The urodeles (salamanders and newts) have long been useful animals for research on the mechanisms of regeneration (Thornton,1968; Stocum,1995; Mescher,1996; Brockes,1997). Like other vertebrates, they regenerate tissues such as blood, bone, muscle, and epithelia by the proliferation and differentiation of reserve stem cells. Their uniqueness lies in their ability to regenerate a wide range of complex structures by creating local stem cells by cellular dedifferentiation. These stem cells are renewable, not in the conventional sense of reserve stem cell self-renewal by asymmetric division but in the sense that they can be produced anew from differentiated tissues in successive rounds of regeneration. Regeneration of the limb has been most extensively analyzed because of its size, ready accessibility, and ease of surgical and chemical manipulation. These features, combined with an emergent molecular biology (Geraudie and Ferretti,1998), make it certain that the urodele limb will remain a premier model for analyzing the mechanisms of vertebrate regeneration.
Limb regeneration follows a typical course in all urodele species (Iten and Bryant,1973; Smith et al.,1974; Tank et al.,1976; Stocum,1979; Young et al.,1983). Reserve stem cells in the basal layers of the epidermis migrate to cover the wound surface within a few hours after amputation (Repesh and Oberpriller,1978). Cartilage, muscle, Schwann cells, and dermal and muscle fibroblasts dedifferentiate to become mesenchymal-like stem cells, which accumulate under the wound epidermis to form a regeneration blastema. The blastema grows by proliferation of these cells, which then differentiate to reproduce the missing structures. In the larva of the salamander Ambystoma maculatum, the new limb parts are restored by 21 days after amputation at room temperature (21–23°C; Stocum,1979).
The mechanisms of dedifferentiation have been reviewed elsewhere (Stocum,1995; Brockes,1997; Brockes and Kumar,2002). Here, we focus on the epithelial and neural factors required for blastema cell proliferation and on how the cells are organized into the structural patterns of the regenerate. The majority of the work cited has been done on Ambystoma larvae, but observations and experiments on regenerating adult newt and Xenopus limbs are included as well. In addition, we incorporate relevant data on amniote limb bud development because, although the cells of the limb bud and regeneration blastema are derived by different means, they appear to use largely similar (although not identical) proliferation and patterning mechanisms.
ROLE OF THE APICAL EPIDERMAL CAP AND THE NERVES IN BLASTEMA CELL PROLIFERATION
As the blastema forms, it is reinnervated by regenerating sensory and motor fibers of spinal nerves III, IV, and V (Van Stone,1955; Piatt,1957), stimulated by neurotrophic factors produced by the blastema cells (Tonge and Leclere,2000). Simultaneously, the wound epidermis thickens to form an apical epidermal cap (AEC) that is the morphologic and functional equivalent of the apical ectodermal ridge (AER) in amniote and anuran limb buds (Saunders,1948; Christensen and Tassava,2000). A basement membrane is not re-established under the AEC until the immediately subjacent blastema cells begin to differentiate. Urodele limb buds lack a morphologic AER (Sturdee and Connock,1975; Tank et al.,1976), but the nonthickened apical epidermis of the growing urodele limb bud has the same outgrowth-promoting function as the AER (Steiner,1928; Balinsky,1935).
When they first form during ontogenesis, limb buds do not exhibit an AER. The AER then forms and is required for further outgrowth. Removal of the AER truncates the limb in the proximodistal (PD) axis, the degree of truncation being greater the earlier the AER is removed (Saunders,1948; Tschumi,1957). Truncation is due to cell death in the apical 200 μm of the mesenchyme (Rowe et al.,1982; Dudley et al.,2002). Late in its development, the amphibian limb becomes dependent on nerves, as well as an AEC, for regeneration. The AEC and a threshold number of nerves are required for both initial blastema formation and further outgrowth. Preventing the formation of a wound epidermis or complete denervation of the limb prevents blastema formation (Singer,1952,1965; Goss,1956a; Mescher,1976; Tassava and Garling,1979; Loyd and Tassava,1980). The problem is not a failure of cells to dedifferentiate but a lack of proliferation of dedifferentiated cells, accompanied by cell death. Terminal deoxynucleotidyl transferase–mediated dUTP nick end-labeling (TUNEL) reactions indicate that, in axolotls, the dedifferentiated cells of denervated limbs undergo apoptosis and are removed by macrophages (Mescher et al.,2000).
Once a regeneration blastema has been established, proliferation of its cells requires continued interaction with the AEC. Removal of the wound epidermis from blastemas of various stages results in PD truncation, the severity of truncation being greater the earlier the epidermis is removed (Polezhaev and Faworina,1935; Goss,1956b; Stocum and Dearlove,1972). Whether pattern truncation in epidermis-free blastemas in vivo is the result of apical cell death, as in amniote limb buds, has not been investigated, although it is known that the pulse labeling and mitotic indices of epidermis-free axolotl and adult newt limb blastemas cultured in vitro in the presence of sensory ganglia are reduced three- to fourfold and the cells differentiate prematurely into cartilage (Globus et al.,1980; Smith and Globus,1989).
Blastemas become nerve-independent for their differentiation and morphogenesis after attaining a critical size (stage), but always remain nerve-dependent for mitosis and thus growth. In Ambystoma larvae, denervation at any time up to the early bud stage results in total regression of the limb. Denervation between early and late bud stages allows variable degrees of regeneration, but blastemas denervated at later stages are all able to undergo patterned differentiation and morphogenesis into miniature regenerates (Schotte and Butler,1944; Maden,1981), even though the mitotic index is drastically reduced and eventually falls to zero (Goldhamer and Tassava,1987). Adult newt limb blastemas seem less susceptible to resorption and can form miniature regenerates in the absence of nerves after attaining the early bud stage (Singer and Craven,1948). When denervation is performed after nerve independence is attained, the cells apparently arrest in G1 (Maden,1979; Oudkhir et al.,1985). It is not known whether there is any significant cell death in these denervated blastemas.
