Biol. Rev, Camb. Philos. Soc. (2003).

[284] Raj, A. & van Oudenaarden, A. Nature, nurture, or chance: Stochastic gene expression and its consequences. Cell (2008).

[285] Rao, C. V., Wolf, D. M. & Arkin, A. P. Control, exploitation and tolerance of intracellular noise. Nature (2002).

[286] True, J. R. & Haag, E. S. Developmental systems drift and ﬂexibility in evolutionary trajectories. Evolution and Development (2001).

[287] Lynch, V. J. Use with caution: Developmental systems divergence and potential pitfalls of animal models. Yale Journal of Biology and Medicine (2009).

[288] Yanai, I., Peshkin, L., Jorgensen, P. & Kirschner, M. W. Mapping gene expression in two xenopus species: Evolutionary constraints and developmental ﬂexibility. Developmental Cell (2011).

[289] Wagner, A. The role of robustness in phenotypic adaptation and innovation. Proceedings of the Royal Society of London B (2012).

[290] J´ımenez, A., Munteanu, A. & Sharpe, J. Dynamics of gene circuits shapes evolvability. Proceedings of the National Academy of Science (2015).

[291] Gibson, G. & Wagner, G. Canalization in evolutionary genetics: a stabilizing theory? BioEssays (2000).

[292] Scharloo, W. Canalization: Genetic and developmental aspects. Annual Review of Ecology and Systematics (1991).

[293] Waddington, C. Genetic assimilation. Adv. Genet. (1961).

[294] Eldredge, N. & Gould, S. Evolution equilibria: an alternative to phyletic gradualism (Cooper and Co., San Francisco, 1976).

88 [295] Theissen, G. Saltational evolution: hopeful monsters are here to stay. Theory Biosci. (2009).

[296] Goldschmidt, R. The Material Basis of Evolution (Yale University Press, New Haven, 1940).

[297] Salazar-Ciudad, I., Jervall, J. & Newman, S. A. Mechanisms of pattern formation in development and evolution. Development (2003).

[298] Salazar-Ciudad, I. Morphological evolution and embryonic developmental diversity in metazoa. Development (2010).

[299] Wang, J.-C. & Thampatty, B. An introductory review of mechanotransduction. Biomechanics and Modeling in Mechanobiology (2006).

[300] Wozniak, M. A. & Chen, C. S. Mechanotransduction in development: a growing role for contractility. Nature (2009).

[301] Ingber, D. Tensegrity: The architectural basis of cellular mechanotransduction. Annu. Rev. Physiol. (1997).

[302] Altman, G. et al. Cell diﬀerentiation by mechanical stress. FASEB Journal (2002).

[303] M¨uller, P., Rogers, K. W., Yu, S. R., Brand, M. & Schier, A. F. Morphogen transport. Development (2013).

[304] Chalancon, G. et al. Interplay between gene expression noise and regulatory network architecture. Trends in Genetics (2012).

[305] Salazar-Ciudad, I. & Jernvall, J. How diﬀerent types of pattern formation mechanisms aﬀect the evolution of form and development. Evolution and Development (2004).

[306] van Eeden, F. & St.Johnson, D. The polarization of the anterior-posterior and dorsal-ventral axes during drosophila oogenesis. Current Opinion in Genetics and Development (1999).

[307] Gerhart, J. et al. Cortical rotation of the xenopus egg: consequences for the anteroposterior pattern of embryonic dorsal development. Development (1989).

[308] Rossant, J. & Tam, P. Emerging asymmetry and embryonic patterning in early mouse development. Develop- mental Cell (2004).

[309] West-Eberhard, M. J. Phenotypic plasticity and the origins of diversity. Annu. Rev. Ecol. Syst. (1989).

[310] Gotthard, K. & Nylin, S. Adaptive plasticity and plasticity as an adaptation: A selective review of plasticity in animal morphology and life history. Oikos (1995).

[311] Metcalfe, N. & Monaghan, P. Compensation for a bad start: grow now, pay later? Trends in ecology and evolution (2001).

[312] Nylin, S. & Gotthard, K. Plasticity in life-history traits. Annual Review of Entomology (1998).

[313] Gilbert, S. F., Bosch, T. C. & Ledon-Rettig, C. Eco-evo-devo: developmental symbiosis and developmental plasticity as evolutionary agents. Nature Reviews Genetics (2015).

[314] de Souza, D. J., Bezier, A., Depoix, D., Drezen, J.-M. & Lenoir, A. Blochmannia endosymbionts improve colony growth and immune defence in the ant camponotus fellah. BMC Microbiology (2009).

[315] Bailly, X. et al. The chimerical and multifaceted marine acoel symsagittifera roscoﬀensis: from photosymbiosis to brain regeneration. Frontiers in Microbiology (2014).

[316] Gilbert, S. F. Mechanisms for the environmental regulation of gene expression: Ecological aspects of animal development. J. Biosci. (2005).

[317] Huang, S. Non-genetic heterogeneity of cells in development: more than just noise. Development (2009).

[318] Balazsi, G., van Oudenaarden, A. & Collins, J. J. Cellular decision making and biological noise: From microbes to mammals. Cell (2011).

[319] Nijhout, H. F. Development and evolution of adaptive polyphenisms. Evolution and Development (2012).

[320] Yntema, C. Temperature levels and periods of sex determination during incubation of eggs of chelydra serpentina. Journal of Morphology (1979).

89 [321] Ayres-Fernandez, C. & Cordero-Rivera, A. Asymmetries as accessory scutes in emys orbicularis from northern spain. Biologia, Bratislava (2004).

[322] Bujes, C. & Verrastro, L. Supernumerary epidermal shields and carapace variation in orbigny’s slider turtles, trachemys dorbigni (testudines, emydidae). Revista Brasileira de Zoologia (2007).

