Analysis of Hox-4.5 and Hox-3.6 expression during newt limb regeneration: differential regulation of paralogous Hox genes suggest different roles for members of different Hox clusters

Following amputation, adult newts are able to regenerate complete limbs. In many respects, the mechanisms under-lying control of regenerative pattern are believed to be the same as those controlling normal limb development. One major difference between the two processes is the origin of the mesenchymal cells that give rise to the limb. The embryonic limb bud is derived from undifferentiated lateral plate mesoderm, with a contribution from myogenic precursors migrating from the myotome of the somite. In the case of limb regeneration, the new limb comes from differentiated adult cells at the cut surface. These are thought to undergo a process of dedifferentiation to form a mass of cells called a blastema. These mesenchymal progenitor cells then pro-liferate, redifferentiate and give rise to that part of the limb that was lost (Wallace, 1981).

One way to conceptualize the patterning of the limb is in terms of positional information. Applying this framework to regeneration, there is a continuum of positional values extending along the length of the proximodistal axis from the shoulder to the tips of the fingers, distal values being lost in amputation. The ability to regenerate precisely the structures that have been amputated then depends on a ‘positional memory’ of the cells at the amputation plane giving them the property of initializing the regenerates at the proper value (Stocum, 1984).

Regeneration can be thought of as generating new posi-tional values distal to the cut surface. It is of great interest that retinoic acid (RA) has the unique ability to change the positional values in the blastema (Niazi and Saxena, 1978; Maden, 1982; Stocum and Crawford, 1987). For example, if a wrist level blastema is treated with an optimal dose of RA, the positional values are altered to those at the level of the shoulder. The regenerate formed is no longer just a hand but a complete new limb. The molecular basis of posi-tional memory in the blastema and the ability of RA to respecify it remain largely unclear.

Urodele amphibians such as the newt and axolotl are the only adult vertebrates able to regenerate their limbs. The cellular and molecular basis for this unique ability is not understood (Brockes, 1989). One critical step for the for-mation of the blastema is the reactivation of genes con-trolling the formation of pattern during development of the limb bud (Muneoka and Bryant, 1984).

One class of molecules that are candidates for playing a role in both providing the capacity for regeneration and for carrying positional information are the products of the Hox genes. Studies on Drosophila and Xenopus homeodomain proteins indicate that these transcription factors control key developmental events (Akam, 1987; Nusslein-Volhard et al., 1987; Wright et al., 1989). Mammalian homeodomain proteins may also regulate differentiation events since they display temporally and spatially restricted expression pat-terns during embryogenesis (Goulding and Gruss, 1989). In Drosophila, homeobox genes are genetically linked to two clusters, termed the Antennapedia complex (Antp-C) and the Bithorax complex (BX-C), which together are known as the Homeotic complex HOM-C (Akam, 1989). Vertebrates have four homologous sets of clustered genes related to ancestral HOM-C, termed Hox-1, Hox-2, Hox-3 and Hox-4 (Duboule and Dolle, 1989; Graham et al., 1989; Kappen et al., 1989; Schughart et al., 1989), which most probably arose during evolution by amplification and duplication events. As a consequence, specific genes in each of the dif-ferent complexes are evolutionary related to each other, forming a subfamily or ‘paralogous’ group (Boncinelli et al., 1989). Fig. 1 illustrates the homology relationship between the vertebrate Hox complexes and the Drosophila HOM-C complex. In both vertebrates and insects, the gene clusters are organized in a physical order colinear with the anterioposterior axis of the developing organism such that the 3′-most genes are expressed in the most anterior domains (Duboule and Dolle, 1989; Gaunt et al., 1988; Gaunt, 1991; Graham et al., 1989). The Hox genes that have the most intriguing pattern described thus far in the devel-oping limb are at the 5′ end of the Hox-4 cluster (Hox−4.4, −4.5, −4.6, −4.7 and −4.8), and the Hox-1 cluster (Hox-1.7, −1.8, −1.9, and −1.10), respectively. The genes in the Hox-4 cluster are expressed coordinately along the anteriopos-terior and proximodistal axis forming a nested set, with the genes located in a more 5′-position expressed in progres-sively more posteriodistal areas (Dolle et al., 1989). Home-obox genes in the Hox - 1 cluster have expression patterns forming domains along the proximodistal axis (Yokouchi et al., 1991). The genes in these two clusters are thus candidates to be involved in the determination of the anterio-posterior, and proximodistal fates.

