The origin of evolutionary novelties raises some of the most fundamental questions of biology. How do new structures arise? Can they evolve de novo or are they generally derived from pre-existing structures? And what is the developmental and genetic basis for their origin and modification1?

The adaptive evolution of vertebrates and arthropods to aquatic, terrestrial and aerial environments was accomplished by the invention of many novel features, especially new types of appendages. Enormous progress has been made in the past few years in understanding appendage development in both phyla. These genetic discoveries can be integrated with palaeontological data to address some of the principal events in the history of animal designs.

We will first examine the origin and evolution of vertebrate limbs and digits and of arthropod legs and insect wings. In both phyla we are confronted with a similar issue, namely the origin and adaptive modification of serially homologous organs. We will integrate palaeontological and developmental evidence that suggests that major innovations are largely derived from pre-existing developmental systems and will illustrate the potential genetic regulatory changes that enabled appendage evolution. Then we will explore the significance of newly discovered genetic similarities between arthropod and vertebrate appendages—similarities that have been retained despite more than 500 million years (Myr) of independent evolution. We will develop the hypothesis that the evolution of successively derived limb types, from lobopods to insect wings, and from agnathan fins to tetrapod limbs appears to be due, in part, to the successive cooption and redeployment of signals established in primitive metazoans. These examples illustrate how comparative developmental genetics can provide a mechanistic explanation of the origin and evolution of structures when palaeontological data are robust and important new hypotheses about evolutionary history when the fossil record is silent.

Origin and diversification of tetrapod limbs

Vertebrate limb diversity was produced by changes in the number, position and shape of structures that can be traced to Ordovician2,3 (463–439 Myr) through Late Devonian2,3,4 (409–362 Myr) fossils. The demands of feeding and locomotion in Ordovician and Silurian seas led to a surprising variability of the earliest known appendages: some forms possessed a continuous anterior fin that ran the length of the body, others had paired fins that projected immediately behind a head shield, and still other primitive vertebrates had no paired fins at all (Fig. 1). A body plan with two sets of paired appendages, pectoral and pelvic, is a derived feature that first appears in later jawed vertebrates (gnathostomes)2,3,4. The number of paired appendages has been highly conserved ever since their origin: the evolution of new gnathostome body plans primarily involved a modification of existing paired appendages rather than the invention of whole new sets (acanthodians are the only exception to this generalization). Therefore, the origin of more recent novelties, such as digits, involved the modification of genetic systems first established in more primitive vertebrates.

Figure 1: Major innovations of vertebrate paired appendages.
figure 1

Basal chordates, such as Amphioxus (not shown), do not possess appendages homologous to those of vertebrates. Unpaired, median fins are the earliest known vertebrate appendages2. Paired appendages are first encountered in Ordovician and Silurian jawless fish as elongate fins that extend laterally along the body wall (for example, Jamoytius; a) or as paired pectoral fins (osteostracans; b). Other basal vertebrates (not shown) do not possess any paired appendages. Multiple sets of paired appendages are a derived characteristic of jawed fish (gnathostomes). In many gnathostomes, pectoral and pelvic fins have often evolved in parallel. This pattern of concerted evolution suggests that pectoral and pelvic appendages shared similar regulatory genes in early stages of gnathostome evolution. A fin axis (blue) is seen in the fins of many gnathostomes, and it is a primitive characteristic for sarcopterygian fish (for example, Eusthenopteron; c) and tetrapods (for example, Ichthyostega; d). Sarcopterygian fins are derived in having a zeugopod and an elaborate endoskeletal fin skeleton. Digits develop within the distal portion of this extensive endoskeleton. The establishment of serially homologous appendages is proposed to result from gene cooption during the evolution of Paleozoic vertebrates. HoxD genes were probably not involved in the origins of body wall outgrowths in basal vertebrates because unpaired fins do not express these genes9. These Hox genes were initially involved in specifying regional identities along the primary body axis, particularly in caudal segments (e). One key step in the origin of jawed fish was the cooption of similar nested patterns of expression of HoxD genes in the development of both sets of paired appendages (f). This cooption may have happened in both appendages simultaneously, or Hox expression could have been initially present in a pelvic appendage and been coopted in the development of an existing pectoral outgrowth8. The reconstructions in a and b are modified from those in ref. 106, that in c from ref. 107, and that in d from ref. 108. The hind limb of Ichthyostega (d) is modified from ref. 5.

Serial homology and adaptive diversification

Primitive genetic systems must have provided a framework for the evolutionary integration of pectoral and pelvic appendages. Digits, for example, arose at the same time in the hand and foot: there is no Devonian tetrapod that has fingers and no toes2,3,4,5. Even in the post-Devonian world many unique designs appeared simultaneously in forelimbs and hindlimbs, as witnessed by chameleons, ungulates and ichthyosaurs. Obviously, serially homologous appendages can also evolve independently, an extreme case being the modification of pectoral appendages into wings in bats, birds and pterosaurs. Even in these extremely modified forelimbs, however, numerous similarities are retained between wing and leg.

The linkage between forelimbs and hindlimbs appears to be an ancient feature that resulted from patterns of gene cooption during the evolution of Palaeozoic fish. Serially homologous paired appendages are seen in Palaeozoic placoderms, acanthodians, chondrichthyians and osteichthyans2,3,4. In addition, pectoral and pelvic fins have evolved in parallel in almost all major gnathostome clades. There are numerous genetic parallels in pectoral and pelvic development that could account for these patterns of concerted evolution6,7. Hox genes, in particular, are likely to have been involved in the evolution of serial homology8. The earliest vertebrate appendages (unpaired fins) presumably did not utilize Hox genes; HoxA and HoxD genes are not expressed during the outgrowth of zebrafish unpaired fins9. Although these Hox genes were probably not involved in the origin of outgrowths in basal vertebrates, their later recruitment in the development of paired appendages was a key step in establishing serially homologous designs. The HoxD genes that came to play a role in appendage development are a subset of those involved in specifying regional identities along the caudal body axis (caudal neural tube, gut, somitic and lateral plate mesoderm)6,7,8,9,10. This suggests one of two situations: either nested patterns of Hox expression were originally present in a caudal set of paired outgrowths, and were later recruited in the development of a cranial set of outgrowths8, or similar Hox genes were recruited in pectoral and pelvic outgrowths at the same time in the evolutionary history of vertebrates (Fig. 1). In either case, pectoral and pelvic appendages were genetically linked early in their history and could have evolved together, presumably because the development of these appendages had already been brought under similar regulatory controls.

