Global analysis of gene expression in Xenopus hindlimbs during stage-dependent complete and incomplete regeneration


  • Matthew Grow,

    Corresponding author
    1. Department of Biochemistry and Molecular Biology, Indiana University School of Medicine, Indianapolis, Indiana
    2. Center for Regenerative Biology and Medicine, Indiana University-Purdue University at Indianapolis, Indianapolis, Indiana
    • Department of Biochemistry and Molecular Biology, Indiana University School of Medicine, 1345 W. 16th Street, Indianapolis, IN 46151
    Search for more papers by this author
  • Anton W. Neff,

    1. Center for Regenerative Biology and Medicine, Indiana University-Purdue University at Indianapolis, Indianapolis, Indiana
    2. Department of Anatomy and Cell Biology, Indiana University School of Medicine, Bloomington, Indiana
    Search for more papers by this author
  • Anthony L. Mescher,

    1. Center for Regenerative Biology and Medicine, Indiana University-Purdue University at Indianapolis, Indianapolis, Indiana
    2. Department of Anatomy and Cell Biology, Indiana University School of Medicine, Bloomington, Indiana
    Search for more papers by this author
  • Michael W. King

    1. Center for Regenerative Biology and Medicine, Indiana University-Purdue University at Indianapolis, Indianapolis, Indiana
    2. Department of Biochemistry and Molecular Biology, Indiana University School of Medicine, Terre Haute, Indiana
    Search for more papers by this author


Xenopus laevis tadpoles are capable of limb regeneration after amputation, in a process that initially involves the formation of a blastema. However, Xenopus has full regenerative capacity only through premetamorphic stages. We have used the Affymetrix Xenopus laevis Genome Genechip microarray to perform a large-scale screen of gene expression in the regeneration-complete, stage 53 (st53), and regeneration-incomplete, stage 57 (st57), hindlimbs at 1 and 5 days postamputation. Through an exhaustive reannotation of the Genechip and a variety of comparative bioinformatic analyses, we have identified genes that are differentially expressed between the regeneration-complete and -incomplete stages, detected the transcriptional changes associated with the regenerating blastema, and compared these results with those of other regeneration researchers. We focus particular attention on striking transcriptional activity observed in genes associated with patterning, stress response, and inflammation. Overall, this work provides the most comprehensive views yet of a regenerating limb and different transcriptional compositions of regeneration-competent and deficient tissues. Developmental Dynamics 235:2667–2685, 2006. © 2006 Wiley-Liss, Inc.


Vertebrate limb regeneration is a process in which reparative events elicited by injury lead into ontogenetic events similar to those of the embryonic limb. The mesenchymal cells of the regeneration blastema arise locally from tissue dedifferentiation rather than from stem cells and interact to re-establish the pattern for an anatomically complete, fully functional new limb (Mescher, 1996; Stocum, 1999). Among adult tetrapods such regeneration is restricted to urodele amphibians, but regeneration involving similar mechanisms also occurs in fins of many teleost fishes and in the developing limbs of larval anuran amphibians. Limbs of adult anurans, like those of other vertebrates, fail to regenerate after amputation or do so without normal patterning (Geraudie and Ferretti, 1998).

The loss of regenerative capacity in developing anuran limbs has been studied most extensively with larvae of Xenopus laevis but is similar in other species. Dent (1962) showed that hindlimbs amputated through the stylopodium regenerate completely if amputated at or before Nieuwkoop and Faber stage 53 (st53; Nieuwkoop and Faber, 1967), when histogenesis of cartilage, muscle, and skin in the thigh has begun, but regenerate with highly incomplete patterning if amputated at st57, shortly before metamorphosis when the hindlimbs are much larger and fully developed, with phalanges undergoing ossification. At st57 or later, amputated hindlimbs form less cellular “pseudoblastemas,” which produce only a tapering rod of unsegmented cartilage, sometimes distally bifurcated, with normal skin but little or no musculature (Dent, 1962). In a careful histological analysis of regenerative decline, Wolfe et al. (2000) confirmed earlier work in Rana that, at any given stage, amputation at a distal level yields more complete regeneration than proximal amputation. They showed further that at st57 and later, amputation through the ossified tarsus produces more poorly patterned regenerates than amputation through either adjacent joint (Wolfe et al., 2000). The nature of the factors mediating the decline of regenerative capacity in anuran limbs is unknown, but that such factors are local and intrinsic to the limb tissues rather than hormonal or systemic is suggested by the regeneration results after reciprocally grafting limb tissues between st53 larvae and postmetamorphic froglets (Sessions and Bryant, 1988).

Information about the molecular basis of the limb regeneration process remains very incomplete. Expression of many genes or other molecular markers in specific cells of regenerating amphibian limbs has been found empirically (Geraudie and Ferretti, 1998). More recently, differential screening approaches for genes expressed in regenerating Xenopus limbs or tails have provided additional insights into molecular events important for regeneration (Ishino et al., 2003; King et al., 2003; Tazaki et al., 2005). We previously had used subtractive hybridization to identify several hundred genes differentially expressed 7 days postamputation in hindlimbs at st 53 when regeneration can go to completion and at st59 when regeneration is incomplete, producing only the cartilaginous spike (King et al., 2003).

We present here the first published use of an Affymetrix GeneChip for the comprehensive study of limb regeneration. To obtain a more global analysis of genes expressed during amphibian limb regeneration, we used the Affymetrix Xenopus laevis Genome GeneChip, which has over 15,000 probe sets representing more than 14,000 gene transcripts ( Gene expression was assayed in regenerating hindlimbs at st53 (regeneration-complete) and at st57 (regeneration-incomplete) to compare normal regeneration and abortive regeneration in limbs of the same species. Moreover, at each of these larval stages, we examined two different phases of the regeneration process: at 1 and 5 days postamputation (1dPA and 5dPA). These time points were chosen to allow direct comparison of genes expressed during the early wound healing phase of limb regeneration, when injury-induced inflammation is present and tissue dedifferentiation has begun and those expressed in the later redevelopment phase when normal patterning, cell proliferation, and growth predominate (Dent, 1962; King et al., 2003; Wolfe et al., 2004).

Clearly, there are many differences in the genes expressed in the hindlimb at st53 when compared to the st57 hindlimb. The st57 hindlimb has differentiated muscle, cartilage, and ossifying bone; a complex vasculature; and a peripheral nerve network; thus, a pattern of gene expression reflective of a more differentiated state than that of the st53 hindlimb. A comparison of the pattern of gene expression in the intact hindlimbs would not reveal as important a set of genes between the two hindlimb stages as would a direct comparison between regenerating limbs at these two stages. Therefore, as indicated above, we chose to focus our array screen to capture differences in gene expression after hindlimb amputation. Our goal was twofold: (1) to identify many more genes whose expression is up-regulated during the initial wound healing, dedifferentiation, patterning, and growth phases of amphibian limb regeneration; and (2) to compare these expression profiles in hindlimbs at two larval stages in which regeneration is either complete or incomplete and pattern-deficient. Molecular analyses of anuran limb regeneration during the period when the capacity for complete organ regeneration is lost provide insights into the failure of limb regeneration in higher vertebrates and a test of the hypothesis that inflammation accompanies the loss of regenerative capacity (King et al., 2003; Mescher and Neff, 2005, 2006).


