Limb positioning and initiation: an evolutionary context of pattern and formation

Abstract

Before limbs, or fins, can be patterned and grow they must be initiated. Initiation of the limb first involves designating a portion of lateral plate mesoderm along the flank as the site of the future limb. Following specification, a myriad of cellular and molecular events interact to generate a bud that will grow and form the limb. The past three decades has provided a wealth of understanding on how those events generate the limb bud and how variations in them result in different limb forms. Comparatively, much less attention has been given to the earliest steps of limb formation and what impacts altering the position and initiation of the limb have had on evolution. Here, we first review the processes and pathways involved in these two phases of limb initiation, as determined from amniote model systems. We then broaden our scope to examine how variation in the limb initiation module has contributed to biological diversity in amniotes. Finally, we review what is known about limb initiation in fish and amphibians, and consider what mechanisms are conserved across vertebrates.

Introduction

We all learn to walk before learning to run. In doing so, we make use of appendages that have been shaped by evolution over millions of years. Those appendages themselves have to be shaped during embryonic development. However, limbs must be induced before they can be shaped. Here, we review past and current advances in our understanding of this fundamental event in limb development and consider how modifications to the process of limb induction have contributed to the evolutionary diversification of limb pattern.

All gnathostomes, or vertebrates with jaws, have two sets of paired appendages. These structures have attracted considerable attention from both a developmental and an evolutionary perspective. The developing limb has proven to be a powerful model system, and its study has provided insights into cell-cell communication, developmental patterning, and cell differentiation.

At a high level, the early steps of limb development have been found to be widely conserved across amniotes (the subset of vertebrates that include reptiles, birds, and mammals). Nonetheless, variations in the limb developmental program have been elucidated, contributing to the diversity of structure and form seen across these animals.¹ Appendage formation in more basal vertebrates, such as fish and frogs, show even greater divergence in the mechanisms of limb formation. While limb initiation has received somewhat less attention than later steps of limb patterning and morphogenesis, it is critical for setting the stage for later limb development.

Initiation is a step in limb development that is occasionally overlooked in favour of pattern formation or outgrowth. Here, we first review what is known about the genes involved in limb initiation and positioning in amniotes, before broadening our view to examine examples where there has been variation in this model. In this context, it is important to bear in mind that animals with paired fins are thought to have originated at least 440 million years ago (Ma),² while crown amniotes didn’t arise until at least 332Ma.³ The myriad species which diverged during the 100 million years of paired fin and limb evolution prior to the emergence of amniotes demonstrate that the patterns seen in amniotes are not universal in all respects. Moreover, the variations in limb initiation explored later in this review remind us that model amniote systems do not represent all amniotes. Finally, the axes of the body in embryos follow different naming conventions from those in humans. For the absence of confusion, we note that the anterior-posterior axis of the embryo corresponds to the rostral-caudal axis in human anatomy.

Proximal Steps of Limb initiation in the Amniote Model

Much of what is currently known about limb initiation comes from two model systems, the mouse (Mus musculus) and the chick (Gallus gallus). Over the past 25 years, study of these two species has allowed us to begin to understand the processes and genes involved in the earliest stages of limb formation. For an in depth analysis of the various gene regulatory networks which continue through later stages of limb bud patterning and outgrowth see Zuniga and Zeller.⁴

The limb forms from the somatopleure, the layer of mesoderm which underlies the ectoderm lateral to the neural tube and somites. The early limb bud is essentially a mass of mesenchymal cells contained within an ectodermal cap. The generation of this mesenchymal mass, and hence the initiation of the limb bud itself, was initially assumed to be due to higher rates of proliferation in the presumptive limb regions compared to the adjacent flank.⁵ However, work in the past 10 years has shown that the first step in limb initiation is an early and sustained epithelial to mesenchymal transition (EMT) in the limb bud forming region, preceding a later and less efficient EMT in the flank.⁶ Prior to limb initiation, the somatopleure is a columnar epithelium, displaying typical epithelial markers such as apical F-actin and basal laminin forming a basement membrane (Figure 1A).⁶ Upon limb initiation, the somatopleure cells lose their strict alignment, and polarisation, and begin to exit the epithelium, forming a loose, disordered ball of mesenchymal cells. This process is marked by a loss of polarised expression in markers such as actin and laminin (Fig. 1B). Overexpression of RhoA, which stabilises the basement membrane and prevents an EMT,⁷ leads to a loss of limb bud formation in the chick.⁶

Figure 1 – Schematic Illustration of Limb Initiation in Amniotes –

(A) The cells of the somatopleure are initially arranged in a regular, columnar, pattern with apical Actin (green) and basal Laminin (red). (B) Tbx5 induces Fgf10 in the limb forming region which causes an epithelial to mesenchymal transition. Actin and Laminin lose their polarized expression patterns. (C) Fgf10 in the mesenchyme induces Fgf8 in the ectoderm of the limb bud, setting up a feedback loop which leads to limb outgrowth. So – Somite, IM – Intermediate mesoderm, NT – Neural tube, No – Notochord. Pink: Somatopleure, Blue: Ectoderm, Red: Actin, Green: Laminin. See text for references.

