How unique and diverse forms of living creatures evolve is arguably one of the most tantalizing questions in biology. Morphological disparities between the adult forms of related species derive from modifications during embryogenesis. As early as the 19th century, the German biologist Ernest Haeckel postulated two ways by which the modifications could be made in ontogeny to alter the species’ morphology: heterochrony, the displacement of developmental stages in time; and heterotopy, the repositioning of embryonic characters in space (Haeckel, 1875). In an endeavor to couple development and evolution, he proposed the so-called “biogenetic law,” the central argument of which is “ontogeny recapitulates phylogeny,” meaning that the developing embryos of higher animals recapitulate the adult stages of more basally branching animals before developing into their final forms (Haeckel, 1875). Heterochrony in this theory refers to those inconsistencies where the order of developmental events does not obey the sequence of their appearance in evolution. Although the biogenetic law is now discredited, the concept of heterochrony has been preserved in modified form, broadened to include not only alterations in developmental sequence, but also the changes in rate and duration of ontogenetic events (Gould, 1985).

The general concept of heterochrony encompasses a number of discrete mechanisms by which evolutionary change can occur (De Beer, 1951). Today, the most accepted way of breaking down heterochrony posits six distinct mechanisms, placed in two categories, reflecting the polarity of the observed change (Buendia-Monreal and Gillmor, 2018). These include mechanisms that extend development (called peramorphosis) and those that truncate development (called paedomorphosis). The former can be achieved in three ways: processes that follow a normal developmental trajectory but continue for an extended period of time relative to an ancestral organism (hypermorphosis), processes that develop to the same temporal end point, but begin earlier than in the ancestor (pre-displacement), and processes that occur during the same embryonic time period as in the ancestor, but develop at a faster rate (acceleration). Similarly, paedomorphosis can be achieved by three different mechanisms: processes starting at the same time as in an ancestor but developing for a shorter period (progenesis), processes initiating at a later time than in an ancestor but terminating them as normal (pre-displacement), and processes taking place in the same developmental window as in an ancestor but proceeding at a slower rate (neoteny).

These ideas were developed in an attempt to understand morphological evolution prior to modern molecular and genetic insights. Developmental processes, as we understand them now, entail sophisticated interactions between genetic players, themselves expressed at specific times and places, representing a multi-dimensional gene regulatory network. The output of such networks establishes cell behaviors within the embryo, including proliferation, differentiation, migration, and in some cases apoptotosis, which collectively drive morphogenesis. Thus, a molecular understanding of the causality between heterochrony and the evolution of form requires that we comprehend the interplay between genetics and cellular mechanisms in resolving tissue morphogenesis.

Limb development is an excellent system to illustrate the principles by which timing mechanisms are deployed in building tissue and in evolving morphologies. Most vertebrate species have appendages such as fins or limbs. Tetrapod species, the majority of which live on land, have two pairs of limbs, the forelimbs (arms) and hindlimbs (legs). Limb morphologies display remarkable diversity, including differences between the forelimb and hindlimb of a single individual as well as sometimes extreme differences between species. The basic processes required to develop a limb, by contrast, are highly conserved. However, changes in the timing of these fundamental processes have played a major role in the evolution of limbs.

Development of the limbs begins with the specification of limb fields at the prospective cervical-thoracic and lumbar-sacral vertebral boundaries. Limb-type specific T-box transcription factors, Tbx5 and Tbx4, are expressed in these locations and specify the forelimb and hindlimb fields respectively (Logan et al., 1998; Rodriguez-Esteban et al., 1999; Takeuchi et al., 1999; Naiche and Papaioannou, 2003; Rallis et al., 2003; Duboc et al., 2021) (Figure 1A). These Tbx genes activate an FGF paracrine signaling ligand in the mesoderm, Fgf10. Fgf10 serves both as a morphogenetic cue and as a cytokine; on the one hand it triggers the loss of the epithelial architecture of the LPM, prompting the transition to a mesenchymal state (Gros and Tabin, 2014), and on the other hand it enhances cell proliferation (Sekine et al., 1999). While Fgf10 transcription is restricted to the LPM, secreted FGF10 protein diffuses to the flanking ectodermal tissue, where it activates expression of another FGF ligand, Fgf8 (Ohuchi et al., 1999). The Fgf8-expressing ectodermal region is shaped as a crescent ridge located at the most distal point of the limb bud, named Apical Ectodermal Ridge (AER). Fgf8 travels backward from the AER to the LPM, where it retro-activates Fgf10, forming an Fgf10-Fgf8 positive-feedback loop that effectively sustains limb outgrowth (Xu et al., 1998) (Figure 1A). This gene regulatory module for limb outgrowth is highly conserved in amniote species. However, the timing of the onset of this module is subject to variation, which creates alterations in limb positioning.

