Regeneration of axolotl hindlimbs bearing surgically created discontinuities in the anterior—posterior axis

The mechanisms underlying pattern regulation in the limbs of urodele amphibians have been studied in detail by tissue level grafting experiments leading to the creation of limb stumps with an abnormal anatomy (see reviews by Wallace, 1980; Tank & Holder, 1981). Two basic types of tissue grafts have been performed; those in which a positional discontinuity is created within the limb (see below for references) and those in which a symmetrical arrangement of tissues is produced in which no such discontinuity exists (see, for example, Holder, Tank & Bryant, 1980; Bryant & Baca, 1978; Krasner & Bryant, 1980; Tank, 1978a). Numerous examples of the first type of experiment have been performed. The most fruitful method involves the contralateral transplantation or ipsilateral rotation of limb blastemas when either both or one transverse limb axis can be positionally misplaced. The general outcome of such grafts is the production of supernumerary limbs, the position of origin, frequency of formation and anatomy of which have been used to retrospectively determine the cellular interactions responsible for their initial formation (see, for example, Bryant & Iten, 1976; Tank, 1978b, 1981; Maden & Mustafa, 1982; Papageorgiou & Holder, 1983; Maden & Turner, 1978; Stock, Krasner, Holder & Bryant, 1980).

Following contralateral blastemal exchange supernumerary limbs of normal anatomy form in predictable positions when either the anterior–posterior or dorsal–ventral axes are opposed. These results have generally been taken to mean that the discontinuities produced in any of the four axial pole positions results in the production of new tissue which is produced as a consequence of some pattern regulation mechanism attempting to iron out the discontinuity. Furthermore, the most detailed framework for pattern regulation, the polar coordinate model (French, Bryant & Bryant, 1976; Bryant, French & Bryant, 1981) assumed that the production of supernumerary limbs following axial misalignment of anterior–posterior or dorsal–ventral axes resulted from a common mechanism in all axial positions (Bryant & Iten, 1976). This assumption underlies much of the subsequent discussion of results in terms of this model (Bryant & Baca, 1978; Bryant, Holder & Tank, 1982). However, this and other conclusions drawn from the polar coordinate model and any model based on the principle of restoration of continuity during pattern regulation (Lewis, 1981; Winfree, 1984) have been brought into question by the results of several recent sets of experiments which analyse pattern regulation in the dorsal–ventral axis.

It is now clear that the establishment of the dorsal–ventral axis does not rely on a principle of continuity. The evidence for this conclusion comes from experiments in which clear anatomical discontinuities are readily maintained in the dorsal–ventral axis. Such anatomical discontinuities are evident in supernumerary limbs produced following 180° blastemal rotation (Maden, 1980, 1982; Maden & Mustafa, 1982; Tank, 1981; Papageorgiou & Holder, 1983), skin grafting and nerve deviation (Reynolds, Holder & Fernandes, 1983; Maden & Holder, 1984) and the surgical creation of mixed-handed limb stumps (Holder & Weekes, 1984). In this last example a range of limb types resulted following amputation of mixed handed stumps, three of which showed anatomical discontinuities in the d–v axis. However, the formation of supernumerary structures was extremely limited. Furthermore, to date, no additional structures have been produced by limbs which are themselves supernumerary, and which bear d–v discontinuities. These results demonstrate that regulation in the dorsal–ventral axis does not lead to the production of extra structures. Two central questions remain in the wake of these results. In the first instance, if anatomical discontinuities are not resolved in the dorsal–ventral axis, why do supernumerary limbs form following contralateral blastemal exchange with misalignment of this axis? Secondly, can it be demonstrated directly that the anterior–posterior axis is regulated by a mechanism based on continuity? The experiments reported in this paper are concerned directly with the second question and the results allow further discussion of the mechanism of pattern regulation in the dorsal–ventral axis.

In order to compare directly the regulatory mechanism in the two transverse axes, it is necessary to perform a comparable operation with the two axes and compare the results. To this end, we have performed a contralateral exchange graft which creates a discontinuity in the anterior and posterior positions of the limb which is the same in design as that used to create mixed-handed limb stumps bearing a dorsal–ventral discontinuity (Holder & Weekes, 1984). The results show clearly that extra structures are produced in a majority of cases suggesting that the anterior–posterior axis is regulated by a mechanism which is basically different to that controlling patterning in the dorsal–ventral axis. The structure of the limbs regenerated was analysed with reference to both the anterior-posterior and dorsal–ventral axes and the results reveal several points which are discussed in terms of possible mechanisms for pattern regulation in the transverse axes.