While spatially distinct, the underlying tissues of the limb forming region and the flank have similar properties prior to induction. In experiments conducted almost 100 years ago, Balinsky observed that implanting the otic vesicle, nasal placode, or pituitary gland in the flank of a developing newt embryo could induce ectopic limb formation.⁸ The important interpretation of this experiment is that the whole flank is competent to form a limb and that there must be an inductive factor which drives limb formation. This result raises two questions: what are the factors which induce limb formation, and what are the mechanisms which restrict the limbs to their correct position, despite the limb forming potential in the flank.

Many different fibroblast growth factors (Fgfs) can induce an ectopic limb in the flank of a chick embryo.^9–12 However, only two Fgfs are expressed in the right place at the right time to be involved in endogenous limb initiation. These are Fgf8, which is detected in the ectoderm of the limb field prior to limb budding,^12,13 and Fgf10, which is restricted to the prospective limb mesoderm.¹¹

Importantly, Fgf10 is expressed prior to the time Fgf8 is detectable. Moreover, while Fgf10 directly triggers downstream events of limb initiation (within 17 hours), Fgf8 only does so (in 28 hours) after it first induces Fgf10 expression.¹¹ Although mice that carry mutations in Fgf8 still initiate formation, mice which are deficient for Fgf10 fail to form limbs or even express Fgf8 in the pre-limb ectoderm.^14,15

The ability of the downstream gene Fgf8 to induce Fgf10 when ectopically applied is explained by a feedback loop that is normally establish in the subsequent step of limb patterning. Once expressed, Fgf10 in the mesoderm also induces Fgf8 in the overlying ectoderm, leading to apical ectodermal ridge (AER) formation.¹¹ In turn, this ectopically-derived Fgf8 reciprocally induces Fgf10 in the mesoderm, setting up a setting up a positive feedback loop which leads to continued proliferation and limb outgrowth (Fig. 1C).^11,16,17

From the standpoint of limb initiation, the key role of Fgf10 is not its relationship with Fgf8, but rather its effect on the process of EMT. Chick embryos with reduced Fgf10 expression show a more stabilised basement membrane and a higher proportion of epithelial cells in the limb forming region when compared to WT embryos⁶ and Fgf10 has been previously implicated in EMTs in both development and metastasis.¹⁸ Snail and Twist are expressed in a similar pattern to Tbx5/4 and Fgf10.¹⁹ Snail is a crucial upstream regulator of EMT in vertebrates.²⁰ Twist is also a regulator of EMT and is involved in mesoderm formation,²¹ but is thought to be redundant with Snail in normal oxygen levels²² and is not crucial for vertebrate development.²⁰ Manipulations in the level of Snail in the chick did not have an effect on the EMT which formed the limb, possibly due to redundancy.⁶

Activation of Fgf10 expression in the presumptive limb forming domains is thus a key event in limb initiation. Critical transcription factors acting upstream of Fgf10 are Tbx5 and Tbx4. Tbx5 and Tbx4 are T-box genes which are expressed in the forelimb and hindlimb respectively and are present in these respective fields before the limb buds themselves have formed.^23–26 Ectopic limbs induced by Fgf proteins will express Tbx5 or Tbx4/Pitx1 depending on the anterior-posterior position of the limb is induced.^24,27

If Tbx5 is deleted prior to, or during, limb initiation then no forelimb forms.^28,29 Depletion of Tbx5 in the chick results in an over-stabilised epithelium in the limb forming region,⁶ mirroring the Fgf10^−/− phenotype, as Tbx5 directly induces Fgf10 in the lateral plate mesoderm.^28,30 Tbx4 expression can rescue Tbx5^−/− phenotypes and induce a forelimb,³¹ suggesting Tbx4 and Tbx5 play mirror roles in the two limb pairs. Indeed, dominant negative Tbx5 and Tbx4 both led to limbless phenotypes and a downregulation of Fgf10 in chick.³² In mice, despite being upstream of Fgf10,³³ and in contrast to the role of Tbx5 in the forelimb, Tbx4 appears to be required for continued hindlimb outgrowth and Fgf10 expression, but not initiation.³³ Fgf10 is induced in the hindlimb at low levels without Tbx4 expression, suggesting there is a co-regulator of Fgf10 expressed in the hindlimb but not the forelimb, at least at the earliest stages of expression.³⁴ Wnts are likely suspects as canonical Wnt signalling is required to maintain Fgf10 expression in concert with Tbx5 during the limb outgrowth period.²⁸ Another potential co-factor is retinoic acid, which is also thought to act cooperatively with Tbx5 and Tbx4 in activating Fgf10.³⁵

Another difference between Tbx5 and Tbx4 is that whilst Tbx5 appears to be directly induced by Hox genes at the forelimb level^36,37 (see below for further details), Tbx4 appears to be downstream of another hindlimb gene, the OTX-related homeobox gene Pitx1.^38–42 Pitx1 binding sites are present in one of the two Tbx4 hindlimb enhancers⁴³ and ectopic Pitx1 can induce Tbx4 in the forelimb of the chick.³⁸ Tbx4 expression is reduced in Pitx1-null mice,^44,45 confirming that Tbx4 expression in the hindlimb is dependent on Pitx1 expression. However, Pitx1 alone is not sufficient to rescue Tbx5 deletion,³¹ perhaps because by the time Tbx4 expression is induced the critical period for T-box expression in limb formation³² is over.

