J. Anat. (2014) 225, pp doi: /joa Generating anatomical variation through mutations in networks implications for evolution

Transcription

1 Journal of Anatomy J. Anat. (2014) 225, pp doi: /joa Generating anatomical variation through mutations in networks implications for evolution Jonathan Bard Department of Physiology, Anatomy & Genetics, University of Oxford, Oxford, UK Abstract Genetic mutation leads to anatomical variation only indirectly because many proteins involved in generating anatomical structures in embryos operate cooperatively within molecular networks. These include generegulatory or control networks (CNs) for timing, signaling and patterning together with the process networks (PNs) for proliferation, apoptosis, differentiation and morphogenesis that they control. This paper argues that anatomical variation is achieved through a two-stage process: mutation alters the outputs of CNs and perhaps the proliferation network, and such changed outputs alter the ways that PNs construct tissues. This systemsbiology approach has several implications: first, because networks contain many cooperating proteins, they amplify the effects of genetic variation so enabling mutation to generate a wider range of phenotypes than a single changed protein acting alone could. Second, this amplification helps explain how novel phenotypes can be produced relatively rapidly. Third, because even organisms with novel anatomical phenotypes derive from variants in standard networks, there is no genetic barrier to their producing viable offspring. This approach also clarifies a terminological difficulty: classical evolutionary genetics views genes in terms of phenotype heritability rather than as DNA sequences. This paper suggests that the molecular phenotype of the classical concept of a gene is often a protein network, with a mutation leading to an alteration in that network s dynamics. Key words: anatomical variation; evolution; molecular networks; mutation; systems biology; variation. Introduction Anatomical variation is, as Darwin (1859) realized, the basis of evolution because it provides the options on which selection acts. Its origins remain hard to explain, but there were two early perspectives: Lamarck (1809, see Gould, 2001) assumed that organisms have innate tendencies to become complex and to adapt, while Huxley (1857) noted that change usually started during embryogenesis. We now know that it derives from the large amounts of mutation within a population (1000 genome consortium, 2010), and this paper explores how. It starts from the view that many proteins involved in tissue construction operate within gene-regulatory networks (Wilkins, 2007a,b; Newman et al. 2009; Davidson, 2010; Peter & Davidson, 2011), probably better called control networks (CNs), and that mutation tinkers (Jacob, 1977) with their functional output. The contention here is that Correspondence Jonathan Bard, Department of Physiology, Anatomy & Genetics, University of Oxford, South Parks Road, Oxford OX1 3QX, UK. E: j.bard@ed.ac.uk Accepted for publication 30 April 2014 Article published online 17 June 2014 the tinkered outputs of CNs change the ways in which the downstream process networks (PNs) for morphogenesis, differentiation, proliferation and apoptosis operate during embryogenesis, and that it is these changes that themselves in principle can be subject to mutational changes that produce novel anatomical variants. This view, that evolutionary change is rooted in the changed properties of protein networks, fits within the general framework of systems biology. This perspective means that, while one can formulate biological concepts at increasing levels of complexity from DNA sequences to whole organisms and even populations and eco-systems (Noble, 2012), the key level for developmental anatomy is that of the protein network (Bard, 2013). Although Zhu et al. (2010) have studied how parameter modulation in network equations can mimic change in limb formation, relatively little attention has been paid to how CNs control developmental anatomy or what effects mutation in such networks might have in initiating variants. The aim of this paper is to investigate the general implications of network mutation for heritable change in animal anatomy. It starts by providing a systems context for tissue formation in embryos; it then looks at examples where one can understand which changes in the developmental networks of a last common ancestor have led to evolutionary change in related organisms. This is followed by an analysis