[323] Caracappa, S. et al. Nonmodal scutes patterns in the loggerhead sea turtle (caretta caretta): a possible epigenetic eﬀect? Can. J. Zoology (2016).

[324] Davy, C. M. & Murphy, R. W. Explaining patterns of deformity in freshwater turtles using macculloch’s hypothesis. Can. J. Zoology (2009).

[325] Dordevic, S. Carapace pecularities of hermann’s tortoises (testudo hermanni) in several balkan populations. North-Western Journal of Zoology (2014).

[326] Douglass, J. Abnormalities of scutellation in a population of gopherus polyphemus (reptilia, testudinidae). Florida Scientist (1977).

[327] Ergene, S., Aymak, C. & Ucar, A. H. Carapacial scute variation in green turtle (chelonia mydas) and loggerhead turtle (caretta caretta) hatchlings in alata, mersin, turkey. Turk J Zool (2011).

[328] Farke, C. & Distler, C. Ontogeny and abnormalities of the tortoise carapace: a computer tomography and dissection study. Salamandra (2015).

[329] Gadow, H. Orthogenetic variation in the shell of chelonia. Willey’s Zool. Results (1899).

[330] Good, H. M. Shell anomalies in the desert tortoise (gopherus agassizii). populations of the beaver dam slope, utah and desert tortoise natural area, california. Proc. Desert Tortoise Council 1984 Symposium (1984).

[331] Hewavisenthi, S. & Parmenter, C. Inﬂuence of incubation environment on the development of the ﬂatback turtle (natator depressus). Copeia (2001).

[332] Hildebrandt, S. F. Duplicity and other abnormalities in diamondback terrapins. Journal of the Elisha Mitchell Scientiﬁc Society (1930).

[333] Kaska, Y., Downie, R. & Furness, R. W. Abnormal development in sea turtle embryos. Proceedings of the Eighteenth International Sea Turtle Symposium (1998).

[334] Kazmaier, R. T. & Robel, R. J. Scute anomalies of ornate box turtles in kansas. Transactions of the Kansas Academy of Science (2001).

[335] Lynn, W. Variation in scutes and plates in the boxturtle, terrapene carolina. The American Naturalist (1937).

[336] Lynn, W. & Ullrich, M. C. Experimental production of shell abnormalities in turtles. Copeia (1950).

[337] MacCulloch, R. D. Variation in the shell of chrysemys picta belli from southern saskatchewan. Journal of Herpetology (1981).

[338] Margaritoulis, D. & Chiras, G. Scalation patterns of loggerhead turtles nesting in laganas bay, zakynthos island, greece. Mar. Turtle Newsletter (2011).

[339] McEwan, B. Bone anomalies in the shell of gopherus polyphemus. Biological Sciences (1982).

[340] Oliver, G. Variabilitéet malformations de l’écaillurede la caouanne, caretta caretta (l 1758) (reptilia cheloni- idae), sur les côtesfrancaises de méditerranée. Bull. Soc. Herp. Fr. (2011).

[341] Ozdemir, B. & Turkozan, O. Carapacial scute variation in green turtle, chelonia mydas hatchlings in northern cyprus. Turk J Zool (2006).

[342] Packard, G. C., Packard, M. J., Miller, K. & Baoardman, T. J. Inﬂuence of moisture, temperature, and substrate on snapping turtle eggs and embryos. Ecology (1987).

[343] Parker, G. H. Correlated abnormalities in the scute and bony plates of the carapace of the sculptured tortoise. The American Naturalist (1901).

[344] Plymale, H. H., Jackson, C. G. J. & Collier, G. Kyphosis in chrysemys scripta yaquia (testudines: Emydidae) and other turtles. The Southwestern Naturalist (1978).

[345] Pritchard, P. Sea turtles of the guianas. Bulletin of the Florida State Museum (1969).

90 [346] Rivera, G. & Claude, J. Environment media and shape asymmetries: a case study on turtle shells. Biological journal of the Linnean Society (2008).

[347] Sugunama, H., Horikoshi, K. & Tachikawa, H. Scute deviation of green turtle hatchlings from a hatchery in ogasawara islands, japan. Proceedings of the Fourteenth Annual Symposium on Sea Turtle Biology and Conservation (1994).

[348] Telemeco, R. S., Warner, D. A., Reida, M. K. & Janzen, F. J. Extreme developmental temperatures result in morphological abnormalities in painted turtles (chrysemys picta): a climate change perspective. Integrative Zoology (2013).

[349] Velo-Anton, G., Becker, C. G. & Cordeno-Rivera, A. Turtle carapace anomalies: The roles of genetic diversity and environment. Plos One (2011).

[350] Wandolleck, B. Eine bucklige testudo graeca. Zoologische Jahrbuecher (1904).

[351] Hay, O. P. Descriptions of ﬁve species of north american fossil turtles, four of which are new. Proceedings of the United States National Museum (1903).

[352] Grant, C. Orthogenetic variation. Proceedings of the Indiana Academy of Science (1936).

[353] Coker, R. Gadow’s hypothesis of ”orthogenetic variation” in chelonia. Johns Hopkins U. Circ. (1905).

[354] Coker, R. Orthogenetic variation? Science (1905).

[355] Gadow, H. Orthogenetic variation. Science (1905).

[356] Agassiz, L. North American Testudinata. In: Contributions to the natural history of the United States. Vol.1 part 2 (Little, Brown and Co., Boston, 1857).

[357] Gaﬀney, E. S. Historical analysis of theories of chelonian relationships. Syst.Zool. (1984).