Besides their roles in the initiation of limb development, Tbx5 and Tbx4 play additional key roles in patterning the limb bud and in imparting distinct morphologies to the fore and hindlimb.^32,46 Interestingly, the two genes are thought to have arisen from a single gene during genome duplication.⁴⁷ This ancestral gene is still present in amphioxus, a basal chordate, and the amphioxus Tbx4/5 gene can rescue forelimb formation in a Tbx5^−/− mouse.⁴⁸ For greater detail on the roles of Tbx5 and Tbx4 in limb identity and morphology, see Duboc and Logan.⁴⁹

Determining Limb Position

The tissue fated to become the limb is specified much earlier than Tbx5 expression becomes apparent, as tissue taken from chick embryos up to 20 hours before Tbx5 expression is apparent will begin to express Tbx5 during explant culture.⁵⁰ In fact, the LPM which will become the limb is specified during gastrulation.⁵¹ Intriguingly, this is the same time as the collinear expression of Hox genes is set up as cells ingress through the primitive streak.⁵² The Hox genes establish regional identity along the anterior-posterior body axis. Limb buds form at the cervical-thoracic boundary and the lumbar-sacral boundary (Fig. 2).⁵³ The correlation between limb position and axial identity suggests the involvement of Hox genes in limb positioning and, consistent with this, some Hox mutants do show subtle shifts in forelimb and hindlimb location.^54,55 The lack of a more dramatic effect, and indeed the lack of any change in limb positioning in most Hox mutants, is likely explained by redundancy. Six Hox genes from paralogous groups 4 and 5 overlap with the earliest forelimb associated Tbx5 expression.³⁶ These same Hox genes have been shown to drive Tbx5 expression by directly binding to an enhancer in intron 2 of the Tbx5 gene.³⁶ In contrast, Hox9 genes have been shown to directly repress Tbx5 by binding at the same locus.³⁷

Figure 2 – The Gene Regulatory Networks Involved in Limb Positioning in Amniotes –

The position of the limb bud is determined by gene regulatory networks which integrate signals from the somites and the lateral plate mesoderm. The above schemes show the major players in determining the position of the limb along the flank. The indications for Wnt genes are specific for experiments carried out in chick embryos. Upregulation and activation are demonstrated by a pointed arrow (→) whereas downregulation and repression are demonstrated by a blunt ended arrow (⊣). See text for references.

While the Hox genes may be responsible for the positional information placing limb bud at the proper axial level, their expression within the LPM is not sufficient to initiate limb formation. A physical barrier inserted between the somatopleure and axial structures can preclude limb formation,^56–58 implying there is additionally a necessary inductive signal emanating from the medial tissues. Fgf8 expressed in the intermediate mesoderm was once thought to be one such inductive signal in the forelimb region.^13,59 However, the finding that blocking signals from, or ablating Fgf8 in, intermediate mesoderm had no effect on limb formation^17,60 finally eliminated that hypothesis. It is now the somites, or paraxial mesoderm, which is considered to be the source of the inductive signal.³⁵

Retinoic acid (RA) is a strong contender for the inductive signal. RA has been proposed to have multiple roles in both the patterning of the LPM and in limb initiation. One of RA’s proposed roles is refining the Hox expression levels to determine the location of Tbx5 expression, as RA has been shown to be upstream of the Hox genes involved in forelimb patterning.^61–64 In the post-gastrulation embryo, Hox genes are regulated by a balance between RA expression anteriorly and Fgf8 expression posteriorly.^65–67 Changes to RA expression during gastrulation can affect axial levels of Hox expression and forelimb positioning.⁵¹ Genes which regulate RA can also have an effect on limb positioning. Cux2 is expressed in the forelimb field and directly regulates Raldh, an RA synthesis gene, and Hoxb genes⁶⁸ thus refining the position of Tbx5 expression.

Gdf11 is a secreted TGFβ family member which is involved in the trunk to tail transition. As the hindlimbs form at the trunk-tail boundary, Gdf11 is thought to play an important role in hindlimb positioning.⁷² Indeed, mutations in Gdf11 are associated with axial shifts in the position of the hindlimb.^71,72 A direct target of Gdf11 is Islet1, which is transiently expressed in the LPM before hindlimb formation. When Islet1 is inactivated, Tbx4 is downregulated and Fgf10 is not expressed,⁷⁴ suggesting it may be the primary regulator of hindlimb positioning, not Hox genes. Other evidence suggests Gdf11 activates posterior Hox genes,^69,70 possibly directly or possibly by modulating RA expression through Cyp26a1, an enzyme which inactivates RA.⁷³