Superimposed on these ancient genetic parallels are secondary differences in gene expression and interaction that may have served as the basis for the independent evolution of pectoral and pelvic appendages. Hox gene expression in extant tetrapod limbs is dynamic and encompasses at least three distinct phases initiated successively in the primordia of the stylopod, zeugopod and autopod7 (Fig. 2). The presence of three distinct phases of Hox expression in limbs may reflect the observation that all tetrapods maintain a standard pattern of organization (Fig. 2), whereas specific differences in expression (or gene interaction) in each phase could result in the independent modification of pectoral and pelvic appendages7. Phase II Hox expression is practically identical in the fore- and hindlimb buds of mice, which possess generally similar skeletal patterns. In contrast, the wings and legs of chicks have very different skeletal patterns and different patterns of phase II Hox expression as well7,11. Surprisingly, phase III expression is very similar in chick wing and leg buds, indicating that aspects of the derived structure of chick wings are established by some other genetic means. Candidates include Hox genes of other clusters that are differentially expressed in the wing and leg buds12 and the T-box genes, another family of putative transcription factors differentially expressed in the forelimb and hindlimb buds13. Combinatorial action between genes may explain different functional requirements for the HoxA and HoxD genes in the forelimb and hindlimb. For example, homozygous deletion of both HoxA-11 and HoxD-11 results in almost complete loss of the zeugopod in the forelimb but not in the hindlimb14, despite the fact that these genes have equivalent patterns of phase II expression in both limbs. A possible explanation for the observed differences between the appendages may be that there is expression of the paralogous gene HoxC-11 during phase II in the hindlimb but not the forelimb, where it may act redundantly with HoxD-11.

Figure 2: Stylopod, zeugopod, and autopod: patterning the limb.
figure 2

The tetrapod limb consists of three distinct compartments: a, the stylopod (upper arm and thigh); b, zeugopod (lower arm and calf); and c, autopod (hand and foot). This subdivision of the limb is supported by phylogenetic comparison4, analysis of gene expression7, and experimental manipulation14,36,37,38,39,40,41,109,110,111. There is a broad correlation between the position of a compartment and its evolutionary history4. The sylopod (a) is the most ancient (possibly of Late Silurian origins) whereas the zeugopod and autopod are the most recent (being first encountered in Devonian sarcopterygians). This same order of appearance of the three limb segments is recapitulated during development. Early removal of the apical ectodermal ridge (AER) results in a limb with only a stylopod; the zeugopod and autopod are produced after successively later surgeries112,113. The Abd-B-related genes of the HoxD cluster are expressed in a complex, dynamic pattern encompassing at least three distinct7, independently regulated phases24,26. In the first phase (a; phase I), two of these genes (HoxD-9 and HoxD-10) are expressed across the entire limb bud7. This expression correlates with the time that the stylopod is specified7,112. Subsequently, a second phase of expression (b; phase II) is initiated in response to the secreted factor, Sonic hedgehog. Here, Hox genes are expressed in a nested set centred around the Sonic-expressing cells, with HoxD-13 being expressed in the most restricted domain, and HoxD-12 and HoxD-11 each encompassing a broader domain7. This pattern of expression coincides with the time of specification of the zeugopod and takes place in cells fated to form this segment. Finally, a third phase of expression (c; phase III) is initiated later during limb development, when these Hox genes are all expressed across the majority of the distal portion of the limb bud7. During this phase, the expression of Hox genes still appears to be a consequence of the Sonic hedgehog signal, but the relative responsiveness of the different genes has changed so that HoxD-13 now has the broadest expression domain and HoxD-12 and HoxD-11 are nested within it7. The phase III expression (c) patterns occur in the presumptive autopod at the time that segment is specified. The combination of the change in relative size of Hox expression domains with a phenomenon known as ‘posterior prevalence’ (the general rule that more 5′ genes in the Hox cluster are phenotypically dominant) results in different Hox genes playing pre-eminent roles in different limb segments: for example, HoxD-9 during phase I in the stylopod (a), HoxD-11 during phase II in the zeugopod (b), and HoxD-13 during phase III in the autopod (c). The expression of the dominant Hox genes in each phase is essential for the formation of the bones in each segment, as seen in their knockout phenotypes. The knockout phenotype of a gene consists of alterations in the pattern and shape of skeletal elements. The location of these modifications depends on the position of the gene within the cluster. For example, mice engineered to be deficient in both HoxD-11 and in HoxA-11 (the paralogous gene of the HoxA cluster) form limbs that are essentially missing the zeugopod14 (b). Phenotypes of Hox D-9-deficient mice, in contrast, are specific to the stylopod109 (a), whereas Hox D-13 deficient mice primarily have defects in the autopod39 (as do mice engineered to be deficient of Hox A-13 paralogues110,111) (c). Knockout data are derived from refs 14, 36,37,38,39,40,41 and 109,110,111.

Learning to crawl: the fin-to-limb transition.