A major issue in limb regeneration is the identification of survival and proliferation factors provided to blastema cells by the AEC and the nerves. To be identified as an AEC or neural proliferation factor for blastema cells, a molecule synthesized by these tissues must satisfy four criteria (Brockes,1984): (1) the molecule must be secreted into the blastemal mesenchyme; (2) removal or destruction of the AEC or nerves should reduce the level of the molecule in the blastema; (3) the molecule should be able to substitute for the AEC or nerves in promoting mitosis in vitro or in vivo and/or promoting regeneration to digit stages when supplied exogenously; (4) neutralizing the synthesis or activity of the molecule in vitro or in vivo should decrease mitosis of blastema cells. Similar criteria would hold for factors synthesized by the AER of the limb bud.
In the amniote limb bud, Fgf-4 and -8 appear to be two factors synthesized by the AER that meet these criteria (Martin,1998; for review, Meyers et al.,1998; Moon and Capecchi,2000; Sun et al.,2000). Fgfs administered in beads to early chick limb buds after AER removal or amputation of the distal limb tip prevent cell death and allow development to digit stages (Niswander et al.,1993; Fallon et al.,1994; Taylor et al.,1994). Fgf-4 and -8 genes can be selectively knocked out in mouse limb buds by inserting the Msx2-cre transgene (Sun et al.,2002). These double knockouts completely prevent Fgf-4 and -8 expression in nascent hindlimb buds. The limb buds are very small due to subnormal recruitment of mesenchymal cells and subsequent cell death in the proximal part of the bud and do not develop further. In the forelimb buds, Msx2-cre inactivates the Fgf-4 and -8 genes somewhat later, allowing an initial burst of Fgf-4 and -8 synthesis to take place. This burst is enough to produce early limb buds of normal size, but proximal cell death then sets in, reducing the size of the limb bud. The final forelimbs possess all three PD segments, but the zeugopodium and autopodium are hypoplastic. The morphology and gene expression of the AER is normal in both hindlimb and forelimb knockouts until late stages of development, when the AER begins to degenerate. Genes normally expressed by the mesenchyme (shh, Hoxd-13, Bmp2, and -4, and others) could not be detected in hindlimb knockouts but were normally expressed in forelimb knockouts. On the basis of these observations, Sun et al. (2002) propose that the role of the AER is to ensure the presence of sufficient mesodermal cells to form each limb segment.
Given the similarity of AER and AEC function, Fgfs are logical candidates to be mitogenic signals of the AEC during regeneration. Fgf-1 and -2, but not Fgf-4, are synthesized by the AEC of the regenerating axolotl limb (Boilly et al.,1991; Mullen et al.,1996; Christensen et al.,2002; Dungan et al.,2002), and FGFR1 (receptor for Fgf-1, -2, -4) is expressed in the blastema stem cells of regenerating newt limbs (Poulin et al.,1993). Both Fgf-1 and -2 appear able to substitute for the AEC in promoting mitosis of blastema cells in vivo and in vitro (Chew and Cameron,1983; Albert et al.,1987), but the ability of these factors to sustain regeneration to digit stages in the absence of the AEC has not been tested. Cultured blastemas release a mitogenic activity into the medium that is inhibited by one-third when antibody to Fgf-1 is added (Zenjari et al.,1997). Thus, Fgf-1 satisfies possibly two of the criteria for mitogen status, and Fgf-2 satisfies possibly one; neither has been tested with regard to the other criteria. Fgf-8 is also expressed by the AEC of the axolotl blastema (Han et al.,2001; Christensen et al.,2002) but has not been tested in this species with regard to satisfying the mitogen criteria. None of the Fgfs synthesized by the AEC of the regeneration blastema have been shown to be secreted into the underlying mesenchyme.
The expression of Fgf-8 in both the AER and the AEC appears to be dependent on Fgf-10 from the limb bud and blastema mesenchyme. Additionally, the expression of Fgf-2 by the AEC is nerve-dependent during nerve-dependent stages of axolotl regeneration (Mullen et al.,1996).
Fgf-10 null mice do not form limbs (Martin,1998; for review, Sekine et al.,1999; Ohuchi et al.,2000). Fgf-10 is expressed in the limb bud and regeneration blastema of the Xenopus tadpole hindlimb, but not in amputated hindlimbs of Xenopus tadpoles approaching metamorphosis, which at best regenerate a cartilage spike (Christen and Slack,1997; Yokoyama et al.,2000). Exogenously supplied human recombinant Fgf-8 promotes a limited improvement of regeneration in these regeneration-deficient limbs. Exogenous human recombinant Fgf-10 promotes greater improvement, restoring the regeneration of foot structures while simultaneously inducing expression of Fgf-8 in the AEC but fails to restore more proximal elements (Yokoyama et al.,2001). FGFR1 and 2 are expressed in the wound epidermis and blastemal mesenchyme of regeneration-competent Xenopus hindlimbs but not in the wound epidermis of regeneration-deficient hindlimbs. Inhibitors of these receptors suppress the regeneration of regeneration-competent limbs (D'Jamoos et al.,1998). These observations indicate that, as in amniote limb bud development, Fgf-10 and Fgf-8 are key signaling molecules in the regeneration of Xenopus hindlimbs and that the expression of Fgf-8 by the AEC is dependent on Fgf-10. That proximal parts are missing in Fgf-10–treated regeneration-deficient animals suggests that other members of the Fgf family, or AEC growth factors outside this family are necessary for complete regeneration. Alternatively, human Fgf-10 may not interact very specifically with amphibian Fgf receptors because of structural dissimilarities, leading to deficient regeneration. The axolotl Fgf-10 protein exhibits only 62.4% identity with human Fgf-10 (Christensen et al.,2002). Fgf-10 is also expressed in the mesenchyme of the axolotl regeneration blastema (Han et al.,2001), inferring the existence of a similar dependency, but this inference has not yet been tested.