Another role for RA may be as a direct regulator of limb induction. Some data suggest RA acts as a permissive factor for forelimb development, repressing pre-limb expression of Fgf8,⁶⁷ and allowing the forelimb bud to initiate.^75,76 Others argue that RA acts to directly induce limb formation. Application of a RA inhibitor at pre-limb bud stages prevents formation of a limb bud in chick embryos.⁷⁷ Similarly, no forelimbs were formed in mice where an RA synthesis gene (Raldh2) was deleted and Tbx5 and Fgf10 expression was also lost.^75,78,79 Maternal RA application was sufficient to rescue hypoplastic forelimbs as well as Tbx5 and Fgf10 expression.^79,80 A dominant negative Fgf receptor was also sufficient to rescue Raldh2^−/− embryos,⁷⁶ suggesting that RA may be antagonising Fgf8 to allow Tbx5 expression. In all of these studies, the hindlimb appeared normal. This could be because RA is required only to repress Fgf8 to permit forelimb development, but another explanation could be that there is a compensatory source of Raldh3 expression near the hindlimb of the mouse.⁸⁰

Possibly the most compelling evidence for a direct requirement for RA is through a classical barrier insertion experiment. An impermeable barrier inserted between the paraxial mesoderm and LPM in the chick was sufficient to abrogate limb initiation and Fgf10 expression, despite strong expression of Tbx5 and Tbx4³⁵. Application of RA to these embryos was sufficient to rescue both Fgf10 expression and limb formation. Inserting the barrier earlier, only a few hours after somite formation has begun, prevents Tbx5 expression completely.³⁵ Tbx5 expression can be rescued by subsequent application of RA soaked beads. Taken together, these results imply there is an early phase of RA which induces Tbx5 and then a later phase where Tbx5 and RA work in concert to induce Fgf10. The Tbx5 forelimb regulatory element³⁶ also contains RA response elements (RAREs) which directly regulate Tbx5 expression in vivo,³⁵ although this was found to be non-essential in mutation studies.⁸¹

The Tbx5 forelimb regulatory element also contains a TCF/LEF site in the mouse.³⁵ TCF and LEF are genes involved in the β-catenin signalling cascade, suggesting a role for Wnt signalling in the limb initiation process. In the chick embryo, Wnt2b is expressed in the LPM at the forelimb level and Wnt8b is expressed at the hindlimb level.⁸² Wnt2b is upstream of Tbx5 in chick.³⁰ Grafting of cells which overexpress Wnt2b or Wnt8c to the flank of a chick embryo is sufficient to induce Fgf10 in the flank and thus an ectopic limb.⁸² Both Wnt2b and Wnt8c signal through the canonical Wnt/β-catenin signalling pathway. Further, β-catenin is sufficient to induce Fgf10 and a limb in the flank of the chick embryo and antagonising β-catenin blocks limb formation.^35,74,82 LEF and TCF knockouts in the mouse lead to significantly smaller limb buds, due to failure of Fgf8 and AER induction.⁸³

Canonical Wnt signalling maintains Fgf10 expression in concert with Tbx5 during the limb outgrowth period of mice.²⁸ However, the mechanism is not fully clear. Agarwal et al.²⁸ demonstrated that Tbx5 is upstream of Wnt signalling as opposed to downstream, while other experiments using a dominant negative Tbx5 demonstrated that Wnt signalling is downstream of Tbx5.³² Dominant negative LEF did not affect Tbx5 expression initially, but Tbx5 levels did decrease after an extended period. An explanation for the delay could be that loss of Wnt2b reduced levels of Fgf10 which subsequently led to the reduction in Tbx5.³² A feedback loop may exist between Tbx5 and Wnt genes to maintain Tbx5 expression in the forelimb.

Whilst Wnt2b may play a role in chick forelimb development, Wnt2b is not expressed in the limb in mouse until digits have formed.⁸⁴ Another canonical Wnt gene involved in limb induction in the chick is Wnt3a, which mediates the induction of Fgf8 by Fgf10 in chick.^82,85 Wnt3a is not present in the AER of mice,⁸⁶ however, Wnt3 is present in the limb ectoderm^84,87 and removal of Wnt3 in the mouse limb ectoderm prevents AER formation and Fgf8 expression.⁸⁷ Thus, Wnt3 and Wnt3a appear to have taken on similar roles in the mammalian and avian lineages. It is not clear which, if either, had this activity in the ancestral form.

Variation in Limb Positioning within Amniotes

Amniotes come in a much wider variety of shapes and sizes than represented by the research above. Accordingly, there are some interesting variations where limb initiation differs from the ‘standard’ pattern described above.

In both birds and mammals, the forelimb is placed at the cervical-thoracic transition. In mammals, the vertebral formula is relatively conserved in the cervical region and so the forelimb is almost always formed at the level of the 8^th vertebra. In contrast, the bird forelimb varies tremendously in position, from the 10^th to the 25^th vertebrae. A shift in the border between Hox4 and Hox9 appears to be responsible for changing the posterior extent of Tbx5 expression and thus forelimb position.⁵¹ These spatial shifts in Hox expression are caused by temporal changes in Hox regulation, resulting in a longer period of expression of Hoxb4 and a concomitant delay in the onset of Hoxb9 expression.