Like other vertebrates, the genomes of newts contain clusters of Hox genes (Belleville et al., 1992). In earlier studies, we and others have described a newt homolog to the Hox-3.3 gene that is expressed in normal and regener-ating newt limbs. Its expression is restricted to the mes-enchyme of the forelimb bud, although it is found in the ectoderm of both the forelimb and hindlimb buds (Savard et al., 1988; Tabin, 1989). In addition, a newt homolog to the Hox-4.6 gene has been described that is expressed in the mesenchymal tissue of regenerating limbs and tails, but not differentiated limbs; it has been termed ‘regeneration-specific’ (Brown and Brockes, 1991). These two Hox genes belong to different homeobox clusters, and also occupy dif-ferent positions within each cluster. We wondered whether the difference in their expression patterns is a consequence of their belonging to the Hox-3 and Hox-4 clusters, respec-tively, or whether it is due to the different position they each occupy within their cluster. We therefore decided to examine the expression pattern of two additional homeobox genes, one from each of these clusters, but from an equiv-alent paralogous position within the clusters.

In this paper, we present two new paralogous newt homeobox genes, the homologues of Hox-3.6 and Hox-4.5. We compare the transcript patterns of these two cognate genes in unamputated (normal) and regenerating tissue, and find that the regulation of Hox-3.6 is similar to that of Hox-3.3 while the regulation of Hox-4.5 is similar to that of Hox-4.6. This has important implications for the roles of these Hox clusters in specifying regenerative limb territories and positional values in axis formation during limb regeneration.

Adult newts (Notophthalmus viridescens) were supplied by M. Tolley, Donelson, Tennessee. Anesthesia, amputation procedures, and the collection and storage of tissue samples were all as described earlier (Tabin, 1989). For RA treatment, groups of about 100 newts were amputated distally at the level of the mid-fore-arm (mid-radius/ulna). 5 days later the animals received a single intraperitoneal injection with 10 μl of either a RA/DMSO solu-tion (10 mg/ml) or DMSO alone. Blastemas were harvested 7 days after RA treatment, at the end of the delay period in regeneration provoked by RA (Maden, 1983), and when they had reached the mid-bud stage: 20 days after RA treatment, or 15 days after DMSO treatment. Control groups were allowed to regenerate completely to verify that proximalization had occurred. For separation of blastemas into epithelial and mesenchymal fractions, blastemas were harvested and treated with 0.1 M EDTA (pH 8) for 2 min-utes on ice. The epidermis was then manually separated from the mesenchymal tissue. All tissue samples were quick frozen in liquid nitrogen and stored at −80°C prior to RNA preparation.

Individual cDNA libraries were made from newt forelimb and hindlimb mid-bud-stage blastema RNA applying the cloning strat-egy described by Aruffo and Seed (1987). In brief, a plasmid vector pBluescript KS(+) (Stratagene) was modified to contain two identical BstXI sites placed in inverted orientation with respect to each other and separated by a short replaceable segment of DNA. Non-self-complementary BstXI adaptors were ligated to the cDNA, size-selected molecules were ligated into the compatible BstXI site of the vector, and the ligation mixture was used to trans-form E. coli XL1-Blue (Stratagene) by electroporation (Dower et al., 1988). Aliquots of the forelimb and hindlimb blastema cDNA libraries were mixed and screened with a completely degenerate oligonucleotide probe to the most conserved 8-amino acid motif (K/Q-I/V/M-W-F-Q/S-N-R-R) in the recognition helix of the homeodomain (Burglin et al., 1989). Hybridization and washing conditions were as described by Burglin et al. (1989). Positive clones were initially sequenced using the degenerate oligonu-cleotides used for screening to verifiy their homology to the home-obox gene family. Further sequencing was performed using vector-derived primers and internal cDNA-specific primers. The clones were sequenced in both orientations by double-stranded sequencing using Sequenase (US Biochemicals). Sequence analy-sis was performed using the sequence analysis software package (version 7.1), Genetics Computer Group, Inc. (Devereux et al., 1984).

Total RNA was isolated from frozen tissue by homogenizing the tissue in 4 M guanidinium thiocyanate, followed by ultracentrifu-gation through a 5.7 M CsCl cushion (Sambrook et al., 1989). Poly (A) ⁺ RNA was purified using the Poly(AT) tract System III (Promega) following the instructions of the manufacturer.