2 124 Generating anatomical variants, J. Bard Table 1 Some major networks whose outputs drive development (from Bard, 2013). Control Networks (CNs) Processes Networks (PNs) Signaling pathways (gene names) ERK/MAPK FGF JAK/STAT Notch-delta SHH SMAD TGFb VEGF Wnt Patterning pathways HOX patterning RTK patterning Notch oscillator system Signalling gradients (e.g. SHH) Differentiation to Haematopoiesis lineage erythroid lineage lymphocyte lineage myeloid lineage epithelium ep-mesenchyme mesenchyme chondrocyte fibroblast muscle osteoblast mesothelium neuron neuron-support cell pigment-producing cell Morphogenesis boundary formation (eph-ephrin) epithelium branching folding migration rearrangement mesenchyme adhesion migration Proliferation Cyclin + down-stream events Apoptosis caspase, fas cellular apoptosis CN, control network; PN, process network. of the effect of mutation on networks. The discussion focuses on the evolutionary implications of this approach. Systems developmental biology This is based on the approach that CNs and PNs drive anatomical development, with the former controlling the latter (Davidson, 2010; Bard, 2011, 2013). A diagram showing how these systems co-operate is given in Fig. 1, types of network are given in Table 1 and an example of a CN is given in Fig. 2; all core information about development is available in Gilbert (2013) and further networks are available at Some control networks (often as directed by further CNs that regulate timing) output the signals that initiate simple change, while others output the gradients of signals that direct patterning (here, different parts of the responding tissue do different things, e.g. the hox coding that patterns anterior posterior axes in vertebrates and insects; Gilbert, 2013). These signals attach to receptors present on the cells of the responding tissue, and ligand-receptor binding activates a signal-response CN (e.g. Fig. 2) whose output is a nuclear activator that binds to a transcription complex and causes it to initiate a new round of gene expression that may in turn activate a PN. This activation may also lead to the formation of whole tissue modules (see below). The Fig. 1 A systems view of how anatomical change takes place in embryos: as a result of signaling, PNs are activated in a competent tissue and these processes generate change. The Adjacent tissue box represents the effect of the local environment, particularly on morphogenesis.

3 Generating anatomical variants, J. Bard 125 Fig. 2 An example of a CN: the EGF signaling pathway includes > 60 proteins and has several alternative pathways (from www. sabiosciences.com/pathwaycentral.php, where many other networks are also shown). various types of CNs thus underpin all changes to embryo phenotypes. The detailed expression that results from activating a standard signaling pathway depends of course on the competence of the cell [e.g. which transcription factors (TFs) are in place] and this derives from the cell s lineage. In vertebrate embryos, for example, the sonic hedgehog signal (Shh) secreted by the notochord induces adjacent somites to form sclerotome and overlying neural tube to form neurons (Gilbert, 2013). While much is known about such signaling networks (Fig. 2; Table 1), less is known about the patterning and timing networks. Process networks are activated by CNs and represent the executive arm of the genome: they implement phenotypic functions that include proliferation, apoptosis, differentiation to a new cell type and morphogenetic activity (movement, adhesion, etc.; Table 1). PNs are used ubiquitously across embryos, and their DNA sequences can, to use a computational metaphor, be considered as tunable genomic modules or subroutines. The differentiation, apoptosis and proliferation PNs are cell-autonomous, while those for morphogenesis often involve groups of cooperating cells (e.g. mesenchymal-condensation formation and epithelial folding), and their activity may well be constrained by the local environment (represented in Fig. 1 by the Adjacent tissue box). We now know the constituents and interactions of many networks (Fig. 2), but still lack the rate constants for their interactions, we also do not know why one internal pathway rather than another might be used, or how a network (e.g. for controlling microfilament activity) might be made tissue-specific. Tissue modules are anatomical features that form as a whole following CN activity. Typical examples are bones, joints, eyes, epithelial tubule bifurcations, insect wings, digits and teeth. What seems to happen here is that activation of a TF mobilizes a set of networks that together produce a tissue module, and this module may be used several or many times, albeit that it is often context-tuned (e.g. teeth