The hindlimbs are located lateral to the sacral vertebrae, which are specified by Hox11 genes.⁸⁸ Overexpression of the secreted factor Gdf11 is sufficient to induce posterior Hox genes, including Hox11 paralogues, and the timing of Gdf11 expression appears to be tightly correlated with the position of the hindlimbs in tetrapods.⁶⁹ This is probably related to the role of Gdf11 in regulating RA expression. In fact, the timing of RA expression was correlated with the position of the forelimb as well, and modulating RA expression levels was sufficient to shift the expression borders of both Hox genes and Tbx5.⁵¹

As stated above, almost all mammals have seven cervical vertebrae. Any variation in length or kinematics of the neck is achieved by morphological changes to the vertebrae, but not through a change in vertebral number. There exist a few notable exceptions to this rule, these are the two-toed (Choloepus) and three-toed (Bradypus) sloths and the manatees (Trichechus) (Fig. 3). Choloepus can possess 5–8 cervical vertebrae, Bradypus 8–10, and Trichechus 6. It is thought that the slow metabolism and lifestyle of these animals allows them to avoid the extreme selection against homeotic transformations in most mammals.^89–91 However, the developmental explanation for this shift in the position of the forelimb, which lies at the cervical-thoracic boundary, is somewhat unclear. One hypothesis suggests that there has been a homeotic transformation of thoracic vertebrae to cervical, or vice versa,⁹⁰ while another suggests that there has been a decoupling of the patterning of the vertebral column (primaxial tissues) from the distal ribs and limb girdles (abaxial tissues).^92,93 The ossification patterns of the last two cervical vertebrae in the long necked Bradypus were similar to thoracic vertebrae, suggesting the distal ribs and limbs had been shifted, whilst the vertebral column remained the same.⁹² However, morphometric analysis suggested that a homeotic transformation is more likely, as the same number of morphometric ‘modules’ are present in the sloth neck, just with displaced boundaries.⁹⁴

Figure 3 – Summary of Amniote Variations in Limb Initiation and Positioning –

The phylogenetic relationships between the species discussed in this review and a summary of the changes which have occurred in the initiation of their limbs. See text for references.

Modifying Limb Pattern by Delaying Limb Formation

There has been convergent flight loss in many paleognaths, an early branching bird lineage, associated with a decrease in limb size and, in many cases, an increase in body size.⁹⁵ In the emu (Dromaius novaehollandiae), the reduction in limb size is associated with a delay in the formation and development of the limb bud.^96,97

A cardiac marker, Nkx2.5, is upregulated in the emu forelimb, possibly repressing forelimb development,⁹⁸ but this expression does not appear until after the emu limb is already delayed in its formation,⁹⁹ suggesting it is not a primary driver of the change in timing of limb formation. Nkx2.5 expressing cells also express Pax3⁹⁹ and so this could be a myoblast specific change.

Tbx5 expression has also been described as being delayed in the LPM of the emu compared to the chick, although it appeared at the expected time in the heart field and Pitx1 is expressed as expected in the hindlimb,¹⁰⁰ consistent with the possibility that emu forelimb might be initiated later than the chick. However, further analysis has demonstrated that the emu limb is initiated at the same stage as in the chick, and the EMT takes place as expected,⁹⁹ but subsequent proliferation and outgrowth is delayed. Indeed, Tbx5 expression is comparable between chick and emu and Fgf10 is also expressed at similar stages. However, the Fgf10 domain is smaller in the emu. Supplemental application of Fgf10 can rescue outgrowth of the limb, showing that the LPM is competent to form a limb at early stages.⁹⁹ This suggests that Fgf10 expression is sufficient to induce the limb but is initially at too low a level to activate the Fgf10-Fgf8 positive feedback loop and form a bud. This causes a the delay in limb formation and thereby contributes to the size reduction in the emu wing.⁹⁹

When Limbs are Not Initiated at All

Many amniote clades have secondarily lost their fore or hindlimbs. Within mammals, whales and dolphins have vestigial pelvises but lack hindlimbs. Developmentally, the hindlimb buds initiate and the AER is set up but Fgf8 expression is not maintained in the AER and the limb buds falter.¹⁰¹

Manatees also have a vestigial pelvis, which is larger on the left-hand side than the right. This phenotype is very reminiscent of the pelvic reduction seen in mice and stickleback fish deficient for Pitx1 (See below). The asymmetric morphology in these mutants is attributed to partial compensation by the related gene Pitx2 which is asymmetrically expressed on the left side of the embryo. The similarity of the rudiment of the pelvis in the manatee to these mutants suggests that reduction in Pitx1 may be implicated in the manatee.¹⁰²

The clade which has by far the highest incidence of limb loss is the squamates, encompassing lizards and snakes. Limb reduction and loss has been documented at least 62 times in 53 lineages in squamates alone.¹⁰³ Limblessness in squamates is correlated with axial extension and increased uniformity in vertebral shape,¹⁰⁴ exemplified most notably in the snakes. Within squamates, forelimb loss is more common than hindlimb loss.¹⁰⁵ All snakes have lost their pectoral skeleton¹⁰⁶ and whilst many limbless lizards and basal snakes initiate a hindlimb bud which then fails, some more derived snakes do not initiate hindlimb buds at all. Pythons have a rudimentary hindlimb. The python hindlimb bud does not form the correct signalling centres and is subsequently severely truncated, similar to the cetaceans.¹⁰⁷