5 μg of poly(A)⁺ RNA was separated by electrophoresis through a 1% agarose-0.66 M formaldehyde gel run in 1×MOPS buffer. The RNA was transferred to Genescreen membrane using a pres-sure blotter (Stratagene) and then crosslinked to the nylon mem-brane using an UV-Stratalinker (Stratagene) under standard con-ditions. For use as probes, gel-purified DNA fragments were randomly primed using the Multiprime DNA labelling system (Amersham). The filter was hybridized with 1×10⁶ cts/minute/ml of the radiolabelled probes at 42°C in 50% formamide; 5×SSPE; 0.5% SDS; 5×Denhardt’s solution; 100 mg/ml salmon sperm DNA. High stringency washes were carried out in 0.2×SSC; 0.5% SDS at 65°C for 3×20 minutes. The filter was exposed to Kodak XAR-5 film for 7 days at −70°C with intensifying screens after hybridization to homeobox probes, or for 1 hour at room tem-perature after hybridization to a newt EF-1α probe isolated by homology to a Xenopus EF-1α clone provided by D. Melton (Krieg et al., 1989). Before being reprobed, the blot was stripped of probe by washing in boiling water for 20 minutes.

RNase protection was done according to a modified procedure described by Melton et al. (1984). Appropriate fragments of the cDNAs were derived by PCR and cloned into Bluescript KS+. Antisense RNA probes were synthesized from the linearized plasmid template using either T3 or T7 RNA polymerase incorporat-ing [α-³²P]UTP. 2×10⁶ cts/min probe (approximately 800 pg radi-olabelled transcript) was hybridized to 5 μg of each total RNA sample in a final volume of 40 μl of hybridization solution (50% formamide; 40 mM Pipes, pH 6.7; 400 mM NaCl; 1 mM EDTA) at 60°C overnight. 300 μl of 0°C RNase solution (10 mM Tris, pH 7.5; 300 mM NaCl; 1 mM EDTA; 40 μg/ml of RNaseA; 2 μg/ml of RNaseT₁) was added to the hybridization reactions and the reaction mixtures were incubated at 30°C for 30 minutes. 50 μl of proteinase K solution (4 % SDS; 1 mg/ml proteinase K; 0.4 mg/ml tRNA) was added to each and the samples were incubated for an additional 15 minutes at 37°C. The solutions were extracted with phenol, precipitated with ethanol and resuspended in 90% formamide dye. The protected fragments were separated on stan-dard sequencing gels. The gels were vacuum dried and exposed to Kodak XAR-5 film for 2-3 days at −70°C with intensifying screens in RNase protections with homeobox probes, or for 15 hours at room temperature in RNase protections with an EF-1α probe. All RNA samples were initially normalized by measure-ment of their absorbance at 260 nm. The autoradiographic inten-sity of the protected fragments was quantitated by use of a Mol-ecular Dynamics densitometer. The signals were normalized with reference to the signals of an EF-1α probe for each sample.

In order to identify homeobox-containing genes in the newt, we screened a mixed forelimb and hindlimb blastema cDNA library with a fully degenerate oligonucleotide probe to the most conserved motif in helix 3 of the homeodomain (see Materials and methods). Southern analysis and sequence analysis of the resulting hybridizing colonies identified four distinct groups of cDNAs containing a home-obox. Two of these cDNA groups were chosen for this study. The predicted amino acid sequence of the home-odomain of one of these cDNA groups was found to have the highest degree of homology to those of the mouse Hox-4.5 and the human HOX 4D genes (98.3% and 100% iden-tical, respectively at the amino acid level) as shown in Fig. 2. Three overlapping Hox-4.5 cDNAs were isolated. Two of them (1.62 kb and 0.82 kb) lack a poly(A) tail whereas the third clone (0.84 kb) represents a polyadenylated cDNA. The composite newt Hox-4.5 cDNA has an open reading frame of 356 amino acids and a 3′ untranslated portion of 581 bp (Fig. 3). A potential initiator methionine is located 36 nt downstream of the 5′ end of the cDNA. Its sequence context fits the consensus sequence (-^A/_GNNATGPu-) established for many eukaryotic translation start sites (Kozak, 1987). This suggested protein start site is in per-fect agreement with the one found for the cognate gene in the mouse (Renucci et al., 1992). The second group of cDNA isolates is represented by two overlapping clones. The analysis of their homeodomain reveals the highest degree of homology to the homeodomain of mouse Hox-3.6 and human HOX 3I genes (98.3% and 96.6% identical respectively on the amino acid level) as shown in Fig. 2. Both clones (1.4 kb and 1 kb) span the homeobox region, but lack a 3′ poly(A) tail and do not cover the complete Hox-3.6 protein as recently reported in the mouse system (Peterson et al., 1992). Thus, for the newt Hox-3.6 only par-tial cDNA clones were obtained (Fig. 4).