4 126 Generating anatomical variants, J. Bard and long bones). The groups of networks required to make modules can be seen as sets of context-dependent genomic subroutines. Although this high-level view pushes much of the complexity of development to the details of the networks, it has three advantages: it is parsimonious (there are relatively few networks as compared with proteins); it naturally links genomic and phenotypic activity; and it provides a simple conceptual framework for analysing the development of tissues. Evolutionary changes This section aims to identify the types of networks that underpin animal evolutionary change. The examples have been chosen for two reasons: first, they demonstrate how variants in a last common ancestor would lead to the differences seen in homologous tissues in related contemporary organisms; second, the data illustrate the nature of the network, and in some cases the molecular changes that underpin variation. Body size This is the most common difference within clades. Although the dwarf gecko and the Solomon Islands skink, for instance, have very similar proportions, the former is ~16 mm long while the latter is ~800 mm (Fleming et al. 2009; Hagen et al. 2012), a difference reflecting about five or six cell divisions. Apart from length, the other obvious morphological difference is that the former is brown while the latter is green, both with dark-brown speckles. From a developmental perspective, the major difference is clearly in when and how the body-wide growth networks operate. A set of mutations in the timing CN and in the proliferation and pigmentation PNs would clearly enable both to come from a common ancestor. Hindlimb diminution in cetaceans At 24 days of development, the early dolphin hindlimb bud is much like its forelimb bud (Bejder & Hall, 2002). At 48 days, when digits are forming on the forelimb, the hindlimb bud has virtually disappeared. It seems that the patterning CN in the hind limb bud has ceased all activity between 24 and 48 days of development; the only limb bones present are those that were forming in that early limb bud (the hipbone and femur), and even they remain vestigial. It is not clear whether the external limb bud has apoptosed or has remained, but is too small to be noticed in the adult. Variation in digit number Although most vertebrates have five digits, birds have three (forelimb) or four (hindlimb). Digit number is primarily controlled by a SHH gradient set up by the zone of polarizing activity in the medial autopod, which activates formation of digit modules in ways only partially understood, although limb bud width is one component (Towers et al. 2008). Likely variants in determining digit number across taxa are autopod size and the kinetics of the SHH signaling CN. Digit length Wings are a characteristic of birds, bats and pterosaurs, but each show different digit and phalange proportions. In pterosaurs, only the fifth digit extends; in bats, the four Autopod differences Webs Aquatic birds often have webs between their toes, reflecting the early embryonic state. Webless birds require an additional step for their development: the interdigit epithelium and mesenchyme of the hindlimb autopod are lost through activation of the apoptosis PN. In the webless chick, FGF2 signaling in the hind autopod elicits expression of the TF msx-2, which leads to BMP-activated apoptosis and the subsequent loss of interdigital tissue (Ga~nan et al. 1998). In the webbed duck, FGF2 signaling does not take place, msx-2 is not expressed and apoptosis does not occur even though BMP is present. The role of msx-2 here is to make the apoptosis pathway ready for BMP activation. Underpinning this change is the activation of an FGF2 signaling CN. Fig. 3 The skeletons of a pterodactyl, bat and bird matched in size. The drawings particularly illustrate the differential growth of the different phalanges in the forelimbs of these three animals (from Romanes, 1910).

5 Generating anatomical variants, J. Bard 127 fingers extend; in birds, digits 2 and 3 are lengthened (Fig. 3). It is clear that CN patterning networks operate differentially in the different digits and these primarily control the activity of the local proliferation network, even down to the level of individual phalanges. Ectopic digits Extra sixth digits have separately evolved in pandas (to shred bamboo leaves; Davis, 1964), in moles (to improve digging; Mitgutsch et al. 2012) and in elephants (where they presumably help spread weight; Hutchinson et al. 2011). This homoplasy has three common aspects: first, it occurs in both the fore and hind autopods, irrespective of whether it has a function in the latter region; second, it involves abnormal growth of the existing radial sesamoid bone, which is adjacent to digit one in mammals; third, this extra digit has neither joints nor a nail it is not a standard digit module. In each of these very different animals, the ectopic bone derives from the same developmental change, the activation of extensive growth in the radial sesamoid. Timing changes In placental mammals and, indeed, most tetrapods, the hindlimb buds start to form very soon after forelimb bud initiation. Marsupials such as wallabies and kangaroos do, however, show a heterochronic change: their forelimbs, which they use to migrate to the pouch in which there are the mammary glands, form much earlier than their hindlimbs. It is clear that this developmental feature reflects a variant in the timing network that controls hindlimb development in other tetrapods that is underpinned by the expression of some key patterning TFs and signals (Chew et al. 2014). that pelvic spine formation