Early analysis suggested that expanded, deregionalised Hox domains would preclude the correct, restricted Hox expression patterns which specify limb position.¹⁰⁷ In fact, snakes show the same collinear Hox expression patterns as seen in other amniotes.¹⁰⁸ In the corn snake (Pantherophis guttatus), a derived snake with no limb elements, the Tbx5 expression domain is expanded, with homogeneous expression throughout the somatopleure.¹⁰⁸ As the Hox genes are expressed colinearly, with regionalised expression patterns, the broad Tbx5 domain is interpreted as a decoupling of Hox patterning from the typical limb induction pathways. Whilst Tbx5 deficient mice fail to form a pectoral girdle at all,^29,100 Fgf10^−/− mice form a sternum and pectoral girdle.^15,100 These phenotypic differences between Tbx5 and Fgf10 mutants recapitulate the differences between the forelimb rudiment in limbless lizards and the complete lack of a forelimb in snakes, as well as the differences between the fore and hindlimb in basal snakes.¹⁰⁶ But, as Leal and Cohn note in their 2018 review on limb loss in snakes, vertebrates have taken many paths to limb loss and so inference from a phenotype in a model system or comparing convergent phenotypes may lead us down the wrong path.

Evolution at the enhancers of limb genes provides an attractive mechanism for modulating limb initiation. Accordingly, there has been accelerated evolution at the ZRS, a limb specific sonic hedgehog enhancer, in advanced snakes.^109,110 This accelerated evolution is only seen in derived snakes with a complete loss of hindlimb elements. Given that complete loss of Shh expression in mice or chicks results in well formed, albeit reduced, proximal limb elements,¹¹¹ it is unlikely that the loss of the ZRS in advanced snakes is responsible for their complete absence of limbs. Rather the accelerated evolution of the ZRS is likely a result of relaxed selection leading to degradation. The boa ZRS appears fully functional, further supporting the idea that there is a separate mechanism which led to limb reduction and loss and the degradation of the ZRS may be incidental.¹⁰⁹ Considering the ZRS does not play a role in limb initiation and advanced snakes do not initiate a limb bud, it is likely that the cause of limblessness in snakes has not fully been determined.

Enhancers located close to other limb initiation genes may provide insights. Tbx4 appears to be involved in development of the genital tubercle in amniotes. A common mesenchymal cell population gives rise to the hemipenes and hindlimbs in squamates.¹¹² This developmental constraint may explain why hindlimb reduction is less common than forelimb reduction. Tbx4 has a limb specific enhancer, HLEB, which is conserved throughout vertebrates,⁴³ and this enhancer drives Tbx4 expression in both the hindlimb and genital tubercle in mice.¹¹⁰ The python HLEB enhancer sequence is able to drive reporter expression in the genital tubercle and weak expression in the hindlimb of the mouse. The corn snake HLEB sequence can only drive reporter expression in the genital tubercle of the mouse.¹¹⁰ These enhancers appear to have progressively lost the ability to drive Tbx4 in the hindlimb, and could be responsible for, or at least contribute to, the loss of hindlimbs throughout snake evolution. This pattern has been seen in other genes expressed in both the hindlimb and genital tubercle.¹¹³

Fish and Fins

Zebrafish are teleost fish, and teleost fish have undergone an extra round of genome duplication beyond that of other vertebrates.¹¹⁴ They consequently have two Tbx5 paralogues, tbx5a and tbx5b, which have evolved distinct roles in teleost development.¹¹⁵ Both paralogues are involved in heart and pectoral fin development, similar to Tbx5 expression in amniotes, but they have separate roles in the development of these organs and cannot substitute for each other.¹¹⁶

Similar to the absence of Tbx5 in amniotes, tbx5a loss leads to a complete loss of pectoral fins in zebrafish.^117,118 In a wild type embryo, tbx5a expression appears as a stripe along the flank which condenses into a fin forming region (Fig. 4). In a tbx5a morpholino treated embryo, where the mRNA is present but not translated, tbx5a expression does not condense but remains dispersed and the fin bud fails to form.¹¹⁷ Further research also showed that tbx5a expressing cells migrated to form the pectoral fin bud and that knock down of tbx5a prevented cells converging.¹¹⁹ tbx5b expression is not seen in the pectoral fin itself,¹¹⁵ but tbx5b loss does result in delayed pectoral fins which are reduced in size.¹¹⁶ This suggests that tbx5b may have a later role in fin growth as opposed to initiation.

Figure 4 – There Exist Fundamental Differences in Limb Initiation Across Vertebrates –

Despite similarities in the gene regulatory networks which lead to limb formation in zebrafish and amniote models, the mechanism of initiation is quite different. Whilst the zebrafish limb forms through collective cell migration, the amniote limb forms by a prolonged EMT. Zebrafish limb initiation also requires Fgf24, whereas amniote limb initiation does not. The mechanism of limb initiation in amphibians remains unclear. The evolutionary time since the divergence of amphibians and amniotes is so vast that one cannot assume that amphibians will follow the amniote model. The lack of a clear outgroup means inference of an ancestral state of limb initiation is unwise. A: Anterior, P: Posterior, Red: Actin, Green: Laminin. See text for references.

Fgf24 is another fibroblast growth factor gene, in the same subfamily as Fgf8. Deletion of fgf24 leads to complete pectoral fin loss in zebrafish embryos.¹²⁰ It appears that fgf24 lies between Tbx5 and Fgf10 in the zebrafish limb induction cascade and is also required for Tbx5 cells to converge in the pectoral fin bud.¹²⁰ Interestingly, fgf24 does not appear to be involved in pelvic fin development.