Several nonsecreted molecules are expressed by the AEC that may play a role in the structure and function of the AEC. An unidentified antigen called 9G1 is expressed in the AEC and is down-regulated by denervation through an effect of the nerves on the blastemal mesenchyme (Onda and Tassava,1981). The Dlx3 gene, a homeobox orthologue of Drosophila distal-less, which is required for leg outgrowth in the fly (Cohen and Jurgens,1989), is strongly expressed in the wound epidermis of regenerating newt and axolotl limbs and tail (Beauchemin and Savard,1992; Mullen et al.,1996). Expression peaks just before redifferentiation and decreases to zero by late digit stages (Mullen et al.,1996). The expression of Dlx3 is down-regulated by denervation. The expression of two helix-loop-helix (HLH) inhibitors of differentiation, Id2 and HES1, is up-regulated in the AEC of regenerating newt limbs, and a third gene, Id3, is expressed equally in the AEC and the blastema stem cells (Shimizu-Nishikawa et al.,1999). The transcription factor Msx-2 is also strongly expressed very early in the wound epidermis of regenerating axolotl limbs (Koshiba et al.,1998; Carlson et al.,1998). The specific functions of these gene products are unknown, but they might play a role in regulating the production of proliferation factors.
Fgf-2, Ggf-2, substance P, and the iron transport protein, transferrin, are all found in the axons or cell bodies of limb nerves and have been implicated as neural mitogens for blastema cells. Fgf-2 and Ggf-2 meet one of the mitogen criteria, that of being able to substitute for the nerves. Fgf-2 elevates the mitotic index of blastema cells in vitro (Albert et al.,1987). Fgf-2 and Ggf-2 impregnated beads implanted into denervated blastemas allow them to develop to digit stages (Mullen et al.,1996; Wang et al.,2000). Substance P meets two of the four criteria. At low concentrations, it promotes blastema cell proliferation in vitro. Conversely, antibody to substance P decreases the effectiveness of spinal ganglia to promote mitosis of blastema cells cultured trans-filter to the ganglia (Globus,1988). Whether denervation decreases the level of Fgf-2, Ggf-2, or substance P in the blastema or whether neutralizing Fgf-2 or Ggf-2 abolishes the effect of the nerve is unknown, nor is it known whether Fgf-2, Ggf-2, or substance P are actually secreted into the blastema.
Transferrin has been shown to satisfy virtually all the criteria for mitogen status. Iron is a cofactor for many enzymes crucial to cell proliferation, including the rate-limiting enzyme for DNA replication, ribonucleotide reductase (Mescher and Munaim,1988; Sussman,1989). Transferrin is axonally transported and released distally from the sciatic nerves of regenerating axolotl hindlimbs (Kiffmeyer et al.,1991). Transection of the brachial nerves lowers the concentration of transferrin in the blastema by 50% (Mescher and Kiffmeyer,1992). Transferrin can stimulate blastema cell proliferation in vitro as effectively as nerve extracts (Munaim and Mescher,1986; Albert and Boilly,1988) and can maintain a significant level of mitotic activity when administered locally to denervated blastemas in vivo (Mescher and Kiffmeyer,1992). The growth-promoting activity of neural extracts on organ cultures of denervated axolotl blastemas is completely removed by anti-transferrin antiserum and restored by purified axolotl transferrin (Mescher et al.,1997). Chelation of ferric ions from the extracts abolishes the mitogenic effect of the extracts; the activity is restored by adding iron back to the extract (Munaim and Mescher,1986).
To summarize, a variety of growth and trophic factors found in the AEC and the brachial nerves appear to be essential for the survival and proliferation of blastema stem cells. There are clear differences in the number and types of factors required by the regeneration blastema as opposed to the limb bud. However, the growth factors implicated (Fgf-1, -2, -8, Ggf-2) have not been as rigorously tested for mitogen status as transferrin or the growth factors of the amniote AER (Fgf-2, -4, -8).
DIFFERENTIATION AND PATTERN FORMATION IN THE BLASTEMA
Self-Organization of the Blastema
Although individual blastema cells have some developmental plasticity beyond their cell type of origin (Brockes,1997), the developmental fate of the blastema as a whole is determined according to origin (Stocum,1984,1996,2000). Numerous experiments have shown that the blastema is a self-organizing system, as opposed to being patterned by a set of inductive signals from the mature tissues adjacent to it. Blastemas cultured in vitro, grafted to the dorsal fin, exchanged between forelimb and hindlimb, grafted to a different PD level, or manipulated to disharmonize the anteroposterior (AP) and/or dorsoventral (DV) axes of blastema and adjacent tissues, always develop according to origin with regard to limb type, limb level, and handedness, even when the cells of the graft are forced to dedifferentiate again by reinjury (Stocum,1968a,b,1978b,1980a; Stocum and Melton,1977). When a distally derived blastema is grafted to a more proximal level, the missing intermediate structures are filled in by dedifferentiation and intercalary regeneration from the host limb level (Stocum,1975; Iten and Bryant,1975; Pescitelli and Stocum,1980). Intercalation does not take place when a proximal blastema is grafted to a more distal level (Stocum and Melton,1977). There is, thus, a preferred polarity to the recognition of a structural discontinuity and intercalary regeneration in the PD axis. This preference would not be predicted a priori, and its basis is unknown.
Likewise, when the AP or DV axis of the blastema is reversed with respect to the adjacent tissues, confronting anterior and posterior, or dorsal and ventral blastema cells, a supernumerary circumference is formed, within which intercalary regeneration along the radii takes place to generate a cross-section that grows out into a supernumerary limb (Bryant and Iten,1976; Cameron and Fallon,1977; Tank,1978a; Holder and Tank,1979; Stocum,1980b). Both host and graft tissues contribute to the supernumerary limb to varying degrees (Stocum,1982). Such supernumeraries are also generated after amputation of limbs in which extensor and flexor muscles or skin cuffs are rotated (Carlson,1974,1975).
The fibroblasts of the dermis carry the information for the self-organizing pattern, as shown by experiments in which unirradiated skin (epidermis plus dermis) was grafted to irradiated limbs in axolotls (Namenwirth,1974; Lheureux,1975; Dunis and Namenwirth,1977). The irradiated host muscle and cartilage tissues cannot contribute to the regeneration blastema that forms after amputation of these limbs. Nevertheless, the regenerate that forms, although lacking in muscle, has a normal skeletal pattern, indicating that the blastema cells derived from the dermal fibroblasts of the unirradiated skin are capable of organizing the pattern. Furthermore, if tail skin is grafted onto irradiated limbs, or limb skin onto irradiated tails, the regenerates formed after amputation are entirely of graft character, demonstrating that the fibroblasts carry an appendage-specific pattern (Trampusch,1958a,b).