All experiments were performed on larval axolotls (Ambystoma mexicanum) which were spawned in the colony at King’s College. Animals were maintained in individual plastic containers in standing tap water throughout the course of the experiment and were fed twice a week on chopped heart. All the animals were between 65 and 90 mm in length. During surgery they were anaesthetised in MS222 (Sigma).

All experiments were carried out on the thigh region of the leg. The surgeries involved cutting out the soft tissues of either the dorsal or the ventral halves of the thigh and exchanging the tissues removed with the corresponding piece from the contralateral limb (Fig. 1). The grafts were then sutured in place with the anterior–posterior axes misaligned and the dorsal–ventral and proximal–distal axes in harmony. The limbs were then amputated immediately at the distal end of the graft and the tissues trimmed to produce a flat amputation plane with at least 2 mm of graft tissue present proximal to the level of amputation. The operated animals were then returned to standing tap water and were observed at regular intervals to check that the grafts had healed in place.

The complete limbs were fixed in Bouin’s fluid 10 weeks after the time of the initial surgery. The fixed limbs were decalcified in EDTA, dehydrated, stained with Victoria-Blue and cleared with methyl salicylate. At this point the skeletal structure of the regenerates was recorded. The limbs were then returned to alcohol and processed for wax histology. Serial transverse 10 μ m sections were cut on a rotary microtome and stained with haematoxylin and eosin to reveal their muscle and epidermal structure (see Holder & Weekes, 1984).

Control operations involved an identical surgical procedure to that described for the experimental cases except the graft tissues were not exchanged contralaterally. Instead, the grafts were sutured back in the position from which they were removed and the limb amputated as before. Following 10 weeks of regeneration the limbs were fixed and processed as described above.

Unlike the muscle patterns of the axolotl forelimb, that of the hindlimb has not previously been described in detail, although the muscles at the metatarsal level have been shown to be similar to that of the metacarpals (Maden & Holder, 1984). In order to assess the basic muscle pattern of the tarsus and shank, a set of normal hindlimbs were fixed and processed for wax embedding as described above. The muscle patterns were established from serial section reconstructions from 10 μm transverse sections. In addition to the muscle pattern, the Leydig cell distributiori in the dorsal and ventral epidermis of the metatarsal region was noted.

The skeleton of the hindlimb has been described in detail previously (see, for example, Stocum, 1978; Maden & Holder, 1984) and is illustrated in Fig. 2A. The key features used for determining axial character in this study are the phalangeal formula (2,2,3,3,2 from anterior to posterior) and the relationship of the metatarsals to the row of basal tarsals. In the normal hindlimb four basal tarsals are found, one articulating with each of the metatarsals of digitsS, 4 and 3, and digits 2 and 1 articulating with the remaining basal tarsal.

The muscle patterns of the shank, tarsus and digit regions were assessed from four unoperated limbs in serial transverse sections and are shown in Figs 2B–D. This analysis was not intended to produce a comprehensive anatomical account such as that given by Grim & Carlson (1974) for the forelimb but rather to identify the position of muscles at specific limb levels to allow the identification of features which characterize the transverse limb axes. The muscle patterns of mid-distal shank, proximal tarsus and metatarsal levels are shown in Fig. 2B–D, and the muscle names are given in the legend to this figure. The key features of the muscle pattern which give reliable indications of axial relations are as follows. In the midshank region on the ventral side of the tibia (anteriorly) no muscles are found, whereas two distinct muscles (mfp and mfam) are found ventral to the fibula (posteriorly). A pronator muscle (mpp) lies between the tibia and fibula ventrally, the muscle fibres of which run obliquely and are cut tangentially in transverse section. Dorsally, two muscles lie above the tibia (meet and meti), one of which (mett) reliably extends around its anterior margin. Two muscles are also located dorsal to the fibula (mectf and med). The med dorsally and the mfp ventrally form widened sheets of muscle which extend to the tarsal level. At this point the mfp forms a clearly identifiable connective tissue sheet, the plantar fascia, which is an excellent ventral marker (see Fig. 2C–D). At the tarsal level the ventral muscles are more complex than the dorsal muscles, and the mpp and the complex med fibres insert tangentially into the plantar fascia. Dorsally the ad I muscle is a clear indicator of the anterior side (dorsal to the tibiale and the most anterior basal tarsal (see Fig. 2A)). Posteriorly the mad 5 sits close to basal tarsal 5. In the metatarsal region the dorsal muscles form semilunar-shaped caps over the skeletal elements whereas the ventral muscles form a complex sheet of 4, the innermost of which run tangentially (see also Maden & Holder, 1984).

