Models for patterning primary embryonic body axes: the role of space and time

Transcription

1 This is a reformatted version of an article appeared in Seminars Cell Dev. Biol doi: /j.semcdb Models for patterning primary embryonic body axes: the role of space and time Hans Meinhardt Max Planck Institute for Developmental Biology Spemannstr. 35, D Tübingen hans.meinhardt@tuebingen.mpg.de Models for the generation and interpretation of spatial patterns are discussed. Crucial for these processes is an intimate link between self-enhancing and antagonistic reactions. For spatial patterning, long-ranging antagonistic reactions are required that restrict the self-enhancing reactions to generate organizing regions. Self-enhancement is also required for a permanent switch-like activation of genes. This self-enhancement is antagonized by the mutual repression of genes, making sure that in a particular cell only one gene of a set of possible genes become activated - a long range inhibition in the gene space. The understanding how the main body axes are initiated becomes more straightforward if the evolutionary ancestral head/brain pattern and the trunk pattern is considered separately. To activate a specific gene at particular concentration of morphogenetic gradient, observations are compatible with a systematic and time-requiring promotion from one gene to the next until the local concentration is insufficient to accomplish a further promotion. The achieved determination is stable against a fading of the morphogen, as required to allow substantial growth. Minor modifications lead to a purely time-dependent activation of genes; both mechanisms are involved to pattern the anteroposterior axis. A mutual activation of cell states that locally exclude each other accounts for many features of the segmental patterning of the trunk. A possible scenario for the evolutionary invention of segmentation is discussed that is based on a reemployment of interactions involved in asexual reproduction. Key words: Pattern formation / gene activation / segmentation / main body axes / interpretation of gradients 1 Development of a higher organism starts, as a rule with a single cell and proceeds, of course, under the control of genes. Since the genetic material is essentially the same in every cell, a central question is how the correct arrangement of differentiated cells is achieved. A most important step on the way from the fertilized egg to an adult organism is the setting up of the primary body axes, anteroposterior (AP) and mediolateral/ dorsoventral (DV). In this process, organizing regions - small nests of cells that act as sources or sinks of signalling molecules - play an important role. The Spemann organizer is a prominent example. This organizer is usually regarded as the only organizer that exists in vertebrates, raising the question how a single organizer can specify the cells along the two major body axes that are oriented perpendicular to each other. Moreover, DV patterning has to occur along the long extended AP axis. This task cannot be accomplished by a patch-like organizer directly. A local signalling source would lead to a conical positional information profile with a constant slope into all directions, which is clearly insufficient to specify the DV axis. Models will be discussed that account for the generation and alignment of the main body axes. The formation of organizing regions requires interactions in which local self-enhancement and longrange inhibition is involved, discussed in detail elsewhere [1, 2, 3]. The interaction of the self-enhancing Nodal with the antagonistically acting Lefty is an example [4, 5]. Such interactions can generate patterns in an initially homogeneous assembly of cells. For embryos that start with a large size such as the amphibian embryo, the employment of maternal determinants is an appropriate strategy. By making only a small part of the embryo competent for organizer formation, localized determinants only allow a single organizer to be formed. Maternal determinants are not required if development starts as a small nest of cells such as it is the case in mouse or chick development. In this case,

2 the self-regulatory features of pattern-forming reactions allow complete development in fragments even if the organizer is removed [6]. The formation of several embryos after early fragmentation of a chicken embryo is an example [7]. The generation of organizing regions requires communication between cells. If based on diffusion, the range of this signalling is restricted to short distances. Thus, these patterns can only be generated at small scales during early stages of development and have to be converted into a pattern of stable cell determinations by activating particular genes. The activities of these genes have to be maintained even if the evoking signals are no longer available. Complementary to models for setting up positional information for the body axes, models for the stable activation of genes under the influence of the resulting signal distributions will be discussed. It will be shown that activation of the correct gene at a particular position is a time-requiring process. 1. Axes formation in two steps: the head and the trunk Understanding of the patterning along the main body axes of vertebrates is much facilitated if it is realized that different mechanisms are involved in patterning the head and in the trunk. This is true for both the patterning along the AP and the DV axes. Several observations suggest that the AP pattern in the brain is under control of a Wnt gradient that is generated at the blastopore/marginal zone [8, 9, 10]; (reviewed in [11]). The region of forebrain formation has the largest distance to the blastopore and emerges at a low WNT concentration, suggesting that the forebrain is the default state. Thus, the first steps in the AP patterning of the brain can be regarded as an example where a morphogen gradient accomplishes posteriorization by gene activation in a concentration- and thus position-depending manner. In contrast, the AP patterning of the trunk is achieved by a sequential posterior elongation. In cells close to the blastopore but with the exemption of the organizer region, new Hox genes become activated in a sequential way, causing specification of more and more posterior structures - a time-dependent process [12]; (Durston and Zhu, this issue). Although both mechanisms overtly look very different, as shown below, modelling revealed that gene activation under the influence of a static gradient has also a strong time-dependent component, suggesting that interpretation of a gradient, e.g., in the brain and the time-dependent posteriorization in the trunk shares common elements. Pronounced differences between brain and trunk patterning also exist for the patterning of the DV axis. For the patterning of amphibian brain it is crucial that cells derived from the Spemann organizer move underneath the ectoderm, forming the prechordal plate and 2 induce neuronal tissue in the overlying ectoderm [13]. The prechordal plate is the precondition to form the midline, the dorsal-most structure from which the distance of the cells is measured, so to say, a reference line. In contrast, for midline formation of the trunk, cells near the blastopore (marginal zone) move towards the organizer (node), causing a conversion a ring perpendicular to the AP axis into a rod-like structure parallel to the AP axis. This process will be discussed further below in more details. The two very different functions of the organizer become established very early by a subdivision into a head- and a tail-organizer [14]. Frequently the Spemann-organizer is assumed to provide the positional information for organizing the AP axis [14]. In the view of the model proposed, this conclusion is partially misleading and results from the fact that most AP-markers are absent if the organizer is missing. However, most of these markers are neuronal markers that disappear if no midline is formed. Thus, the loss of anterior AP markers in the absence of the organizer is caused by a non-functional DV patterning. In this relation a very instructive set of experiments has been done by Ober and Schulte-Merker in the zebra fish [15]. By removing required maternal components, they obtained embryos reliably devoid of any organizer. To visualize the AP patterning they removed all BMP signalling, allowing in this way the expression of neuronal markers. Genes like Otx and Krox20 were expressed at nearly normal positions but in a completely radially-symmetric way, illustrating that the activation of the anterior AP genes do not require the organizer. However, as discussed further below, the organizer plays a crucial role in the AP patterning of the trunk by terminating the time-dependent posteriorization. 2. The separation of axes formation into a brain- and a trunk-part has an evolutionary justification Coelenterates are assumed to represent a basal branch of the metazoan evolutionary tree. A comparison of the gene expression patterns in the radial-symmetric freshwater polyp Hydra - a muchinvestigated Coelenterate - and homologous genes in higher organisms suggested that the body column of an ancestral sac-like creature with a single opening evolved into the brain of higher organisms; the ancestral oral-aboral pattern evolved in the head/brain APpattern [16]. Wnt and Brachyury are expressed at the tip of the hypostome and at the most posterior region in higher organisms, suggesting that, in contrast to a naïve expectation, the so-called Hydra head is the most-posterior structure, corresponding to the blastopore in higher organisms (Fig. 1). This view is sup-