A Model for Self-Organization
The development of the blastema according to origin, in concert with intercalary regeneration to eliminate discontinuities, suggests a model for self-organization based on intercalation between boundaries (Stocum,1980b,1984,1996) (Fig. 1). The model assumes that limb cells have positional identities in three-dimensional space that constitute a “normal neighbor map” (Mittenthal,1981). During initial dedifferentiation, before there is even a visible blastema, the essential outline of what is to be regenerated, consisting of circumferential, proximal, and distal boundaries, is present at the amputation surface. All dedifferentiated cells inherit the circumferential and proximal boundaries. We postulate that the distal boundary is conferred on the distal-most blastema stem cells in contact with the AEC. There is no hard evidence that the early wound epidermis functions in establishing a distal boundary, but it does confer a distal direction of PD outgrowth on the blastema. Shifting the AEC to an eccentric location by removing a patch of skin at the base of the blastema in Ambystoma larvae results in outgrowth of the regenerate at an angle to the stump; grafting an AEC to the base of the blastema results in production of a supernumerary limb (Thornton,1960,1962; Thornton and Thornton,1965).
The intermediate positional identities are then intercalated by local cell interactions until a normal neighbor map is restored. The initial interaction is between the proximal and distal boundary stem cells, which recognize the discontinuity between these boundaries. This recognition triggers cell division, which is promoted by growth and trophic factors from the AEC and nerves, accompanied by the progressive assignment of the intermediate positional identities to the progeny cells until a complete map is reconstituted (French et al.,1976; Maden,1977; Mittenthal,1981; Stocum,1978b,1980b,1982,1996). The first PD identities to be established are those for each segment. This is a very early event, so that all the segments are represented by the time the initial blastema has formed. Each segment then expands by proliferation, accompanied by the intercalation of intrasegmental identities. Proximal segments of the regenerate might expand faster than distal within the blastema, or vice versa, or all could expand at equal rates.
Intercalation follows three rules. First, blastema cells and their progeny can change their positional identities only in a distal direction along the PD axis and in any direction within the circumferential boundary to intercalate the missing parts of the pattern (“rule of distal and radial intercalation”). Second, given a choice of intercalating the longer or shorter number of positional identities that will restore a complete pattern, cells always choose the shorter route (“rule of shortest intercalation,” French et al.,1976). Third, when all positional identities have been filled in, intercalation stops (“rule of normal neighbors,” Mittenthal,1981).
The distal pattern truncation of established blastemas that are deprived of their AEC could then be explained in the same way as that for the amniote limb bud deprived of its AER, by the death of cells destined to form the distal structures. This presumably would contrast with the lack of pattern truncation in blastemas deprived of nerves, in which blastema cell division would be diminished, but apical cell death would be absent, thus preserving the PD pattern or allowing morphallactic regulation to produce miniature regenerates. The validity of these ideas, however, awaits a detailed analysis in vivo of cell death and proliferation in blastemas deprived of epidermis vs. those deprived of nerves.
A similar model has been proposed for development of the embryonic urodele limb bud (Stocum and Fallon,1982) and may be applicable to amniote limb buds as well. The initial limb bud cells would have a proximal default state that is converted to distal in those cells in contact with the AER, followed by intercalation of intermediate positional identities. Labeling studies using [3H]thymidine and DiI indicate that cells representing all three segments of the limb are present very early in chick limb bud development (by stage 18 or 19) and then expand at different rates, proximal faster than distal (Stark and Searls,1973; Dudley et al.,2002; Saunders,2002, for review). Furthermore, the distal-most cells of stage 19 wing buds self-organize into wing digital elements when grafted to the coelom or to a hind limb bud stump, indicating that they are already destined to become wing digits (Dudley et al.,2002). They are not yet determined to do so, however, because stage 20 apical mesenchyme cells (destined to form digits) form all three segments when dissociated, repacked into an ectodermal jacket, and grafted to a host embryo. Apical cells from stage 22 formed zeugopodium and autopodium, whereas stage 24 cells formed only autopodium, indicating that segmental determination takes place in a proximal-to-distal sequence (Dudley et al.,2002). These results are consistent with the findings of Chiang et al. (2001) and Litingtung et al. (2002) of an early PD prepattern in the mouse limb bud that can be expressed in the absence of AP or DV signaling centers.
Maximum Circumferential and Radial Positional Diversity Is Required for Normal Pattern Formation
There is a relationship between the amount of pattern regenerated and the number of positional identities in the circumference and radii of the regenerating limb. This relationship was discovered in experiments that restricted the number of positional identities at the amputation surface. Irradiated newt limbs wrapped around their circumference with longitudinal strips of unirradiated skin from a small arc of a single quadrant, failed to regenerate after amputation through the graft, whereas irradiated limbs that received grafts of un-irradiated skin strips taken from four quadrants of the limb regenerated perfectly (Lheureux,1975). Similarly, blastema formation is inhibited in amputated double half posterior and ventral limbs treated with retinoic acid (RA) to completely posteriorize and ventralize circumferential and radial positional identity (Kim and Stocum,1986; Ludolph et al.,1990; see below). Forelimb and hindlimb stylopodia constructed from two anterior half circumferences (double anterior limbs) form a blastema, but regenerate what appears to be only a tapered extension of the humerus or femur (Bryant,1976; Stocum,1978a; Tank,1978b; Tank and Holder,1978; Holder et al.,1980). Double anterior zeugopodia, on the other hand, regenerate a PD-complete but distally convergent symmetrical pattern that ends in one or two digits (Stocum,1978a; Krasner and Bryant,1980).
Interpreted in terms of our model, dedifferentiation and the establishment of distal and proximal boundaries would take place after amputation of these limbs. However, in the case of limbs with only one or two circumferential and radial identities (skin strips or RA-treated limbs), there would be immediate radial convergence of the cross-sectional pattern and no PD intercalation. In amputated double anterior limbs, there are two half cross-sections containing anterior, dorsal, and ventral positional identities. Thus, circumferential, radial, and PD intercalation can all take place, although to a more limited extent than normal. The cross-sectional pattern of the regenerate becomes progressively converged distally, due to the fact that circumferential and radial interactions are taking place in a conical compartment, which forces cells into a smaller radius of interaction toward the tip. There is less convergence in double anterior zeugopodia, because only the autopodium must be regenerated, but it is more severe in double anterior stylopodia due to the greater amount of pattern to be regenerated distally. PD limb segments cannot be discerned in the regenerate, because radial convergence masks the jointed PD pattern, thus giving the illusion of a truncated stylopodium.