In the epidermis of the metatarsal region many more Leydig cells are located dorsally than ventrally. The Leydig cells are relatively large spherical cells and, consequently, the dorsal epidermis is considerably thicker than the ventral epidermis, as it is in the metacarpal region of the forelimb (Holder & Weekes, 1984). In addition, as in the forelimb, the differential distribution of Leydig cells at the metatarsal level is not evident more proximally.

A total of 26 successful grafts were initially produced, 16 of which were dorsal exchanges and 10 of which were ventral exchanges. The anatomy of the 26 regenerates was assessed from both Victoria-blue-stained preparations and serial sections. A range of limb types were produced and these have been categorized principally using features of their skeletal patterns. The structure of each regenerate is recorded with reference to these skeletal characteristics in Tables 1 and 2. In the subsequent analysis, all limbs are given a number which refers to those used in these tables. Prior to describing the main features of each category of regenerate, it is necessary to mention two general points.

The most important point to be considered in the analysis of the results is that supernumerary structures can form with respect to both the transverse limb axes of the stump. In addition, in some limbs, axial disturbances are evident at some limb levels and not others. Thus extra structures may be present in the shank with the tarsus and digits normal or vice versa.

These are normal or almost normal regenerates and the group is subdivided according to the degree of abnormality (Table 1). None of the nine limbs in this category have extra structures which are obvious results of interactions caused by the discontinuity induced by the initial graft. The only extra structures seen are single additional tarsal elements. In all of the type 1 regenerates, the muscle patterns were essentially normal and this, in combination with the phalangeal formula, allowed the handedness of the limbs to be readily identified. In all cases in this category the limbs were of stump handedness.

Five limbs form group 1A and all are considered perfectly normal in both a–p and d–v axes. In each of the four limbs of type IB the general organization of the skeletal pattern was not as perfect as those in 1 A. Limbs 6 and 7 had only four digits. In limb 6, metatarsals ‘3 and 4 fused at the level of the first phalange and had a phalangeal formula of 2,2,3,1 from anterior to posterior and the tarsal and shank muscle pattern allowed the clear identification of anterior and posterior poles. The same general pattern was evident in limb 7 which had a phalangeal formula of 1,2,3,2 from anterior to posterior. Limb 8 had five digits but digit 1 was skewed to the anterior side and articulated directly with an abnormally wide tibia. However, no alterations in orientation of the two transverse axes were evident. The final example (limb 9) was the most structurally malformed limb in this category. The tibia and fibula did not articulate with the femur and proximally these elements ended free on the posterior side of the limb. The five digits and muscle pattern allowed its handedness to be established as that of the stump.

Four limbs were placed in this group (10–13, Table 1). All four showed clear skeletal variations with respect to the dorsal-ventral axis of the original stump. Limbs 10,11 and 12 all showed a clear ventral extension to the tibia although the fibula was of normal shape and position (Fig. 3A). In all of these cases the muscle patterns were normal and the limbs were of stump handedness. The digits of 10 and 11 both had phalangeal formulae of 2,2,3,3,2 from anterior to posterior although limb 10 had branched phalanges on digits 2 and 3. Thus limbs 10–12 had clear axial variations on the anterior side of the shank which were not evident at more distal levels.

The final limb in this category had three clear skeletal elements in the shank. The extra element was located on the anterior, ventral side beneath the tibia (Fig. 3B). The tarsal elements were almost normal except that basal tarsals 4 and 5 were fused to give an overall total of 8, and the muscle pattern in this region was clearly normal.

By contrast the muscles and epidermis at the metatarsal level showed a mixed handed pattern with anterior digit 1 having a reversed polarity to the remainder, which were stump-handed (Fig. 3C).