The proximodistal axis of the limb is organized through the activity of two signaling centers, the AER producing Fgf10 distally and the flank producing RA proximally. The Fgf10-Fgf8 positive feedback loop is the most active in the distal end of the limb bud underneath the AER, from which it decays progressively towards the proximal end (Mariani et al., 2008). By contrast, RA is synthesized and emanates from the trunk, from which it diffuses distally. The distal region also expresses the RA degrading enzyme, Cyp26b1, which helps shapes the RA gradient from proximal to distal (MacLean et al., 2001). The anti-parallel gradients between FGF and RA create varying positional inputs to cells along the PD axis, ultimately establishing three distinct segments marked (from proximal to distal) by transcription factors Meis1/2, Hoxa11, and Hoxa13, respectively (Boulet and Capecchi, 2004) (Cunningham et al., 2013) (Fromental-Ramain et al., 1996) (Mariani et al., 2008) (Mercader et al., 2000; Tabin and Wolpert, 2007) (Figure 2A). The three regions later become the stylopod, (the proximal limb segment, extending from the shoulder/hip to the elbow/knee), zeugopod (the middle limb segment, extending from the elbow/knee to the wrist/ankle), and autopod (the distal limb segment).

The evolutionary loss of digits has occurred repeatedly in tetrapod lineages. In a classic study, Alberch and Gale (1985) (Alberch and Gale, 1985) put this evolutionary pattern into a developmental context. First, they established the order in which digits form during ontogeny in both frogs and salamanders, and found that the sequence of digit formation was different in these two clades. They then categorized the evolutionary order in which digits were lost in each group. They found an inverse relationship where, in both frogs and salamanders, the first digit to be lost evolutionarily, was the last digit to form developmentally (in the pentadactyl species). Moreover, they showed that experimental reduction of the limb bud size, using mitotic inhibitors, yielded the same pattern of digit loss as seen in the variation between natural population (first digit lost corresponding to the last one to form developmentally) (Alberch et al., 1985). This was an important study for establishing the concept of “developmental constraint” (there being a constraint on the order in which reduced digits could be selected). Mechanistically, in the absence of mitotic inhibitors, the authors suggested two ways in which smaller limb fields could have occurred, through “global developmental truncation” (corresponding to progenesis) or through “a slowdown in the rate of proliferation” (corresponding to neoteny), both modes of heterochrony. But how this is actually achieved at a developmental level had to wait until the genetic mechanisms of digit specification were at least partially worked out.

While cis-regulatory sequences upstream of Shh activity do not seem to be involved in the evolution of digit loss in skinks, they do appear to have played a central role in the evolutionary loss of the entire limb in snakes. Snake-like body plans evolved multiple times in the squamate lineage. In some snake-like lizards the forelimbs are lost, in others the hindlimbs are lost, whereas some lineages such as the glass lizard are completely limbless, like snakes themselves (Brandley et al., 2008).

The evolutionary loss of a structure, such as the limb, in and of itself, would not necessarily be viewed as a heterochronic change. This applies to the loss of the forelimb in snakes, which is believed (based on the fossil record) to have preceded the loss of hindlimbs (Apesteguia and Zaher, 2006). Neither fossil extinct snakes nor extant forms display any remnant of pectoral appendages, nor are there any known cases where a modern snake embryo initiates a forelimb bud (Raynaud, 1985). The forelimb has simply been lost, and how this may have occurred is still poorly understood.