3 Fig. 1: Generation of the primary body axes and a near-cartesian coordinate system. (A-C) The body pattern of an ancestral radial-symmetric organism (A), generated under control of a Wnt-driven organizer (green) at the oral opening (blastopore), is assumed to represent the ancestral AP axis, corresponding to the brain-part of the AP axis in contemporary higher organisms. The pattern of the freshwater polyp Hydra (B) is assumed to be a living fossil of this ancestral pattern. This ancestral body pattern evolved into the pattern as seen in the early vertebrate gastrula, giving rise to fore- and midbrain (Otx, blue) and the heart (Nkx2.5, pink). (D-J) The secondary DV axis requires a stripe-like line of reference along the entire AP axis. Nature found different solution for this intricate patterning problem that is intimately linked to the induction of mesoderm (pink). In vertebrates (E-G), mesoderm is induced at the most posterior position, i.e., at the oral opening that was enlarged to a huge ring. The induction of the Spemann-Organizer on this ring is necessarily connected with a symmetry break and the generation of bilaterality. The midline for the brain is generated dorsally with the movement of organizer-derived cells underneath the ectoderm (yellow) [13]. The midline of the trunk results from a ring-to-rod transformation (see Fig. 2). The simulation (G) shows a stripe-like midline that is induced by a moving patch-like signaling center. A second organizer (blue), induced but repelled by the midline, can lead to a left-right symmetry break [108]. (H-J) In contrast, in insects and in the spider, signaling from the dorsal side (yellow) represses mesoderm and midline formation. Ventrally a narrow signal with a stripe-like AP-extension induces first mesoderm and then the ventral midline that specifies the DV-patterning. The simulation (J) shows a stripe-like ventral midline (pink) that results from the repression emanating from a dorsal signaling center (green). These different mechanisms are assumed to be at the core of the DV-VD reversal in vertebrates and insects [35]. 3

4 ported by the expression patterns of several genes. Otx, a gene characteristic for the fore- and midbrain in vertebrates, is expressed in the whole body column of hydra except of the hypostome and the foot [17]. The posterior border of Otx expression, located in hydra between the tentacles and the hypostome, became an important secondary organizer in vertebrates, the midbrain-hindbrain border. In other words, the posterior end of ancestral organism was at a position that corresponds in today s animal to the midbrain/hindbrain border. This is in agreement with the expression of an Aristaless-related gene in Hydra at the tentacle zone [18] and in Xenopus in the telen- and diencephalon [19]. In this view, the aboral side of Cnidarians, e.g., the foot in hydra evolved into the most anterior part of the brain. This fits the expression of six3/6 in Nematostella [20] on the one hand and in vertebrates [21] and insects [22] on the other. Usually, heart formation in higher organisms is initiated at a very anterior position. The (anterior) hydra foot is under the control of the same master gene as the vertebrate heart, Nkx2.5, suggesting a common ancestry for these two so differently looking organs [23]. Indeed, the hydra foot acts already as a pump for circulating the gastric fluid. This scenario provides a rationale why brain formation of insects and vertebrates is under control of closely related genes [24, 25,26], although their common ancestor certainly did not had an evolved brain; these expression patterns are the relicts from an ancestral body pattern. The trunk is an evolutionary later invention and, as the rule, is also formed later during individual development of an organism. This view is supported by the observation that Hydra does not posses the 3-5 Hox gene sequence typical for trunk patterning [27]. With the insertion of the trunk, the posterior organizer realized by Wnt/Brachyury obtained an increasing distance from the original position at the posterior Otx border. The midbrain/hindbrain organizer can be regarded as a replacement that remains at the position of the original organizer. In his gastrea concept, Ernst Haeckel [28] proposed already that development of higher organisms generally proceed through a gastrula-like stage. The scenario I propose modifies this view by postulating that this gastrea stage is equivalent not to the whole body but only to the brain. 3. The generation of the two main body axes and their correct alignment as a self-organizing process Although frequently regarded as the only existing organizer, the Spemann-type organizer in vertebrates, based on the Chordin-BMP interaction, is neither the only nor, evolutionary, the earliest organizer. The setting up of the oral-aboral axis in the radially-symmetric Hydra by the Wnt-Brachyury system [29, 30] and its conservation in higher systems [31, 32] for the organization of the AP axis suggest that the WNT pathway was involved in organizing an ancestral axial system. Its self-organizing properties are well investigated in the Cnidarian system and are theoretically well understood [33]. This originally small organizer evolved in vertebrates into a large ring, the blastopore. A possible reason for this enlargement could be that this signalling system became also engaged in mesoderm formation (Fig. 1). Thus, the geometry of the early AP organizer in vertebrates is a large ring, not a localized patch as usual. This could be the reason why this ancestral organizer was often not regarded as such, in contrast to the patch-like Spemann organizer. To specify positional information in a twodimensional field, two reference lines are required to set up a near Cartesian coordinate system, so to say, an X and a Y axis. Especially for the patterning of the DV- (or mediolateral-) axis, a stripe-like array of signalling cells have to generated that extends along the entire AP axis. The generation of such an organizing line is an intricate pattern-forming problem. Stripe-like patterns can be generated if the self-enhancing component of a pattern-forming reaction saturates at high concentrations [34]. Due to this upper bound, the activation has a lower peak height but obtains a larger extension. Stripes are the preferred pattern since the activated regions are large but, nevertheless, non-activated regions are nearby into which, for instance, the inhibitor can be dumped. On its own, however, this mechanism would lead to a periodic stripe pattern as seen in the proverbial zebra stripes. The formation of a single solitary stripe requires cooperation with a second pattern-forming system that makes sure that only a single stripe can emerge. It was proposed that different mechanism for midline formation evolved (Fig. 1) [35]. In vertebrates, the secondary Spemann-type organizer, localized on the much-enlarged blastopore, initiates and elongates midline formation in two directions, towards anterior by giving rise of the prechordal plate and towards posterior due to the ring-to-rod conversion (Fig. 2). Thus, unusually, for the positional specification of a cell not the distance to the organizer but the distance to the stripe-like midline induced by the organizer, i.e., to the notochord and floor plate is decisive [36]. Since the organizer is dorsally located, the midline emerges also dorsally, has from the beginning a narrow DV extension but becomes elongated in the course of time. Since posterior elongation of the midline requires time, the reference line for DV patterning also emerges in the course of time with the same pace. This view provides a straightforward explanation why AP and DV patterning in the trunk 4