results from local Pitx1 activity; thisisatfgenewhosewidespreadexpressioninthefishis controlled by four enhancer regions. In the freshwater sticklebacks, the enhancer region for the pelvic region has been lost and the gene is not expressed there (Chan et al. 2010). As a result, the downstream module responsible for spine production is not activated. Vertebrate skin patterns There are obvious patterning differences in the hair pigmentation of cats, zebras and giraffes, and the scale pigmentation of fishes. It has long been suspected that Turing (reaction-diffusion) kinetics underlies these diverse patterns (Bard, 1977, 1981), a view strengthened by the richness of pattern types that can be generated by parameter variation and non-linear analysis (Barrio et al. 1999). Moreover, different zebra species have different striping patterns, and embryological analysis suggests that these differences derive from the developmental age when patterning is activated (Bard, 1977). Pattern variation clearly derives from heritable changes in both patterning and timing CNs. In these examples, which seem typical of animal diversity, the range of phenotypes mainly derives from variation in the signaling, patterning and timing CNs, and this in turn modulates the activity of the downstream differentiation, apoptotic and morphogenetic networks. These examples do not make it clear whether the proliferation networks are completely regulated by CNs or whether they include pathways that regulate their own output proliferation rates (see below). What is perhaps surprising is that it has proved very difficult to find an example where change has unambiguously derived from an alteration from a PN other than that for proliferation. Three non-limb examples of evolutionary change Beak size The Galapagos ground finches have broad, strong beaks for tearingatcactusrootsandeating insect larvae, while cactus finches have narrow beaks allowing them to punch holes in cactus leaves and eat the pulp. Abzhanov et al. (2006) and Mallarino et al. (2011) have shown that beak morphology is controlled by BMP4 signaling and calmodulin expression: the amount of BMP4 signaling controls beak breadth and depth, while calmodulin upregulation facilitates beak elongation. It is clear that variation in upstream patterning CNs determines the amounts of BMP4 and calmodulin expression in the anterior head, and these in turn control the growth pathways. Stickleback spines Marine stickleback fishes have dorsal and pelvic spines for protection; freshwater sticklebacks that evolved from their marine forebears some years ago lack pelvic spines and have small dorsal ones. Molecular analysis has shown How mutation affects protein networks Mutation can affect the genome in many ways (Shapiro, 2011) and, although downstream effects on network output are hard to predict quantitatively, the possibilities are clear. In homozygotes particularly, mutation will cause some networks to fail and this will probably be lethal. In many cases, however, the effects of mutation will be buffered and the output will not fall outside normal variation (Davies, 2009). A greater degree of output change can happeninatleasttwoways:first,analternativepathwaycan be activated within the network that excludes the mutated protein; and second, the network output can be altered because the mutation affects internal rate constants, and this is probably the most common effect as each protein can independently mutate. In complex sets of protein protein interactions, output is determined by slow reactions as the faster ones equilibrate rapidly (Smith & Crampin, 2004). In such cases, a change in output kinetics will only occur if a mutation leads to a slow rate constant being altered for some interaction.

6 128 Generating anatomical variants, J. Bard Control networks Data on the effect of mutation on CNs come mainly from analysing the genetic causes of congenital abnormalities and disorders resulting from gene targeting. Signaling CNs mainly work as switches that activate gene expression and often result in PN activation. Mutations in such networks, particularly in heterozygotes, lead to many disorders, both congenital and deriving from somatic mutation (e.g. cancers; for review, see Gilbert, 2013). Patterning through signal gradients is more complicated and little is known about the effects of mutation, other than for the Shh CN in the autopod that generates five digits in mammals and three or four in birds. Although we know little about how this gradient produces digit modules, or even if the Shh works alone here (de Bakker et al. 2013), there is a well-defined set of digit-number abnormalities and these clearly derive from upstream patterning by the CNs that control digit modules: in almost every case except for sesamoid digits, the extra modules have normal joints, nails and histology (Malik, 2013). We have little experimental information about other patterning systems, and theoretical analysis has mainly focused on Turing