[358] Cope, E. On the homologies of some of the cranial bones of the reptilia, and on the systematic arrangement of this class. Proc. Am. Assoc. Adv. Sci. (1871).

[359] Geigenbaur, H. Vergleichende Anatomie der Wirbelthiere. Erster Band. (Engelmann, Leipzig, 1898).

[360] Berry, C. T. A freak or evolution? The Scientiﬁc Monthly (1935).

[361] Yntema, C. & Mrosovski, N. Sexual diﬀerentiation in hatchling loggerhead (caretta caretta) incubated at diﬀerent controlled temperatures. Herpetologica (1980).

[362] Ewert, M., Jackson, D. R. & Nelson, C. E. Patterns of temperature-dependent sex determination in turtles. Journal of Experimental Zoology (1994).

[363] Ewert, M. A. & Nelson, C. E. Sex determination in turtles: Diverse patterns and some possible adaptive values. Copeia (1991).

[364] Frazier, J. Analisis estadistico de la tortuga golﬁna lepidochelys olivacea (eschscholtz) de oaxaca, mexico. Ciencia Pesquera (1983).

[365] Varriale, A. Dna methylation, epigenetics and evolution in vertebrates: facts and challenges. Int. J. Evol. Biol. (2014).

[366] Head, J. Patterns of dna methylation in animals: an ecotoxicological perspective. Integr. Comp. Biol. (2014).

[367] Matsumoto, Y., Buemio, A., Chu, R., Vafaee, M. & Crews, D. Epigenetic control of gonadal aromatase (cyo19a1) in temperature-dependent sex determination of red-eared slider turtles. Plos One (2013).

[368] O’Steen, S. Embryonic temperature inﬂuences juvenile temperature choice and growth rate in snapping turtles chelydra serpentina. Journal of Experimental Biology (1998).

[369] Yntema, C. Eﬀects of various temperatures on the embryonic development of chelydra serpentina. Anatomical Record (1960).

[370] Fox, W. Eﬀect of temperature on development of scutellation in the garter snake, thamnophis elegans atratus. Copeia (1948).

91 [371] Zhadanova, N. & Zakharov, V. M. Developmental stability of sand lizard (Lacerta agilis L.) under optimal and non-optimal conditions. In: Mainland and Insular Lizards: A Mediterranean Perspective (Firenze University Press, Firenze, 2006).

[372] Idrisova, L. The eﬀect of incubation temperature on deviations of pholidosis and malformations in grass snake natrix natrix (l. 1758) and sand lizard lacerta agilis (l.1758). Knowledge E Life Sciences (2018).

[373] Bell, B., Spotila, J. R. & Congdon, J. High incidence of deformity in aquatic turtles in the john heinz national wildlife refuge. Environmental Pollution (2005).

[374] Bishop, C. A. et al. Environment contamination and developmental abnormalities in eggs and hatchlings of the common snapping turtle (chelydra serpentina serpentina) from the great lakes - st.lawrence river basin (1989-91). Environmental Pollution (1998).

[375] Stockard, C. Developmental rate and structural expression: An experimental study of twins, ’double monsters’ and single deformities, and the interaction among embryonic organs during their origin and development. Am. J. Anat. (1921).

[376] Kajii, T., Kida, M. & Takahashi, K. The eﬀect of thalidomide intake during 113 human pregnancies. Clinical Teratology (1973).

[377] Heyne, G. W. et al. Deﬁnition of critical periods for hedgehog pathway antagonist-induced holoprosencephaly, cleft lip and cleft palate. Plos One (2015).

[378] Pourqui´e,O. Vertebrate somitogenesis: a novel paradigm for animal segmentation? International Journal of Developmental Biology (2003).

[379] Angielczyk, K. D., Feldman, C. R. & Miller, G. R. Adaptive evolution in plastron shape in emydine turtles. Evolution (2011).

[380] Myers, E. M., Janzen, F. J., Adams, D. C. & Tucker, J. Quantitative genetics of plastron shape in slider turtles (trachemys scripta). Evolution (2006).

[381] Savriama, Y. et al. Bracketing phenotypic limits of mammalian hybridization. BioRXiv (2018).

[382] Harjunmaa, E. et al. Replaying evolutionary transitions from the dental fossil record. Nature (2014).

[383] Schroer, K. & Wood, B. Modelling the dental development of fossil hominins through the inhibitory cascade. Journal of Anatomy (2015).

[384] de Visser, J. A. G. & Krug, J. Empirical ﬁtness landscapes and the predicability of evolution. Nature Reviews Genetics (2014).

[385] Dyson, F. A meeting with enrico fermi. Nature (2004).

[386] Osborne, J. M., Fletcher, A. G., Pitt-Francis, J. M., Maini, P. K. & Gavaghan, D. J. Comparing individual-based approaches to modelling the self-organization of multicellular tissues. PLOS Computational Biology (2017).

[387] Sharpe, J. Computer modeling in developmental biology: growing today, essential tomorrow. Development (2017).

[388] Tanaka, S. Simulation frameworks for morphogenetic problems. Computation (2015).

[389] Wolfram, S. Cellular automata as models of complexity. Nature (1984).

[390] Deutsch, A. & Dormann, S. Cellular automaton modeling of biological pattern formation: characterization, applications and analysis (Birkh¨auser,2007).

[391] Graner, F. & Glazier, J. A. Simulation of biological cell sorting using a two-dimensional extended potts model. Physical Review Letters (1992).

[392] Glazier, J. A., Balter, A. & Poplawski, N. J. II.1. Magnetization to Morphogenesis: A Brief History of the Glazier-Graner-Hogeweg Model. In: Single-Cell-Based Models in Biology and Medicine, Mathematics and Bio- sciences in Interaction (Birkh¨auserVerlag Basel/Switzerland, 2007).