Microarray Results: Comparisons of Microarray Results

Xenopus laevis limb blastema RNA was obtained from two different developmental stages (st53 and st57) at two different time points (1dPA and 5dPA). After normalizing and filtering the gene expression data (as described in the Experimental Procedures section), a strategy for comparing each of the four different sample types was adopted as shown in Figure 1. The four different comparisons (labeled A, B, C, and D) measure the differences in gene expression between the different tissue sources as described below:

Figure 1.

Strategy for comparisons. RNA from four different tissue sources (with four replicate samples each) were compared to one another using Welch's t-tests. AD: The diagram illustrates the four comparisons used, as described in the Experimental Procedures section. 1dPA, 1 day postamputation; 5dPA, 5 days postamputation.

A: St53 1dPA Versus st57 1dPA.

This comparison measures the differences in transcript levels between st53 blastemas and st57 pseudoblastemas during the early response to amputation (day 1 postamputation).

B: St53 5dPA Versus st57 5dPA.

This comparison measures the differences in transcript levels between st53 blastemas and st57 pseudoblastemas during regeneration of the limb (st53) or during spike formation (st57) at 5 days postamputation.

C: St53 1dPA Versus st53 5dPA.

The comparison of these samples measures the changes in transcript levels from day 1 to day 5 during the regeneration of the st53 limb.

D: St57 1dPA Versus st57 5dPA.

The comparison of these samples measures the changes in transcript levels from day 1 to day 5 during st57 spike formation.

A Summary of Comparisons

To make a general assessment of the microarray results, we have summarized the total number of hits obtained when the data were filtered by P value and fold differences. Table 1 displays the number of probe hits observed for each comparison when the results are filtered based on statistical significance (by P value from Welch's t-test), and when the results are filtered by both a value of P < 0.05 and either a 2.5- or 5-fold difference. We believe that biologically significant fold changes can be relatively small (less than twofold), and for the presentation of our data, we have only filtered data for P < 0.05 and have not implemented the fold change filters. All raw data, P values, fold changes, and false discovery rates are available for review in the Master Table in the Supplementary data, which can be viewed at

Table 1. Summary of Microarray Resultsa
  • a

    For each of the four comparisons (A–D), this table summarizes the total number of probe sets passing a filter for statistical significance (based on a Welch's t-test), or passing both a filter for significance (with a value of P < 0.05) and a fold difference of either >2.5× or >5×.

P < 0.000152990
P < 0.00137167651
P < 0.00518251828710
P < 0.0129984549518
P < 0.05110824281677115
>5X total143215331
>2.5X total42075721815

Additional Annotation

The Affymetrix Probe Set ID (or Xl Probe Set ID) was used as the key identifier of each Xenopus Genechip probe set (details available at To better identify hits (genes significantly expressed differentially in comparisons at a t-test P value of < 0.05) and compare with other published results, the most likely human ortholog(s) for each of the genes of interest were obtained through one of three methods. The methods described below are listed in the order of preference (i.e., the first method was used whenever possible, then the second method, and so on). In each case, additional annotation for each of the putative human orthologs was then obtained (using Gene ID) through the Source Web site ( (Diehn et al., 2003). The process is represented in Figure 1.

Preferred Method.

Through TIGR Resourcerer (Tsai et al., 2001), the Eukaryotic Gene Ortholog database release 9.0 (EGO; Lee et al., 2002) was used to match the Xenopus laevis Genome Genechip probe set IDs to the best Human orthologs (a representative human GenBank accession number provided).

Second Method.

The Affymetrix-provided annotation (17 September 2005) for the Xenopus laevis Genome Genechip provides gene symbols, when identified. These symbols were compared with all human gene symbols and aliases available from the NCBI Gene database (Wheeler et al., 2005). Conflicting matches and multiple Gene IDs for the same transcript were resolved manually through close examination of the probe set (in the manner described below).

Third Method.

The NCBI Unigene cluster (Wheeler et al., 2005) assigned to a Xenopus laevis probe set (from Affymetrix annotation) was viewed, and either direct links to NCBI Homologene ( were followed to identify the human ortholog (and Gene ID) or Unigene links were followed back to either Genbank nucleotide or protein entries. In some situations, the Unigene cluster had been retired, in which case the Affymetrix-provided Xenopus target sequence was used as a base nucleotide sequence for searching. A BLASTX search (Altschul et al., 1990) was performed for nucleotide sequences, which sometimes led to the identification of a likely orthologous human proteins, or matching Xenopus proteins. A BLINK search was performed on any Xenopus protein entry in an attempt to identify any likely human orthologs. Whenever a human protein was identified, its corresponding Gene ID was then recovered from the NCBI Gene database (Wheeler et al., 2005; Fig. 2).

Figure 2.

Method for enhanced annotation of the Xenopus laevis Genechip. This diagram illustrates the process used to identify putative human orthologs for Xenopus laevis Genechip probes.

Overview of the Results Based on Gene Ontology Terms of Interest

Statistically significant probe sets (P < 0.05) with putative human homologs (see the Experimental Procedures section) for each of the four comparisons were used for an analysis of the results based on Gene Ontology (GO). Resulting Uniprot IDs for human orthologs were used to generate a Gene Ontology dataset using the GOToolBox ( (Martin et al., 2004). The GO-Stat tool was used to analyze the data set for both process and function terms. GO terms of interest were selected manually and linked back to the microarray results through the human ortholog gene symbol. The numbers of hits identified for each selected GO term were then used to generate the distributions shown in Figure 3.

Figure 3.

The distribution of hits for Select Gene Ontology (GO) terms of interest. The top row shows the number of hits on a scale of 60 genes, whereas the bottom row has a scale of 12 genes. AD: The histograms show the distribution of genes for each comparison, differentially represented by color as described in the results. The individual genes and their associated GO term(s) used to compile this figure can be found in the Master Table in the Supplementary Materials. [Color figure can be viewed in the online issue, which is available at]

Note that, in many of the tables and figures, coloration has been used as a visualization aid. For comparisons A and B, green indicates transcript at higher levels in st53 samples, whereas red indicates transcript at higher levels in st57 samples. For comparisons C and D, green represents higher expression in day 1 relative to day 5, while red represents lower expression in day 1 relative to day 5.

The Most Statistically Significant Hits From Comparisons A, B, and C

For Tables 2–4, the top 50 most statistically significant hits (those with the lowest P values) were selected from comparisons A, B, and C, respectively. The results are ranked by fold difference in detected expression in each of their respective comparisons. Putative human orthologs are provided as available. The “Notes” column indicates the following: identified as up-regulated in Xenopus tail regeneration in Tazaki et al. (2005); identified as down-regulated in Xenopus tail regeneration in Tazaki et al. (2005); found differentially expressed in st53/st59 Xenopus laevis hindlimbs in King et al. (2003); and found to be up-regulated in skeletal myogenesis by Summan et al. (2003).