5 Fig. 2: Different modes for the ring-to-rod conversion in amphibian and chick development [36]. (A-C) The early amphibian gastrula corresponds to an ancestral body pattern (see Fig. 1). During trunk formation a sequential posteriorization takes place in cells near the blastopore by activating HOX genes (1, 2, 3..) [12]; (Durston and Zhu, this volume). Cells move towards the organizer (red arrow), leaving the most posterior zone in which the sequential posteriorization takes place and obtain therewith their final determination. Note that the DV or mediolateral determination can only take place after the corresponding part of the midline is formed; accounting for the observation that AP and DV determination is under the control of the same developmental clock [37]; (Mullins et al., this volume). Cells antipodal to the organizer (blue arrow), classically assigned as being ventral, form posterior structures since they remain longest in the zone of posterior transformation. (D-F) In the chick the ancestral sac-like arrangement became distorted to a near flat disk. The entire outer border (red) resembles the blastopore, a view supported by the early β-catenin expression there [109]. As in amphibians, the organizer forms on this blastopore. As shown by Gräper [110], cells at the left and at the right of the organizer move in a polonaise movement over the yolk, a movement that can be regarded as a partial epiboly [36]. This leads to a hair-pin like deformation of the blastopore, the primitive streak. In contrast to the notion in many textbooks, there is no anterior movement of the organizer [110]. Later the organizer (node) moves over the deformed blastopore like the handle of a zipper. Thus, in amphibians, the ring-to-rod transformation occurs in parallel with posteriorization, relative movement of the organizer and shrinkage of the blastopore. In the chick, the ring-to-rod transformation occurs first; the movement of the node starts only afterwards. 5

6 is under control of the same developmental clock [37], (Mullins et al., this issue). At a particular AP level, the DV pattern can only be generated after the corresponding part of the midline is generated (Fig. 2). This also solves some confusion concerning ventral in the early amphibian embryo. Frequently the region antipodal to the dorsal organizer is declared as ventral. Fate mapping has shown, however, that these cells form the most posterior structure [38, 39]. In terms of the model, cell antipodal to the organizer remain longest in the zone where sequential posteriorization takes place (Fig. 2 B, C). In insects and spiders the situation is completely different. An inhibition spreading from a dorsal signaling center restricts midline formation to the ventral side. The midline has from the beginning the full AP extension of the embryo but sharpens in the course of time to a narrow ventral line. The sharpening of the Dorsal transcription in Tribolium [40] is an impressive example for this theoretically predicted mode [41]. In a spider, a clump of BMP-expressing cells, the cumulus, move from the center of the germ disk, the blastopore, towards the periphery - a posterior-to-anterior movement - defining in this way the dorsal side. The midline proper, however, is not formed dorsally behind the moving cumulus but ventrally due to a cumulus-bmpmediated inhibition of Chordin. Chordin expression and thus midline formation is focused to a narrow ventral stripe at maximum distance from the cumulus [42]. The much discussed DV-VD reversal between vertebrates and insects [43] was proposed to have its origin in these different modes of midline formation, invented during early evolution of bilateral-symmetric body patterning [35]. In protostomes, a dorsal organizer repels the midline that appears, therefore, ventrally. In contrast, in deuterostomes, the dorsal organizer elongates the midline that appears, therefore, at the dorsal side. Most remarkably, these different modes of midline formation are intimately connected with the invention and positioning of mesoderm. In vertebrates, the mesoderm is formed around the large blastopore at a most posterior position. A reason for this enlargement could be that a sufficient number of cells can be specified as mesodermal. In contrast, in insects, mesoderm is initiated by a newly generated stripe-like signal that stretches from anterior to posterior; it precedes the formation of the ventral midline and occurs under control of a different set of genes. These basically different mechanisms for mesoderm and midline formation suggest that different mechanisms evolved for generating a bilaterally-symmetric body plan in originally more or less radial-symmetric organisms. These early differences are in contrast to the concept of an urbilaterian as a unique bilaterally-symmetric ancestor. It should also be noted that this view is very different 6 from the amphistomy concept according to which the blastopore is closed in a slit-like manner [44]. According to this view, in protostomes the mouth, in deuterostomes the anus remained open. However, in insects and spiders the blastopore is not involved in the localization of the slit at which mesoderm invaginates. In vertebrates, the blastopore does not close in a slit-like manner but resembles more a ring that shrinks due to the preferential removal of material at the site of the organizer, a closing noose. For a system that was regarded as a prototypical example for a closing slit, the onychophoran blastopore, it was recently shown that the posterior end of the slit is not coincident with the blastopore [45]. 4. The role of Wnt antagonists in the view of the proposed mode of axes formation There is a surprising diversity of Wnt antagonist, inspiring Brown and Moon to entitle a review Wnt signaling: why is everything so negative? [46]. In terms of the model, Wnt antagonists are required in several steps for proper cell specification [36]; (see also Sokol, this volume).the Wnt pathway is crucially involved in amphibians to trigger the Spemann organizer in the marginal zone. However, due to the posteriorizing nature of the WNT signal, it is necessary to rapidly switch off Wnt signaling in the organizer shortly after organizer formation. If not, organizer-derived prechordal plate cells would carry a posteriorizing activity into the region that should form the most anterior structure, the forebrain, which requires low Wnt signalling. Indeed, forebrain formation is lost if the Wnt inhibition in the organizer by Dkk-1 is non-functional [47]. Further, it is necessary to maintain Wnt-signaling restricted to the marginal zone although these cells also ingress; otherwise there would be no localized source to set up the Wnt gradient. In the fish, for instance Dkk-1 is expressed first in the organizer and extends later throughout the germ ring [48]. Thus, Dkk-1 could be involved in restricting Wnt to parts of the marginal zone in spite of the ingression. Analogously, in the ancestral hydra system Dkk seems also to be involved keeping Wnt signalling localized at the oral opening [33]. A third expected function of Wnt antagonists is to accomplish a necessary delay in the in the sequential posteriorization; this will be discussed further below. 5. From totipotent to differentiated cells: not as in Waddington s morphogenetic landscape Waddington [49] used a well-known analogy to illustrate his view how cell fates become more and more restricted on the way from the totipotent fertilized egg to a terminally-differentiated cells: a sphere rolls down in a valley; new hills within existing valleys lead to branch