[393] Ouchi, N. B., Glazier, J. A., Rieu, J.-P., Upadhyaya & Sawada, Y. Improving the realism in the cellular potts model in simulations of biological cells. Physica A (2003).

92 [394] Mar´ee,A. F., Grieneisen, V. A. & Hogeweg, P. II.2. The Cellular Potts Model and Biophysical Properties of Cells, Tissues and Morphogenesis. In: Single-Cell-Based Models in Biology and Medicine, Mathematics and Biosciences in Interactions (Birkh¨auserVerlag Basel/Switzerland, 2007).

[395] Steinberg, M. Diﬀerential adhesion in morphogenesis: a modern view. Current Opinion in Genetics and Development (2007).

[396] Foty, R. A., Pﬂeger, C. M., Forgacs, G. & Steinberg, M. S. Surface tensions of embryonic tissues predict their mutual envelopment behavior. Development (1996).

[397] Mar´ee,A. F. & Hogeweg, P. How amoeboids self-organize into a fruiting body: Multicellular coordination in dictyostelium discoideum. Proceedings of the National Academy of Science (2001).

[398] Peskin, C. S. The immersed boundary method. Acta Numerica (2002).

[399] Mittal, R. & Iaccarino, G. Immersed boundary methods. Annu. Rev. Fluid Mech. (2005).

[400] Pathmananthan, P. et al. A computational study of discrete mechanical tissue models. Phys. Biol. (2009).

[401] Newman, T. J. Modeling Multicellular Structures Using the Subcellular Element Model in: Single-Cell-Based Models in Biology and Medicine (Springer, 2007).

[402] Alt, S., Ganguly, P. & Salbreux, G. Vertex models: from cell mechanics to tissue morphogenesis. Philosophical Transactions B (2016).

[403] Fletcher, A., Osterﬁeld, M., Baker, R. & Shvartsman, S. Vertex models of epithelial morphogensis. Biophysical Journal (2014).

[404] Kauﬀman, S. The Origins of Order: Self-organization and Selection in Evolution (Oxford University Press, 1993).

[405] Shmulevich, I., Dougherty, E. R., Kim, S. & Zhang, W. Probabilistic boolean networks: a rule-based uncertainty model for gene regulatory networks. Bioinformatics (2002).

[406] Vijesh, N., Chakrabarti, S. K. & Sreekumar, J. Modelling of gene regulatory networks: A review. L Biomecical Science and Engineering (2013).

[407] DiVentura, B., Lemerle, C., Michalodimitrakis, K. & Serrano, L. From in vivo to in silico biology and back. Nature (2006).

[408] de Jong, H. Modeling and simulation of genetic regulatory systems: A literature review. INRIA (2000).

[409] Conrad, E. D. & Tyson, J. J. Modeling Molecular Interaction Networks with Nonlinear Ordinary Diﬀeren- tial Equations in: System Modeling in Cellular Biology. From Concepts to Nuts and Bolts (The MIT Press, Cambridge, MT, London, GB, 2006).

[410] Salazar-Ciudad, I. & Jernvall, J. A computational model of teeth and the developmental origins of morphological variation. Nature (2010).

[411] Raspopovic, J., Marcon, L., Russo, L. & Sharpe, J. Digit patterning is controlled by a bmp-sox9-wnt turing network modulated by morphogen gradients. Nature (2014).

[412] Holzhuetter, H.-G., Drasdo, D., Preusser, T., Lippert, J. & Henney, d. M. The virtual liver: a multidisciplinary, multilevel challenge for systems biology. WIREs Syst Biol Med (2012).

[413] Hogeweg, P. Evolving mechanisms of morphogenesis: on the interplay between diﬀerential adhesion and cell diﬀerentiation. Journal of theoretical Biology (2000).

[414] Izaguirre, J. et al. Compucell, a multi-model framework for simulation of morphogenesis. Bioinformatics (2004).

[415] Delile, J., Doursat, R. & Peyri´eras,N. Computational modeling and simulation of animal early embryogenesis with the mecagen platform. Computational Systems Biology (2013).

[416] Starruss, J., de Back, W., Brusch, L. & Deutsch, A. Morpheus: a user-friendly modeling environment for multiscale and multicellular systems biology. Bioinformatics (2014).

[417] Peralta, X. et al. Upregulation of forces and morphogenetic asymmetries in dorsal closure during drosophila development. Biophysical Journal (2007).

93 [418] Brunet, T. et al. Evolutionary conservation of early mesoderm speciﬁcation by mechanotransduction in bilateria. Nature Communications (2013).

[419] Yntema, C. A series of stages in the embryonic development in chelydra serpentina. Journal of morphology (1968).

[420] Kondo, S. An updated kernel-based turing model for studying the mechanics of biological pattern formation. Journal of Theoretical Biology (2017).

[421] Henderson, J. & Carter, D. Mechanical induction in limb morphogenesis: The role of growth-generated strains and pressures. Bone (2002).

[422] Heisenberg, C.-P. & Bella¨ıche, Y. Forces in tissue morphogenesis and patterning. Cell (2013).

[423] Bazelli`eres,E. et al. Control of cell-cell forces and collective cell dynamics by the intercellular adhesome. Nature Cell Biology (2015).

[424] Tambe, D. T. et al. Collective cell guidance by cooperative intercellular forces. Nature Materials (2011).

[425] Campinho, P. et al. Tension-oriented cell divisions limit anisotropic tissue tension in epithelial spreading during zebraﬁsh epiboly. Nature Cell Biology (2013).

[426] Wyatt, T. P. et al. Emergence of homeostatic epithelial packing and stress dissipation through divisions oriented along the long cell axis. Proceedings of the National Academy of Science (2015).