Table 2. The Most Statistically Significant Hits in Comparison Aa
  • a

    This table lists the 50 probe sets with the lowest P values from comparison A. A fold difference is shown for each of the four comparisons (where P < 0.05), as well as the Xl gene title from Affymetrix annotation, and putative human identity when available.

inline image
Table 3. The Most Statistically Significant Hits in Comparison Ba
  • a

    This table lists the 50 probe sets with the lowest P values from comparison B. A fold difference is shown for each of the four comparisons (where P < 0.05), as well as the Xl gene title from Affymetrix annotation, and putative human identity when available.

inline image
Table 4. The Most Statistically Significant Hits in Comparison Ca
  • a

    This table lists the 50 probe sets with the lowest P values from comparison C. A fold difference is shown for each of the four comparisons (where P < 0.05), as well as the Xl gene title from Affymetrix annotation, and putative human identity when available.

inline image

Comparisons to Previously Published Results

The results of eight previously published regeneration-related genomic analyses were selected for comparison to the microarray results (Ishino et al., 2003; King et al., 2003; Summan et al., 2003; Katogi et al., 2004; Swamy et al., 2004; Wolfe et al., 2004; Reddien et al., 2005; Schnapp et al., 2005; Tazaki et al., 2005). Due to the variety of organisms studied in these articles and the information provided, different approaches were taken to compare them to the Xenopus data. The sequences by King et al. were compared with Xenopus Genechip targets using BLAST. For data by Reddien et al., Planaria names were searched using the Gene database for homologous human names. For the data by Swamy et al., the provided Genbank nucleotide ID was used to link the Human Entrez Gene ID. For all other data sets, BLASTX was used on the sequences derived from provided Genbank nucleotide accession numbers to identify the most likely human orthologous sequence and associated Human Entrez Gene ID.

Those genes with probe sets exhibiting significant (P < 0.05) differential expression in our comparisons were matched to eight sets of previously published results. Hits that match select previously published genes are noted in Table 5.

Table 5. Select Genes Previously Reported in High-Throughput Publicationsa
  • a

    Unless otherwise shown with the Xenopus laevis gene identity in solid brackets, the orthologous human gene identity is given.

inline image

Qualitative Polymerase Chain Reaction Analysis of Select Genes

To validate results of expression levels obtained from the array screen, we carried out reverse transcription-polymerase chain reaction (RT-PCR) and quantitative PCR (qPCR) analysis of several select genes (Fig. 4). Our decision on which genes to assay for validation was made based upon two criteria. First, we have hypothesized that the immune system may be involved in the regulation of regenerative capacity, and several genes in this category were identified by the array screen and, due to their levels of differential expression, we chose to analyze SOCS3, galectin-1, MyD88, gp96, FGL2, C3, and nNOS. Second, as expected, numerous genes known to be involved in patterning were identified in the screen, and we chose to examine shh, FGF8, Msx2, and Tbx3 due to both their levels of differential expression and because of their importance in overall limb patterning.

Figure 4.

Reverse transcription-polymerase chain reaction (RT-PCR) and quantitative PCR (qPCR) analysis of select genes. Expression levels of a select set of genes found from the array screen were assayed by RT-PCR and qPCR to validate the array results. RT-PCRs were carried out for 25 cycles, except for those genes identified with an asterisk (30 cycles). Gene names or symbols are indicated to the left of each panel. Numbers below each panel represent the qPCR values for expression levels when normalized to ODC and with 1 day postamputation (1dPA) samples arbitrarily set to a value of 1. Tables to the right of each panel compare the ratios of gene expression determined by qPCR with those determined from the array screen. Columns A, B, and C refer to the designations outlined in Figure 1 and the number coloring is the same as described, e.g., green numbers for A indicate that the expression ratio is higher in stage 53 (st53) 1dPA compared to st57 1dPA. A dash indicates that there was no statistically significant ratio for that value from the array screen. ODC, ornithine decarboxylase; SOCS3, suppressor of cytokine signaling 3; FGL2, fibrinogen-like protein 2; C3, complement factor 3; nNOS, neuronal nitric oxide synthase.

As development of the hindlimb proceeds, there is reduced expression of genes involved in patterning, along with potentially altered expression of immune-related genes as adaptive immunity develops. Although our focus was to compare expression of genes between st53 and st57 at both 1dPA and 5dPA, we believed that it would be useful to include RNA from intact limbs in the RT-PCR validation studies. It is important to note that the intact limb refers to the entire limb from the hip to the toe and does not constitute a 0dPA control in these experiments, which would be the limb tissue collected immediately from the distal end of the limb stump after amputation.

We assessed the level of expression for each gene in intact limbs and compared these levels to expression at 1 (1dPA) and 5 days (5dPA) after limb amputation. Relative levels of expression were compared to those of the invariant housekeeping gene ornithine decarboxylase (ODC). Unless otherwise indicated, each endpoint PCR was carried out for 25 cycles. The level of expression of each of these genes, determined by both endpoint PCR and qPCR, was assayed in a minimum of two independent pools of RNA with several being assayed in three to five pools of RNA. Each RNA pool consisted of from 20–25 st53 blastemas and 15–20 st57 pseudoblastemas. Although absolute levels of expression were slightly variable from one pool of RNA to another, the pattern of expression was consistent. After normalization to ODC, the level of expression of each gene in the st53 1dPA tissue was arbitrarily set as the calibrator (value set to 1), and expression in the other blastema samples was normalized to this value.

To directly compare the level of expression of each gene determined by qPCR to the levels determined from the array screen, we calculated the ratio of expression for each of the three comparisons outlined in Figure 1. These values are presented in the tables to the right of each gene in Figure 4. Generally, good agreement in the relative expression ratios for each gene was found between qPCR and array results for comparisons A, B, and C.

Expression of SOCS3 exhibits a unique pattern relative to the other immune modulators we examined. The level of expression of SOCS3 in intact limbs is significantly lower than in postamputation tissues. After amputation, expression is highest in st53 blastemas, relative to st57 pseudoblastemas, early after amputation (1dPA) but exhibits a reciprocal pattern of expression by 5dPA. Although expression of SOCS3 is rapidly up-regulated 1dPA at st53, its expression by 5dPA is reduced 14.3-fold. Whereas, expression of SOCS3 is 2.3-fold lower at 1dPA comparing st57 to st53, its expression is 4.3-fold higher by 5dPA even though there is no significant change in expression from 1dPA to 5dPA at the regeneration-incomplete stage.

Significant expression of galectin-Ia (LGALS1) is only seen at the regeneration-incomplete stage in both intact limbs and pseudoblastemas. Although we can detect LGALS1 expression in regeneration-complete blastemas, the levels are 18.4- and 140-fold lower at 1dPA and 5dPA, respectively, compared to pseudoblastemas. MyD88, gp96, and FGL2, each has a pattern of expression similar to LGALS1 in that expression of these genes is higher in the regeneration-incomplete pseudoblastema relative to the -complete blastema.

Expression of the complement factor, C3 is higher in regeneration-complete compared to -incomplete tissues making the pattern of expression of this innate immune system gene reciprocal to those of the other immune system modulators we have examined. Another interesting observation is that, in contrast to the other 11 genes analyzed, C3 is the only gene whose expression was not detected in either the st53 or st57 intact limbs.