7 7 points at which the sphere has to make a decision; it can only take the one or the other path. At the end of its journey the sphere will end up in one of the many valleys, one of the possible alternative stable states. This analogy, however, is misleading. It suggests that the decisions are made at unstable points; the decision for the one or the other possibility would occur essentially by chance and is essentially irreversible. Such a mechanism would not be robust. If the position of a new valley-separating hill would be slightly shifted, all cells would end up in one valley only. Cell differentiation is achieved by activation of genes under the influence of transcription factors. At such a delicate situation, a minute change in the ability of transcription factors to activate the one or the other gene would have dramatic effects. In Waddington s analogy, the majority of cells would end up at the same side; there would be no chance to obtain differently determined cells in the correct ratio. Also the situation that precedes the decision would have a decisive impact; a slight deviation form the correct initial situation would lead to a catastrophic imbalance. It is not only important that eventually differently determined cells are formed in a correct proportion; they have to emerge at the correct positions. The frequently observed ability to regenerate lost structures emphasizes that in many situations later corrections are possible, in contrast to Waddington s analogy. Nature solved this sensitivity problem by making decisions not at instable points where minute shifts would have dramatic and irreversible consequences. To use Waddington s analogy, one strategy is that for a particular determination process, all cells start in a particular valley. By morphogenetic signals, a certain fraction of cells at particular positions are shifted over the hill into an adjacent valley. In other words, a determination process can start with the activation of a default gene. A morphogenetic signal, if high enough and available for a sufficient time, can lead to the activation of a higher gene, which is usually connected with a repression of the previously-active gene. Mechanisms that allow such gene activation under the control of a graded morphogen distribution will be discussed in the next section. A second mechanism involves feedback mechanisms. To use Waddington s analogy once again, if too many spheres went to the right valley, the probability increases that other spheres go preferentially to the left. A corresponding mechanism requires that determining gene activities not only exclude each other locally (e.g., within the same cell) but mutually support each other on a longer range. This interaction allows balanced and patterned distributions of differentiated cells with strong self-regulatory capabilities. The engrailedwingless-hedgehog interaction involved in segmentation is a typical example of such an interaction; it will be discussed further below. This list is not exhaustive. In many cases, fate determination is closely connected to cell division; in each daughter cell a different set of transcription factors becomes activated - clearly a non-random decision mechanism. 6. Switch-like gene activations by positive autoregulatory feedback loops A stable switch-like activation of a single gene can result from a non-linear saturating autocatalytic feedback of a gene product on the activation of its own gene (Fig. 3) [50, 51, 2]. The condition of non-linearity is satisfied if gene activation is not accomplished by the transcription factor itself but by a dimer. This requirement is easy to understand. At low concentration the chance of finding a partner for building a dimer is low. Therefore, the normal first-order decay is dominating and gene activation will decrease further. If the morphogen signal has an additional activating influence on this gene activation, with increasing signalling strength, the rate of dimerization will increases too. From a certain threshold level onwards, the rate of the nonlinear auto-activation becomes larger than the first order decay rate; gene activation will become stronger until a saturation level is reached. In other words, under morphogen control gene activation can switch from an OFF- into an ON-state. This switch requires time since a certain concentration of the gene product has to be reached before the activation becomes irreversible. Due to the positive feedback, the activation can (but need not) remain in the ON state even if the signal is no longer available (Fig. 3). The required autoregulation can be direct but also can be realized by an inhibition of an inhibition of two transcription factors [2]. 7. Gene activation by a graded distribution: a time-requiring process According to a classical view and most clearly formulated by Wolpert in his positional information concept [52], a graded distribution of a morphogen causes an ordered activation of several genes in a concentrationdependent response. This raises the question how cells can measure the local concentrations with such a precision. At particular positions different genes should reliably be activated in adjacent cells although the differences in the morphogen concentrations are minute. An analysis of ligation experiments made by Klaus Sander in the early seventies with non-drosophila insects [53] suggested that cells do not obtain their final determination in a single step; instead, they are stepwise promoted from one state to the next until the signal concentration is insufficient to accomplish a further

8 Fig. 3: Model for stable activation of a gene under the influence of a morphogen signal. (A) A gene with a non-linear saturating feedback on its own activation. (B) Such feedback can lead two stable steady states (dg/dt = 0), one at a low and one at a high activation of the gene (red and blue circles). Whenever the gene activation is higher than the instable steady state (yellow circle), the change is positive; the activation increases further until the upper steady state is reached. In the presence of an inducing morphogen that causes an additional production of the gene activator (green curve), only the high steady state may exist (pink circle) leading to the switch from low to high activation that will be maintained even after the signal is gone [50, 2]. (C) Simulation of gene activation (red) under the control of a gradient (green). All cells that are above a certain threshold switch into the activated state and remain there even if the gradient is later removed (blue arrow). Note that cells just above the threshold level need much longer to accomplish the switch. (D) If the saturation of the self-enhancement occurs at a lower level of activation (higher κ in equation 1, BOX1), the high steady state in the absence of the signal may no longer exist; the switching remains but the activation ceases after signal removal. The activation of Monopterus under Auxin control [111] is an example. 8

9 Figure 4: Model for the space-dependent activation of several genes under the control of a morphogenetic gradient. Genes are assumed whose gene products have a positive non-linear feedback on the activation of their own gene. They compete with each other for activity (Box 1; equation 2). In a given cell, only one of the alternative genes can be fully active [51, 55, 2]. (A) Starting with activation of a default gene 1 (blue), the genes 2, 3 and 4 become activated in the course of time. Regions with sharp borders are formed. A later reduction of the morphogen (green to red distribution) remains without effect. The sequential activation of genes proceeds faster in regions of high signal concentration, which leads to the apparent wave-like movement of gene activities as observed in neural tube determination [56]. Activation of new genes occurs only in one direction (distal or posterior transformation). (B) Upon a later increase, gene activation adapts to the new level. (C) After a premature drop of morphogen level, the activation of the gene that is usually activated close to the source (yellow) may fail since the available time was insufficient; an example can be found in ref [59]. The pixel density indicates activation of particular genes. (D) The unidirectional promotion allows arbitrary growth without that cells lose their achieved determination (local gene activation indicated by the height of the bars). (E, F) At an early stage, cells could be too close to the source and the signal so high that activation of the default gene (blue) is lost (E). A fading maternal antagonist such as Cerberus [112, 113, 114] (yellow,) could delay the promotion until a sufficient extension is achieved; the activation of the default gene is maintained (F). 9