Pattern convergence is relieved if double anterior or normal blastemas from the distal zeugopodium are grafted to double anterior stylopodia (Stocum,1980b,1981). In this case, a distal stylopodium and a single symmetrical zeugopodial element representing the two anterior bones fused in the midline are intercalated from the stylopodium. This result, too, is consistent with the evidence of Chiang et al. (2001) and Litingtung et al. (2002) that PD patterning is not dependent on a posterior signaling center, such as the zone of polarizing activity (see below). The relief of pattern convergence in these constructs might be because PD intercalation is taking place within a cylindrical compartment, as opposed to a conical compartment. It would be instructive to repeat these experiments, analyzing the expression of molecular markers associated with axial patterning, such as the Hoxa and d genes (see below), particularly to see whether the tapered regenerates of double anterior stylopodia express markers associated with each of the limb segments.
Use of Retinoids to Probe the Physical Nature of Positional Identity
The signals that recognize discontinuity, trigger proliferation, and specify positional identities in either regeneration blastemas or limb buds are unknown, but most likely would be short range diffusible molecules or signals generated by contact between cell surface molecules of adjacent cells or their extracellular matrix. In vitro engulfment and in vivo displacement assays in fact indicate that positional identity of blastema stem cells is encoded in the cell surface. Proximal blastema mesenchymes engulf distal ones when pressed together and cultured in hanging drops (Nardi and Stocum,1983). Consistent with Steinberg's (1978) differential adhesion hypothesis, this result indicates that blastemas derived from distal levels are more adhesive than those derived from proximal levels. The differential adhesivity of blastemas derived from different PD levels was also demonstrated by an “affinophoresis” assay in vivo (Fig. 2). In this assay, axolotl blastemas derived from the wrist or elbow levels of the forelimb were grafted to the blastema-stump junction of a hindlimb host regenerating from the mid-thigh (Crawford and Stocum,1988a; Egar,1993). The grafted blastemas were displaced distally (comparable to sorting out in vitro) to their corresponding levels on the host regenerate (ankle and knee), where they differentiated. Furthermore, double half anterior wrist blastemas grafted to the femur level of an amputated hindlimb were too small to evoke intercalary regeneration; instead, they were displaced distally as the host hindlimb regenerated, winding up on the anterior side of the host foot (Stocum,1980b), demonstrating differential adhesivity in the circumference, as well as in the PD axis.
RA has been used in conjunction with affinophoresis and intercalary regeneration assays to define the physical nature of positional identity. RA administered during the stage of initial blastema formation proximalizes, posteriorizes, and ventralizes the positional identities of blastema cells in a concentration-dependent way (Niazi and Saxena,1978; Maden,1982,1997; Thoms and Stocum,1984; Niazi et al.,1985; Kim and Stocum,1986; Ludolph et al.,1990,1993; Stocum,1991; Monkmeyer et al.,1992; Niazi,1996). That is, cells adopt positional identities more proximal, posterior, and ventral to the identities of their position of origin. At the maximum concentration of retinoid, the cells of a wrist-level blastema will be proximalized to the level of the shoulder and a whole limb, complete with shoulder girdle, will regenerate from the level of the wrist. A direct correlation between positional identity and blastema PD level-specific affinity was demonstrated by using RA-treated medium bud stage blastemas. When the affinophoresis assay was performed using donor wrist and elbow blastemas proximalized by RA, distal displacement was abolished (Crawford and Stocum,1988a). Similarly, grafting an RA-treated wrist-derived blastema to a more proximal level abolished the intercalary regeneration that would normally have taken place between distal and proximal levels (Crawford and Stocum,1988b).
These results suggest that positional memory is a property of the cell surface and that RA alters gene activity affecting the qualitative and/or quantitative molecular composition of the cell surface. Consistent with this idea is that tunicamycin, which inhibits biosynthesis of the oligosaccharide component of N-linked glycoproteins on the cell surface, blocks the proximalization of positional identity by RA (Johnson and Scadding,1992). The gene for Prod1, the newt orthologue of CD59, a surface protein that regulates the later steps of the complement pathway, has been implicated in encoding positional memory of limb regeneration blastema cells (da Silva et al.,2002). Prod1 is expressed at higher levels in proximal blastemas vs. distal ones and is up-regulated by RA. Staining with antibodies to two peptides of Prod1 showed that it is anchored to the plasma membrane of blastema cells by glycosylphosphatidylinositol (GPI) and is released by bacterial phosphatidylinositol-specific phospholipase C (PIPLC). PIPLC and affinity-purified antibodies to the two peptides of Prod 1 inhibit proximal blastemas from surrounding distal ones in the in vitro engulfment assay, thus implicating this protein in local cell–cell interactions mediated by PD positional identity (da Silva et al.,2002).
Ephrin A and N-cadherin receptors and ligands have been implicated in defining PD positional identity in chick limb bud cells. Both ephrin A receptors and ligands and N-cadherin are expressed on the surface of chick limb bud cells, and ephrin A ligands are anchored to the plasma membrane by GPI. Proximal chick limb bud cells normally sort out from distal cells in vitro. PIPLC and antibodies to the ephrin A receptor or N-cadherin inhibit this sorting (Wada et al.,1993,1998; Wada and Ide,1994; Ide et al.,1994; Yajima et al.,1999). These molecules could also play a role in encoding positional identity in regeneration blastema cells, but this possibility has not been tested. Further studies will hopefully reveal other specific cell surface proteins that are correlated with positional identity in the axes of the regeneration blastema and limb bud.