In assaying the expression of the patterning genes, Msx2, shh, nrp-1 (Musashi-1), Tbx3, and FGF8, we found that, whereas each is expressed in intact limbs, differential expression is observed when comparing regeneration-complete blastemas to -incomplete pseudoblastemas. The differential level of blastema expression of shh was higher than that of any other patterning gene detected in the array screen, and by qPCR, we find that its expression is 125-fold higher in st53 1dPA blastemas relative to st57 1dPA pseudoblastemas. In comparison A, shh is expressed at higher levels in the st53 samples vs. st57, while a target of its inhibition, Wnt11 (Marcelle et al., 1997), is expressed at lower levels (see Supplementary Materials). BMP4, PAX3, GLI3, and Noggin are also among the patterning genes enriched in the st53 tissue compared to st57 (see Supplementary Materials). Increased expression of such patterning genes in regeneration-competent limbs relative to -incompetent limbs is expected given the pattern-deficient nature of the spike that will form on st57 limb stumps.

Expression of nNOS is significantly higher in st53 intact limbs relative to st57 intact limbs. In response to amputation, expression of nNOS is induced (10.3-fold) only at the regeneration-complete stage from 1dPA to 5dPA. Of the 12 genes we analyzed by qPCR, the differential level of induced expression of nNOS was the greatest, being 206-fold higher in st53 5dPA blastemas relative to st57 5dPA pseudoblastemas.

In the use of qPCR to validate our array results, we found only one gene whose expression pattern was an exception to the results obtained from the array screen: MyD88. Expression of MyD88 was shown to be higher in the regeneration-incomplete stage 1dPA, relative to the regeneration-complete stage when assayed by qPCR (ratio is in red for comparison A) but was shown to be lower from the array screen data (ratio is green for this comparison). Although the overall value of the qPCR determined ratios for comparisons A, B, and C may differ from those determined from the array screen, these data demonstrate that results obtained from the array screen represent an accurate assessment of the differences in gene expression after limb amputation at the complete and incomplete regeneration stages of Xenopus development.

Signal Consistency From Probe Sets for Same Genes

It is possible to study the internal variation between two or more probe sets that should detect products from the same gene. While some differences may be expected due to probe sets detecting different transcripts (that are differentially expressed), we expect the bulk of such comparisons to be in agreement with one another. This agreement serves as a loose internal validation of the results. We identified many genes that are represented by more than one probe set on the Xenopus laevis Genome Genechip. From these genes, we selected those for which a human ortholog was identified and there was a significant differential expression (P < 0.05) for at least one of the probe sets in at least one of the four comparisons. The signals from these probe sets were then analyzed for consistency between multiple signals for individual genes. Table 6 summarizes the results of this analysis. A conflict was scored as any instance in which a significant (P < 0.05) hit obtained from one probe set differs in direction (±) from that of another probe set for the same gene in the same comparison. The lack of significant differential expression from one or more probe sets was not considered a conflict due to the wide variety of reasons why values may not pass the statistical and magnitude filters.

Table 6. Consistency Between Multiple Probe Sets for Individual Genes
Total probe sets per geneNumber of genesComparisons with >1 “hit”/geneTotal with conflicting data% Correct

PathwayAssist Relational Mapping of Results

As described in the Experimental Procedures section, several different relationship pathways were generated using PathwayAssist 3.0 software (Stratagene, Inc.) by loading select sets of significant hits from the microarray data (see Supplementary Materials). In the PathwayAssist figures, proteins/genes are colored red or green to represent differential expression in one or more of the four comparisons (A–D). These PathwayAssist diagrams provide visual maps of the relationships between products of one set of functionally related genes (e.g., those related to immunity and inflammation shown in Supplementary Figure S3) and thus, serve two important functions. First, by noting the pattern of the color differences between each of the four panels, it can be easily determined that, for example, many more immune genes are expressed at higher levels at 1dPA at the regeneration-incomplete stage than at the -complete stage. In addition, and we believe equally significant, these maps clearly demonstrate that few differences in gene expression occur from 1dPA to 5dPA at the regeneration-incomplete stage, whereas, numerous changes occur during this same period at the regeneration-complete stage. Red and green lines between protein nodes indicate negative and positive relationships, respectively. Proteins encoded by genes that were significantly different (P < 0.05) in expression in each comparison are colored red or green as described previously.


The study of regeneration at the molecular level in amphibians has been valuable but limited due to the scarcity of high-throughput genomics tools. With the recent availability of the Xenopus Genechip, it is now possible to carry out an analysis of global gene expression during the process of tissue/organ regeneration in this organism. The anuran amphibian Xenopus laevis has the capacity to regenerate hindlimbs after amputation only during tadpole development but not in the peri- and postmetamorphic periods. This phenomenon affords a unique model system within which to begin to identify genes whose functions may either promote regenerative capacity early in development or conversely inhibit the process later in the adult organism. We have undertaken a comprehensive screen of gene expression during the early period of hindlimb responses to amputation at the regeneration-complete and -incomplete stages of Xenopus laevis tadpole development. Specifically, we have chosen to compare gene expression profiles 1 and 5 days after hindlimb amputation. Our results demonstrate that not only are expected gene expression profile differences observed (e.g., patterning genes) but newly identified profile differences (e.g., immune response genes) allow us to more completely characterize the genetic events that may be driving or inhibiting tissue/organ regeneration.

Expanded Annotation and Microarray Data Interpretation

One of the biggest challenges in interpreting our microarray results is that the current annotation is limited to a few hundred known Xenopus genes, several hundred genes with putative identities based on sequence homology, and thousands of Unigene expressed sequence tag (EST) cluster sequences with little or no convincing annotation (with putative identities sometimes conflicting with one another in different public databases). Compounding the analysis difficulties is that most bioinformatics tools are not yet compatible with Xenopus identifiers, and most are limited to entries from just a few organisms (i.e., mouse, rat, or human). As the human genome is currently one of the most well-annotated and compatible with most tools, it was decided that the identification of human orthologs for the microarray hits should be identified and the resulting human gene ID or Uniprot ID be used for cross-comparisons to other microarrays and for input into automated gene ontology and analysis tools (see the Experimental Procedures section). It should be made clear that, although an effort has been made to assign the best annotations possible for each of the microarray hits, future studies using these data should refer back to the original Xenopus laevis Genome Genechip probe set target sequences, not our assigned putative identities.

There are many reasons why a gene may not appear as differentially expressed in our comparative analyses. Absence of a particular gene in our results is not proof that it is not expressed differentially or that it does not play an important role in regeneration. Some possible reasons for exclusion include (1) it had no representative probe set on the Genechip; (2) it did not pass the Present/Absent filter of 0.5; (3) its values did not have a P value of <0.05 in any of the t-test comparisons; (4) a matching human ortholog was not identified for that gene; (5) for the PathwayAssist relational maps, the ResNet 3.0 database used did not recognize the human-LocusLink ID (Gene ID) for the human ortholog.

It is also important to consider the time points selected for this work. Given that the overall differentiation states of st53 and st57 limbs are dramatically different from one another, we expect to see more markers of differentiated tissue (muscle, nerve, vascular, etc.) in the st57 samples. In addition, it is possible that regeneration-enabling genes or regeneration-inhibiting genes are expressed in the intact limbs and do not change significantly in the time frame of our assay. Under such circumstances, a comparison of 0dPA tissues with the blastemas/pseudoblastemas from the same stage would show few or no differences, whereas differences in expression may be identified in comparisons between the two stages. This possibility can be tested through the profiling of 0dPA limb tissue, which we plan for future experiments.