Although there is some evidence for cell movement into the limb bud in amniotes,¹²¹ this occurs only after the limb bud has been initiated and Tbx5 expression is not associated with cell movement in amniotes. This suggests Tbx5 may have acquired a new role between zebrafish and amniotes.²⁹ It is unclear whether EMT or cell migration is the ancestral mode of limb initiation as no outgroup state is known. Tbx5 expression is seen in the pectoral fin of the two sharks studied, the dogfish (Scyliorhinus canicula)¹²² and the skate (Leucoraja erinacea),¹²³ but its mode of action is unknown. Further research in sharks is required before an ancestral mode of appendage initiation can be determined.

Tbx4 and the Loss of Pelvic Fins

The threespine stickleback (Gasterosteus aculeatus) is a fish with a broad geographic and ecological distribution. It is deemed a species complex, rather than a species, because many thousands of isolated or semi-isolated populations exist which show extreme phenotypic diversity.¹²⁴ These phenotypic differences are heritable and include the number of spines, the presence of body armour, body size and behavioural traits. Some populations have pelvic spines, modified from the pelvic fin, and others lack pelvic spines and the pelvic girdle entirely.

Whilst spined morphs form a pelvic fin bud as normal, the spine-deficient morphs never initiate a pelvic fin bud.¹²⁵ The spine-deficient morphs also do not express Tbx4 in the pelvic region, whilst the spined morphs express Tbx4 as normal. Pitx1 also shares this expression pattern.¹²⁵ A genetic mapping experiment, which crossed spined and spine-deficient fish to examine the phenotype of the progeny, found that most of the variation in spine length in the progeny mapped to the Pitx1 gene.¹²⁶ No consequential coding mutations were found in the Pitx1 gene, implying that the spine-deficient phenotype was probably a result of regulatory changes which removed the pelvic region-specific enhancer for Pitx1. Indeed, Pitx1 is expressed identically in the thymus and cranial and trunk neuromasts of both the spined and spine-deficient fish, but is absent in the pelvic region of the spine-deficient morphs.¹²⁶ Later work identified the pelvic spine specific enhancer which lies upstream of Pitx1 and is inactivated in spine-deficient lineages.¹²⁷ Sequence features of the Pitx1 locus in the stickleback genome, specifically its flexibility and susceptibility to double-stranded DNA breaks, may predispose this region to mutations.¹²⁷ This could explain why similar genetic changes underlie the loss of pelvic spines in multiple, independent spine-deficient lineages.^125–127

In the pufferfish (Takifugu rubripes), Tbx4 and Pitx1 expression are also absent in the pelvic region.¹²⁸ However, the mode of Pitx1 loss appears to be different from the sticklebacks. In fish, pelvic fins form during metamorphosis rather than during embryogenesis. In the sticklebacks, Hoxd9 expression reappears in the pelvic region during metamorphosis, whereas in the pufferfish it does not.¹²⁸ It’s hypothesised that the pufferfish have lost their hindlimbs due to a loss of peri-metamorphosis expression of posterior Hoxd9 genes,¹²⁸ whereas the sticklebacks lost the ability of Pitx1 to respond to the patterning by Hox genes.

Amphibians

Frogs, or anurans, are probably the most easily recognised group within the amphibians, and are known for their highly specialised morphology and locomotory style, as well as their drastic metamorphosis. The embryo first develops into a primarily aquatic tadpole, which uses tail-driven locomotion and requires gills for respiration and filter feeding. After this transient larval stage, the frog metamorphoses into its typically terrestrial adult form with powerful, hindlimb-driven locomotion. This life history is ancestral for extant amphibians, but is not thought to represent the ancestral state for tetrapods.¹²⁹ A major consequence of this biphasic life history is that limb buds develop just before metamorphosis.¹³⁰ Formation of the limb long after embryogenesis presents a striking contrast to early, embryonic formation of the amniote limb.

Most work on frog limb buds has been focussed on Xenopus laevis. The X. laevis limb bud is first visible during the tadpole stage, before metamorphosis, as ‘a slight concentration of mesenchyme cells dorsal and lateral to the anal tube’.¹³⁰ At this stage, embryogenesis is complete, the tadpole has begun feeding, and the somatopleure, the embryonic tissue which gives rise to the limb bud in amniotes, has already differentiated. The mechanism of limb initiation in frogs therefore differs from the ‘standard’ pattern described in amniotes. This is similar to the pelvic fins of fish, which arise during metamorphosis rather than during embryogenesis.¹²⁸

Reinterpretation of early work on the frog Rana pipiens suggests an EMT occurs in the epithelium surrounding the proctodeum followed by migration of these mesenchymal cells to the limb bud,^131,132 but this has not been re-examined since 1971, when conclusive cellular markers were unavailable.¹³³ There is also evidence that the lateral plate mesoderm is the ultimate origin of the limb mesenchyme in X. laevis.^69,133,134 Neither the EMT, as seen in limb initiation in amniotes, nor the migration of presumptive limb cells, as seen in zebrafish, is ruled out by existing data (Fig 4).^6,117,131 Further research into the mechanism of limb formation in X. laevis may help to determine which limb formation mechanism is ancestral and which is derived. The mechanisms that allow the limb initiation module to be delayed are yet to be resolved.