Hox-3.6 and Hox-4.5 are paralogous Hox genes, mean-ing they are evolutionary homologues in the Hox-3 and Hox-4 gene clusters. When we compared the predicted amino acid sequence of the newt Hox-4.5 and Hox-3.6, we found their homeodomains to be highly conserved (58 out of 60 amino acids and 96.6%, respectively) as shown in Fig. 2. Further evidence for this relationship can be found in a conserved tryptophan (W) residue at position −6 from the N-terminal end of the homeodomain. Like most mem-bers of the AbdB-like subfamily, the mouse, human and newt Hox-4.5 and Hox-3.6 isolates have this amino acid (Izpisua-Belmonte et al., 1991).

The composite sequence of the newt Hox-4.5 gene appears to represent the complete coding region (Fig. 3). The alignment of the entire newt Hox-4.5 and its mouse cognate gene revealed a high identity of the protein around the aminoterminus that extends for the first 60-70 amino acids, and across the carboxyterminal homeodomain down to the termination codon (16 amino acids 3′ of the home-odomain), which is further extended 5′ of the homeobox for about 35 amino acids. Although we find a Pro/Tyr motif in the newt gene at a similar position upstream of the home-odomain as reported for its mouse Hox-4.5 homolog and other AbdB genes (Izpisua-Belmonte et al., 1991; Renucci et al., 1992), their sequences diverge markedly between the aminoterminus and the homeodomain. Besides a different amino acid composition, we found deletions and insertions of amino acid residues in the newt gene; most striking is an insertion of 7 glutamines (Q) at position −67 with respect to the homeodomain. The comparison of the newt Hox-3.6 and its mouse cognate gene is not as complete because the cDNA clones that we obtained do not represent the entire coding region (Fig. 4). Nonetheless, the conservation observed between the newt and mouse Hox-3.6 home-odomains is extended on the carboxyterminal portion of the protein flanking the homeodomain down to the termination codon (16 amino acids 3′ of the homedomain), and on the aminoterminal side of the homeodomain for about 23 amino acids. Although the predicted newt and mouse proteins are virtually identical at their carboxyterminus, the peptide sequence 5′ of the homeodomain appears to be very diverged.

To obtain an initial indication of the transcriptional regu-lation of the Hox-4.5 and Hox-3.6 genes, northern analysis was performed using poly(A)⁺ RNA from unamputated (normal) forelimbs, blastema forelimbs, unamputated (normal) hindlimbs, blastema hindlimbs and liver. After hybridization with a Hox-4.5 probe (C8.2), a single band of ∼ 1.8 kb was visible in RNA from forelimb and hindlimb blastemas, but no signal was detected in RNA from unam-putated forelimbs, unamputated hindlimbs and liver (Fig. 5A). Thus, like Hox-4.6, Hox-4.5 appears to be regenera-tion specific. A Hox-3.6 probe (A6.1) revealed a band of ∼ 2.3 kb in RNA from unamputated hindlimbs and blastema hindlimbs. No transcripts were found in RNA from unam-putated forelimbs, blastema forelimbs, or liver. Thus, like Hox-3.3, Hox-3.6 is expressed in unamputated and regen-erating tissue but is specific to one limb territory (the hindlimb in the case of Hox-3.6). The RNA samples were normalized by reference to transcripts detected with a newt translation elongation factor 1-α (EF-1α) probe. EF-1α was found to be equally expressed in all newt tissues tested (H.-G. Simon, unpublished data).