Comparisons of Samples Provide Different Perspectives on Transcriptional Differences

Interpreting the results of our Xenopus Genechip assay measuring gene expression in four developmentally different tissues, i.e., amputated hindlimbs in the immediate postamputation period (day 1) or after the onset of growth (day 5) that leads to complete limb regeneration (st53) or formation of cartilaginous spikes only (st57), requires a more detailed understanding of the comparisons.

Comparison A examines the gene expression differences between the regeneration-complete st53 and -incomplete st57 hindlimbs during the initial response to limb amputation (at 1 dPA). The significant differentially expressed genes in this comparison may reasonably be expected to include late wound healing response genes and the initial transcriptional programs associated with blastema formation such as dedifferentiation, early regenerative growth and patterning. GO term distribution of significant hits (Fig. 3), shows that identified muscle-, vascular-, and neural-associated genes are expressed at higher levels in the st57 tissue, while known patterning transcripts are disproportionately represented in the st53 tissue. These results are not unexpected in that it reflects the differences in cellular composition and activities of the limbs at these stages.

At 5 days postamputation, growth is well under way, yielding either a blastema for the new limb (st53) or a pseudoblastema leading to spike formation (st57). Comparison B, therefore, serves to measure the differences in transcript composition of cells involved in these different processes, with the st53 tissue presumably more transcriptionally complex than the st57 forming spike. This comparison yields significant expression differences, as expected, because by this time point, both the st53 blastema and st57 pseudoblastema have undergone growth that will lead to different outcomes.

Comparison C provides an assessment of transcriptional changes associated with a st53 regenerating Xenopus hindlimb from 1dPA to 5dPA. The distribution of hits with select GO terms (Fig. 3), shows that there is a net up-regulation of genes involved in patterning, myogenesis, neurogenesis, and skeletal development. There is also significant transcriptional activity of genes associated with inflammation, immunity, and stress.

The most striking observation from comparison D is the essentially static gene expression in the incomplete regenerating st57 tissue from 1dPA to 5dPA. This finding suggests that the overall tissue composition undergoes little or no change in this period and that the transcriptional changes one would expect to see with controlled regenerative growth are absent. This result is consistent with histological studies after amputation of st57 hindlimbs: except for fibroblast proliferation, there is little change in cellular activity in the st57 limb stump between days 1 and 5 (Dent, 1962; Wolfe et al., 2004).

Validation of Results

As an independent test of the validity of our microarray results, we confirmed the expression profiles of 12 selected genes using qPCR. These results (Fig. 4) show agreement in 11 of 12 cases (either the same direction change or inconclusive/insignificant differences) except for MyD88, which in comparison A, showed an enrichment in the st57 tissue by array, whereas the qPCR results showed an enrichment in the st53 tissue. Overall, this strong correlation using samples obtained separately and measured with two different methods allows for general confidence in the microarray results.

In a second independent assessment of the validity of our results, we examined the signals obtained for individual genes represented by multiple probe sets on the Genechip (Table 6). For all such genes, the frequency of conflicting data is less than 7% (except for the single gene represented by nine probe sets). This finding indicates that there is consistency from the Genechip results for multiple probes designed for the same transcript. Additionally, this consistency increases confidence that the method of identifying human orthologs is valid because the genes for this test were identified not through the Xenopus annotation but through the human Gene ID number assigned to each.


Activity of Immune Response/Inflammation-Related Genes During Limb Regeneration

Recent evidence has begun to shed light on the potential importance of both the innate and adaptive immune responses on cellular dedifferentiation and how these responses may affect the onset of tissue regeneration as well as regenerative capacity itself (Harty et al., 2003; Mescher and Neff, 2005, 2006). The influence of the immune system on regeneration was considered over 20 years ago (Sicard, 1985), but a detailed examination of the expression of immune system regulatory and regulated genes had yet to be undertaken. Our screen has shown that many genes previously identified as being important for immune responses are indeed differentially expressed during regeneration-complete and -incomplete stages of limb development (see Master Table in Supplementary Materials). We find that expression of several select immune genes is elevated from day 1 to day 5 postamputation during the regeneration-incomplete stage (Fig. 4, and Supplementary Master Table). The adaptive immune system undergoes considerable development in the perimetamorphic period in Xenopus (Rollins-Smith, 1998), and it is possible that the observed expression differences are reflective only of this maturation process and not directly related to regenerative capacity. To address this possibility, we chose to examine expression of these genes in the intact limb. Although expression may be seen in the intact limb, this finding may not necessarily indicate that the expression domain is in the midzeugopodial region where amputations were performed. This latter possibility can be addressed with 0dPA tissue. However, as we have hypothesized (Harty et al., 2003; Mescher and Neff, 2005, 2006), the increased complexity of the adult Xenopus immune system may be directly correlated with regenerative capacity. We have focused on several immune-related genes in our validation studies, either because previously published data implicated these genes in regeneration or because of their striking patterns of expression during the process of regeneration.

Several genes known to be involved in the immune response and/or inflammation that have been reported previously as playing a role in regeneration were found to be differentially expressed in our assay. These genes include FKBP4 (Avramut and Achim, 2003), FN1 (Christensen and Tassava, 2000; Tazaki et al., 2005), SOCS3 (Campbell et al., 2001), Vimentin (VIM; Ribotta et al., 2004), C3 (Del Rio-Tsonis et al., 1998), LTF (Tazaki et al., 2005), and ANXA2 (Tazaki et al., 2005). In the regeneration-complete (st53) tissues, there is a widespread down-regulation of most immune response markers, including SOCS3, VIM, FN1, EGR1, ETS1, MMP13, ANXA2, RAF1, and PTAFR, from 1dPA to 5dPA. However, some genes, including THBS1, INHBA, TIMP2, and CXCL12, are up-regulated in the st53 blastemas (see Supplementary Figure S3, including high-resolution images). We selected several immune/inflammation related genes for further analysis by RT-PCR and qPCR, and we discuss the potential implications of their expression patterns below.

The complement system, with its classic, lectin-mediated and alternative activation pathways, plays a central role in the inflammatory response and tissue/organ injury response. Traditionally, the function of complement has been linked to the innate immune system's role in the recognition and elimination of pathogens through direct killing and stimulation of phagocytosis. Complement factor 3 (C3) is a central player in each of the complement activation pathways. C3 is synthesized locally by cells of regenerating urodele limbs (Del Rio-Tsonis et al., 1998) and was shown to be upregulated during Xenopus tail regeneration (Tazaki et al., 2005). Importantly, we have extended this observation by showing that C3 is expressed at higher levels in regeneration-complete blastemas compared to regeneration-incomplete pseudoblastemas. Moreover, the following stimulatory as well as inhibitory components of the complement system were found to be up-regulated in our microarray screen: C4 beta chain, C8 beta chain, C1s, C1q, C9, complement component factor H, and complement factor D (see Supplementary Materials). Other components of the complement system have also been associated with tissue and organ regeneration: C5 (Del Rio-Tsonis et al., 1998), complement factor B (Tazaki et al., 2005), and CD59 (da Silva et al., 2002). The expression of so many components of the classic, lectin-mediated and alternative activators and inhibitors of the complement pathway in several different regeneration systems suggests that the complement system in these cases is doing more than defense against pathogens and may be playing an active role in establishing a permissive or tolerant immune environment influencing epimorphic regeneration (Mescher and Neff, 2005, 2006).