This category covers the remaining 13 limbs (14–26, Table 2) and includes those with the most conspicuous supernumerary structures. They are divided into a further three sub-classes (A–C) depending on the axial organization of the extra structures. These can be explained with reference to the initial amputation plane where a set of four possible a–p axes can be established (Fig. 4). These include the original a–p axis in the ungrafted region which will lie in the dorsal side of the limb when the ventral half was grafted and the ventral side when the dorsal half was grafted (see also Fig. 1). The following description of the four axes assumes that a dorsal graft has been placed into a right hindlimb and this convention is followed in Fig. 4. Apart from the a–p axis of the ungrafted ventral tissue, a reversed a–p axis resides in the grafted dorsal tissue. The additional two axes lie at the original anterior and posterior sides of the limb but have the orientation of the host d–v axis. On the host anterior side, the a–p axis lies with posterior dorsally and anterior ventrally whereas on the host posterior side anterior lies dorsally and posterior, ventrally.

Each of the four a–p axes are assumed to be integrated at the poles so that a total of 16 digits will be produced in a square (Fig. 4A.iii). In addition, using the same criteria, four shank elements would be predicted which are arranged in a square (Fig. 4A.ii) with a complex set of tarsals arranged in a similar manner. This extreme result was not produced; however the cases in category 3 show combinations of this pattern where specific parts of the possible square regenerate. It is this set of combinations that is used to sub-divide the remaining 13 limbs.

The six limbs in group 3A (14–19, Table 2) share two basic characteristics.

1. They all have clear axial disturbances in the shank which involve three or four of the four possible a–p axes evident at the initial amputation plane. At least one of these axes must, therefore, have the same orientation as the host d–v axis.

2. In each case the a–p axial relationships within the regenerates alters at more distal levels resulting in the formation of digits arranged predominantly in the host a–p axis. Thus the experimentally produced a-p axis or axes which were located coincident with host d–v polarity at the shank level progressively disappear more distally and merge into the remaining host a–p axis. In four of the six cases in this group the digits are arranged with posterior at the lateral and medial extremes and anterior in the centre.

Two of the limbs (14 and 15, Table 2) showed sets of four shank elements which were arranged in a distinct diamond such that the two original anterior axial poles shifted to host dorsal and ventral positions. This arrangement of shank elements is clear in Victoria-blue-stained preparations and sections (Fig. 5A,B). At more distal levels the anterior poles of the regenerate situated in the host dorsal and ventral positions merged toward the host a-p axis to produce a set of digits arranged in this axis. In one case (Fig. 5A) the digit arrangement was 54321112345, where the central digit was raised slightly into a dorsal position (Fig. 5C) and was double dorsal in character. The remaining digits showed clear d–v asymmetry with stump handedness. The central dorsal digit 1 is the only pattern effect in the digits which reflects the presence of four a-p axes at the shank level. The second limb with four shank elements had 13 digits, all of which were of stump handedness. The digits were arranged as three identifiable a–p sets in three sides of a square (Fig. 6 and refer to Fig. 4a,iii). On the host anterior side the digit sequence is 12 3 4 5 from host dorsal to host ventral. On the host posterior side the sequence is 1234 from host ventral to host dorsal. Lying between these sets in the third sequence running from host anterior to host posterior as 4 3 21, that is, with the polarity of the grafted a–p axis. The missing fourth set, which would have completed the square, if present, would lie on the host dorsal side with the a–p polarity of the host limb. The overall sequence is, therefore, 1234543211234 (Fig. 6). In both of these cases the muscle patterns at the shank level were very complex, although in the 11-digit limb (case 14) the dorsal shank element appeared to be double dorsal reflecting the double dorsal digit 1 in the centre of the digital array in addition to being double anterior (refer to Fig. 5B). The tarsal patterns were not interpretable in terms of identifying individual tarsals but clear abnormalities in the host d–v axis were evident.