reaction-diffusion kinetics. As to timing mechanisms operating during development, our knowledge is still weak:we have some idea of the molecular nature of the clocks that run heart beat and circadian rhythms (Noble, 2011a; Anderson et al. 2013), but our knowledge of developmental clocks is limited to that underpinning somite formation (Riedel-Kruse et al. 2007) and some of the proteins involved in the transition from larva to adult in Caenorhabditis elegans (Hada et al. 2010; Horn et al. 2014). Although the complex CNs that regulate timing remain elusive, the anatomical comparisons discussed above suggest that their functional outputs can be modulated by mutation. Process networks Process networks fall into two groups: those whose activity leads to a qualitative change of state such as differentiation; and those whose output can be varied quantitatively. Differentiation seems highly sensitive to mutation: any change that altered a differentiation choice would probably be lethal (apart from pigmentation), as would incomplete differentiation (e.g. the runx2 mutation that blocks ossification; Takarada et al. 2013). Congenital abnormalities affecting such PNs seem rare, although there are many due to the effectsofmutationonproteinsthatreflectsomeaspectof the differentiated state (e.g. the collagen-associated diseases). An example of where the rate of differentiation may be altered by mutation is craniosynostosis (discussed below). The various morphogenetic networks also fall into the second group, but it is not easy to identify phenomena that highlight clear rate changes in any of them. In contrast, there is some evidence that the rates of proliferation and perhaps apoptosis PNs may be under local control: Hornbruch & Wolpert (1970), for example, have found that mitotic rates within the developing chick limb can vary over a factor of five, while Godlewski et al. (2005) have shown that environmental constraints can modulate the rates of apoptosis in the functioning gut, albeit to a lesser degree. It thus seems possible that the proliferation PNs at least are context tunable rather than being under the direct control of a CN; if so, size variation may be achieved from mutations affecting proliferation PNs. Discussion The key point made in this paper is that much of the anatomical variation that underpins evolutionary change derives from the downstream effects of mutation on the CNs for switching, patterning and timing that drive the PNs for differentiation, apoptosis, morphogenesis and proliferation. This systems-biology approach carries wider evolutionary implications for understanding how new features can be generated and how they can be inherited. These points are not easily explained by the modern evolutionary synthesis (Huxley, 1942), with its emphasis on phenotypes rather than genotypes, and on selection and population genetics rather than molecular detail. Although Fisher (1930) showed that continuous or quantitative change could be explained on the basis of interactions between independent genes, the theory was not designed to cope with the reality of complex networks. It is of course also true that systems biology cannot yet do this quantitatively either, but it does provide a molecular framework within which these problems can be discussed at a qualitative level and in a way that meshes with the complexity of the molecular data (Huang, 2012). Traits, genes and networks The classical gene was originally defined as the heritable underpinning of a phenotypic trait by Johannsen in 1909 (see Noble, 2011b; Fisher, 1930). While this top-down definition is still useful for population studies of evolutionary change, it does not mesh with the modern, bottom-up view of genetic heritability in which a gene is a DNA sequence with some function. Neither definition really helps understand the genetic basis of anatomical variation: the former is not defined in terms of the genome, while the proteins from individual genes (in the modern sense) do not, as this paper makes clear, generate anatomical features on their own: they operate within complex networks, an intermediate level of activity (Fig. 4). There is a further problem: while it is obvious how a mutated gene is inherited at the genomic level, the molecular basis of trait inheritance is rather less clear. Some of these difficulties are clarified if gene function, in the modern sense, is viewed in terms of the contribution of