The signaling of cytokines inside cells is regulated, in part, by suppressors of cytokine signaling (SOCS). SOCS3 is a member of this family that is a potent regulator of signaling by proinflammatory cytokines such as IFN-γ, TNF-α, and IL-6 (Alexander, 2002). Expression of SOCS3 is rapidly and transiently induced after limb amputation and is significantly higher 1dPA at st53 than at st57. Suppression of cytokine signaling by SOCS3 early in the postamputation regeneration-complete limb may be a critical event allowing for immune tolerance of dedifferentiating cells and regeneration to proceed. Consistent with this possibility, other models have shown that SOCS3 may play a role in both wound healing and regeneration. Xenopus SOCS3 is rapidly and transiently induced in response to tadpole epithelial wounds and mitogen-activated protein kinase inhibitors did not block the induction (Kuliyev et al., 2005). Additionally, SOCS3 is rapidly up-regulated after partial hepatectomy where it appears to be regulated by IL-6 (Campbell et al., 2001). This outcome may be a necessary response, allowing for liver regeneration to proceed.

Consistent with Katogi et al. (2004) who found SOCS3 (identified as olrf30i08) expressed at higher levels at day 3 postamputation vs. day 10 postamputation in regenerating Medaka fins, we find that the induced expression of SOCS3 declines 14-fold from 1dPA to 5dPA at the regeneration-complete stage. However, at the regeneration-incomplete stage, the decline in expression is only 1.3-fold (Katogi et al., 2004). Continued expression of SOCS3 at the regeneration-incomplete stage compared to the regeneration-complete stage may indicate an attempt to negatively regulate the chronic cytokine-induced inflammation of st57 5dPA pseudoblastemas relative to st53 blastemas (Schreiber et al., 2002).

Like SOCS3, the other modulators of immune responses that we have examined, galectin-Ia (LGALS1), MyD88, gp96, and FGL2, each show similar patterns of expression in our model. These four genes are expressed at higher levels in st57 intact limbs relative to st53 limbs, consistent with the maturation of the immune system between these two stages. In addition, all four genes are expressed at higher levels in the regeneration-incomplete pseudoblastema relative to the -complete blastema.

LGALS1 is a member of the galectin family of β-galactoside–binding endogenous lectins (Barondes et al., 1994). LGALS1 appears to play a role in a broad array of cell functions, including cell adhesion, migration, apoptosis, differentiation, nerve and muscle regeneration, and immunoregulation (Almkvist and Karlsson, 2004; Watt et al., 2004; Jiang et al., 2005; Liu, 2005). LGALS1 has also been associated with an anti-inflammatory function (Chung et al., 2000; La et al., 2003). LGALS1 could be part of the general anti-inflammatory mechanism in st57 pseudoblastemas to counter the effects of proinflammatory components such as MyD88, gp96, and FGL2. Interpretation of the positive correlation of LGALS1 expression and incomplete limb regeneration will have to await localization of both the message and protein in normal stage 57 limbs and pseudoblastemas.

MyD88 is an adapter protein involved in signal transduction initiated by IL-1 and Toll-like receptors (TLRs), resulting in the production of many proinflammatory cytokines in a MyD88-dependent manner. TLR10, which signals by means of MyD88 (Brown et al., 2006; Obhrai and Goldstein, 2006) was shown to be differentially expressed in st57 1dPA pseudoblastemas in our array (see Supplementary Materials).The early expression of the MyD88/TLR pathway is essential for normal liver regeneration (Seki et al., 2005) and repair of acute lung injury (Jiang et al., 2005).

The endoplasmic reticulum chaperone gp96 (also designated as Grp94—glucose regulated protein-94 and TRA1) is a member of the HSP90 family that is instrumental in initiation of both innate as well as adaptive immunity (Yang and Li, 2005). Gp96 has been localized to the cell surface in Xenopus lymphocytes (Robert et al., 2004) and shown to have proinflammatory properties that are dependent on MyD88 (Liu et al., 2003).

The prothrombinase fibroleukin (fibrinogen-like protein 2, FGL2) is related to the fibrinogen beta and gamma chains, which can directly cleave prothrombin to thrombin (Chan et al., 2002). FGL2 appears to exist in two forms: a membrane-linked FGL2 with prothrombinase activity, and a soluble FGL2 that has potent immunosuppressive activity (Chan et al., 2003). FGL2 is expressed in several subsets of T-lymphocytes, including gut mucosal lymphocytes (Ruegg and Pytela, 1995) as well as activated hepatic endothelial cells and macrophages (Marsden et al., 2003). FGL2 is induced in type 1 lymphocytes by IFN signaling pathways, including irf1 (Hancock et al., 2004), which we found up-regulated in our array screen (see Supplementary Materials). Conflicting data from FGL2-deficient mice experiments show that FGL2 may or may not contribute to immunologically mediated thrombosis (Marsden et al., 2003; Hancock et al., 2004). Thrombin, proposed to be activated by tissue factor (TF) appears to couple injury and regeneration (lens and limb) in urodele amphibians (Imokawa et al., 2004). Because FGL2 bypasses the TF/factor VII extrinsic thrombin activation pathway (Marsden et al., 2003), we speculate that FGL2 may lead to thrombin activation in anuran amphibians by a TF-independent pathway. Given that expression of FGL2 and gp96 in amputated hindlimbs is similar and that gp96 is also induced by IFN, it is tempting to speculate that these expression patterns are indicative of IFN stimulation of other proinflammatory signals.

Patterning Pathways and Regeneration

Several patterning and growth-related genes that are known to play crucial roles in the specification of the axes of the limb are differentially expressed during limb regeneration (e.g., SHH, MSX2, TBX3, FGF8). As might be expected, our data demonstrate that each of these genes is expressed in regeneration-complete limbs and blastemas. By contrast, these patterning genes show little or no expression in regeneration-incomplete limbs or pseudoblastemas. Amputation of a st57 hindlimb is known to result in a pattern-deficient spike of cartilage; thus, the observed pattern of expression of this family of genes is not surprising.

Sonic hedgehog (Shh) is a critical limb morphogen that plays a role in anterior–posterior patterning of the limb, particularly the autopod (Zakany et al., 2004). Our data confirm the observation that Xenopus Shh is only expressed in blastemas at the regeneration-complete stage (Endo et al., 1997). It is not known if failure of Shh expression in regeneration-incomplete pseudoblastemas is responsible for complete regeneration failure. However, the observation that st57 amputated limbs give rise to a single spike and that Shh−/− mice produce a single digit (Chiang et al., 2001) suggests that the principal role of this gene is in the patterning process rather than in creating conditions that allow for regeneration.

The homeobox-containing genes Msx1 and Msx2 are expressed in multiple tissue–tissue interactions during development, including limbs (Lallemand et al., 2005), digit tips, mouse digit tip regeneration (Han et al., 2003), and during axolotl wound healing and limb regeneration (Carlson et al., 1998). Unlike Xmsx1, which is expressed in both regeneration-complete blastemas and regeneration-incomplete pseudoblastemas and pseudoblastemas (Endo et al., 2000; our unpublished observations), Xenopus Msx2 is shown to be preferentially expressed in regeneration-complete 1dPA and 5dPA blastemas.