Two of the remaining four limbs in group 3A (17–18, Table 2) had three shank elements arranged in a triangular pattern (refer to Fig. 4Bi) and one limb had two shank elements, one of which was clearly double posterior in character, although the pattern was also clearly triangular (16, Table 2 and Fig. 7B). In all three limbs the central shank element was the tibia which denotes the anterior limb region. In limbs 16 and 17 the tibia was located dorsally (see Fig. 7A) whereas the element was situated ventrally in limb 18 (see Fig. 8A). In each case the central tibia bore the symmetry of the region in which it was located (see Figs 7B and 8B). As with the two cases in group 3A described above, the elements in the central position coincident with the host d–v axis merged into the host a–p axis at more distal levels. In the tarsus an extra set of tarsals were located in the dorsal or ventral positions as appropriate for the shank pattern and the digits were arranged almost completely in the host a–p axis with posterior at the medial and lateral extremes and anterior in the centre. With the two cases with a central dorsal shank element, the digital arrays were 5432112345 (Fig. 7A), and one of the copies of digit number 1 showed clear double dorsal symmetry (Fig. 7D) and was located dorsal to the remaining digits which were all asymmetrical with host d–v polarity. The limb with a ventral shank element had a digital array of 543212345 (Fig. 8A) where digit 1 was situated ventrally and had double ventral symmetry (Fig. 8D). All of the remaining digits were asymmetrical with host d–v polarity.

The remaining case in group 3A (19, Table 2) did not show such an extreme duplication of skeletal elements. At the shank level the tibia was extended in the host d–v plane, although the muscle pattern was not clear enough for confident identification of dorsoventrality. The tarsals were clearly arranged in host a–p and d–v axes with the dorsoventral tarsal set corresponding to the host anterior side which is consistent with the d-v extended tibia. The six digits were arranged with host a-p orientation and were identified as 12 3 3 4 5 from host posterior to anterior (Fig. 9A). The a–p polarity of the digits was thus reversed with respect to the stump and an extra digit 3 was inserted in the midline of the pattern. In addition the dorsoventral symmetry of the digits was abnormal with digits 1, 2 and 3 being double dorsal and 3, 4 and 5 being asymmetrical with a polarity corresponding lo that of the original stump (Fig. 9B).

Group 3B contains just one limb which produced ten digits in two separate arrays, both of which were coincident with the host d–v axis (20, Table 2 and Fig. 10A). Thus two sides of the possible square of a–p axes had regenerated with the missing axes being those parallel to the host a–p axis. At the shank level three elements were present, two of which were located on the host posterior side and the other on the host anterior side. The muscle pattern was not clear enough to allow accurate assessment of dorsoventrality. The tarsus was very complex but the tarsals were clearly arranged in two sets running from dorsal to ventral (Fig. 10B). The sequence of the digits was only clear on the set running from dorsal to ventral on the host anterior side of the limb. As the host limb was right and the grafted tissue was dorsal, the foot located on the host anterior side would be expected to run from posterior dorsally to anterior ventrally (Fig. 4), and this was clearly the case. The six digits on the host posterior side were malformed and it was not possible to identify the sequence. The final point of interest about this limb was that four of the five digits making up the well-formed foot on the host anterior limb margin were double dorsal (Fig. 10C). This was also the case for at least two of the poorly formed digits in the foot located on the host posterior side although the plane of section did not allow the symmetry of all of these digits to be assessed. The digits on the ventral side in both the anterior and the posterior digit sets appeared to be asymmetrical.

The six limbs in category 3C (21–26, Table 2) have digital patterns arranged in both host d–v and a–p axes which are maintained to the digit level. This group differs from 3A in that the 3A limbs show a-p axes in the host d–v axis predominantly at the shank level. In the three cases involving dorsal grafts (21, 22, 23, Table 1) the digits in the host d–v axis were located on the host anterior side whereas in the three involving ventral grafts the digits in the host d–v axis were located on the host posterior side. In each of the six limbs the digits arranged in the host d–v axis were more disorganized than those in the host a–p axis (see, for example, Fig. 11) and the digits in this latter axis, which were identifiable in four of the six limbs, were of stump a–p polarity (see for example, Fig. 11).

With regard to their detailed structure, the limbs will be briefly described individually, beginning with the three dorsal grafts. Case 21 had an essentially normal shank with identifiable muscles and the first abnormalities were evident on the tarsus where the muscle pattern became very complex. On the host anterior side the tarsals were present through the host d–v axis whereas they were present only in the host a–p axis posteriorly. A total of nine digits were present which were identified as 5432112345 with the host anterior set curling to the dorsal side (Fig. 11). The digits in the host a–p axis were asymmetrical and stump handed. Case 22 showed much the same set of characteristics except that the tibia was extended in the host d–v axis. The four digits in the host d–v plane were not identifiable, although they curled clearly to the host dorsal side. The three digits in the host a–p axis were asymmetrical and stump handed. The third limb (23, Table 2), resulting from a dorsal graft also had an essentially normal shank, a complex tarsus with tarsals spreading in the host d–v axis on the host anterior side and 11 digits, all of which were poorly formed and unidentifiable.