7 Generating anatomical variants, J. Bard 129 Fig. 4 The ways in which the modern evolutionary synthesis and genome-based view of evolution see the levels at which change occurs. In the classical approach, the molecular phenotype is a variant network, whereas the modern view sees this as being based on a set of variant proteins that derive from genomic mutation. its expressed protein to network activity, while the molecular phenotype of a classical gene can often be seen as the network itself (it is hard to think of a general meaning for the classical gene in terms of an individual DNA sequence). This in turn implies that variant phenotypic traits are the result of modified network activity deriving from mutation in one or more of its proteins. This intermediate role of networks is clarified by the example of craniosynostosis, a disorder characterized by the premature closure of the sutures between skull bones. Normally, there is a balance between ossification at the suture margins and proliferation in the adjacent regions. In craniosynostosis of the coronal suture, the underlying cause of premature closure is the increased ossification rate of sutural mesenchyme between the frontal and parietal bones, leaving inadequate amounts of mesenchyme for proliferation (Iseki et al. 1999; Morriss-Kay & Wilkie, 2005). Genetic analysis of the disease in humans has shown that it can be generated by different classes of genes (Holmes, 2012) that include TFs (TWIST, MSX-2), receptors (FGFRs, EFNB1) and signals (EPHA4, FGF3/4). Although we do not yet understand the precise roles of these genes here, it is clear that their mutations disturb the normal differentiation/proliferation balance through their effects on these networks. It is equally important that there is no unique gene for craniosynostosis; a set of proteins is needed for normal development and a substantial mutation in any one of them can lead to disturbance of the CNs regulating differentiation. While it is hard to see that craniosynostosis itself would have any selective advantage, much of the anatomical evolution, as the examples discussed earlier illustrate, can be traced back to variants whose mutated networks did create anatomical novelties with such an advantage. Generating novel anatomical traits The possible effects of mutation on networks are to alter output rates, timing switches and module choices. Given the range of genomic variation, one can expect a spread of network outputs within a population, and this is the basis of natural variation. While each network protein is subject to variation, buffering will initially maintain outputs within fairly narrow limits. Networks are, however, amplifiers of variation (Wilkins, 2007a), and, as minor mutational changes accumulate within a population, it becomes harder for the system to buffer their increasing effects. Eventually, a protein, making what might seem an innocuous change, will have a major effect that leads to the offspring having a phenotype outside the normal range. Such naturally occurring variants are known as sports among breeders, and examples include roses with unexpected shapes (Vosman et al. 2004), spotted zebras (Bard, 1981) and of course the variants whose selective breeding led to the pigeon varieties that so interested Darwin (1859). The best experimental evidence for this comes from the classical experiments of Waddington (1953): he took an outbred population of Drosophila with normal anterior wings and posterior halteres, and followed up the very small proportion that, when their embryos were exposed to ether, switched the haltere to a hind-wing (a phenocopy of the bithorax mutant phenotype). He subjected a small group of such flies to further rounds of interbreeding and their embryos to ether exposure. After just 20 generations, a period far too short for a novel mutation to spread through the population, the resultant populations of bithorax flies were found to breed true without ether exposure Waddington had bred into the Drosophila line a novel phenotype resembling that of the four-winged, dragonflylike ancestor on the basis of extant mutation in the population. More recent work has shown that this phenotype is due to mutation in the ultrabithorax gene complex and its associated CN (Hersh et al. 2007). The implication of both Waddington s work and breeding observations is that the extent of genetic variation within a population can produce a wider range of phenotypes than is often supposed. Heritability Waddington s experiments show that abnormal phenotypes can just be extreme outliers of the spectrum of possibilities that mutation within the normal population can generate (Waddington noted that he could not do these experiments using inbred populations). This implies that, because abnormal phenotypes can come from essentially normal networks, there is no reproductive barrier to outliers mating with normal members of the population, albeit that the offspring may not have the abnormal phenotype to the same extent (Bard, 2010). In terms of allopatric and other such modes of speciation, this makes it straightforward for the essential imbalance in the spread of variants in the separated population to create organisms with novel phenotypic characters that initially remain part of the wider breeding population. A similar conclusion can be drawn from the speciation of ring species such as the