Analysis of Tbx3 was carried out because it is required for normal development of the posterior structures of the limb (Davenport et al., 2003) and appears to be involved in the positioning of the limb along the craniocaudal axis (Rallis et al., 2005). The higher level of expression of Tbx3 in st53 limbs and st53 1dPA and 5dPA blastemas, relative to st57 tissues, is consistent with its role in patterning.

FGF8 is expressed throughout the apical ectodermal ridge during normal limb development, where it is involved in the maintenance of limb bud outgrowth and patterning (Moon and Capecchi, 2000). FGF8 and FGF4 are also expressed in the intermediate mesoderm, where they are involved in regulating limb bud initiation and outgrowth (Boulet et al., 2004). FGF8 is expressed in axolotl limb blastemas (Christensen et al., 2002) and in Xenopus blastemas at the regeneration-complete stage as well in pseudoblastemas at the regeneration-incomplete stage (Endo et al., 2000). Consistent with the published data, FGF8 was expressed in both regeneration-complete and -incomplete 1dPA and 5dPA blastemas. However, it is expressed at a higher level in regeneration-complete blastemas relative to regeneration-incomplete pseudoblastemas.

Nrp-1 (Musashi-1) encodes an RNA-binding protein originally thought to be a neurogenesis marker (Richter et al., 1990) and plays a major role in controlling stem cell maintenance in the brain and other tissues (Okano et al., 2005). Consistent with its function as a stem cell marker, it is expressed in regeneration-complete developing limbs at a much higher level than in the more differentiated regeneration-incomplete limbs. Nrp-1 is also expressed at much higher levels in regeneration-complete 1dPA and 5dPA blastemas than regeneration-incomplete pseudoblastemas. It is tempting to speculate that nrp-1–positive cells represent multipotential stem cells within the developing limb and blastemas.

Neuronal nitric oxide synthase (nNOS or NOS1) is highly expressed in some central and peripheral neuronal cells and is also widely expressed elsewhere, including epithelial cells and smooth and skeletal muscle cells (Forstermann et al., 1998). In contrast to some of the other limb growth/patterning genes such as FGF8, TBX3, nrp-1, SHH, and MSX2, nNOS is expressed at much lower levels in st53 one dPA blastemas, but like the other genes, its expression is up-regulated in st53 five dPA blastemas. Because nNOS knockout mice show no gross limb abnormalities and most of the defects are related to nervous system development (Nelson et al., 1995), it is unclear what the function of nNOS is during st53 limb development or in regeneration-complete st53 5dPA blastemas. However, the observation that nNOS is an essential factor for peripheral nerve regeneration (Keilhoff et al., 2003) and that it is up-regulated during mouse skin repair (Boissel et al., 2004) indicates that nNOS may not be involved in patterning of the regenerate, but instead in the regeneration response itself in the st53 limb. Of interest, during mouse skin wound healing, nNOS expression is down-regulated at 1 day after skin injury, followed by dramatic up-regulation at 3 days (Boissel et al., 2004).


Over the past few years, advanced genomic techniques have been used increasingly for global analyses of the large numbers of genes involved in regeneration. This work has been conducted using many different organisms, including Planaria, fish, frogs, salamanders, mice, and humans (Ishino et al., 2003; King et al., 2003; Summan et al., 2003; Katogi et al., 2004; Swamy et al., 2004; Wolfe et al., 2004; Reddien et al., 2005; Schnapp et al., 2005; Tazaki et al., 2005). Through various approaches (see the Experimental Procedures section), we compared our microarray results to the reported results of eight other published works (Tables 2–5).

Leucine Rich Repeat Neuronal 1 (LRRN1) was the sole matching hit to those genes reported by Wolfe et al. (2004), who found that LRRN1 was up-regulated in regenerating Xenopus lens and hindlimb. In our results, a difference in LRRN1 expression was observed in st53 five dPA blastemas.

Of the genes identified through RNAi knockdown as having functional significance in Planaria regeneration (Reddien et al., 2005), we have identified 14 putative matching genes. Six of those genes (MyoD1, TUBB6, PABPN1, UQCRC2, CAP2, and BMP1) were found to be expressed at higher levels in the 1dPA st57 samples, vs. the regeneration-complete st53 samples (comparison A). Five genes (ARPC2, ETF1, CAP1, RPL3, and NARG1) were found with a net decrease in expression from 1dPA to 5dPA during regeneration of the st53 amputated limb, whereas three genes (HADH2, KARS, and SLIT2) displayed an increased expression.

A characteristic difference between regeneration-complete and regeneration-incomplete hindlimb regeneration in Xenopus is that the st57 spike contains a skeletal framework of cartilage but essentially no skeletal muscle (Satoh et al., 2005). Therefore, we looked into nonepimorphic muscle regeneration outside the context of epimorphic limb regeneration to see if data from this work may shed some light on the issue. The work of Summan et al. (2003) focused on gene regulation associated with regenerating skeletal muscle. VIM was identified in that work as up-regulated in skeletal myogenesis and is found in our results to be enriched in the regeneration-incomplete st57 1dPA tissue vs. the regeneration-complete st53 tissue. Of note, cyclase–associated actin monomer binding protein, CAP1, identified by Reddien et al. (2005) as important for Planaria regeneration, was found to be down-regulated during skeletal myogenesis by Summan et al. (2003). In Comparison C, one CAP1 probe set reports a significant decrease in CAP1 expression in the st53 regenerating limb from 1dPA to 5dPA. This finding is additional evidence that CAP1 plays an important role during regeneration, possibly in the regulation of myogenesis.

Of all of the comparisons to other published work, the largest number of matching genes occurs with the study of Xenopus laevis tail regeneration reported by Tazaki et al. (2005), which may serve to further validate the work reported here. Most matching genes (with the exception of VDAC1 and PFKFB1) reported by Tazaki et al. (2005) to be down-regulated during tail regeneration are found in our results to be expressed at much lower levels in the regeneration-complete st53 1dPA blastema vs. st57 pseudoblastema (comparison A). It is worth noting, however, that many of the genes are muscle related, and they may owe their appearance as significant hits in comparison A to differences in the greater extent of muscle development at st57 vs. st53.

Several genes reported by Tazaki et al. (2005) to be up-regulated during Xenopus tail regeneration, undergo a net down-regulation from 1dPA to 5dPA during st53 limb regeneration (comparison C). These genes include GADD45G, C21orf33, KRT6IRS, FN1, KRT8, Annexin2, ATP6V1G1, and Unigene cluster Xl.2503. It is important to keep in mind that, while our comparison C measures the change in expression in the limb blastema from day 1 to day 5 postamputation, Tazaki et al. (2005) initially identified these genes through comparisons of day 1.5 and day 3 regenerating tails compared to tails immediately after amputation (Tazaki et al., 2005). Therefore, although direct comparison of these experiments is useful, they do not provide enough evidence to draw hard conclusions about the function of these genes during regeneration. That they co-occur as differentially expressed genes in two separate studies of tail and limb regeneration invites closer examination.

We have provided the first comprehensive examination of the global pattern of gene expression occurring during the early stages of amphibian limb regeneration and a comparison of these genes after limb amputation when regeneration fails. In agreement with our hypothesis on the role of the immune system in regeneration, we have identified the differential expression of a large number of genes involved in the modulation of both pro- and anti-inflammatory processes. Expression of anti-inflammatory modulators may be especially important to the establishment of an immunotolerant environment in the regenerating blastema. These findings, in conjunction with results in mammalian systems, suggest that regulating inflammation and conditions leading to scar formation is crucial if regeneration with normal patterning is to follow amputation.