Mechanism by Which Retinoids Modify PD Positional Identity
The biological effects of retinoids are mediated by nuclear receptors of the steroid/thyroid/vitaminD3 superfamily of transcription factors, the retinoic acid receptors (RARs) and the retinoid X receptors (RXRs; Manglesdorf et al.,1994). All-trans RA is the specific ligand for the RARs and 9-cis-RA is the specific ligand for the RXRs. When activated by RA, these receptors bind to retinoic acid response elements (RAREs) in the regulatory regions of target genes. Three major RAR isoforms, α, β, and δ, have been identified in the regenerating newt limb (Ragsdale et al.,1989,1992a,b; Hill et al.,1993; Maden,1997). The α and β isoforms are expressed uniformly throughout the blastema (Giguere et al.,1989). Delta has two isoforms, of which δ1 is the most strongly expressed in the blastema (Ragsdale et al.,1992b).
Despite that δ1 is the more strongly expressed RAR isoform, the δ2 isoform has been identified as the RAR that mediates RA-induced proximalization of positional identity. This finding was shown by an experiment grafting a distal blastema transfected with chimeric RAR receptor constructs to a more proximal level of a host limb. The receptor constructs consisted of the DNA binding regions of individual RARs linked to the ligand binding region of the thyroid hormone (T3) receptor (Pecorino et al.,1996; Fig. 3). This finding allows the experimenter to selectively activate the receptor by administering T3. The chimeric receptor and an alkaline phosphatase reporter gene coupled to a RARE were biolistically transfected into an ankle-level hindlimb blastema. The transfected blastema was then grafted to the mid-humeral level of the forelimb, and the animals were treated with T3 to activate the chimeric receptor, which in turn activated the reporter. As expected, grafting a control ankle-level blastema resulted in a regenerate consisting of a foot derived from the graft and intermediate structures derived by intercalary regeneration from the host forelimb. Transfected blastemas gave the same result as controls when the α, β, and δ1 DNA binding regions were used to make the chimeric receptor; i.e., the cells expressing the reporter were confined to the regenerated foot and the rest of the regenerate was derived from the host forelimb. This result indicated that these RARs did not mediate proximalization. By contrast, when the DNA binding region of the δ2 RAR was used to construct the chimeric receptor, many alkaline phosphatase-expressing cells were found toward the base of the intercalated intermediate structures, indicating that their adhesivity (positional identity) had been proximalized by T3 activation. It is important to point out that a completely proximalized ankle blastema (i.e., one in which every cell had been transfected) would have elicited no PD intercalation. However, the transfection efficiency in this experiment was 1 out of 100 cells. Thus, 99% of the graft cells were still distal and evoked intercalary regeneration. The 1% that were proximalized behaved as in an affinophoresis assay, migrating proximally to a position more compatible with their new positional identity (adhesivity). The result of this experiment also shows that blastema cells are individually proximalized by RA.
Transcription Factors and Signaling Molecules Involved in Specifying Positional Identity
At the level of the gene regulation mediated by the cellular interactions of limb ontogenesis and regeneration, patterning in the PD axis is associated with expression of the Hoxa and Hoxd transcription factors, known to be central mediators of axial patterning in vertebrates (McGinnis and Krumlauf,1992). Hoxa10, 11, and 13 are expressed in overlapping domains from proximal to distal in chick limb buds. At the same time, the Hoxd9–13 genes are expressed in an anteroproximal to posterodistal order, in overlapping and progressively smaller domains so that all of these genes are expressed together only in the posterodistal tip of the limb bud. Production of supernumerary digits after manipulations that repolarize anterior tissue results in activation of the Hoxd genes in a mirror-image pattern to the normal, suggesting that their expression may be directly or indirectly dependent on shh expression. These expression patterns create combinatorial domains correlated with PD segmentation (Stocum,1995, for review). Hoxa9 (proximal) and Hoxa13 (distal) expression has been analyzed in developing axolotl limb buds by Gardiner et al. (1995) and Gardiner and Bryant (1996). Hoxa9 is expressed throughout the limb bud mesoderm as it begins outgrowth at stage 36 and in the adjacent flank mesoderm. Expression is subsequently lost from the flank mesoderm but continues into the skeletal differentiation stages of development of the limb bud, except for a region at the very base of the bud. Hoxa13 expression is first detected in the distal mesenchyme at stage 37, when the limb region has just budded out. This distal restriction is maintained at later stages and is associated with development of the autopodium.
In the stem cells of the regeneration blastema, the Hoxd/a genes are expressed from proximal to distal in 3′ to 5′ order (Brown and Brockes,1991; Simon and Tabin,1993; Gardiner et al.,1995; Gardiner and Bryant,1996; Torok et al.,1998). Thus, Hoxd10 and A9 expression is associated with proximal structures and Hoxd13 and A13 expression is associated with distal structures. Hoxd10 is expressed at a two- to threefold higher level in proximal vs. distal blastemas, and proximalization of distal blastemas by RA is associated with up-regulation of Hoxd10 (Simon and Tabin,1993). Hoxa9 and Hoxa13 are both expressed from 1 day after amputation until redifferentiation begins in regenerating axolotl limbs (Gardiner et al.,1995). Each is expressed uniformly throughout the blastema until early bud, when expression of Hoxa13 starts to become restricted toward a more distal domain. By the late bud stage, Hoxa13 expression is restricted to the region that will form the autopodium, suggesting that this gene is involved in patterning of autopodal structures. Hoxa13 expression in distal blastemas is 30% higher at the medium bud stage than in proximal blastemas. Proximalization of a distal blastema by RA reduces the expression of Hoxa13 by 55%, again suggesting that this gene is involved in distal patterning (Gardiner et al.,1995). Hoxa11 expression has been examined in regenerating newt limbs (Brown and Brockes,1991; Simon and Tabin,1993; Beauchemin et al.,1994). There is a three- to fivefold higher level of expression in mid-upper arm blastemas vs. distal lower arm blastemas. Expression of the gene is unaffected by RA administered at either 5 days after amputation or at the medium bud stage.
The expression of both proximal and distal Hox genes within 1 day after amputation, the subsequent restriction of distally expressed Hox genes to the prospective autopodium, and the down-regulation of these distal genes by RA, is consistent with self-organization by means of intercalary regeneration within distal and proximal boundaries that are present from the start of regeneration. This restriction may correspond to determination of the PD segmental pattern. The mechanism of restriction of Hoxa13 to distal cells is unknown. We do not know whether HoxA9 and HoxA13 are expressed in the same cells of the early blastema, or in different cells. If in different cells, a self-organizing PD axis might be set up by the sorting of cells expressing Hoxa13 to a distal location, because they have an adhesive affinity for the AEC. If every cell is expressing both sets of genes, those cells in contact with the AEC might suppress the expression of Hoxa9 as a result of that association.