To analyse further the expression of Hox-4.5 and Hox-3.6 in a variety of organs, we performed RNase protection experiments. The RNase protections were done with prepa-rations of total RNA. The individual samples were nor-malized by optical density and by RNase protection analy-sis with a probe for the newt EF-1α. As shown in Fig. 5B, we could not detect protected fragments of the Hox-4.5 andHox-3.6 probes in any organ except kidney. While the expression of these homeobox genes in adult tissues is very limited, it is striking that the genes are expressed at com-parable levels in the kidney.

Because it is quantitative and more sensitive than northern analysis, we used RNase protection to evaluate the expres-sion profile of Hox-4.5 and Hox-3.6 during regeneration (Figs 6A, 7A). The differences in expression of Hox-4.5 and Hox-3.6 in unamputated and blastema tissues from forelimbs and hindlimbs was confirmed by the RNase protection analysis. With a Hox-4.5 probe (Fig. 6B), we did not detect a protected band in unamputated forelimbs. After amputat-ing forelimbs at a proximal level (mid humerus), we found a strong induction of the message when the regenerates reached mid-bud stage blastemas (BFL(P)) (when pattern is believed to be specified). In early digit stage regenerates of forelimbs (DFL(P)), we saw a much weaker expression and, in fully regenerated forelimbs (RFL(P)), no Hox-4.5 mes-sage was detectable. When we examined the expression pro-file of Hox-3.6 (Fig. 7B), a protected band was found in unamputated hindlimbs. After proximal amputation (mid femur) of the hindlimbs, an increased level of the Hox-3.6 message was visible in mid-bud stage blastemas (BHL(P)). In digit stage regenerates (DHL(P)) and fully regenerated hindlimbs (RHL(P)), the level of expression was found to be significantly reduced and equivalent to the expression level detected in unamputated hindlimbs.

In order to analyze the spatial expression of Hox-4.5 and Hox-3.6 in newt appendages, we obtained tissue samples following amputation at various proximodistal levels (Fig. 8A for Hox-4.5 and Fig. 9A for Hox-3.6). Fig. 8B shows the RNase protection analysis with the Hox-4.5 RNA probe. No protected fragment was observed in total RNA prepa-rations derived from distal (NFL(D)) and proximal (NFL(P)) segments of unamputated (normal) forelimbs, although after long exposures a faint signal was detectable in the proximal segments of unamputated forelimbs. When we compared the Hox-4.5 expression in distal (BFL(D)) and proximal (BFL(P)) blastemas, we found in repeated exper-iments a 4-to 5-fold higher level in the proximal blastemas. In normal tail, we could not detect a protected band and, in tail blastemas, only after long exposures was a faint signal visible. It has been shown by transplantation exper-iments that the information for producing limb pattern resides in the mesodermally derived ‘blastema’ cells rather than in the wound epidermis that covers them (Stocum and Dearlove, 1972). It was therefore important to ascertain whether Hox-4.5 is expressed in the mesenchymal blastema cells. Proximal hindlimb mid-bud blastemas (BHL(P)) were dissected into epithelial (Epi) and mesenchymal (Mes) frag-ments. When the two preparations were analyzed by RNase protection, expression of Hox-4.5 was observed only in the blastema mesenchyme.

Fig. 9B shows the RNase protection analysis with the Hox-3.6 RNA probe. In unamputated (normal) hindlimbs, we observed a higher level of Hox-3.6 expression in prox-imal than distal limb segments. RNA from proximal (BHL(P)) and distal (BHL(D)) blastemas revealed a pro-tected fragment similar in intensity. Hox-3.6 is expressed in normal tail and in regenerating tail as well, and the level of expression in the tail blastema seemed to be slightly higher than in the normal tail. As with Hox-4.5, we analyzed the expression of Hox-3.6 in dissected epithelial and mes-enchymal fractions of proximal hindlimb blastemas. In the hindlimb blastema, Hox-3.6 is expressed exclusively in the mesenchymal but not in the epithelial tissue. The northern analysis indicated a restricted expression of Hox-3.6 in the hindlimb territory of the newt. To test how far the expres-sion of Hox-3.6 extends anteriorly along the flank, we included surgically dissected flank tissue into the spatial expression analysis. No protected fragment was detected in anterior or posterior flank tissue. The very faint signals for Hox-4.5 and Hox-3.6 observed in the epithelial tissue is most likely due to slight contamination of mesenchymal tissue in the preparation of the two tissue fractions.