Some regions of vertebrate appendages are more variable than others4,15,16. The invention of flippers, wings and other specialized limbs often involved significant changes in the pattern of distal structures rather than proximal ones15. Two broad notions of the homology of distal structures have emerged over the past 130 years: one that sees digits as being unique to tetrapods17,18 and another that sees antecedents of digital structure in the fins of sarcopterygian fish19,20. Both genetic and fossil data support the hypothesis that digits are evolutionary novelties21,22 (Fig. 3).

Figure 3: The origin of digits.
figure 3

A fin axis (grey) is present in the fins of chondrichthyians (a, Cladoselache), basal actinopterygians (not shown), and sarcopterygians (be). In sarcopterygians such as Neoceratodus (b), Sauripteris (c), Panderichthyes (d) and Tulerpeton (e), a single element (stylopod) articulates with the pectoral or pelvic girdle; all other proximal bones of the fin have been lost. The autopod is considered to be a synapomorphy of tetrapods because their nearest relatives do not have any apparent homologues of digits. Box: we propose that the origin of digits correlates with a novel pattern of Hox expression. f, The axis (red) of the fin of Panderichthyes is short and radials branch preaxially. Patterns of phase II Hox expression (f; Hox D-11 in light blue, Hox D-13 in darker blue) were most likely already in place in sarcopterygian fins. g, The proximal portion of the axis (red) of tetrapods compares with the entire axis of Panderichthyes. Phase III Hox expression reflects a reversal of the nested domains of expression of HoxD-11 (g, light blue) and HoxD-13 (g, dark blue) during the time when the autopod is specified. This reversal in the polarity of Hox expression is considered to be correlated with the origin of the autopod; digits differ from more proximal structures in that they lie on the postaxial side of the axis. Because this hypothesis relies on genetic comparisons between phylogenetically disparate taxa (for example, teleosts and tetrapods), the shift from phase II to phase III collinear expression may have evolved in more basal sarcopterygians. Analogues of digits are seen in the fins of other sarcopterygian fish (c, Sauripteris); the expanded endoskeletons of rhizodontids include as many as eight branched preaxial radials. Different lineages of sarcopterygians appear to be inventing similar solutions to life in shallow freshwater ecosystems. The reconstruction of Tulerpreton (e) is modified from ref. 114.

The origin of digits is associated with the evolution of new temporal and spatial patterns of gene expression and regulation7,9,21. In extant tetrapods, the development of digits correlates with a reversal in the anteroposterior order of expression of Hox genes in phase II and phase III7,23 (Figs 2, 3). Recent studies of teleosts (zebrafish) have revealed patterns of Hox expression that are similar to patterns seen in proximal regions of tetrapod limbs9,21. Phase III Hox expression is not seen in the zebrafish and appears to be unique to the digital region of tetrapod limbs7,9 (the expression of other sarcopterygians or more basal actinopterygians is not known). In addition, the different phases of Hox expression are not only discrete in tetrapod limbs, but they are regulated by separate cis-regulatory enhancer elements in each phase24,25,26. During phase II Hox expression in tetrapods, a complex set of enhancer elements is used within the regulatory region of each Hox gene of the cluster (as in the regulation of the Hox genes along the main body axis)25,26. However, regulation of all the HoxD genes in phase III depends upon a single enhancer upstream of the entire cluster24,26. The utilization of this distinct enhancer is consistent with the hypothesis that digits are evolutionary novelties because the development of the autopod is regulated differently from that of the rest of the limb.

The presence of phase III Hox expression in tetrapod limbs, and its absence in teleost fins, suggests that this pattern may be an apomorphy for tetrapods or a more inclusive group. In addition, the presence of a uniquely tetrapod enhancer for phase III Hox expression implies that this regulatory element is also more derived relative to conserved phase I and II enhancers. The shift from phase II to phase III collinear expression involves multiple genes expressed at different times and in different regions of the limb. If these changes were genetically independent then they would have required the joint evolution of numerous regulatory elements. If only a single enhancer was involved, then this shift could have produced a change in the expression of multiple genes in a small number of evolutionary steps. Furthermore, the utilization of the same enhancer in forelimbs and hindlimbs provides a developmental explanation of the observation that fingers and toes arose simultaneously in the fossil record.

We propose that the temporal and spatial shift in the expression of Hox genes during limb development correlates with transformations inferred from the fossil record. Devonian fossils provide morphological links between structures in fins and limbs. Sarcopterygian fins are dominated by an axis of segmented endoskeletal elements that extends from proximal to distal2,3,4,27 (Figs 1, 3). This axis is most similar to tetrapod limbs proximally, where the humerus, radius and ulna (femur, tibia and fibula) can readily be compared between taxa2,3,4,5,15,22. Embryological and palaeontological data suggest that the axis of fins was developmentally bent during the origin of tetrapod limbs27. This scheme holds that there is a dramatic difference between the autopod and the zeugopod because the branching of the axis shifts from the anterior (preaxial) to the posterior (postaxial) compartment of the limb7,27. Proximal elements, such as the radius, project anteriorly from the axis, whereas distal elements, such as the digits, project from the posterior side of the axis. We propose that the reversal of morphological polarities in the appendages of Devonian vertebrates correlates with the reversal of Hox gene expression seen in phase III (Fig. 3f, g). As Hox expression in phase III is driven by a novel enhancer element, the axis was not bent per se; rather, a novel extension (with reversed morphological polarities) is considered to be added to it in the Late Devonian. This hypothesis is supported by a comparison of panderichthyid fins and tetrapod limbs. The fins of this sister-group of tetrapods (Fig. 3d, f) are highly reduced in comparison to other sarcopterygians (Figs 1e, 3b, c, c) and this reduction is most prominent distally. No potential homologues of digits, wrist or ankle bones are preserved in these fish22.