10 BOX 1: Models for gene activation under morphogen control A switch-like activation of a gene is possible if the gene product g has a non-linear self-activating feedback on the activation of its own gene [50] (Fig. 3): g t = cg2 1 + κg 2 rg + m 1 Equation 2 describes an interaction that allows the activation of several genes under the influence of a graded signal m (Fig. 4). A set of genes whose gene products feed back on the activation of their own gene compete with each other for activity; the sum term in the denominator is responsible for the local exclusion; each active gene has an inhibitory influence on the other genes, including a self-inhibition that leads to an upper bound in the self-activation. A particular gene i, i = 1...n becomes activated by the preceding gene i 1 in collaboration with the signal m: g i t = c ig 2 i + b ig i 1 m n i=1 c ig i r i g i 2 There are several possibilities to make sure that the activation of each subsequent gene requires a higher morphogen concentration. One possibility is that genes which are less sensitive for the morphogen are better in the autoregulation. In the equation above, this requires c i+1 > c i ; (i = gene number; i = 1 corresponds to the default gene; the gene with a higher number requires a higher morphogen concentration for its activation). Due to this condition, with each activation of a subsequent gene, the denominator increases. This has the consequence that the activation of each further gene requires an even higher signal concentration. Such an increasing negative feedback has been observed during neural tube patterning [56]. Under this condition the influence of the signal m in the activation of the subsequent gene, bi, can remain essentially the same. Alternatively, the self-enhancement can remain the about the same for all genes, but the activating influence of the signal becomes lower with the activation of a subsequent gene, i.e., b i+1 < b i, causing also that the signal m has to be higher to achieve a further step. In both cases, although the signal is smoothly graded, there is an all-or-nothing response in the activation of the particular genes. This stepping trough requires time. However, a too-low signal concentration cannot be compensated by a longer exposure of the responding cell. 10 step [54]. The situation may be compared with a barrel at the base of a staircase that is lifted up by a flood. The level at which the barrel comes to rest depends on the highest level of the flood. A later fading of the flood remains without effect; a later even higher flood can deposit the barrel at an even higher level. A corresponding model for gene activation [51, 55, 2] predicted the following components (Fig. 4): 1. A set of genes exists that could be active at a particular stage. The genes have a positive non-linear feedback on their own activation. The auto-activation may be realized by a mutual repression of two genes. 2. The activations of such genes are locally exclusive. For instance, in a particular cell only one gene of the set can be active. 3. Gene activation starts with a particular default gene. 4. The morphogenetic signal leads to a sequential activation of higher genes. Each further step requires a higher signal concentration and requires a certain time to activate the subsequent feedback loop. This time-consuming promotion comes to rest if the local signal concentration is insufficient to accomplish a further step. Such a mechanism has regulatory properties as experimentally observed. A once achieved gene activation can be stable against a lowering of the signal concentration (Fig. 4). This is a most important feature since embryos grow (or are subdivided in more and more cells). Thus, due to the increasing distance between a particular cell and the signaling source, the local concentration at a particular cell will decrease. Which of the ever-changing concentration a cell should measure? The answer given by this model: a cell measures the highest concentration to which it was exposed at least for a certain time. Due to time averaging and the restriction to unidirectional changes, this mechanism reduces inherently possible perturbations by noise in the gradient or in the interpretation machinery. The predicted stepwise activation has a remarkable consequence [51, 55] that is now well-known for gene activation under hedgehog control in the neural tube [56, 57]. Those genes that are eventually activated at a distance from the signaling source, i.e., the gene that requires the lowest signal concentration, become first activated close to the source at the ventral center, i.e., close to the hedgehog-signaling notochord. Later, new genes become activated at this position that quenches the activity of the previously activated genes. Gene activation appears to become shifted in a wave-like manner. This shifting reflects only the increasing time required to reach the finally stable activation; it does not depend on signaling between adjacent cells: the cells listen only to the local morphogen concentration (Fig.

11 11 4). The correct neighborhood depends solely on the interpretation of the graded signal. This has the consequence that mismatches caused by transplantation at later stages might be neither detected nor repaired. A systematic change in the threshold level was predicted in order that the stepwise activation of genes comes to rest if the morphogen concentration is insufficient to accomplish a further step [51, 55]. Possible mechanisms are outlined in BOX 1, including that the activation of a subsequent gene has a negative effect on the sensitivity. Recently such an increasing negative feedback has been described; the sensitivity for the hedgehog signal become reduced by upregulation of patched [56]. For Nodal signaling observations are available that support both a threshold model and timerequiring dose model (reviewed in [58]). According to the model proposed, these are just two sides of the same coin. A certain threshold concentration is required to activate a particular gene, but this activation is not achieved at once but requires substantial time. A too-low signal level cannot be compensated by elongating the time of exposure. Also evidence for the predicted behavior that a premature removal of the morphogen source causes that the higher genes may not be activated (Fig. 4C) has been observed, in this case experimentally achieved by an inhibition of Nodal signaling [59]. A characteristic feature of such systems is that a once obtained determination can be changed only in a unidirectional way (distal or posterior transformation). Therefore, upon transplantation from a region of high to a region of low concentration, the cells maintain their already achieved determination. In contrast, after a low-to-high transplantation, the cells change their determination according to the new level. Strong evidence for such a unidirectional promotion exists for the hindbrain [60, 61], in the commitment of CNS progenitors along the dorsoventral axis of Drosophila neuroectoderm [62], and for the response to activin signalling in the early amphibian gastrula - called there a ratchet - like behavior [63]. A stepwise posterior transformation was proposed for the AP-specification in the anterior neural tube [64]. The involvement of one gene in the activation of the subsequent gene - a crucial feature for time-dependent gene activation - has been also shown in the specification of neural crest cells [65]. The described stepwise promotion is not the only mechanism that allows the interpretation of a gradient. An alternative possibility works as follows: If the morphogen has an activating influence on a particular gene at low and an inhibitory influence at high concentrations, there is an optimal concentration for the activation of that gene. Other genes of the set have other optimal levels for activation. Although these activating profiles are smooth, sharp borders are formed due to the mutual repression of the genes. A characteristic feature of such a mechanism is that if one gene is missing due to a mutation, the expression regions of both adjacently expressed genes expands, indicating that there is no lower threshold. This is the situation in the activation of gap genes in Drosophila [66, 67]. This model for the activation of a particular gene has an interesting formal similarity to pattern formation in space. In the latter, e.g., for organizer formation, a particular region has to become activated; the remaining region has to be suppressed. In gene activation, one gene should become activated while alternative genes should be repressed. Thus, the activation of a particular gene can be regarded as a patterning process in the gene space. Reproducible development is achieved by a specific coupling between pattern formation in space and pattern formation between alternative genes. 8. Models for segmentation and the sequential activation of HOX genes in insects With a head alone, rapid swimming is impossible. Thus, the addition of a long extended trunk was a major evolutionary achievement. Characteristic for many phyla is a segmented trunk. Segments are clearly a periodic structure, manifested, for instance, by the alternation of anterior and posterior compartments. Superimposed is a sequential activation pattern of HOX genes, enabling that individual segments undergo specific developments. Both patterns are precisely in register. In contrast to Drosophila, in many species the periodic pattern emerges in the course of time during posterior outgrowth, suggesting that segmentation is based on genuine pattern-forming reactions. Already an inspection of morphological structures suggests that the periodic pattern consists of adjacent narrow stripes with a short extension along the AP but a long DV extension. To account for a periodic pattern of narrow stripes we proposed the mechanism of Mutual activation of locally exclusive cell states [68, 2]. A particular cell type needs cells of another cell type in close proximity. Both cell types provide mutual support for each other. Narrow adjacent stripes are the preferred pattern since the long common border allows an efficient mutual stabilization (Fig. 5). The shortly later discovered engrailed-wingless-hedgehog interaction in segmentation provided direct support. The gene engrailed (en), the key gene for posterior compartmental specification, is autocatalytically activated. Via the diffusible molecule hedgehog (hh), en activates the gene wingless (wg) that is crucial for the anterior compartment (reviewed in [69]). The gene cud, on which wingless-expression depends, is involved in the local exclusion of en and wg expression [70]. The gene sloppy paired contributes to the wg-autoregulation. The wg protein can reach adja-