In contrast, there are vestiges of the hindlimb in both fossil snakes and occasionally in modern basal snakes (boas and pythons), although the hindlimb is completely missing in more advanced snakes. Adult pythons and boas retain a pelvic girdle as well as a variable remnant of a femur. However, embryonically all individuals of these clades form transient limb buds that form chondrogenic condensations corresponding to the tibia, fibula, and footplate in addition to the femur, prior to degenerating (Leal and Cohn, 2016). Thus, hindlimb loss in basal snakes corresponds to a truncation of a developmental process, or progenesis, a form of heterochrony. Advanced snakes appear to have taken this one step further, truncating the limb development program at its onset. At a molecular level these evolutionary changes in hind limb formation appear to have been achieved through the sequential degradation of the cis-regulatory sequences controlling Shh expression in the limb bud.

Specific expression of Shh in the posterior limb bud is regulated by the ZRS (ZPA Regulatory Sequence) enhancer, located within an intron of another gene, Lmbr1, more than 1 Mb away (Lettice et al., 2002). Deletion of this sequence results in a complete absence of Shh expression in the limb (Sagai et al., 2005), while point mutations within it can result in anterior ectopic Shh expression and polydactyly; or reduction in, or even ablation of, Shh expression (Lettice et al., 2008). These results are explained by the fact that the ZRS has a dual function, and corresponding bipartite organization, where the 5’ domain of the ZRS is responsible for the proper spatial and temporal activation of Shh in the posterior limb bud, while the 3’ domain is important for the looping of chromatin to the Shh promoter and is necessary to prevent inappropriate activation of Shh in the anterior (Anderson et al., 2014).

Genomic comparison indicates that there have been multiple mutations within the ZRS in the python lineage (Kvon et al., 2016; Leal and Cohn, 2016). Tests of python, and limbed lizard ZRS activity were carried out with reporter constructs in transgenic mice and show that while the lizard ZRS drives robust activity in the mouse ZPA, the python ZRS is much weaker, and drives expression in only a small subdomain of the ZPA (Leal and Cohn, 2016). Moreover, when the mouse ZRS is replaced by the python ZRS, using CRISPR-Cas9 technology, the resultant mice exhibited limb truncation highly reminiscent of the structure of the python hindlimb (Kvon et al., 2016). The mechanistic basis by which the python mutations alter Shh expression have not been formally tested, but the ZRS changes in python include mutations in putative binding sites for key Shh-regulating transcription factors, including HoxD, ETS, and Hand1 binding sites, potentially explaining their altered expression profile (Kvon et al., 2016; Leal and Cohn, 2016).

In advanced snakes, the heterochronic truncation of the limb development program is pushed even further back to the initiation of the limb bud. In these animals the ZRS is barely identifiable, as the enhancer is much more degenerated than in pythons and boas (Kvon et al., 2016; Leal and Cohn, 2016). The alterations of the ZRS may have been critical in the evolution of total limb loss in advanced snakes, as it seems to have been in the limb reduction in pythons and boas. Alternatively, the complete loss of limbs could have been achieved by a different mechanism, and the subsequent degeneration of the ZRS simply reflecting neutral mutation in the context of a lack of functional selection to maintain the enhancer once limb buds were lost.

The evolution of digit reduction in Hemiergis and hind limb loss in basal snakes can both be viewed as having arisen through paedomorphosis, the truncation of developmental trajectories. Limbs have also evolved through the opposite mode of heterochrony, peramorphosis, or the extension of developmental trajectories. An example of this is the evolution of hyperphalangy, or an increased number of phalanges, in the digits. Although relatively small increases in the number of phalanges are occasionally seen in terrestrial tetrapods, dramatic hyperphalangy has repeatedly evolved in the flippers of secondary aquatic taxa (those descended from land-dwelling species) such as extinct ichthyosaurs, mosasaurus, and modern cetaceans (whales and their relatives) [reviewed in (Fedak and Hall, 2004)].