Little is known about the genetics of limb initiation in frogs, but many of the limb patterning genes are expressed in conserved patterns later in development.¹³² Tbx5 is expressed in the prospective forelimb region just prior to limb bud initiation in Xenopus tropicalis,¹³⁵ hinting that similar regulatory networks may form the limb in frogs as in other vertebrates. It is also worth noting that limb initiation in the metamorphosing tadpole is independent of thyroid hormone, the hormone which triggers metamorphosis in frogs.¹³⁶

Despite the similarities to other vertebrates, X. tropicalis appears to have a regulatory pathway which has not yet been observed in amniotes or fish. The charmingly named xenopus de milo (xdm) mutation results in frogs which completely lack all forelimb elements.¹³⁵ This mutation lies within the nephronectin gene, an integrin gene involved in epithelial-mesenchymal interactions in the mouse kidney.¹³⁷ The forelimb never initiates in these mutants, and Tbx5 is absent from the LPM, suggesting nephronectin lies upstream of Tbx5 in frogs.¹³⁵ The role of nephronectin in Xenopus limb initiation is particularly interesting as the limb initiation signal in amniotes was originally thought to be Fgf activity originating from the intermediate mesoderm, the embryonic tissue which forms the kidney.^12,13,138 However, the intermediate mesoderm was later shown to express Fgf8 after Tbx5 is already expressed in the presumptive forelimb,²⁸ and blocking posterior migration of the mesonephros via barrier insertion had no effect on limb formation.⁶⁰ The kidney is unaffected in xdm mutants,¹³⁵ suggesting nephronectin may have a novel role in anuran metamorphosis, perhaps related to its role in cell migration.^139,140 Abnormalities in the kidney in direct developing frogs have also been reported in conjunction with forelimb loss, but this is likely due to RA deficiency.¹⁴¹ The role of nephronectin, or integrins more widely, in amniote or fish limb initiation has not been studied.

Direct Developers

Not all frogs have a tadpole stage. Some frog species have lost the tadpole stage and develop an adult-like head and limbs during embryogenesis, mirroring amniote development ¹⁴². It is unclear whether these direct developing frogs have reverted to a more ancestral patterning process or evolved a new mechanism of limb initiation and development. One species which has served as a quasi-model system for direct developing frogs is Eleutherodactylus coqui, a species native to Puerto Rico that has spread to California and Hawaii ¹⁴³. E. coqui has allowed researchers to learn more about developmental conservation and novelties and has also led to hypotheses as to how the amniotic egg evolved ¹⁴⁴.

The limb buds of E. coqui form just after the neural folds close and neurulation ends, similar to the timing of limb bud formation in amniotes.¹⁴³ Once the limb bud has formed, the patterning of the limb bud is largely conserved when compared to other model systems, with a few variations.¹⁴⁵ But, similar to the metamorphosing frogs, very little is known about the initiation of the limb bud. Pitx1 expression is seen in the E. coqui hindlimb once it has formed,¹⁴⁶ suggesting conservation of the hindlimb initiation pathway, but there is no definitive evidence. More compelling perhaps is the observation that a lack of retinoic acid (RA) during E. coqui development leads to a loss of forelimb buds and abnormalities in the pronephros.¹⁴¹ Treatment with a compound that inhibits Raldh at early neurulation stages leads to forelimb loss completely, whereas at later stages only digits are lost. This suggests that a Raldh gene is required in E. coqui during early neurulation to initiate forelimbs, as is the case in amniotes.³⁵ Indeed, Raldh2 is expressed in the E. coqui gastrula in a pattern which supports this hypothesis.¹⁴⁷ Examining the E. coqui limb initiation pathway may tell us about whether the early initiation of the limbs is a novel or rederived ancestral trait in these frogs and other direct developers.

Conclusions

The process of limb initiation seems relatively conserved and well understood in amniotes, with only minor variation seen in the emu or the advanced snakes. This process seems to be less constrained in earlier branching groups, with different cellular mechanisms for limb bud formation and variation in the timing of limb initiation. With limited reference points on the phylogeny, it is hard to make statements about which mechanism of limb formation is ancestral or derived.

Shared mechanisms or regulatory networks may tell us about the fundamental pathways needed to form a paired appendage. Tbx5 and Tbx4 appear to be a necessity for limb initiation in all classes of animals studied. Another required gene for pectoral appendage initiation is RA, but its lack of effect on the pelvic fin or limb reminds us that the two sets of appendages cannot be considered identical in their formation. In fact, the significant delay between fish pectoral and pelvic fin development is an interesting area which has not received much attention. How the fish delays its pelvic fin formation relative to its pectoral fins may help answer the question of how frogs delay all limb formation until metamorphosis. Whilst limb morphogenesis remains one of the best understood aspects of vertebrate embryogenesis, the lack of knowledge of limb initiation outside of amniote model systems shows us that nature still has a lot to reveal to us. Therefore, before we close the chapter on limb development and evolution, we need to learn to walk before we run.