Are digits, or their functional equivalents, unique to tetrapods? Fins of rhizodontid fish have stunning similarities to tetrapod limbs20(Fig. 3). The fins of these Devonian fish contain up to eight endoskeletal radials that project distally; in Sauripteris (Fig. 3c), six of these rods terminate at the same proximodistal level. Either these radials are directly homologous to the six to eight digits of Devonian tetrapods (a hypothesis not supported by phylogenetic inference) or they are functional analogues. In either case, these rods reflect a site at which morphological polarities are reversed (for example, the radials of Sauripteris branch postaxially20), suggesting that phase III Hox expression may have arisen in this clade. Several different lineages of Devonian sarcopterygians appear to have evolved the same morphological solution to life in shallow freshwater environments. The tetrapod clade evolved true digits, whereas rhizodontids developed functional analogues. In both cases, the genetic shifts may well have been similar.

Adaptive diversification: how many fingers? Digit reduction is a dominant theme of tetrapod limb evolution; deviations from a pentadactyl pattern virtually always involve the loss of fingers or toes5,28. Polydactylous hands and feet have almost never been fixed in phylogeny, despite the presence of polydactylous variants within populations (individuals of many species including cats, dogs, mice, chickens and humans carry mutations that cause the formation of extra digits). This paradox can be explained in terms of evolutionary constraints by postulating a genetic limitation to digital evolution. One approach holds that the genetic mechanisms that determine the number of digits are distinct from those that regulate morphology, and that there are currently only five discrete genetic programs for specifying unique digit morphology28. The primary genetic limitation is on the number of kinds of digits, not their absolute number. A specific prediction of this hypothesis is that polydactyly can arise, but at least two of the digits will have the same identity (that is, morphology). In this regard, it is easier to modify other carpal or tarsal bones to new functions than to create a new digit. Supporting this notion is the observation that the additional ‘digits’ in extant polydactylous taxa are typically modified carpal or tarsal bones (as in frogs and panda bears, for example). Unfortunately, we cannot as yet evaluate the genetic basis of this constraint because we do not understand the genetic mechanisms that regulate the differences between digits.

Many classical morphologists were interested in defining ‘laws of form’—common trends that appear in widely different groups. Comparative analysis of diverse taxa now offers the promise of fundamental insights into these long dormant questions. The 360-Myr history of tetrapod limbs is witness to dramatic regularities in digital reduction4,16,29. One notion, ‘Morse's law of digital reduction’, contrasts the stability of the inside digits (III, IV) with the lability of outside ones (V, II, I)30,31. In virtually every known example of digital reduction, digits V, II and I are among the first to be lost and digits III and/or IV are typically retained in tetrapods that have the most extreme patterns of digital reduction. This pattern is widespread and has evolved independently in lizards32,33, dinosaur and bird feet34, and in mammals35 (for example, ungulates). The major exception to this trend lies in the hands of theropod dinosaurs (that lose postaxial digits); theropod feet conform to Morse's law, as do the hands and feet of other dinosaurs34.

Are regularities of digital evolution the product of developmental constraints upon variation? Knockouts of different Hox genes (HoxD-11, HoxD-12, HoxD-13, HoxA-13) lead to changes in the shape and number of bones in affected mice and these different genes often have overlapping effects14,36,37,38,39,40,41. One common result is the stability of the internal digits (IV, III) in knockouts of single genes or combinations of genes41. The parallels between the expectations of Morse's law and the results of experimental manipulation suggest that trends of digital evolution may have a developmental basis. The morphological effects of different gene knockouts may reflect the sequence of digital formation: the first digits to be affected are typically the last to form in development37. Although evolutionary patterns of digital reduction are unlikely to involve coding mutations of Hox genes, limb reduction may involve changes in the regulation of Hox genes or the genes that they control. Comparative analysis of gene expression and function in representative taxa could elucidate the mechanisms behind these general evolutionary trends. The lizard genus Lerista, for example, has species with five, four, three, two, and no toes—the range of states of digital reduction in this genus parallels that seen in virtually all other tetrapods32.

Origin and diversification of arthropod limbs

“Exites and endites of the proximal limb segment of arthropods, from trilobites to insects, have an extraordinary history. They have furnished most of the remarkable tools of the phylum: mandibles and other mouth parts; gills of trilobites and Crustacea; swimming and grasping appendages; gill-plates of Ephemeroptera … Their evolutionary potential is comparable with that of the vertebrate limb …” Wigglesworth (1972)42.

The adaptive radiation of arthropods started much earlier than that of tetrapods. The Cambrian fossil record abounds with trilobites, arachnomorphs and crustacean-like forms, and contains many bizarre animals with spectacular appendages and body armour (Fig. 4). The ‘arms race’ of the Cambrian explosion may well have been a ‘limbs race’ among arthropods to evolve better sensory, locomotory, feeding, grasping and defensive appendages.

Figure 4: The lobopod–arthropod transition and the diversification of arthropod limb patterns.
figure 4

Several innovations occurred in the lobopod–arthropod transition43. Various lopopods that may represent different degrees of ‘arthropodization’ are depicted. Opabinia is shown in partial cutaway view43 to reveal the lateral lobes and ventral lopopodia. The relationships among the major arthropod groups is an unresolved polychotomy. The most basal euarthropod was probably fully sclerotized with jointed, biramous limbs115, and a homonomous trunk. Uniramous limbs evolved in the terrestrial arthropods. The subdivision of the trunk and differentiation of individual limbs in modern insects involved regulatory changes in Hox gene domains along the many body axis from an ancestor in which trunk appendages and Hox gene domains were mostly identical (Hox scheme is adapted from ref. 53).