When we analyzed the expression of Hox-4.5 along the proximodistal axis of the regenerating limb, a reproducibly 4-to 5-fold higher level of expression was observed in RNA from proximal blastemas. This raised the possibility that Hox-4.5 might be involved in specifying proximodistal positional information to a regenerating limb. RA is said to ‘proximalize’ the blastema’s positional values because a limb treated with exogenous RA will regenerate as if it had been amputated at a more proximal level. We therefore examined the effect of RA on the level of Hox-4.5 RNA. Newts were amputated distally at the level of the mid fore-arm (mid-radius/ulna) (Fig. 10A). 5 days later the animals were injected with a proximalizing dose of RA dissolved in DMSO or with DMSO alone. The blastemas were har-vested at an early time (day 7), and at a later time when they had reached the mid-bud stage (day 20/day 15). As shown in Fig. 10B, a 2-to 3-fold increased level of expres-sion is observed in the early blastemas of RA-treated ani-mals compared to the DMSO-treated control animals. RA had no effect on the expression level of Hox-4.5 in the mid-bud stage blastemas. For comparison, the expression level of Hox-4.5 in proximal and distal mid-bud blastemas of untreated animals is shown. Thus, RA proximalizes the expression of Hox-4.5 at an early time point in regenera-tion, when blastema cells are recruited from underlying stump tissue and start proliferation.

This result is in contrast to the reported expression of Hox-4.6, whose expression was found not to be affected by RA treatment (Brown and Brockes, 1991). Since both are adjacent genes in the same Hox cluster, and considering the low level of effect that we observed with Hox-4.5 (2-to 3-fold), we wanted to verify this difference when compared under identical experimental conditions. Identical RNA samples from each time point were used in RNase protec-tions with a Hox-4.6 probe (5′22) obtained from R. Brown and J. Brockes. No differences were observed between RA-treated and DMSO controls at any time point (data not shown) consistent with earlier reports.

We have used the newt Hox-4.5 and Hox-3.6 genes as a model system for studying the expression of paralogous homeobox genes during limb regeneration. Upon amputa-tion, expression of Hox-4.5 is reactivated and Hox-3.6 is upregulated in the regenerating limb. The expression of both is highest in the undifferentiated mid-bud blastema, the time when the pattern is thought to be respecified. Both newt homeobox genes described in this paper are expressed in the mesenchymal blastema cells, which are known to determine pattern during regeneration (Stocum and Dearlove, 1972), and not in the overlying ectoderm. Thus, Hox-4.5 and Hox-3.6 are expressed in the correct cell type and in a time window to be expected of a gene involved in the specification of pattern during regeneration. Besides these common expression features, there are significant dif-ferences in the regulation of the patterns of Hox-4.5 and Hox-3.6 expression in the regenerating newt.

Our expression data support the existence of a posteriorly restricted expression domain along the longitudinal body axis for Hox-3.6 similar to that shown in the mouse embryo (Peterson et al., 1992). Because the level of expres-sion of Hox-3.6 and Hox-4.5 in the newt blastema is extremely low, we have not been able to localize their mRNA by in situ hybridization. Based on RNase protec-tion, Hox-3.6 is expressed in the adult hindlimb and tail of the newt, in addition to the regenerating blastema. A similar persistent expression in adult newt tissue has been described for Hox-3.3 ((NvHbox1) Savard et al., 1988; (FH-2) Tabin, 1989), and it has been speculated that the expres-sion of this gene in adult tissue might be related to the regenerative ability of urodele amphibians since the Hox genes are not expressed in other non-regenerating adult ver-tebrates (Savard et al., 1988). The data presented here might suggest that the continued expression may be a general property of genes of the Hox-3 cluster in adults capable of regeneration.

In contrast to Hox-3.6, the Hox-4.5 gene is not expressed in unamputated appendages of the newt. After amputation, Hox-4.5 gene expression is detected in regenerating fore- and hindlimbs. The expression of Hox-4.5 is significantly higher in a proximal mid-bud blastema than a distal one. A very similar expression pattern in limb regeneration has been reported for Hox-4.6, the adjacent gene in the Hox-4 cluster (Brown and Brockes, 1991). Like Hox-4.5, Hox-4.6 is expressed in the limb blastema, but not in the normal limb. In addition Hox-4.6 is clearly expressed in the tail blastema; in contrast, we did not find evidence of Hox-4.5 expression in regenerating tail. Thus, the A/P domain of expression of Hox-4.6 seems to be posterior to Hox-4.5. Since Hox-4.5 is 3′ of Hox-4.6 within the Hox-4 cluster, this is in agreement with the general finding of a colinear relationship between the structural order of Hox genes within their respective cluster and the order of anterior expression limits along the longitudinal body axis (Gaunt et al., 1988; Duboule and Dolle, 1989).