Two homeobox transcription factor genes that may play a role in newt limb identity are Hoxc6, which is expressed exclusively in forelimb blastemas; and HoxC10, which is expressed only in hindlimb blastemas (Tabin,1989; Simon and Tabin,1993). NvTbox1, a member of the T-box family of genes, is expressed exclusively in the mesenchyme of adult newt forelimb regeneration blastemas, suggesting that it, too, might play a role in specifying limb identity (Simon et al.,1997). NvTbox1 expression is twofold higher in proximal vs. distal forelimb blastemas, and RA increases expression of the gene twofold in distal blastemas, suggesting that it has a role in specifying PD positional identity as well (Simon et al.,1997).
In amniote and Xenopus limb buds, a posterior mesodermal signaling center, the zone of polarizing activity (ZPA), plays a role in specifying AP pattern. As the limb bud emerges, the ZPA is located at the base of the bud and moves distally as the limb bud elongates. When grafted to the anterior edge of the bud under the AER or when juxtaposed to anterior tissue by rotation of limb bud tips, the ZPA reverse polarizes the adjacent anterior tissue and evokes supernumerary posterior digit formation (Macabe et al.,1973; Tickle et al.,1975; Fallon and Crosby,1975,1977; Cameron and Fallon,1977; Dvorak and Fallon,1987). The signal responsible for the polarization is the amino terminal fragment of the Sonic hedgehog protein. The expression of sonic hedgehog (shh) is restricted to the ZPA in amniote limb buds (Riddle et al.,1993). Induced expression of shh by RA, ectopic expression of Shh by means of transfection with shh/retroviral constructs, or implantation of Shh-impregnated beads in the anterior mesenchyme under the AER, causes the formation of supernumerary posterior digits (Riddle et al.,1993; Lopez-Martinez et al.,1995). The amino terminal fragment of Shh does not diffuse outside the ZPA and is believed to specify AP digital pattern by regulating the expression of a cascade of bone morphogenetic proteins (BMPs) across the limb bud (Dahn and Fallon,2000; Drossopoulou et al.,2000). Consistent with this idea, knockouts of shh and gli show that Shh is required primarily for limitation of digit number and determination of digit identity (Litingtung et al.,2002).
Shh is expressed on the posterior margin of axolotl, newt, and Xenopus limb buds as well. Expression moves distally with outgrowth until it is restricted to the posterodistal region as digits begin to develop (Imokawa and Yoshizato,1997; Endo et al.,1997; Torok et al.,1999). Two sets of supernumerary digits are evoked by reversing the AP axis of stage 53 Xenopus limb bud tips, one on each side where anterior and posterior tissues are confronted. The foot of the Xenopus hindlimb bud is just beginning to differentiate at this stage. Shh expression is observed not only in the posterior mesenchyme of the grafted limb tip; it is also induced in the posterior mesenchyme of the limb stump on the opposite side, from which expression has previously disappeared (Endo et al.,1997).
Shh is expressed in early to medium bud stages of regenerating newt, axolotl, and Xenopus tadpole limbs, but expression is lost in amputated Xenopus hindlimbs as the tadpoles approach metamorphosis, coincident with the loss of regeneration (Imokawa and Yoshizato,1997,1998; Endo et al.,1997; Torok et al.,1999). Implantation of RA-soaked beads into anterior axolotl blastema cells induced shh expression and supernumerary digit formation (Torok et al.,1999), as did viral shh constructs (Roy et al.,2000). Two patches of Shh expression are observed after rotation of the newt or axolotl limb regeneration blastema. One is the original patch expressed by the posterior tissue of the primary blastema, which contributes to the supernumerary blastema on the anterior side of the limb. The other is a new patch that arises in the posterior edge of the new blastema evoked on the posterior side of the limb (Imokawa and Yoshizato,1997,1998; Torok et al.,1999).
The function of Shh in AP patterning of the regeneration blastema is not clear. Tip rotation in Xenopus distal hindlimbs continues to generate supernumerary digits after shh expression has ceased, even though, at these stages, the limb has lost the capacity to regenerate anything other than hypo- or heteromorphic spikes (Nye and Cameron, unpublished observations). The relatively late expression of the gene (at medium bud) may mean that it is not required for AP patterning of the proximal limb bud or regeneration blastema, but rather is involved in distal AP patterning and specification of digit number and identity. Consistent with this idea, the mouse stylopodium develops normally in the absence of shh, whereas the zeugopodial and autopodial elements are malformed and lack normal identity (Chiang et al.,2001). Furthermore, treatment of regenerating axolotl limbs with the Shh signaling inhibitor cyclopamine results in limbs containing all the segments distal to the amputation plane, but the autopodium is digitally incomplete (Roy and Gardiner,2002).
Another hedgehog gene expressed in large areas of the early newt limb regeneration blastema is Notophthalmus viridescens banded hedgehog (Nv-Bhh) (Stark et al.,1998). Bhh is the orthologue of Indian hedgehog (Ihh) in other vertebrates (Zardoya et al.,1996). Ihh is known to be essential in signaling mechanisms that control cartilage proliferation, hypertrophic differentiation, and ossification in developing limbs (Vortkamp,2001), but its very early appearance during regeneration suggests that it has other (as yet unknown) roles.