The most obvious feature of arthropod diversity is the number, morphology and function of their appendages (Figs 4, 5). Antennae, mouth parts, walking legs, grasping and swimming appendages are all modifications of a basic jointed limb structure that defines the phylum. It is generally though that jointed legs evolved from simple unjointed appendages such as the lobopodia found in the probable sister group of the arthropods, the Onychophora (Fig. 5a). The diversity of Cambrian arthropods and lobopodans (Fig. 4) indicates that the transition from lobopods to jointed appendages occurred before the Cambrian (570–510 Myr). Fossils from this period are scarce and the reconstruction of these transitions mostly relies on the comparative analyses of later, Cambrian fossils43,44.

Figure 5: The evolution of the arthropod limb and the origin of the insect wing.
figure 5

ae, Some of the major transitions in limb architecture. a, A simple unjointed, annulated lobopodium; b, separate lateral lobes may have served a gill-like function; c, a jointed biramous limb in which the two limb branches are joined at the base, an upper branch is often derived from lateral lobes; d, a multibranched limb found in the branchiopod crustacean Artemia; e, the insect wing and uniramous leg appear to derive from a polyramous limb in an aquatic ancestor. Box: left, a developing polyramous limb and corresponding differentiated structure. The apterous gene is expressed (red) in a dorsal respiratory lobe, whereas the Distal-less gene is expressed (blue) in the other limb branches54. An ancestor–descendent relationship between the limbs in the box is not implied. Right, separation of the dorsal respiratory lobe from the ventral limb primordium in a primitive pterygote such as a Paleodictoptera nymph. The proto-wing at this stage was probably a gill-like structure on all trunk segments and still attached to the base of the limb. The apterous and Distal-less genes play critical roles in wing and leg formation in Drosophila.

Morphological studies of Cambrian lobopodans have investigated the series of innovations that led to the basal arthropod design. These novelties include: the evolution of external segmentation; sclerotization; and, most important to our discussion, the origin of jointed, biramous appendages. Whereas the taxonomic relationships between Cambrian forms and extant arthropods are uncertain, Cambrian taxa with different types of ‘arthropodization’ and limb morphology can be identified43. For example, the Cambrian Aysheaia and the Recent Peripatus, possess simple, unjointed uniramous limbs (Fig. 4), but lack the annulation and armour typical of more derived lobopodans such as Hallucigenia43 (Fig. 4). Other lobopodans, such as Opabinia, are externally segmented and possess lateral lobes above the ventral lobopods43 (Figs 4, 5b). Fusion between the gill-like lateral lobes and ventral lobopodia may have given rise to the biramous limb43,45,46,47 (Fig. 4c). In forms such as Anomalocaris, this fusion was accompanied by limb segmentation and sclerotization (Figs 4, 5). Full cuticular sceloratization, then, arose in primitive arthropods.

Serial homology and adaptive diversification. The adaptive radiation of primitive arthropod limbs entailed considerable changes in their number, pattern and function. Two trends are evident. First, in arthropods in general and certain lineages in particular, there have been increases in the number of different limb types. For example, advanced lobopods had perhaps four types of appendages (frontal, jaw, trunk and tail fan) and only one type of trunk appendage, whereas insects possess up to ten distinguishable limb-derived appendages and four or five different trunk appendages (Fig. 4). A second trend has been the dramatic diversification of homologous appendages—not just in size or morphological detail. Major changes in limb organization have evolved, such as the evolution of unbranched walking legs in insects and changes in mandible architecture in myriapods, crustacea and insects (Fig. 4).

The developmental genetics of limb formation and identity in Recent arthropods may help to shed light on the diversification of arthropod limbs. Hox genes have played a different role in arthropod evolution from that seen in tetrapods. Studies of Drosophila melanogaster have revealed that each type of appendage is typically specified by a single or a pair of Hox genes acting in the individual body segment that gives rise to a particular appendage48. For example, the Antennapedia gene acts in all three pairs of walking legs but the distinct morphology of the first, second and third pair of walking legs is determined by the Sex combs reduced, Antennapedia, and Ultrabithorax Hox genes49, respectively. In the antenna, no Hox gene is active. If Hox gene function is lost or ectopically activated in individual segments, the identity of the corresponding appendage is transformed. Thus, loss of Antennapedia transforms second leg to antennal structures50 and expression of Antennapedia in the antenna transforms it to a leg51. Importantly, loss of all Hox gene functions in insects results in a dead embryo bearing antennae on all segments52. This demonstrates that the potential to form a limb exists in all segments, but the type of limb formed is determined by individual Hox genes.

The specification of different limb types by different Hox genes or combinations of Hox genes differs from the nested pattern of Hox genes expressed in the limbs of vertebrates and has important implications for the pattern of morphological evolution in the arthropods. Different, but serially homologous, arthropod limbs are distinguished by the action of different Hox genes that modify the interpretation of a common set of positional signals. Thus, the increase in appendage diversity between lobopods, primitive crustacea and insects must have involved the diversification of Hox gene regulation and function in the arthropod trunk. Comparative studies of Hox gene expression between crustacea (for example, branchiopods) and insects support this notion. In the branchiopod thorax, the expression of the Antp, Ubx and abd-A Hox genes is coincident and the morphology of thoracic limbs is uniform53. In insects, these same Hox genes differentiate the middle thorax, posterior thorax and anterior abdomen (Fig. 5). The adult hexapod abdomen is legless and this is due to the direct repression of limb formation by products of the Ubx and abd-A Hox genes54. These gene products do nor repress limb formation in branchiopods55.

The diversity of the architecture of putatively homologous appendages has fuelled many debates about arthropod relationships. For example, walking legs and mandibles can differ so much between taxa that it has been suggested that the arthropods had multiple ancestors (that is, they are polyphyletic)56, but phylogenetic57,58 studies have refuted this. In addition, developmental studies suggest that different limb architectures arise through modifications of a common genetic program. For example, the Distal-less (Dll) gene controls the development of the distal portion of Drosophila limbs59 and is expressed in the distal domains of limbs in all arthropods studied so far50,60,61. However, Dll is not expressed in insect or developing adult crustacean mandibles55 but is expressed in myriapod mandibles62. These data agree with fossil evidence suggesting that crustacean and insect mandibles were reduced from the primitive whole-limb mandible by truncation of the mandibular proximodistal axis.