Key to understanding hyperphalangy is the recognition that the individual autopod phalanges do not have specific developmental identities. More proximal skeletal elements such as the humerus, and radius and ulna, can be viewed as having distinct identities, being defined by the expression of selector genes, such as Meis1/2 in the stylopod, and Hoxa11 in the zeugopod. However, the autopod elements are not specified individually, but rather are generated through a cyclic program of chondrogenesis and segmentation (Richardson et al., 2004), with every species having defined developmental parameters specifying the ratio of decreased size from one phalanx to the next within a digit (Kavanagh et al., 2013). Hyperphalangy, thus, results from the prolonging of the reiterative program of phalanx generation (peramorphosis).

Prolonging the process of phalanx generation in a digit, with a consequent generation of an extra segment, can be achieved experimentally through manipulations that maintain expression of Fgf8 in the AER overlying a forming digit (Sanz-Ezquerro and Tickle, 2003). Based on this, it was proposed that duration of Fgf activity emanating from the AER, coupled with a characteristic periodicity of segmentation, is responsible for establishing the appropriate number of phalanges in each digit. Consistent with this model, hyperphalangy in cetaceans is accompanied by specific maintenance of the AER (Richardson and Oelschlager, 2002), and Fgf8 expression (Cooper et al., 2007), above the digits forming additional phalanges (and not above adjacent digits that do not exhibit hyperphalangy).

The time of limb bud development is not always congruent in forelimb and hindlimb. In fact, one might expect that the hindlimb should develop somewhat later than the forelimb, as the timing of posterior Hox gene expression is later than that of the anterior Hox genes (Deschamps and Duboule, 2017; Moreau et al., 2019b). However, examination of embryogenesis in different phylogenies shows that this is not always the case. A relative delay in formation of the posterior appendage relative to the anterior appendage is indeed observed between the rostral and caudal fins in zebrafish, and the fore- and hind limbs of certain types of lizards and mammals. However, in most bird embryos, the formation of forelimb and hindlimb buds appears almost simultaneously; and in amphibians, the hindlimb structures, which develop before metamorphosis, are more advanced than the forelimb, the outgrowth of which only becomes apparent after metamorphosis (Richardson et al., 2009; Royle et al., 2021). This indicates multiple aspects of heterochronic regulation, exceeding the scale that can be accounted for by variation in timing of Hox expression. When referring to the timing of limb development, several aspects must be considered, including the time of initial formation of the limb buds, the time to establish a fully matured skeletal pattern, and the growth rate in the post-patterning process.

At the initial phase of limb bud development, nearly all mammalian clades display a delay in the hindlimb outgrowth compared with the forelimb (Richardson et al., 2009; Werneburg and Sánchez-Villagra, 2011). In the non-placental mammals, such as marsupials, where birth is altricial, the delayed hindlimb growth is the most severe (Richardson et al., 2009). The level of heterochrony is species-specific. In the grey short-tailed opossum embryo, for example, at birth, the hindlimb has an established pattern of digits, but the overall size is reduced, and the cartilage structure is poorly differentiated (Keyte and Smith, 2010; Smith and Keyte, 2020) (Figure 4A). The fat-tailed dunnart, on the other hand, shows a more extreme hindlimb retardation where at a time of birth the hindlimb remains in an undifferentiated limb bud state. In the case of the grey short-tailed opossum, the hind limb expression of Tbx4 appears more than half a day later than Tbx5 in the forelimb indicating that the heterochronic delay is already present at the limb initiation stage (Keyte and Smith, 2010). A similar scale of delay can be found in another marsupial species, the Tammar Wallaby (Chew et al., 2014) (Figure 4). Interestingly, in grey short-tailed opossum embryos, the anterior-posterior heterochrony not only exists in limb development but also in somitogenesis; a 4-fold time difference has been identified for anterior vs. posterior somite generation rate, and the newly born neonates lack differentiated cartilage in the posterior somites (Keyte and Smith, 2012). It is worth noting that the length of the tail of the grey short-tailed opossums appears similar to that of the mouse at the time of birth, therefore it is likely that the heterochrony of somitogenesis not only comes from the delay of posterior somite generation but also from an acceleration in anterior somite development.