[427] Matthews, B. D., Overby, D. R., Mannix, R. & Ingber, D. E. Cellular adaptation to mechanical stress: role of integrins, rho, cytoskeletal tension and mechanosensitive ion channels. Journal of Cell Science (2006).

[428] Galle, J., Loeﬄer, M. & Drasdo, D. Modeling the eﬀect of deregulated proliferation and apoptosis on the growth dynamics of epithelial cell populations in vitro. Biophysical Journal (2005).

[429] Nelson, P. J. & Daniel, T. O. Emerging targets: Molecular mechanisms of cell contact-mediated growth control. Kidney International (2002).

[430] Hall, P. A., Coates, P. J., Ansari, B. & Hopwood, D. Reguation of cell number in the mammalian gastrointestinal tract: the importance of apoptosis. Journal of Cell Science (1994).

[431] Stoker, M. G. P. The leeuwenhoek lecture 1972: Tumor viruses and the sociology of ﬁbroblasts. Proceedings of the Royal Society of London B (1972).

[432] Sennett, R. & Rendl, M. Mesenchymal-epithelial interactions during hair follicle morphogenesis and cycling. Seminars in Cell and Developmental Biology (2012).

[433] Chuong, C.-M., Widelitz, R. B., Ting-Berreth, S. & Jiang, T.-X. Early events during avian skin appendage regeneration: Dependence on epithelial-mesenchymal interaction and order of molecular reappearance. Journal of Investigative Dermatology (1996).

[434] Clark, C. et al. Evidence for the neural crest origin of turtle plastron bones. Genesis (2001).

[435] Moustakas, J. E. The evolution and development of skeletogenic mesenchyme in reptiles (ProQuest Dissertations Publishing, University of California, Berkley, 2007).

[436] Chen, B.-S. & Wang, Y.-C. On the attenuation and ampliﬁcation of molecular noise in genetic regulatory networks. BMC Bioinformatics (2006).

[437] Arcure, P. & Murray, J. Pattern sensitivity to boundary and initial conditions in reaction-diﬀusion models. Journal of theoretical Biology (1986).

[438] Lepp¨anen,T., Karttunen, M., Barrio, R. & Kaski, K. The eﬀect of noise on turing patterns. Progress of Theoretical Physics (2003).

[439] Spemann, H. Uber¨ organisatoren in der tierischen entwicklung. Naturwissenschaften (1924).

[440] Hamburger, V. Hans spemann and the organizer concept. Experientia (1969).

[441] Harland, R. & Gerhart, J. Formation and function of spemann’s organizer. Annu. Rev. Cell Dev. Biol. (1999).

[442] Spemann, H. & Mangold, H. Uber¨ induktion von embryonalanlagen durch implantation artfremder organisatoren. Archiv f¨urmikroskopische Anatomie und Entwicklungsmechanik (1923).

94 [443] Von Dassow, G. & Munro, E. Modularity in animal development and evolution: elements of a conceptual framework for evodevo. The Journal of experimental Zoology (1999).

[444] Townsend, G., Harris, F., Edward, Lesot, H., Clauss, F. & Brook, A. Morphogenetic ﬁelds within the human dentition: A new, clinically relevant synthesis of an old concept. Archives of Oral Biology (2009).

[445] Zimmermann, L., De Jesus-Escobar, J. & Harland, R. The spemann organizer signal noggin binds and inactivates bone morphogenetic protein 4. Cell (1996).

[446] Mandler, M. & Neub¨user,A. Fgf signalling is required for initiation of feather placode development. Development (2004).

[447] Aberg, T. et al. Runx2 mediates fgf signaling from epithelium to mesenchyme during tooth morphogenesis. Developmental Biology (2004).

[448] Cooper, R. L. et al. An ancient turing-like patterning mechanism regulates skin denticle development in sharks. Science Advances (2018).

[449] Olivera-Martinez, I., Thelu, J. & Dhouailly, D. Molecular mechanisms controlling dorsal dermis generation from the somitic dermomyotome. Int. J. Dev. Biol. (2004).

[450] Salazar-Ciudad, I. & Jernvall, J. A gene network model accounting for development and evolution of mammalian teeth. Proceedings of the National Academy of Science (2002).

[451] Suzuki, K. et al. Embryonic hair follicle fate change by augmented beta-catenin through shh and bmp signaling. Development (2009).

[452] Srivastava, A. K. et al. The tabby phenotype is caused by mutation in a mouse homologue of the eda gene that reveals novel mouse and human exons and encodes a protein (ectodysplasin-a) with collagenous domains. Proceedings of the National Academy of Science (1997).

[453] Houghton, L., Lindon, C. & Morgan, B. A. The ectodysplasin pathway in feather tract development. Develop- ment (2005).

[454] Pispa, J. et al. Cusp patterning defect in tabby mouse teeth and its partial rescue by fgf. Developental Biology (1999).

[455] Klein, O. D. et al. Sprouty genes control diastema tooth development via bidirectional antagonism of epithelial- mesenchymal fgf signaling. Developmental Cell (2006).

[456] Harris, M. P. et al. Zebraﬁsh eda and edar mutants reveal conserved and ancestral roles of ectodysplasin signaling in vertebrates. Plos Genetics (2008).

[457] St-Jacques, B. et al. Sonic hedgehog signaling is essential for hair development. Current Biology (1998).

[458] Botchkarev, V. A. Bone morphogenetic proteins and their antagonists in skin and hair follicle biology. Progress in Dermatology (2002).

[459] Niswander, L. & Martin, G. R. Fgf-9 and bmp-2 have opposite eﬀects on limb growth. Nature (1993).