According to a model on how positional information is encoded in blastemas, cells along the adult limb maintain positional values that were set up during embryonic devel-opment (Stocum, 1984). In regeneration new positional values are generated distal to the cut surface, which means that the structures of the regenerate depend on the inherent positional values at the amputation plane. Since there is a difference in Hox-4.5 expression between proximal and distal blastemas, Hox-4.5 expression could play a role of providing positional information. To test this, we examined whether the level of expression in a distal blastema could be altered by a dose of RA that proximalized the blastema’s positional identity. In several previous experiments, RA was not found to alter Hox gene expression during regen-eration. However, some of these analyzed expression levels at a single time point, when the blastemas had reached a mid-bud stage (Savard et al., 1988; Tabin, 1989). In this study, we in addition analyzed the effect of RA on distal blastemas very early in regeneration, and were indeed able to detect an increased Hox-4.5 transcript level in RA-treated blastemas relative to control blastemas at this early time point but not at the later one (mid-bud-stage blastemas). This result is striking because it is the first time that any blastemal cDNA has been isolated for a gene whose expres-sion is regulated by RA in a direction consistent with it providing axial positional information. When Brown and Brockes (1991) examined the effect of RA on the newt Hox-4.6 gene in a very similar experimental design, they could not detect any change of expression level. The difference in responsiveness of the two Hox genes is unexpected since they are directly adjacent in the same Hox-4 cluster. It was possible though, that the contrary results were related to minor differences in the experimental protocols. When Hox-4.6 expression was analyzed under identical conditions, we saw no evidence for regulation by RA, confirming Brown and Brockes’ results. They speculated on a possible tran-sient change of Hox-4.6 provoked by RA, which they might have missed in their analysis. We in fact provide evidence for a very early transient effect of RA on Hox-4.5 gene expression in a defined time window, when blastema cells are still being recruited by dedifferentiation and just start proliferation. A slightly different time window might reveal RA regulation of Hox-4.6. The finding that RA proximal-izes the expression of Hox-4.5, at least transiently, suggests that endogenous RA levels might play a role in specifica-tion of pattern along the proximodistal (PD) axis. Regard-less of whether or not this turns out to be correct, the fact that the pharmacological RA treatment of regenerating limbs results in concurrent shifts in Hox-4.5 expression and regenerative pattern supports the model that Hox-4.5 plays a role in the specification of limb axis.

The comparison of the expression patterns of the two par-alogous genes Hox-3.6 and Hox-4.5 in limb regeneration demonstrates that the regulation of Hox-3.6 is similar to that of Hox-3.3 while the regulation of Hox-4.5 is similar to that of Hox-4.6. In view of this, and considering the cur-rent information on the expression of Hox-3 and Hox-4 genes in limb development, it seems likely that different Hox clusters address different functional roles in develop-ment. The genes from different parts of the Hox-3 cluster (AbdB-like and Antp-like genes) might be spatially regu-lated to specify the morphogenesis of specific body parts, whereas the 5′ Hox-4 genes (AbdB-like genes) appear to be involved in the generation of positional values in axis for-mation. The fact that the expression of Hox-3 genes is main-tained in adult newt limbs suggests that the positional memory they provide, differentiating forelimbs from hindlimbs, must be retained to initiate the proper regener-ation of appendages. Conversely, the fact that Hox-4 genes are only reactivated upon amputation implies that the posi-tional values they specify in both limbs can be autonomously established within the regenerating limb field.

We are grateful to Drs J. Sheen, B. Seed and T. Burglin for helpful discussions and the availability of their reagents. We also thank Drs D. Melton and J. Brockes for generously providing a Xenopus EF1-α and a newt Hox-4.6 (p5′22) cDNA clone, respec-tively, A. Iyer for technical assistance, and Drs D. Fekete, B. Riddle and T. Burglin for comments on the manuscript. H.-G. S. was supported by an Boehringer Ingelheim Fonds postdoctoral fel-lowship. The research was supported by a grant to C. J. T. from the National Institute of Health (HD 26842).