Another interesting aspect of the heterochrony in marsupial embryos is the compensatory growth of hindlimb after the onset of lactation. After the embryos are born and can access the milk in the pouch, the hindlimb grows and differentiates quickly and it soon catches up to the stage of forelimb (Cook et al., 2021). It is possible that this acceleration of hindlimb growth is regulated by a nutrient sensing mechanism. Indeed, an “energy trade-off” hypothesis has been postulated to explain the anterior-posterior heterochrony of marsupial embryos, as the level of growth heterochrony correlates inversely with the length of the gestation time (Smith and Keyte, 2020). The post-natal growth rate seems to be regulated by maternal milk supply as the cross-fostering between young and old age neonates allows the growth rate to adjust to the new maternal lactation age (Trott et al., 2003; Kwek et al., 2009). It is unknown whether it is the presence of the nutrient supply that stimulates the hindlimb-specific increase in growth or whether other hormone-related factors come into play. It is also worth noting that different aspects of heterochrony may be regulated by distinct mechanisms, such that the heterochrony in limb initiation may not be regulated by the same mechanisms as those for post-patterning. Overall, the heterochrony in anterior and posterior development, partially manifested by the timing differences needed to develop forelimb and hindlimb appendages, is a fascinating question and yet the molecular and cellular basis has been poorly explored. The understanding of mechanisms accounting for this striking heterochronic phenotype will be instrumental to understanding the regulation of the rate of growth and differentiation.

As noted above, limb bud formation begins with localized EMT in the presumptive limb fields, triggered by expression of Tbx5/4 (Gros and Tabin, 2014). It therefore makes sense that the heterochronic delay in hindlimb development seen in marsupials would be controlled at the very first step of Tbx4 expression. However, this does not turn out to be the case in all examples of developmental delay in limb initiation. The wing of the flightless emu is greatly reduced in size, in part due to a delay in the outgrowth of the wing buds. Surprisingly, however, a recent study showed that the initial steps of forelimb development, including Tbx5 activity and consequent EMT, occur at equivalent stages in the emu and chick (Young et al., 2009). Rather, it is a subsequent step in the process, the induction of the downstream gene FGF10, that is modulated, resulting in a failure to activate target genes necessary for proliferation of the limb mesenchyme.

The changes in time, tempo, or duration of developmental processes are a prevailing phenomenon during evolution. There is a long history of characterizing developmental heterochrony, yet mechanistic insight is only beginning to emerge. Understanding heterochrony requires an understanding of the genetic and cellular transformations that shape tissues. As a comprehensive developmental system where, genetic drivers are largely revealed and cellular events are characterized, limb development is one of the best-suited systems to study how timing variations cause morphological adaptations and to yield principles of how timing mechanisms can drive changes in tissue formation. As supported by the many examples discussed here, timing mechanisms have been extensively used in establishing the correct pattern of the limb. As a result, evolutionary cases can be found in which alterations in timing cause different anatomical changes in the limb. Many of the case studies we discuss here essentially reveal the molecular or cellular basis of the heterochronic events. However, a systematic solution is still needed to fully comprehend how heterochrony causes the observed phenotypes.

Understanding the precise modulation of the timing of gene expression is still a daunting task, and currently the phenotypes that result from temporal perturbations of non-linear interactions are difficult to interpret. The development of new quantitative genetic approaches, advances in imaging, and modelling hold the potential to better solve the problems. In addition, although many limb heterochrony issues have been addressed, there are still many that await investigation. Examples include the various aspects of anterior-posterior heterochrony in mammals and the heterochrony in chondrogenesis in different segments of the limb, which have been previously reviewed (Richardson et al., 2009). Moreover, in nearly all the cases we discuss here, heterochrony results from alterations in gene expression. The extent to which environmental adaptation or physiological changes also contribute to the timing of tissue development will be an interesting topic to explore.