[460] Andl, T., ReReddy, S. T., Gaddapara, T. & Millar, S. E. Wnt signals are required for the initiation of hair follicle development. Developmental Cell (2002).

[461] Neub¨user,A., Peters, H., Balling, R. & Martin, G. R. Antagonistic interactions between fgf and bmp signalling pathways: A mechanism for positioning the sites of tooth formation. Cell (1997).

[462] Chen, J., Lan, Y., Baek, J.-A., Gao, Y. & Jiang, R. Wnt/beta-catenin signaling plays an essential role in activation of odontogenic mesenchyme during early tooth development. Developmental Biology (2009).

[463] Sarkar, L. et al. Wnt/shh interactions regulate ectodermal boundary formation during mammalian tooth development. Proceedings of the National Academy of Science (2000).

[464] Michon, F., Forest, L., Collomb, E., Demongeot, J. & Dhouailly, D. Bmp2 and bmp7 play antagonistic roles in feather induction. Development (2008).

[465] Chiang, C. et al. Essential role for sonic hedgehog during hair follicle morphogenesis. Developental Biology (1999).

[466] Chen, Y., Bei, M., Woo, I., Satokata, I. & Maas, R. Msx1 controls inductive signaling in mammalian tooth morphogenesis. Development (1996).

95 [467] Lin, C.-M., Jiang, T. X., Widelitz, R. B. & Chuong, C.-M. Molecular signaling in feather morphogenesis. Current Opinion in Cell Biology (2006).

[468] Alibardi, L. & Minelli, D. Sites of cell proliferation during scute morphogenesis in turtle and alligator are diﬀerent from those of lepidosaurian scales. Acta Zoologica (2014).

[469] Cho, S.-W. et al. Interactions between shh, sostdc1 and wnt signaling and a new feedback loop for spatial patterning of the teeth. Development (2011).

[470] Devenport, D. & Fuchs, E. Planar polarization in embryonic epidermis orchestrates global asymmetric morphogenesis of hair follicles. Nat Cell Biol. (2008).

[471] Renvois´e,E. et al. Mechanical constraint from growing jaw facilitates mammalian dental diversity. Proceedings of the National Academy of Science (2017).

[472] Huelsken, J., Vogel, R., Erdmann, B., Cotsarelis, G. & Birchmeier, W. beta-catenin controls hair follicle morphogenesis and stem cell diﬀerentiation in the skin. Cell (2001).

[473] Kishimoto, J., Burgeson, R. E. & Morgan, B. A. Wnt signaling maintains the hair-inducing activity of the dermal papilla. Genes and Development (2000).

[474] N¨arhi,E., Katja, Birchmeier, W., Taketo, M. M., Mikkola, M. L. & Thesleﬀ, I. Sustained epithelial beta-catenin activity induces precocious hair development but disrupts hair follicle down-growth and hair shaft formation. Development (2008).

[475] J¨arvinen,E. et al. Continuous tooth generation in mouse is induced by activated epithelial wnt7beta-catenin signaling. Proceedings of the National Academy of Science (2006).

[476] Liu, T. et al. Wnt/beta-catenin signaling directs multiple stages of tooth morphogenesis. Developmental Biology (2008).

[477] Kratochwil, K., Dull, M., Farinas, I., Galceran, J. & Grosschedl, R. Lef1 expression is activated by bmp-4 and regulates inductive tissue interactions in tooth and hair development. Genes and Development (1996).

[478] Thesleﬀ, I. & Hurmerinta, K. Tissue interactions in tooth development. Diﬀerentiation (1981).

[479] Dassule, H. R., Lewis, P., Bei, M., Maas, R. & McMahon, A. P. Sonic hedgehog regulates growth and morphogenesis of the tooth. Development (2000).

[480] Cobourne, M., Hardcastle, Z. & Sharpe, P. Sonic hedgehog regulates epithelial proliferation and cell survival in the developing tooth germ. J. Dent. Res. (2001).

[481] Nohno, T. et al. Involvement of the sonic hedgehog gene in chick feather formation. Biochem. Biophys. Res. Commun. (1995).

[482] du Cross, D. L. Fibroblast growth factor and epidermal growth factor in hair development. Journal of investigative Dermatology (1993).

[483] Schmidt-Ullrich, R. & Paus, R. Molecular principles of hair follicle induction and morphogenesis. BioEssays (2005).

[484] Botchkarev, V. A. et al. Modulation of bmp signaling by noggin is required for induction of the secondary (nontylotrich) hair follicles. Journal of Investigative Dermatology (2002).

[485] Botchkarev, V. A. et al. Noggin is a mesenchymally derived stimulator of hair-follicle induction. Nature Cell Biology (1999).

[486] Tucker, A. S., Al Khamis, A. & Sharpe, P. T. Interactions between bmp-4 and msx-1 act to restrict gene expression to odontogenic mesenchyme. Developmental Dynamics (1998).

[487] Vainio, S., Karavanova, I., Jowett, A. & Thesleﬀ, I. Identiﬁcation of bmp-4 as a signal mediating secondary induction between epithelial and mesnechymal tissue during early tooth development. Cell (1993).

[488] Noramly, S. & Morgan, B. Bmps mediate lateral inhibition at successive stages in feather tract development. Development (1998).

[489] Zhang, M., Brancaccio, A., Weiner, L., Missero, C. & Brissette, J. L. Ectodysplasin regulates pattern formation in the mammalian hair coat. Genesis (2003).

96 [490] Charles, C. et al. Distinct impacts of eda and edar loss of function on the mouse dentition. Plos One (2009).

[491] Crampin, E. Pattern formation in reaction-diﬀusion models with nonuniform domain growth. Bulletin of Mathematical Biology (2002).