12 Fig. 5: Mutual activation of locally exclusive cell states: a basic mechanism in segmentation. (A) Prototypical reaction scheme: two feedback loops locally exclude each other but activate each other on a longer range [68]. The engrailed-wingless-hedgehog system has turned out to be of this type [76]. (B) Such a system has pattern-forming capabilities, forming preferentially narrow stripes. (C) Small asymmetries such as a gradient from a posterior organizer can orient the emergent stripes. (D) With growth at the posterior pole (right), the periodic pattern can be elongated. Whenever the extension of a particular specification becomes too large, it will switch to the other specification (blue arrow). Thus, the most posterior cells oscillate - a feature that important for a model of somite formation (Fig. 7). cent cells via vesicle transport and is required there to stabilize en [71, 72]. As expected from the theory, the en gene activity requires an active wg gene in its neighbourhood and vice versa, although both genes are transcribed in non-overlapping regions. Such a mechanism has interesting features appropriate to describe segmentation (Fig. 5). Narrow stripes can emerge spontaneously. To obtain stripes with an orientation perpendicular to the AP axis, almost any asymmetry resulting from the posterior organizing region is sufficient. During posterior outgrowth, more stripes will be added. This can be connected with an oscillation at the posterior pole. For example, whenever the size of cells with anterior compartmental specification exceeds a critical limit, a switch to a posterior specification will occur in the posterior region, and vice versa. In other words, posterior outgrowth leads to a periodic change between two (or more) specifications at the most posterior position (Fig. 5D). The on-offon activation of the hairy gene in the Spider is an example [73]. If stripes obtain a certain width due to an overall growth, the central part of a thicker stripe can change its activation, leading to a split. For instance, in Tribolium the initial even-skipped stripes split at a later stage [74]. A small diffusion of the self-enhancing components makes stripes more robust against disintegration into patches or into a salt-and pepper pattern. In this respect it is interesting that engrailed and some other homeobox genes can be actively exchanged between cells [75]. In contrast to a simple alternation between two states...apapap... segments have an internal polarity. For this reason I proposed that, in addition to the anterior and posterior compartment, at least one additional element, termed S, must be present, such that a sequence...sap/sap/sap/... results. The region S separates one A-P pair from the next [2]. Now it is generally assumed that the AP pattern of each parasegment is founded by four differently determined cells (see [76]). Four cells also establish the parasegments in crustaceans [77]). A subdivision into three or more compartments has the consequence that only one A/P border exists per segment. This is important since the A/P border was proposed to generate a precondition to form imaginal disks and thus legs or wings [78]. 9. Formation of a precise number of different segments during terminal outgrowth Segments are not only a periodic pattern, but they obtain different specifications, known to be under control of the HOX gene (reviewed in [79]). Classical observations have shown that the number of segments is precisely controlled. The polychaet Clymenelly torquata, 12

13 Fig. 6: Model for HOX gene activation in insects in register with the compartmental specification during posterior outgrowth. Assumed are three compartmental specifications, A (red), P (green) and S (blue) that activate each other in a cyclic way. During outgrowth, a regular sequence...a,p,s,a,p,s... emerges (see also Fig. 5). With an at least threefold subdivision, each periodic unit has an intrinsic polarity and only one AP border per (para-)segment, the precondition that only one wing or leg per segment is formed [78]. To achieve Hox gene activation in register, a component is assumed to be produced in the anterior compartment that activates of next HOX gene, but its actual activation is blocked. After a switch to the posterior specification, this activation is no longer blocked but the corresponding component is no longer produced; the available component concentration allows only a single switch. Thus, a transition from one HOX gene to the next can only occur with an A-to-P transition in the compartmental specification [2]. (A-D) snapshots at subsequent time points. In (B), the system is ready to switch from A to P at the posterior position. Shortly later (C), this transition is achieved and the next HOX gene is active (blue arrows) [2]. 13

14 for instance, has 22 segments. If posterior segments are removed, regeneration occurs such that eventually 22 segments will be present independent of the number of segments removed [80]. In the leech, initially more than the final 32 segments might be formed. The surplus segments are later removed by programmed cell death [81, 82]. These observations indicate that some sort of counting mechanism exists. In an early model I proposed that the oscillation between compartmental specifications at the outgrowing posterior pole is used to drive the activation of new specifying HOX-genes [2], similar as the periodic movement of a pendulum drives the sequential advancement of the pointer of a grandfather s clock. Each full cycle leads to the advancement by one and only one unit - a stop and go mechanism. Fig. 6 shows a simulation. According to our present knowledge, this early model has to be modified and extended because particular HOX genes are known to be active in several segments. However, there are segment-specific cis-regulatory units on the chromosome, which are separated by chromosomal insulators [83,84, 85] In terms of the model, each oscillatory cycle may cause the advancement from one such insulator to the next, suggesting that a certain number of insulators have to be passed until the activation of the subsequent HOX gene takes place. 10. Somite formation: the conversion of a periodic pattern in time into a periodic pattern in space In contrast to insects, segmentation in vertebrates takes place in the mesoderm. In the hemichordate Amphioxus somites form similar as in short germ insects in a growth zone at the posterior pole [86]. In contrast, in higher vertebrates, somite formation occurs at a considerable distance from the posterior pole by a sequential separation from a non-segmented presomitic mesoderm (PSM) in an anterior-to-posterior sequence. Somite formation has been reviewed extensively [87, 88] (see also the article of Oates in this volume). Therefore, the following remarks are restricted to some historical notes in somite modeling and to inherent parallels to insect segmentation. The first model for somite formation has been proposed by Cooke and Zeeman [89]. Their clock and wave front model was name-giving for many later models. It has an interesting background. In careful experiments Jonathan Cooke found that removal of material from the ventral side of an amphibian embryo leads to shorter embryos. Thus, most remarkable, the AP, not the DV extension of the somites became smaller, although the material was removed from the ventral side. In an attempt to find an explanation, Cooke and Zeeman considered a model according to which a wave moves from anterior to posterior that initiates somite formation. In addition, an oscillation was assumed; each full activation causes for a short period the suppression of somite formation and thus the separation of one somite from the next. The frequency of the oscillation was assumed to depend on the overall size of the embryo, the smaller the embryo the higher the frequency. Assuming a constant velocity of the wave, this would lead to a regulation of A-P extension of the somites in relation to the total size of the embryo. Cooke found shortly later that the frequency of the oscillation is not size-dependent, but the clock and wave front model was born. Cooke and Zeeman did not formulate their model in a mathematical way. However, such a model can be easily provided by the toolkits of pattern-forming reactions (Fig. 7A). Already in the early eighties it became clear that the clock and wave front model was not tenable in this form. Important information in this pre-molecular time came from heat shock experiments [90]. After a short heat shock - much shorter then the time required to from a somite - about fife additional somites formed in an unperturbed manner, indicating that these somites were already determined. In later-formed somites, characteristic perturbations occurred such as partial Y-shaped splits of somites into two or to an incomplete border between two somites (Fig. 7B). Such an irregularity is frequently followed by a second irregularity that compensates partially the first, allowing that subsequent somite formation can return to normal. In my view, this is a strong indication that a component of spatial patterning is involved that attempts to maintain a certain spatial wavelength. This feature has been disregarded in many contemporary models but has found recently direct experimental support. Somite formation has been observed in cell culture in the absence of any waves [91]. To maintain the posterior oscillation in a model for somite formation I proposed that an oscillation between two states takes place in the posterior portion of the vertebrate embryo [2, 92]. In contrast to the insect case (Fig. 5) the oscillation was proposed to be no longer restricted to the posterior pole but to extend anteriorly up to region in which somite formation occurs. The mechanism explained above for segmental patterning in space (Fig. 5) also can act as an oscillator (Fig. 7). For instance, if only A cells are present, the P-cell state will get strong support, while the support of the A state by cells in the P state is missing. Thus, cells will switch from A to P and, for the same reason, later back to A. In other words, the cells will oscillate between the two states (Fig. 7C). How can the transition from a posterior oscillating into an anterior stable spatial pattern be enforced? Imagine that initially only one A region exists at the anterior end, the remaining cells are all in the P 14