The three remaining type 3C limbs resulted from ventral tissue grafts (24,24 and 26, Table 2). Case 26 had a normal shank skeleton and muscle pattern and a set of tarsals in which the elements on the host posterior side were extended in the host d–v axis. Of the nine digits, three were well formed and these were located on the host anterior side, the remaining digits were poorly formed although clearly in the host d–v axis on the host posterior limb margin. The digit muscle patterns were not clear enough to allow determination of dorsoventrality.

Both the tibia and fibula were extended in the host d–v axis of limb 25 (Fig. 12B). This was the only limb that showed such a pattern in the shank and although the muscle pattern at this level was abnormal the presence of mpp on one side and med on the other suggests a basically normal dorsoventrality. At the tarsal level the tarsal elements were arranged in three rows from host dorsal to host ventral and, consequently, 19 tarsals were present. Again, at this level the presence of a clear plantar fascia and med suggests a normal dorsal–ventral distinction. This disappeared in the digits, however, where a set of three toes was present in the host a–p axis on the host posterior limb margin. The remaining five digits were arranged in the host dorsal-ventral axis and showed abnormal symmetry (Fig. 12A). The two digits on the host ventral side were double ventral and the two found dorsally were asymmetrical with stump handedness. Thus three sets of ventral muscles were seen in parallel (Fig. 12D).

A similar set of characteristics were evident in the final limb to be described (case 24). In the shank the tibia was extended in the host d–v axis and the muscle pattern was normal. At the proximal tarsal level an additional set of tarsal bones appeared on the host anterior side in correspondence with the d–v extended tibia in the shank. The tarsals on the host posterior side were in the a–p axis (Fig. 13B). The muscle pattern at this proximal tarsal level was essentially normal. At the level of the basal tarsals, this pattern had altered. The tarsals in the d–v axis now emerged from the host posterior side to reveal a clear set running ventrally (Fig. 13C). The plantar fascia and tangentially cut med muscle fibres were located on the outer (ventral) edge of the additional set of basal tarsals and a second plantar fascia appeared on the host anterior side where the additional tarsals existed more proximally. Thus the a–p axis coincident with the host d–v axis on the host anterior side had disappeared within the tarsals to be replaced by an a–p axis coincident with the host d–v axis on the host posterior side. This reversal was maintained in the digits where a set of seven toes were formed (Fig. 13A). The sequence of five toes from host posterior to host anterior was 54321 with the remaining two toes running towards the host ventral side being uninterpretable. However, these two toes were double ventral, whereas the sequence of five had host d–v polarity (Fig. 13D).

3. Controls Six control grafts were performed for both the dorsal and ventral tissue exchanges. Following immediate amputation each limb produced regenerates with normal anatomy as seen in Victoria-blue-stained preparations and serial sections.

The purpose of creating limbs with discontinuities in the a–p axis was to compare the regeneration from such stumps with regeneration of stumps bearing discontinuities in the d–v axis (see Holder & Weekes, 1984). It is clear from the results presented here that the regenerates from the two types of stump show different characteristics. Discontinuities in the a–p axis are not maintained and extra limb parts are produced in the majority of regenerates (Tables 1 and 2), whereas discontinuities are readily maintained in the d–v axis. The detailed structure of the regenerates in the present study also provide clues as to the mechanisms for pattern regulations in these two transverse limb axes.