[492] Newman, S. A. & Bhat, R. Activator-inhibitor dynamics of vertebrate limb pattern formation. Birth Defects Research. Part C (2007).

[493] Menshykau, D., Kraemer, C. & Iber, D. Branch mode selection during early lung development. Plos Compu- tational Biology (2012).

[494] Menshykau, D., Blanc, P., Unal, E., Sapin, V. & Iber, D. An interplay of geometry and signaling enables robust lung branching morphogenesis. Development (2014).

[495] Alibardi, L. Proliferation in the epidermis of chelonians and growth of the horny scutes. Journal of Morphology (2005).

[496] Raﬀ, R. A. & Wray, G. A. Heterochrony: Developmental mechanisms and evolutionary results. J. evol. Biol. (1989).

[497] Hamada, H., Meno, C., Watanabe, D. & Saijoh, Y. Establishment of vertebrate left-right asymmetry. Nature Reviews (2002).

[498] Gabel, C. V., Antonie, F., Chuang, C.-F., Samuel, A. D. & Chang, C. Distinct cellular and molecular mechanisms mediate initial axon development and adult-stage axon regeneration in c.elegans. Development (2008).

[499] Simon, A. & Tanaka, E. M. Limb regeneration. Wileys Interdisciplinary Reviews (2013).

[500] Grill, C. P. & Moore, A. J. Eﬀects of a larval antipredator response and larval diet on adult phenotype in an aposematic ladybird beetle. Oecologica (1998).

[501] Moczek, A. P. Horn polyphenism in the beetle onthophagus taurus: larval diet quality and plasticity in parental investment determine adult body size and male horn morphology. Behavioral Ecology (1998).

[502] Boggs, C. L. & Freeman, K. D. Larval food limitation in butterflies: effects on adult resource allocation and fitness. Oecological (2005).

[503] Davis, A. K. Intraspeciﬁc variation in wing colour is related to larval energy reserves in monarch butterﬂies (danaus plexippus). Physiological Entomology (2014).

[504] Zangerl, R. The turtle shell. In: Biology of the Reptilia. Vol.I Morphology A (Academic Press, London, 1969).

[505] Krieg, M. et al. Tensile forces govern germ-layer organization in zebraﬁsh. Nature Cell (2008).

[506] Salbreux, G., Charras, G. & Paluch, E. Actin cortex mechanics and cellular morphogenesis. Trends in Cell Biology (2012).

[507] Rhen, T. & Lang, J. Temperature during embryonic and juvenile development inﬂuences growth in hatchling snapping turtles, chelydra serpentina. Journal of Thermal Biology (1999).

[508] Moore, A. et al. Eﬀects of temperature shift on cell cycle, apoptosis and nucleotide pools in cho cell batch cultures. Cytotechnology (1997).

[509] Watanabe, I. & Okada, S. Eﬀects of temperature on growth rates of cultured mammalian cells (l5178y). The Journal of cell biology (1967).

[510] Yntema, C. Incubation times for eggs of the turtle chelydra serpentina (testudines: Chelydridae) at various temperatures. Herpetologica (1978).

[511] Baneyx, F. & Mujacic, M. Recombinant protein folding and misfolding in eschrichia coli. Nature Biotechnology (2004).

[512] Somero, G. N. Proteins and temperature. Annu. Rev. Physiol. (1995).

[513] Goldberg, A. Protein degradation and protection against misfolded and damaged proteins. Nature (2003).

[514] Rugh, R. & Marsland, D. A. The eﬀect of hydrostatic pressure upon the early development of the frog’s egg (rana pipiens): I. macroscopic observations. Proceedings of the American Philosophical Society (1943).

97 [515] Horner, V. L. & Wolfner, M. F. Mechanical stimulation by osmotic and hydrostatic pressure activates drosophila oocytes in vitro in a calcium-dependent manner. Developmental Biology (2008).

[516] Neidlinger-Wilke, C. et al. Regulation of gene expression in intervertebral disc cells by low and high hydrostatic pressure. Eur Spine J (2006).

[517] Myers, K. A., Rattner, J. B., Shrive, N. G. & Hart, D. A. Hydrostatic pressure sensation in cells: integration into the tensegrity model. Biochem. Cell Biol. (2007).

[518] Rivero, F. et al. The role of the cortical cytoskeleton: F-actin crosslinking proteins protect against osmotic stress, ensure cell size, cell shape and motility, and contribute to phagocytosis and development. Journal of Cell Science (1996).

[519] Acevedo, A. D., Bowser, S. S., Gerritsen, M. E. & Bizios, R. Morphological and proliferative responses of endothelial cells to hydrostatic pressures: Role of ﬁbroblast growth factor. Journal of Cellular Physiology (1993).

[520] Ozturk, S. S. & Palsson, B. O. Eﬀect of medium osmolarity on hybridoma growth, metabolism and antibody production. Biotechnology and Bioengineering (1991).

[521] Alphen, W. & B., L. Inﬂuence of osmolarity of the growth medium on the outer membrane protein pattern of escherichia coli. Journal of Bacteriology (1977).

[522] Boyd, L. M. et al. Osmolarity regulates gene expression in intervertebral disc cells determined by gene array and real-time quantitative rt-pcr. Annals of Biomedical Engineering (2005).

[523] Sugiyama, Y., Ota, Y., Hara, M. & Inoue, S. Osmotic stress up-regulates aquaporin-3 gene expression in cultured human keratinocytes. Biochimica et Biophysica Acta (2001).

[524] Brigotti, M. et al. Eﬀects of osmolarity, ions and compatible osmolytes on cell-free protein synthesis. Biochem. J. (2003).