The branching of arthropod limbs has been very important to their functional evolution. Different limb branches can be specialized for respiration, locomotion and a variety of other functions. Chelicerates, trilobites and aquatic crustacea have biramous or polyramous limbs and first appear in the Cambrian, whereas the terrestrial myriapods, insects and crustacea have unbranched (uniramous) limbs and appear much later in the Silurian63 and Devonian64, respectively. It has been argued that the ancestral arthropod was biramous43,45,46,47. Comparisons of Dll expression and limb outgrowth in various types of crustaceans and insects reveal that all limb types arise from the same relative anteroposterior position within the body segment but differ in their dorsoventral branch points55. This suggests that uniramy, biramy and polyramy are the products of shifts in signals along the dorsoventral axis of the body wall (or appendage) and that additions or reductions in branch number may evolve readily. This flexibility was crucial to the later evolution of perhaps the most significant invention by any arthropod—wings.

Learning to fly: the leg-to-wing transition. Early in the Devonian, before tetrapods arose, one major animal group had already invaded land—the insects. The subsequent evolution of flight presaged the enormous radiation of insects: these taxa now comprise more than two-thirds of all known animal species. The evolution of insects from an as-yet uncertain arthropod ancestor and the emergence of winged (pterygote) forms involved major transitions in limb architecture and function.

Few evolutionary mysteries have inspired more theories than the origin of insect wings. It is not certain when wings first arose because the Early Devonian insect fossil record is scanty64. All of the major pterygote groups appear by the Carboniferous (362–290 Myr) and are assumed to have arisen earlier. Fundamentally, discussion on the origin of wings has focused on whether they are novelties or whether they are modified versions of ancestral structures. If wings were derived from existing structures, what were their anatomical origins and initial functions?

One of the longest-held models is the ‘paranotal’ theory, which holds that wings are novelties derived from hypothetical rigid extensions of the body wall of a terrestrial ancestor65. A second hypothesis, the ‘limb-exite’ model proposes that insect wings evolved in a series of transitions beginning with the polyramous exite-bearing legs of an aquatic pterygote ancestor42,66. According to this model, proximal limb elements were modified to flap-like structures and adapted to spiracular or movable gill covers to facilitate respiration. These sack-like pro-wings were found on all thoracic and abdominal segments and became stronger as ancestral pterygote nymphs (similar to Recent mayfly larvae) used them presumably for propulsion. As insect lifestyles became more amphibiotic, some insects might have used a rudimentary proto-wing for surface skimming, as seen in extant stoneflies67. Finally, wings acquired the mechanical strength and flexibility (with corrugation and veins) and the supporting musculature to support active flight.

Recent studies of the developmental regulatory mechanisms controlling wing formation and number in Drosophila suggest a close ontogenetic and evolutionary relationship between legs and wings. For example, in Drosophila68 and other Diptera, the wing arises in close association with the leg. Cells that give rise to the wing field in Drosophila actually migrate out of the developing ventral limb field68. There are important regulatory differences that distinguish the development of the sheet-like wing from the tubular structure of the leg. One key difference is that adult and developing wings are divided into discrete dorsal and ventral compartments whereas legs are not. In Drosophila, the definition of dorsal versus ventral cell fates is orchestrated by the apterous gene which is expressed only in dorsal cells and is necessary for wing but not leg formation69. Clearly, apterous, which regulates several crucial downstream signalling components, and is involved in a conserved dorsal pattern of expression in insect wings70, was coopted into a distinct role in dorsoventral patterning at some stage of wing evolution. apterous primitively could have specified a dorsal compartment (or branch) of an ancestral polyramous appendage or an evolving proto-wing. Remarkably, this is exactly what has been found in a recent study71. The dorsal branch of a branchiopod crustacean respiratory epipodite specifically expresses the apterous gene and one other developmental marker of the insect wing field (Fig. 5d). This suggests that Recent wings evolved from the respiratory lobe of an ancestral polyramous limb, probably first appearing in the immature aquatic stages as gill-like structures, such as those found on all trunk segments of extinct Paleodictyoptera or extant mayfly larvae (Fig. 5d). Wings subsequently emerged as adult appendages and acquired greater strength and flexibility for sustained flight (Fig. 5e).

Interestingly, if the respiratory epipodite origin of insect wings is correct, then wings may have an even deeper origin, not just in the aquatic ancestor of pterygotes, but in lobopodans. The origin of the biramous limb has been postulated to involve the fusion of the gill-like structure of lateral lobes (such as those in Opabinia) with the ventral lobopod43,45,46,47 (Fig. 5c). If this is true, then the wing may indeed be derived from lateral lobes, not of a terrestrial insect as once thought, but of a much more distant lobopodan ancestor.

Deep homology and origin of appendages

It is clear from the fossil record that chordates and arthropods diverged at least by the Cambrian. The appendages of these two groups are not homologous because phylogenetically intermediate taxa (particularly basal chordates) do not possess comparable structures. The most surprising discovery of recent molecular studies, however, is that much of the genetic machinery that patterns the appendages of arthropods, vertebrates and other phyla is similar. These findings suggest that the common ancestor of many animal phyla could have had body-wall outgrowths that were organized by elements of the regulatory systems found in extant appendages. We now describe these similarities and use them to consider the origin of animal limbs.