The extra structures which form following amputation of stumps bearing a–p discontinuities can be categorized most simply with reference to the four a–p axes present at the amputation surface (Fig. 4). The presence of four a–p axes was identified structurally only in the shank where two limbs had four shank skeletal elements arranged clearly in a square (see, for example, Fig. 5B). This extreme anatomy was never maintained to the digit level where the maximum number of digits found was thirteen (Fig. 6). Nonetheless, the limbs bearing supernumerary structures could be categorized as having subsets of the four possible a-p axes at the digit level and these subsets involved outgrowth of a–p axes in both host a–p and d–v orientations. Such a pattern suggests that each a–p axis can behave independently within the single field of blastema cells which originally dedifferentiate from a single stump. The exact anatomy of the regenerates appears to represent the maintenance of some of the four possible established a–p axes and the disappearance of others as outgrowth proceeds. Thus the axial relationships within the eventual limb alter with proximodistal level. This variation is evident in many of the limbs and ranges from the presence of a d–v extended tibia in an otherwise essentially normal pattern (group 2, Table 1) to the presence of four a–p axes in the shank and three in the digits (case 15, Table 2). Perhaps the most prominent example of such an alternative is limb 24 (Fig. 13) in which the a–p axis present with host d–v orientation begins proximally on the host anterior side and switches to one on the host posterior side in the tarsus. Such variations in presence and extent of each a–p axis within the blastema support the notion that each is initially behaving independently.

The results also support the considerable evidence from previous experiments that the mechanism of pattern regulation in the a-p axis is based on a principle of continuity. Consistent with this now established view is the observation that, when identifiable, the digits were always arranged in a continuous pattern. No digital sequence was found in which a digit was located next to another digit which was not a normal neighbour.

The analysis of serial sections also allows some comment concerning the mechanism of pattern regulation in the d–v axis. The evidence for the maintenance of discontinuities within this axis is now overwhelming (Maden, 1980, 1982, 1983; Maden & Mustafa, 1982, 1984; Tank, 1981; Papageorgiou & Holder, 1983; Reynolds et al., 1983; Maden & Holder, 1984; Burton, Holder & Jesani, 1984). The presence of limb regions symmetrical in the d–v axis in the present study is particularly important because it provides a clue as to the relationship between the a–p and d–v axes. Symmetrical double-dorsal and double-ventral regions were particularly clear when a–p axes were present within the regenerates with host d–v orientation. In some cases symmetry was evident at the shank and tarsal levels, especially in group 3A when three shank elements were present with the central element being present in a clear dorsal or ventral position (see Figs 4, 7 and 8). Symmetry was most clearly seen within the digits, however, and examples are described in the results in each of the group 3 categories. It is evident from the analysis that digits present within an a–p axis established in the host d–v axis show the symmetry appropriate for the dorsal or ventral positions. It seems then that dorsal symmetry occurs when part of the a–p axis is established within a blastemal region containing dorsal cells and ventral symmetry occurs when part of the a–p axis involves ventral cells. This conclusion is consistent with the notion proposed by Meinhardt (1983) that dorsal and ventral subfields exist within the limb blastema and that the cells within these subfields behave in a mosaic fashion during regeneration (Holder & Weekes, 1984; Maden & Mustafa, 1984). The exact relationship between the d–v and a–p axes does not have to be Cartesian, however, with the possible a–p axes arranged with two at right angles and two parallel to the original host dorsal and ventral regional boundary. For example, in some cases, notably those in category 3, anterior appeared to be located in either a host dorsal or host ventral position or both at least at the shank level (see also Fig. 4B.i, and Figs 5, 7 and 8). The position of the a–p axis with respect to the line of d–v discontinuity therefore is not set. In those cases where the a–p axes had shifted relative to the host d–v axis, the anterior pole may become coincident in position with either the dorsal or ventral poles of the regenerate. What is not clear at present is whether this movement of the anterior pole is due to respecification of positional values within the blastema with respect to the existence of two posterior extremes as has been suggested from other experiments in the amphibian limb (Slack, 1980).

It is also the case that the a–p axes which have host a–p orientation are asymmetrical in the d–v axis and invariably show stump handedness. This observation implies that dorsal and ventral subfield cells contribute to the a–p axis forming in this orientation and demonstrates that the a–p axis in this orientation maintains the polarity of the host limb region rather than that of the grafted tissue, which would be of opposite handedness (see Fig. 1). It seems likely, as in other studies where graft cell contribution has been discussed (see for example Tank & Holder, 1978; Holder, 1981; Holder & Weekes, 1984), that the grafted tissue contributes fewer cells to the blastema than the host tissue. The extreme case of this would be no contribution of grafted cells to the blastema which may then regenerate a normal limb. This explanation may account for the limbs in groups 1 and 2 (Table 1).

It is a pleasure to thank Rosie Burton, Nigel Stephens and Peter Wigmore for comments and criticisms throughout the course of this work and Malcolm Maden for discussing the muscle patterns. The research was supported financially by the SERC.