[525] Lecuit, T., Lenne, P.-F. & Munro, E. Force generation, transmission, and integration during cell and tissue morphogenesis. Annual Review of Cell and Developmental Biology (2011).

[526] Mao, Y. et al. Diﬀerential proliferation rates generate patterns of mechanical tension that orient tissue growth. The EMBO Journal (2013).

[527] Guillot, C. & Lecuit, T. Mechanics of epihelial tissue homeostasis and morphogenesis. Science (2013).

[528] Pouille, P.-A., Ahmadi, P., Brunet, A.-C. & Frage, E. Mechanical signals trigger myosin ii redistribution and mesoderm invagination in drosophila embryos. Science Signaling (2009).

[529] Evstifeeva, A., Kremnyov, S. & Beloussov, L. Changes in topology and geometry of the embryonic epithelium of xenopus during relaxation of mechanical tension. Russian Journal of Developmental Biology (2010).

[530] Shraiman, B. Mechanical feedback as a possible regulator of tissue growth. Proceedings of the National Academy of Science (2005).

[531] Puliaﬁto, A. et al. Collective and single cell behavior in epithelial contact inhibition. Proceedings of the National Academy of Science (2011).

[532] Abercrombie, M. Contact inhibition in tissue culture. In Vitro (1970).

[533] Chiquet, M., Matthisson, M., Koch, M., Tannheimer, M. & Chiquet-Ehrismann, R. Regulation of extracellular matrix synthesis by mechanical stress. Biochem. Cell Biol. (1996).

[534] Swartz, M., Tschumperlin, D., Kamm, R. & Drazen, J. Mechanical stress is communicated between diﬀerent cell types to elicit matrix remodeling. Proceedings of the National Academy of Science (2001).

[535] Thorpe, S. et al. Dynamic compression can inhibit chondrogenesis of mesenchymal stem cells. Biochemical and Biophysical Communications (2008).

[536] Lee, C. H. et al. Nanofiber alignment and direction of mechanical strain affect the ecm production of human acl fibroblast. Biomaterials (2005).

[537] Brun-Usan, M., Marin-Riera, M. & Salazar-Ciudad, I. On the eﬀect of phenotypic dimensionality on adaptation and optimality. Journal of Evolutionary Biology (2014).

98 [538] Adamska, M. et al. The evolutionary origin of hedgehog proteins. Current Biology (2007).

[539] Adamska, M. et al. Wnt and tgf-beta expression in the sponge amphimedon queenslandica and the origin of metazoan embryonic patterning. Plos One (2007).

[540] Bertrand, S., Iwerna, T. & Escriva, H. Fgf signaling emerged concomitantly with the origin of eumetazoans. Mol.Biol.Evol. (2013).

[541] Croce, J. C. & McClay, D. R. Evolution of the wnt pathways. Methods Mol. Biol. (2008).

[542] Davis, G. K. & Patel, N. H. Short, long and beyond: Molecular and embryonal approaches to insect segmentation. Annu. Rev. Entomol. (2002).

[543] Sucena, E., Vanderberghe, K., Zhurov, V. & Grbic, M. Reversion of developmental mode in insects: evolution from long germband to short germband in the polyembryonic wasp macrocentrus cingulum brischke. Evolution and Development (2014).

[544] Schierenberg, E. Embryological variation during nematode development. Wormbook (2006).

[545] Elinson, R. P. & del Pino, E. M. Developmental diversity of amphibians. Wiley Interdisciplinary Reviews: Developmental Biology (2012).

[546] Milinkovitch, M. C. Escaping the mouse trap: the selection of new evodevo model species. Journal of Experi- mental Zoology (Mol. Dev. Evol.) (2007).

[547] Jenner, R. A. & Wills, M. A. The choice of model organisms in evo-devo. Nature Reviews (2007).

[548] Urdy, S., Wilson, L. A., Haug, J. T. & Sanchez-Villagra, M. R. On the unique perspective of paleontology in the study of developmental evolution and biases. Biological Theory (2013).

[549] Scheyer, T. M., Werneburg, I., Mitgutsch, C., Delﬁno, M. & Sanchez-Villagra, M. R. Three Ways to Tackle the Turtle: Integrating Fossils, Comparative Embryology , and Microanatomy. In: Morphology and Evolution of Turtles. Vertebrate Paleobiology and Paleoanthropology (Springer, Dordrecht, 2012).

[550] Duboule, D. Temporal colinearity and the phylotypic progression: a basis for the stability of a vertebrate bauplan and the evolution of morphologies through heterochrony. Development (1994).

[551] von Baer, K. Ueber Entwicklungsgeschichte der Thiere: Beobachtung and Reﬂektion (1828).

[552] Hall, B. Phylotypic stage or phantom, is there a highly conserved embryonic stage in vertebrates? Trends Ecol. Evol. (1997).

[553] Raﬀ, R. The Shape of Life: Genes, Development and the Evolution of Animal Form (Univ. Chicago Press, 1996).

[554] Kalinka, A. T. et al. Gene expression divergence recapitulates the developmental hourglass model. Nature (2010).

[555] Irie, N. & Kuratani, S. The developmental hourglass model: a predictor of the basic body plan? Development (2014).

[556] Bininda-Emonds, O., Jeﬀery, J. & Richardson, M. Inverting the hourglass: quantitative evidence against the phylotypic stage in vertebrate development. Proceedings of the Royal Society of London B (2003).

[557] Huang, S., Eichler, G., Bear-Yam, Y. & Ingber, D. E. Cell fates as high-dimensional attractor states of a complex gene regulatory network. Physical Review Letters (2005).

[558] Huang, S. Multistability and multicellularity: Cell fates as high-dimensional attractors of gene regulatory networks. Computational Systems Biology (2005).