15 Fig. 7: Different concepts in oscillation-driven models of somite formation. (A) A mathematically-formulated version of the clock and wave front model of Cooke and Zeeman [89]. Similar as in an infection wave, a wave is generated by a self-enhancing activator (dark green) and a non-diffusible inhibitor (red) that has a longer half life. The diffusion of the activator leads to the wave-like spread [115], showing up in the time record as an oblique line. The wave triggers the irreversible activation of a gene causing somite formation (brown) in a switch-like manner (see Fig. 3). A similar assumption was made for the oscillation; a more rapid spread of both components (blue and light green) leads to a global synchronization. A burst in the oscillation blocks gene activation for somite formation, leading to a separation between two somites. (B) Experiments of Elsdale and Pearson [90] have shown that characteristic perturbations can occur after a short heat shock. If a somite becomes too large, a partial border will be inserted or an existing border can split, suggesting that a pattern-forming mechanism is involved that has an inherent spatial wave length. (C) The mutual activation mechanism (Fig. 5) can work as an oscillator. An A region (red) next to a P region (green) is stabilized, the remaining P cells switch to the A state, forming a new stable A-region, and so on. With each full cycle, one new stable A-P pair is added. During later terminal growth, further A-P pairs can be added without oscillations, allowing a smooth transition between both modes. (D) Model of somite formation based on posterior oscillations between two cell states [2, 92]. Oscillation becomes arrested and stable patters are formed whenever the level of a gradient (pink) drops below a certain level, determining the AP level at which stable patterns emerge. The gradient is now known to be FGF [96]. The mechanism, proposed in 1982, predicted correctly many unexpected features that were later experimentally observed [93]. 15

16 BOX 2: Time-dependent gene activation For the purely time-dependent gene activation it is assumed that each active gene i produce a component pi that has an activating influence on the activation of the subsequent gene: The region in which posterior transformation can take place is restricted, for instance, due to the requirement of a high BMP/Brachyury concentration [100]. This signal s in cooperation with the signal pi produced by presently active gene drives the activation of the subsequent gene: p i t = αg i β i p i 3 The region in which posterior transformation can take place is restricted, for instance, due to the requirement of a high BMP/Brachyury concentration [100]. This signal s in cooperation with the signal pi produced by presently active gene drives the activation of the subsequent gene: g i+1 t = c i+1g 2 i+1 + bp is n i=1 c ig i r i+1 g i+1 4 Although models for gene activation under the influence of a morphogen gradient and the time-dependent gene activation share many formal similarities, the actual mechanisms are presumably more different. The first process depends on a direct activation of transcriptional feedback loops. In the second, the activation of HOX genes, on the opening of chromatin regions play a decisive role [116]. In this model, a newly activated gene suppresses completely the previously activated gene (Fig. 8). In reality, the switching off is not as abrupt. state. The direct P-neighbors of the A-cells are stabilized, while the other will switch to the alternative state. With each complete cycle one new pair of A/P specifications is added. In the course of time the region of stable periodic patterning would enlarge on the expense of regions in which cells still oscillate. To obtain a first border and to get a regular and size-regulated pattern, I assumed that a gradient with a high posterior point exist; in a region of high concentration the oscillation is maintained. Thus, formation of a stable pattern needs a low gradient level as given at a certain distance from the posterior pole. The posterior-spreading front where somite formation occurs is an indirect consequence of the sequential stabilization, not of an independent wave-forming mechanism as in the Clock and wave front mechanism. Although the posterior oscillation and spread of this 16 activities towards anterior appeared counter-intuitive, it was the only mechanism I found to be compatible with classical observations and to have some relations to the model derived for insect segmentation. It took fifteen years until essential points of this model were shown to be correct [93, 94]. It turned out that the model predicted correctly that there is 1. An out of phase oscillation between two activities (one turned out to involve components of the Wnt pathway [94]). 2. Each of these alternating activities gives rise to one half-somite. The existence of half-somites was also a prediction since this feature has been shown only two years later [95]. 3. A gradient assumed to control the posterior maintenance versus the anterior block of oscillation; it turned out that this gradient is realized by FGF [96]. Although the model predicted the essential and unexpected elements correctly, it was frustrating for me that in the seminal experimental paper the model was quoted as follows [93]: Our results also argue against models in which a reaction-diffusion mechanism patterns the rostral PSM into two states that lead to the segregation of alternating anterior and posterior somitic compartments (Meinhardt, 1986) [92]. There was no discussion that the main features were correctly predicted. In the model I proposed there is only a moderate slowing down of the waves on their way towards anterior; there is a rather abrupt separation between still oscillating cells and cells that obtained their anterior/posterior half-somitic specification (Fig. 7D). This is in contrast to what was later observed [97]. However, as expected from the model, it was found that cells stop oscillation in discrete groups, not in a continuous wavelike process that moves towards posterior [98]. This is not surprising since in nonlinear oscillations it is to be expected that cells either go through another cycle or stop. Any model according to which an oscillation can be arrested at an arbitrary phase of the cycle is molecularly not realistic. An interesting feature not yet understood is that the slowing down of the wave is connected with an increase in the amplitude [98]. It should be emphasized that the 1982-model [2] was very different from the original clock and wave front model [89]. In the latter there are no waves moving from posterior towards anterior that come to rest in the somite-forming zone; the oscillation was assumed to separate two somites, not being involved in the formation of half-somites. Thus, the original clock and wave front model predicted a completely different mechanism with little resemblance to what was later found.