In Drosophila, the anteroposterior (AP) axis of the leg or wing imaginal disc (the larval precursor to the adult appendages) is divided into two compartments (reviewed in ref. 49). The posterior half of the disc expresses the gene hedgehog72,73, which encodes the key signal that initiates AP patterning (Fig. 6). In response to hedgehog, a thin layer of cells running along the border of the anterior and posterior compartments is induced to produce another secreted protein encoded by the gene decapentaplegic (dpp)74. dpp, in turn, is a long-range signal providing positional information, and hence differential AP fates, to cells in both compartments75,76,77,78,79. Misexpression of either hedgehog or dpp in the anterior of the disc results in AP mirror-image duplications of limb structures74.

Figure 6: A cladogram of selected metazoans shows the distribution of major genes involved with appendage development.
figure 6

Homologous signals are deployed in similar locations in the limb primordia of arthropods (a, Drosophila) and vertebrates (b, chick). Equivalent orientations (dorsal up, anterior left) of a chick left wing bud and a Drosophila left wing imaginal disc are shown. Sonic hedgehog (red) in the chick, and its homologue hedgehog (red and purple, where coexpressed with apterous) in the fly are produced in a posterior domain. These factors induce the expression of secondary patterning signals in the appendages: overlapping expression of Bmp-2 (green) in the chick and adjacent expression of dpp (green), the Bmp-2-homologue, in the fly. Dorsal cell fates are specified in both systems by LIM homeodomain transcription factors expressed throughout the dorsal half of the appendage primordia: Lmx-1 (blue) in the chick, and apterous (blue and purple, where coexpressed with hedgehog) in the fly. The outgrowth of both appendages is driven by a specialized group of cells (the AER in the chick and the wing margin in the fly) running along the anteroposterior axis at the junction of the dorsal and ventral compartments (yellow). These key groups of cells are specified in both the chick and the fly by the border between dorsal cells expressing the gene fringe and ventral cells not expressing fringe. For simplicity in viewing, conserved genes in signal transduction (such as the hedgehog receptor, patched) and other parallels between the two systems (such as the expression of Wnt genes) are not shown. Distal-less homologues (Dll in Drosophila and Dlx in the chick) are drawn in orange. Distal-less orthologues are expressed in a wide variety of animal appendages, including the lobopods of onycophorans, the tube feet of echinoderm, and the wings of birds and flies (orange). The limbs of these taxa are not homologous as appendages because phylogenetically intermediate groups do not possess comparable structures. This suggests at least two phylogenetic possibilities: either similar genetic circuits were convergently recruited to make the limbs of different taxa, or these signalling and regulatory systems are ancient and patterned a different structure (presumably another type of outgrowth) in the common ancestor of protostomes and deuterostomes99.

The AP axis of the vertebrate limb is set up in a very similar manner (Fig. 6). A key organizing signal is Sonic hedgehog (Shh), one of the three direct homologues of the Drosophila gene hedgehog80,81,82,83. Like hedgehog, Shh is localized posteriorly in the limb bud. Misexpression of Shh anteriorly causes AP mirror-image duplications analogous to those caused by hedgehog misexpression in the fly imaginal disc74,83. In addition, Bmp-2, one of the two vertebrate homologues of the arthropod dpp signalling protein, is expressed in the limb bud in response to Shh23. Unlike Drosophila dpp, Bmp-2 does not have the ability to cause full limb duplications. However, it clearly functions as a secondary signal in the Shh pathway, polarizing the overlying ectoderm84.

Signals organizing proximodistal (PD) outgrowth in vertebrate and insect appendages also operate similarly. In the insect imaginal wing disc, PD outgrowth is organized by a specialized set of cells running the length of the dorsoventral (DV) border, the ‘wing margin’ (Fig. 6). The dorsal compartment of the wing is characterized by the expression of the transcription factor apterous. apterous specifies dorsal-specific cell fate69 and controls the expression of a secreted protein called fringe85. The interface between cells expressing and cells not expressing fringe becomes the wing edge or margin85. A key downstream effector of fringe activity is encoded by Serrate86. In response to fringe, Serrate is induced, leading to the activation of several downstream effector genes87,88,89,90 and the production of a signal at the margin which organizes the growth of the wing blade87,88,89.

Unexpectedly, outgrowth of vertebrate limbs appears to be established by a very similar genetic cascade. Outgrowth of the limb bud is driven by signals from a specialized ectodermal structure, the apical ectodermal ridge91 (AER), which, like the wing margin, runs along the DV border of the limb. Remarkably, a vertebrate homologue of fringe, called Radical-fringe, is expressed in the dorsal half of the limb ectoderm prior to formation of the AER92. At the border between cells expressing Radical-fringe and cells not expressing Radical-fringe, a homologue of Serrate, Ser-2, is induced and the AER forms. A Radical-fringe boundary is required to form the AER and ectopic radical fringe can induce an additional AER on the ventral surface93.

There are also parallels between the regulation of the DV axis in vertebrate and arthropod appendages (Fig. 6). Genes specifying DV polarity in both groups have been identified. In Drosophila, the early ventral expression of the gene wingless, a member of the Wnt family of secreted factors, is necessary for the proper DV patterning of the wing94,94. Subsequently, the expression of the transcription factor apterous defines the dorsal compartment and specifies dorsal cell fates69. In the vertebrate limb, the early expression of a different Wnt family member is also required for DV patterning. Wnt7a is specifically expressed throughout the dorsal ectoderm and is necessary and sufficient for many aspects of dorsal patterning96,97,98. Wnt7a acts by inducing mesodermal expression of Lmx-197,98. Like apterous, Lmx-1 is a related member of the LIM-homeodomain family of transcription factors. As with apterous (Drosophila), Lmx-1 (vertebrates) expression defines a dorsal compartment, being expressed early throughout the dorsal half of the limb bud, and it is sufficient to convey dorsal cell fate97,98.