8 130 Generating anatomical variants, J. Bard greenish warbler populations that surround the Himalayas (Martins et al. 2013). The networks that underpin tissue formation can include scores of proteins (see tral.php), each of which can mutate independently but whose effects are cumulative. Changes in phenotype can thus occur far more rapidly than would be expected were mutation in a single gene alone responsible generating a new phenotype. Waddington found that only 20 generations were required to generate a novel phenotype under laboratory conditions if selection was strong enough. Analysis on the recent rapid speciation of sticklebacks and cichlid fish species flocks in the wild suggests that little over years, or about the same number of generations, is enough for dispersed subgroups of a parent species to develop anatomical novelties appropriate to their new habitats (Chan et al. 2010; Danley et al. 2012; Martins et al. 2013). Conclusions This systems-level approach discussed here suggests that normal tissue variation within a population essentially derives from the range of network outputs generated by the spread of rate constants in the interactions that drive the networks, and that these in turn derive from mutations in the proteins constituting that network. Each individual will have its own set of variant proteins and consequent tissue phenotypes, which will usually be within the normal range but may occasionally be outside it. In the latter case, one would expect to find an underlying network with an abnormal output due to one or more of its constituent proteins being mutated. There is, however, unlikely to be any block on the inheritance of that variant, even in those rare cases where the cumulative load of mutations leads to a phenotype well outside the normal range, as the Waddington experiments illustrate. There are thus several advantages in this approach to understanding the development of anatomical traits. It shows that the link between the classical and molecular views of genes is at the level of the network. It explains the basis of normal variation, and suggests how novel traits can form through mutation and still be heritable. Moreover, because a novel phenotype can be generated through changes in network properties that derive from mutation in one of several genes whose loads are additive, it explains how evolutionary change can be more rapid than is often thought possible. Acknowledgements The author is grateful to Gillian Morriss-Kay for discussion and for commenting on manuscript drafts. The author also thanks Denis Noble and other members of the Conceptual Foundations of Systems Biology Seminar in Oxford, where these ideas were first developed. References 1000 Genomes Project consortium (2010) A map of human genome variation from population-scale sequencing. Nature 467, Abzhanov A, Kuo WP, Hartmann C, et al. (2006) The calmodulin pathway and evolution of elongated beak morphology in Darwin s finches. Nature 442, Anderson G, Beischlag TV, Vinciguerra M, et al. (2013) The circadian clock circuitry and the AHR signaling pathway in physiology and pathology. Biochem Pharmacol 85, de Bakker MA, Fowler DA, den Oude K, et al. (2013) Digit loss in archosaur evolution and the interplay between selection and constraints. Nature 500, Bard JBL (1977) A unity underlying the different zebra striping patterns. J Zool 183, Bard JBL (1981) A model generating aspects of zebra and other mammalian coat patterns. J Theor Biol 93, Bard J (2010) A systems biology view of evolutionary genetics. BioEssays 2, Bard J (2011) A systems biology formulation of developmental anatomy. J Anat 218, Bard J (2013) Driving developmental and evolutionary change, a systems biology view. Prog Biophys Mol Biol 11, Barrio RA, Varea C, Aragon JL, et al. (1999) A two-dimensional numerical study of spatial pattern formation in interacting Turing systems. Bull Math Biol 61, Bejder L, Hall BK (2002) Limbs in whales and limblessness in other vertebrates, mechanisms of evolutionary and developmental transformation and loss. Evol Dev 4, Chan YF, Marks ME, Jones FC, et al. (2010) Adaptive evolution of pelvic reduction in sticklebacks by recurrent deletion of a Pitx1 enhancer. Science 327, Chew KY, Shaw G, Yu H, et al. (2014) Heterochrony in the regulation of the developing marsupial limb. Dev Dyn 243, Danley PD, Husemann M, Ding B, et al. (2012) The impact of the geologic history and paleoclimate on the diversification of East african cichlids. Int J Evol Biol 2012, Darwin C (1859) On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life. London: John Murray. Davidson EH (2010) Emerging properties of animal gene regulatory networks. Nature 468, Davies JA (2009) Regulation, necessity, and the misinterpretation of knockouts. BioEssays 31, Davis DD (1964) Skeleton. In: The Giant Panda. A Morphological Study of Evolutionary Mechanisms: Fieldiana Zoology Memoirs, Vol.3, pp Chicago: Chicago Natural History Museum. Fisher RA (1930) The Genetical Theory of Natural Selection. Oxford, UK: Clarendon Press. Fleming PA, Verburgt L, Scantlebury M, et al. (2009) Jettisoning ballast or fuel? Caudal autotomy and locomotory energetics of the Cape dwarf gecko Lygodactylus capensis (Gekkonidae). Physiol Biochem Zool 82, Ga~nan Y, Macias D, Basco RD, et al. (1998) Morphological diversity of the avian foot is related with the pattern of msx gene expression in the developing autopod. Dev Biol 196,