17 Fig. 8: Simulation of space- and time-dependent gene activation for the regionalizing the AP axis. (A-C) Genes in the anterior part are activated by a promotion of under the influence of a gradient (red) generated by an organizer (green), as described in Fig. 4. Assumed is an activator-inhibitor system; the inhibitor (red) has a double function, keeping the organizer localized and act by its longer range as signal for gene activation. Gene activations are indicated by pixel densities. To achieve a time-dependent gene activation of the more posterior genes 4 to 8, whenever the terminal signal is high enough, each gene produces a component that activates the subsequent gene (BOX 2). The accumulation of this component is decisive when the next switch will occur. Thus, the time-based posteriorization is restricted to a region near the organizer; in the more anterior region the determination is fixed. (D) If growth would be arrested in a stage as shown in (B), the posterior transformation would proceed autonomously until the most posterior gene is activated. (E) If the time-dependent posteriorization would not work, only the anterior (head-) genes can be activated. 11. Time-dependent activation of HOX-genes Ironically, one of the reasons to predict the oscillation - the possibility to achieve a precise sequential activation of HOX genes in relation to the periodic pattern - seems no longer tenable. In mutants that have a longer oscillation frequency, the next HOX gene becomes activated after a certain time, not after a particular number of oscillations [99]. Also both processes take place in different regions of the developing embryo. New HOX-genes are activated the horseshoe-shaped region around the blastopore excluding the midline and organizer [100]. In contrast, oscillations and waves occur at the dorsal midline, i.e., in a complementary region. The sequential promotion model as proposed for the interpretation of a gradient (Fig. 4) can be extended in a straightforward manner to achieve a sequential promotion in the course of time. For the simulation Fig. 8 it was assumed that the anterior genes are activated under gradient control as described above. For the more posterior genes, each (HOX-) gene produces a component that has an activating influence on the sub- 17 sequent HOX-gene (BOX 2). It is only produced in the region where the signal from the AP-organizer is high, i.e., close to the organizer (the restriction to the nonorganizer mesoderm is ignored in this one-dimensional simulation). While models of this kind can be easily constructed, so far it is unknown how both the periodic pattern of the somites and the sequential pattern of Hox-gene activation can be brought so precisely into register. What terminates the time-dependent posteriorization? Although mainly involved in the DV patterning by midline formation, the Spemann-type organizer plays a crucial role also in the AP patterning of the trunk by being involved in the termination of the time-dependent posteriorization. This is an unusual function of an organizer since, as the rule, the strength of an organizerderived signal determines the fate of a cell. In amphibians, cells of the non-organizer mesoderm move towards the organizer and join the elongating midline (Fig. 2). Due to this movement, cells leave the zone in which posterior transformation takes place, obtaining in this way their final AP determination. As mentioned above, after removal of the organizer, the anterior neuronal genes were almost normally expressed in fishes mutant for BMP [15]. The AP pattern of the trunk, however, was completely missing. This is in agreement with the view that, due to the absence of the organizer, cells were unable to escape from the posteriorization until the most terminal determination is reached. 12. A possible scenario for the evolutionary invention of segmentation: giving up complete separation during asexual reproduction Segmentation, the metameric repetition of body parts, was certainly a fundamental evolutionary invention, for instance, to allow rapid swimming. Many proposals for its evolutionary origin has been put forward [101]. Although segmentation in insects and vertebrates share some common features, it is still controversially discussed whether both processes have a common evolutionary origin [102, 103]. Segmentation emerged and was lost several times during evolution, suggesting that segmentation was derived from mechanisms that were available early in evolution in non-segmented animals for a different purpose. Here I propose that one of the preceding mechanisms was asexual reproduction. Primitive organism are usually organized by two organizing regions, one at each pole, head/tail or aboral/oral. One way of asexual reproduction is the insertion of a new tail (T) - head (H) pair, (T/H), somewhere in the body column, converting H T pattern into a pattern H T/H T pattern, followed by a separation at the new T/H border; leading two organisms with identical polarity. Such a separa-

18 Fig. 9: A hypothetical scenario for the evolutionary invention of segmentation by modification of asexual reproduction. (A) The anteroposterior axis of ancestral organisms is organized by two organizing regions, one for anterior (head, pink, left) and one for posterior (tail, green), realized e.g., by the WNT-pathway. To reproduce asexually, a new pair of tail/head organizers becomes inserted, followed by a separation that gives rise to two complete animals. (B) For segmentation, the insertion of the Wnt-expressing regions was maintained that forms, not accidentally, the most posterior part of each parasegment. In contrast, the head signal was replaced by a signal beginning of the trunk (e.g., engrailed, red). The separation into individual animals became replaced by the generation of metameric structures that no longer separate from the parental body. The sequential insertion of Wnt signals in short germ insects [106] follows the proposed scheme. (C, D) A system at the borderline between asexual reproduction and segmentation, the strobilation of Aurelia [107]. In the polyp stage (C) a repetitive structures is generated that leads to the shedding off of juvenile Medusas (D); (figure C and D kindly supplied by Barbara Siefker). tion occurs spontaneously, for instance, in the flatworm Stenostonum incaudatum [104]. According to this view, crucial for the transition from asexual reproduction to segmentation was the replacement of the tail/head insertion by another pair that didn t led to a separation. Wnt is involved in both the formation of the posterior terminal end of animals and in the formation of the most posterior part of parasegments [105]. This suggests that the Wnt pathway was originally involved in the determination of the posterior pole. Whenever asexual reproduction should occur by insertion of a new tail-head pair, a Wnt activity has to be triggered somewhere between head and tail (Fig. 9A). The insertion of new Wnt regions during growth of short germ insects [106] is proposed to be conserved from this ancestral process. It is the expected pattern of a terminal gene that was once involved in asexual reproduction and that became subsequently adopted for segmentation (Fig. 9). For segmentation, the insertion of a new WNT activity occurs whenever sufficient space is available between the terminal Wnt expression and the already established Wnt regions [106], a feature that is explicable in a straightforward manner by our toolkit of pattern-forming reactions. In insects the most anterior part of each parasegment is specified by the engrailed gene. engrailed (see 18 [76] for review) has no function in the determination of the anterior-most structures in any organism. In vertebrates it is activated at the mid/hindbrain border, i.e., at a region that can be regarded as the most anterior part of the trunk. This suggests that the insertion of tail/head pairs for asexual reproduction became replaced by pairs of signals for end/begin of primordial trunks. The physical separation of the animals was replaced by the formation of metameric units; the integrity of the anteroposterior axis became maintained. A possible link between asexual reproduction and a segmentation-like process can be seen in the strobilation process of asexual reproduction in some Cnidaria. Aurelia aurata (Scyphozoa) forms first a polyp. Starting at the tentacle region a repetitive pattern is formed that is overtly very reminiscent of a segmented structure (Fig, 9C) [107] A substantial part of the polyp around the foot remains non-segmented. Later, these metameric units are released as individual ephyra (Fig. 9D), the juvenile form of the medusa (jellyfish). Extrapolating the situation from hydra [16] the non-segmented part around the foot is the anterior part of the organism. It is a common feature of many segmented animals that only the posterior part becomes segmented. The tentacle-bearing part at which the segmentation starts is posterior. In this case, the segmentation does