The dorsal ectoderm is thought to be a signaling center controlling DV pattern of the vertebrate limb bud. DV reversal of the early chick limb bud ectoderm with respect to the underlying mesoderm reverses the DV axial polarity of the mesoderm to conform to that of the ectoderm (Akita,1996). The Wnt-7a gene is expressed in the dorsal ectoderm and in turn induces expression of the gene Lmx-1 in the dorsal mesoderm. Dorsal limb mesoderm is ventralized by loss of function of Wnt-7a or Lmx-1 and misexpression of these genes in ventral mesoderm converts it to a dorsal fate (Parr and McMahon,1995; Cygan et al.,1997; Chen et al.,1998; Rodriguez-Esteban et al.,1998). Conversely, the gene En-1 is expressed in the ventral ectoderm. Loss of its function leads to expression of Wnt-7a in the ventral ectoderm and dorsal pattern duplication (Cygan et al.,1997; Loomis et al.,1998; Christen and Slack,1998), but ectopic expression of En-1 does not result in dorsalization of mesoderm (Logan et al.,1997). These results suggest that En-1 normally represses Wnt-7a expression in ventral ectoderm and that expression of these two genes is the basis of dorsoventral polarity.
Investigations of gene expression associated with DV axial polarity have so far been carried out only in regenerating Xenopus limb buds. Xenopus Lmx-1 is expressed in the dorsal mesenchyme of limb buds at stages 51–53 (regeneration-competent) and of regeneration blastemas formed after amputation through the zeugopodium at these stages. Expression is not seen in stage 55 blastemas, which form only symmetrical cartilage spikes (Matsuda et al.,2001). The DV axial relationship between the epidermis and mesodermal tissues was reversed by stripping the epidermis from the zeugopodial segment of the limb bud and reversing the DV axis of this segment with respect to the limb stump. Fresh epidermis of normal DV polarity grew over the reversed segment, which was then allowed to regenerate. When this operation was performed at stage 52, Lmx-1 expression and structural pattern of the regenerate conformed to the epidermal polarity, i.e., it was reversed. But when the operation was performed at stage 55, the original Lmx-1 expression and structural polarity of the segment were maintained in the regenerate (Matsuda et al.,2001). These results suggest that DV polarity in the limb regeneration blastema is determined by the wound epidermis by means of expression of Lmx-1. Before stage 52 or 53, DV polarity may be specified but labile; by stage 55, the polarity is determined. The expression patterns of Wnt-7a and En-1 in the regeneration blastema have not yet been reported.
Although many insights into the mechanisms of limb regeneration have been obtained over the preceding century, there remain numerous questions that insure the continuing vitality and significance of limb regeneration research, including the application of findings to the nascent field of regenerative medicine. Some of the most exciting and important unanswered questions are as follows.
Do Conventional (Reserve) Stem Cells Contribute to the Blastema?
Reserve stem cells reside in the bone marrow stroma and periosteum (mesenchymal stem cells) and muscle (satellite cells) of amphibian limbs for use in the repair of fractures or muscle injuries. Although a contribution of these cells to the blastema is not ruled out, and in fact would logically be expected, there is no direct evidence for it. Testing for such a contribution is difficult, because a method is required to selectively label mesenchymal stem or satellite cells so they can be tracked during limb regeneration. Muscle explants might possibly be used to label satellite cells. The nuclei of satellite cells in newt muscle explants are selectively labeled by [3H]thymidine during the first 6 days of culture (Cameron et al.,1986). Such labeled muscle explants might be transplanted back into an amputated limb, and the labeled cells subsequently followed. Salamanders with marked mesenchymal stem cells might be made by repopulating the marrow of irradiated animals with marked marrow cells (triploid, transgenic for LacZ or green fluorescent protein, opposite sex chromosomes, lipophilic tracker dyes). The limbs of the marked animals could then be amputated and the blastema examined for the presence of marked cells.
What Are the Mechanisms of Growth and Pattern Formation in Regenerating Limbs, and How Are They Linked?
We do not yet understand what factors govern proliferation of blastema stem cells, nor do we understand the interactions required for blastema cells to reconstruct the set of positional identities that constitutes the PD axial pattern of the limb. How many mitogenic factors are produced by the AEC and the nerves, and what are their roles? Why do blastema cells require nerves for proliferation, whereas limb bud cells require only the AER? Do the wound epidermis and AEC establish a distal boundary for an intercalation mechanism, as we have proposed? Is the PD pattern truncation observed in epidermis-free blastemas due to distal cell death or failure of distal prospective regions to expand? Why is PD discontinuity recognized only when distal is juxtaposed to proximal and not vice versa? What is the relationship between nerves, AEC, and requirement for disparate circumferential positional identity in blastema formation and outgrowth? How many signaling (cell surface and diffusible molecules) and transcription factors are involved in patterning and what are their roles? Do the signaling molecules exert their effects over a long or short range?
A urodele molecular biology is being rapidly developed, which should help answer some of these questions. cDNA libraries of axolotl and newt limb regeneration blastemas are available (Malacinski and Duhon,1996), and several genes active in regeneration have been cloned and their expression patterns analyzed (Geraudie and Ferretti,1998). We also need to make more use of the in vitro culture methods available for blastema cells (Stocum,2000) and further develop these methods. Highly efficient ways to transfect urodele cells, for example by electroporation, that will complement current viral transfection methods (Roy et al.,2000) are desirable. Finally, transgenic urodeles would be useful for tracking cells in a variety of experiments (for example, LacZ or GFP), or for performing gain-of-function experiments.
Does Limb Regeneration Use the Same Mechanisms as Limb Embryogenesis?
Differences beyond the mechanism of providing stem cells for regeneration are to be expected between embryonic and regenerating limbs. These differences probably reflect the different state of physiology of the tissues, rather than fundamentally different developmental mechanisms. Limb buds and regeneration blastemas appear to use similar cell interactions for pattern formation, as shown by an experiment in which the tips of undifferentiated axolotl limb buds and regeneration blastemas were exchanged, with reversal of the AP axis of the grafts (Muneoka and Bryant,1982). Supernumerary limbs composed partly of graft and partly of host tissues were evoked, suggesting that the limb bud and blastema cells use similar mechanisms to detect discontinuities in positional identity, trigger mitosis, and intercalate missing positional identities. However, amputating the limb buds in this experiment would likely cause dedifferentiation that would render the limb bud cells equivalent to those of the dedifferentiated cells of the larval limbs, thus eliminating any intrinsic differences that might exist. Whatever differences exist in the developmental mechanisms of limb buds and regeneration blastemas can be fully defined only by comparing molecular markers and their temporal and spatial patterns of expression.
E.A.G.C. received funding from the NSF, and D.L.S. was funded by the Indiana 21st Century Research and Technology Fund.