9 Generating anatomical variants, J. Bard 131 Gilbert SF (2013) Developmental Biology, 10th edn. Sunderland, MA: Sinauer Press. Godlewski MM, Słupecka M, Wolinski J, et al. (2005) Into the unknown the death pathways in the neonatal gut epithelium. J Physiol Pharmacol 56(Suppl 3), Gould SJ (2001) A tree grows in Paris. In: The Lying Stones of Marrakech, pp , London: Random House. Hada K, Asahina M, Hasegawa H, et al. (2010) The nuclear receptor gene nhr-25 plays multiple roles in the Caenorhabditis elegans heterochronic gene network to control the larvato-adult transition. Dev Biol 344, Hagen IJ, Donnellan SC, Bull CM (2012) Phylogeography of the prehensile-tailed skink Corucia zebrata on the Solomon Archipelago. Ecol Evol 2, Hersh BM, Nelson CE, Stoll SJ, et al. (2007) The UBX-regulated network in the haltere imaginal disc of D. melanogaster. Dev Biol 302, Holmes G (2012) The role of vertebrate models in understanding craniosynostosis. Childs Nerv Syst 28, Horn M, Geisen C, Cermak L, et al. (2014) DRE-1/FBXO11-dependent degradation of BLMP-1/BLIMP-1 governs C. elegans developmental timing and maturation. Dev Cell 28, Hornbruch A, Wolpert L (1970) Cell division in the early growth and morphogenesis of the chick limb. Nature 226, Huang S (2012) The molecular and mathematical basis of Waddington s epigenetic landscape: a framework for post-darwinian biology? BioEssays 4, Hutchinson JR, Delmer C, Miller CE, et al. (2011) From flat foot to fat foot, structure, ontogeny, function, and evolution of elephant. Science 33, Huxley TH (1857) Letter of July 7, 1857, quoted in W. E. Friedman and P. K. Dingle, Charles Darwin and the origin of plant evolutionary developmental biology. Plant Cell 23, Huxley JS (1942) Evolution, the Modern Synthesis. London: Allen & Unwin. Iseki S, Wilkie AO, Morriss-Kay GM (1999) Fgfr1 and Fgfr2 have distinct differentiation- and proliferation-related roles in the developing mouse skull vault. Development 126, Jacob F (1977) Evolution and tinkering. Science 196, Malik S (2013) Polydactyly, phenotypes, genetics and classification. Clin Genet 85B, Mallarino R, Grant PR, Grant BR, et al. (2011) Two developmental modules establish 3D beak-shape variation in Darwin s finches. Proc Natl Acad Sci 108, Martins AB, de Aguiar MA, Bar-Yam Y (2013) Evolution and stability of ring species. Proc Natl Acad Sci 110, Mitgutsch C, Richardson MK, Jimenez R, et al. (2012) Circumventing the polydactyly constraint, the mole s thumb. Biol Lett 8, Morriss-Kay GM, Wilkie AO (2005) Growth of the normal skull vault and its alteration in craniosynostosis, insights from human genetics and experimental studies. J Anat 207, Newman SA, Bhat R, Mezentseva NV (2009) Cell state switching factors and dynamical patterning modules: complementary mediators of plasticity in development and evolution. J Biosci 34, Noble D (2011a) Successes and failures in modeling heart cell electrophysiology. Heart Rhythm 8, Noble D (2011b) Neo-Darwinism, the Modern Synthesis and selfish genes, are they of use in physiology? J Physiol 589, Noble D (2012) A theory of biological relativity, no privileged level of causation. Interface Focus 2, Peter IS, Davidson EH (2011) Evolution of gene regulatory networks controlling body plan development. Cell 144, Riedel-Kruse IH, M uller C, Oates AC (2007) Synchrony dynamics during initiation, failure, and rescue of the segmentation clock. Science 317, Romanes GJR (1910) Darwin, and after Darwin. Chicago: Open Court Publishing. Shapiro J (2011) Evolution, a View from the 21st Century. Upper Saddle River, NJ: FT Press Science. Smith NP, Crampin EJ (2004) Development of models of active ion transport for whole-cell. Mol Biol 85, Takarada T, Hinoi E, Nakazato R, et al. (2013) An analysis of skeletal development in osteoblast-specific and chondrocytespecific runt-related transcription factor-2 (Runx2) knockout mice. J Bone Miner Res 28, Towers M, Mahood R, Yin Y, et al. (2008) Integration of growth and specification in chick wing digit-patterning. Nature 452, Vosman B, Visser D, van der Voort JR, et al. (2004) The establishment of essential derivation among rose varieties, using AFLP. Theor Appl Genet 109, Waddington CH (1953) Genetic assimilation of the bithorax phenotype. Evolution 10, Wilkins AS (2007a) Genetic networks as transmitting and amplifying devices for natural genetic tinkering. Novartis Found Symp 284, Wilkins AS (2007b) Between design and bricolage, genetic networks, levels of selection, and adaptive evolution. Proc Natl Acad Sci 104(Suppl 1), Zhu J, Zhang YT, Alber MS, et al. (2010) Bare bones pattern formation: a core regulatory network in varying geometries reproduces major features of vertebrate limb development and evolution. PLoS ONE 5, e doi: /journal.pone