The simplest phylogenetic implication to draw from these comparisons is that individual genes that are expressed in the three orthogonal axes are more ancient than either insect or vertebrate limbs (Fig 6). Indeed, several of the regulatory systems seen in arthropod and vertebrate limbs are also involved in the development of other organs in a variety of taxa. The phylogenetic distribution of regulatory circuits and morphological structures presents two major interpretations: either similar genetic circuits were convergently recruited to make the limbs of different taxa or a set of these signalling and regulatory systems are ancient and patterned a structure in the common ancestor of protostomes and deuterostomes99.

The first model holds that genes and/or genetic circuits were convergently recruited for limb development during the evolution of vertebrates and arthropods. These genes would not be involved in appendage development in the common ancestor of vertebrates and arthropods; each gene or circuit was involved in other developmental events. This notion would require the parallel cooption of members of similar gene families, acting along different developmental axes to pattern an outgrowth of the body wall in at least two taxa. The evolution of limbs in each group would, then, have involved the convergent recruitment of numerous genes to define similar developmental axes. If confirmed, this hypothesis would provide a stunning case of convergent evolution.

The second model is that some of these genes or circuits were components of an ancestral genetic regulatory system that was used to pattern a structure in the common ancestor of vertebrates and arthropods. This ancestral structure need not have been homologous to arthropod or vertebrate limbs; the regulatory system could have originally patterned any one of a number of outgrowths of the body wall in a primitive bilaterian for example. The genes themselves were initially involved in other developmental events; the key step in animal limb evolution was the establishment of an integrated genetic system to promote and pattern the development of certain outgrowths. Once established, this system provided the genetic and developmental foundation for the evolution of structures as diverse as wings, fins, antennae and lobopodia.

The evaluation of these two alternative models requires consideration of several factors that could affect the identity and deployment of appendate-patterning genes. First, given the independent histories spanning more than 500 Myr of the lineages being compared, one should expect regulatory differences among patterning systems. Even the insect wing and leg, which are both probably derived from an ancestral polyramous appendage, have acquired different patterning mechanisms. Because individual regulatory components appear to have been gained, lost and modified during insect wing and leg evolution there is no apriori reason to expect that either structure would be more genetically similar to vertebrate limbs. Second, one member of a gene family may substitute for another during normal development. Indeed, some of the genes that are deployed similarly in arthropods and vertebrates are not strictly orthologous. Such substitution could be the product of either convergent evolution or descent with modification (by substitution) among redundant genes. Third, it could be argued that the presumed inversion of the DV body axis during deuterostome evolution would imply a corresponding inversion of appendage DV axis patterning signals, an expectation that is not met by the observed expression of apterous- and fringe-related genes in vertebrate limb buds. A parallel in reversal of the DV axes of the body and the limbs would be expected if vertebrate and insect limbs were structurally homologous to a common ancestral appendage which predated the reversal. However, we know that the modern vertebrate limb evolved after the DV inversion of the body axis. Thus, a DV inversion in the patterning of the limbs would not necessarily be predicted, whether the axis patterning genes were independently coopted for appendage formation according to the first model, or whether they were coopted as a unit, but regulated independently of the body axis, according to the second model.

One can argue many ways from the comparison of only two taxa: the alternative phylogenetic hyopotheses need to be tested by additional comparative data. Evidence in support of an ancient common mechanism for the formation of outgrowths of the body wall comes from phylogenetic comparison of the expression of the transcription factor Distal-less (Dll)100 (Fig. 6). Dll is expressed at the distal end of growing insect limbs55,60,61, and is essential for appendage outgrowth59. Dll orthologues are expressed in the distal portion (AER) of the embryonic limb buds of vertebrates, the ampullae and siphons of tunicates, the tubefeet of echinoderms, the parapodia of annelids, as well as Onychophoran lobopodia100. The expression of Dll-related genes could represent convergent utilization of the gene. However, the fact that out of the hundreds of transcription factors that potentially could have been used, Dll is expressed in the distal portions of appendages in six coelomate phyla makes it more likely that Dll was already involved in regulating body wall outgrowth in a common ancestor of these taxa (Fig. 6). The additional parallels between vertebrate and arthropod limbs suggest that this ancestral outgrowth may have also been patterned along the three orthogonal axes.

If a conserved outgrowth patterning system was available for co-option in the evolution of vertebrate limbs, then it must have been used in patterning non-limb outgrowths in basal taxa. Genetic studies provide an example of at least one secondary outgrowth patterned along these axes that predated the evolution of vertebrate limbs: the branchial arches. As the branchial arches grow out from the cranial region of the chick embryo, they express important components of the limb patterning system in similar developmental regions101,102. Like arthropod and vertebrate limbs, the branchial arches contain localized, posterior expression of a hedgehog gene, in this case Shh101,102. Furthermore, Shh is coexpressed with Bmp posteriorly101. Yet more similarities lie in the DV and AP axes: fringe-expressing cells are initially confined to the dorsal ectoderm and later are restricted to distal regions of the outgrowth where Distal-less orthologues are also expressed93. The ectopic deployment and modification of an existing patterning program, such as that of the branchial arches, may have given rise to the predecessors of vertebrate appendages.

Determination of whether two structures are homologous depends on the hierarchical level at which they are compared103,104,105. For example, bird wings and bat wings are analogous as wings, having evolved independently for flight in each lineage. However, at a deeper hierarchical level that includes all tetrapods, they are homologous as forelimbs, being derived from a corresponding appendage of a common ancestor. Similarly, we suggest that whereas vertebrate and insect wings are analogous as appendages, the genetic mechanisms that pattern them may be homologous at a level including most protostomes and deuterostomes. Furthermore, we propose that the regulatory systems that pattern extant arthropod and vertebrate appendages patterned an ancestral outgrowth and that these circuits were later modified during the evolution of different types of animal appendages. Animal limbs would be, in a sense, developmental ‘paralogues’ of one another; modification and redeployment of this ancient genetic system in different contexts produced the variety of appendages seen in Recent and fossil animals.