Of significance to our reported results are our comparisons to numerous other regeneration studies in a wide array of organisms. Taken together, it is clear that, although some differences can be detected between expression profiles in different regeneration systems, we have compiled a significant database for future regeneration studies. These data provide a strong foundation upon which an understanding of the complex genetic programs promoting or repressing regenerative capacity can be generated.


Limb Amputation and Blastema Collection

Larval Xenopus laevis were raised in the laboratory or obtained commercially. Hindlimbs were staged according to Nieuwkoop and Faber (1967). All surgical procedures were performed after anesthesia in 0.002% benzocaine. Hindlimbs at either st53 (regeneration-complete) or st57 (regeneration-incomplete) were amputated unilaterally or bilaterally at the mid-zeugopodia level. The 1dPA and 5dPA tissues were collected 1 mm proximal to the original level of amputation. Total RNA was isolated using RNaqueous micro kit (Ambion, Inc.).

Microarray Sample Preparation

Total RNA samples from Xenopus laevis 1dPA and 5dPA blastemas at st53 and pseudoblastemas at st57 were submitted to the Center for Medical Genomics (CMG) microarray facility for processing (as described at The quality of the total RNA samples was assayed for purity and integrity by means of 1% agarose gel and spectrum scan from 200–360 nm using a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies, A starting amount of 1 μg of total RNA was used as the input quantity of RNA for each sample analyzed. Four replicate samples (obtained independently) for each tissue type was used.

Microarray Target Synthesis and Hybridization

One microgram of total RNA from each tissue type was amplified and the cRNA target synthesized according to standard protocols (see CMG Web site and Affymetrix protocols) using the Affymetrix 2-cycle amplification kit ( and Hybridizations were performed on the Affymetrix GeneChip Xenopus laevis Genome Array ( using standard, Affymetrix-recommended protocols ( Genechips were washed and scanned using an Affymetrix wash station and scanner, respectively.

Microarray Analysis

Scanned results in the form of .dat, .chp, and .cel files were generated using Affymetix Microarray Analysis Suite 5 software ( The absolute data were then loaded into the CMG's Microarray Data Portal (MDP, for further statistical analysis. A Principal Component Analysis was performed on all samples in a given experiment. The expected grouping of the samples was observed in at least one of the first three resulting Eigengenes. This finding serves as evidence that the samples are correctly grouped with no substantial outlying chips ref (Alter et al., 2000). Initial probe set IDs for the Xenopus laevis Genome Genechip were matched to the results using the MAS 5.0 software. The results were then linked by probe set ID to the Affymetrix annotations for the Xenopus laevis Genechip (September 17, 2005 update). The complete data set has been deposited in the NCBI Gene Expression Omnibus under the accession no. GSE4738 (

Two-Way Comparisons

Two-way comparisons of experimental groups were performed using a Welch's t-test. The results were filtered for present/absent call (generated by the MAS 5 software) average value of 0.5 for each group (numeric values: A = 0, M = 0.5, P = 1), Additionally, the results were filtered for statistical significance based on P value. For the remainder of this article, the term “hit” will refer to a transcript whose probe set, in a given comparison, passed the above present/absent filter and had a P value of P < 0.05.

Pathway Generation

As shown in Supplementary Figures 1–3, the biological pathway mapping software, PathwayAssist 3.0 was used to create visual maps of biological relationships between gene products (proteins). For each of the various pathways presented, different sets of genes were used as the starting point for pathway generation. Typically, the pathway was initially generated using the shortest direct path between proteins in a list, using only proteins as “nodes” and allowing any type of “control” to be mapped. The resulting pathways were then manually curated to make the diagrams more readable (sometimes through the reduction of the total number of nodes). Although an effort was made to manually curate some of the relationships, the PathwayAssist software relies on a database (ResNet 3.0) created through the robotic scanning and analysis of scientific literature (bibliometrics) and, therefore, should only be considered hypothetical in nature. The resulting pathways from our work, however, appear to offer a fair reflection of the transcriptional differences in a diverse set of different biological relationships.


Analysis of the expression of several hits from the array screen was carried out using RT-PCR essentially as described (King et al., 1998). Total RNA samples were extracted using the RNAqueous system (Ambion, TX). Reverse transcription reactions were carried out using 1μg of total RNA purified from indicated sources. The RNA was first denatured for 3 min at 70°C then chilled on ice and the remaining reagents were added such that the reaction contained 1× reverse transcriptase buffer, 1 mM MgCl2, 0.5 mM all 4 dNTPs, 0.5–1 μl RNase inhibitor (Promega, WI), and 100 pmol of random hexamers and comprised a volume of 60 μl. The primers were allowed to anneal to the RNA at room temp for 5–10 min. Lastly, 200 U of SuperscriptRT (Invitrogen, Inc.) was added and the reactions incubated for 60 min at 37°C. The RNA template was degraded by incubation with 1 μg of RNaseA at 37°C for 15 min.

For endpoint PCR reactions, an amount of the RT reaction equivalent to 16.7 ng of input RNA was subjected to the PCR. The reaction volume was 25 μl containing, 1× PCR buffer, 250 μM all 4 dNTPs, 2 mM MgCl2, 10 pmol of each specific PCR primer, and 2–3 U of Taq polymerase. All primers were 24-mers, with no more than a 50% G-C content. PCRs were carried out with an initial 5 min 95°C denaturation followed by 25 cycles (unless otherwise noted) of 95°C for 30 sec, 62°C for 30 sec, 72°C for 30 sec. After PCR, 10–15 μl of each reaction was analyzed by agarose gel electrophoresis and photographed by ultraviolet transillumination. As a control for RNA loading into the RT reaction, expression of Xenopus laevis ODC was assayed. When expression was to be quantified by qPCR (see below), the 60-μl RT reaction was first diluted sixfold and 1μl of the diluted RT was used as the template for each qPCR.


The qPCR was performed using the Mx3000P PCRmachine (Stratagene, La Jolla, CA). Fluorescence detection chemistry involved utilization of SYBR green dye master mix (Bio-Rad, Richmond, CA). The qPCR primers were HPLC purified 24-mers composed of no more than 50% G-C content. Optimal primer concentrations were determined using a fivefold reciprocal dilution series starting with a final primer concentration of 50 nM and ending with 250 nM. Each reaction consisted of optimal primer concentrations (generally found to be 150 nM for each primer), SYBR green master mix, and an amount of RT reaction consisting of approximately 2.5 ng of input RNA from the RT reaction. All qPCR reactions were carried out in triplicate and used a 40-cycle reaction whose time and temperature parameters were the same as for endpoint PCR. Melting-curve analysis of all products demonstrated a single peak, indicating that each set of primers produced a single product. Each RT reaction was equalized for RNA input by assessing the level of expression of the relatively invariant housekeeping gene ODC. To determine quantitative values, standard curves were generated with each primer pair using a dilution series ranging from 16.7- to 0.027-ng RNA equivalents from an RT. Expression of each gene of interest was then normalized to the level of ODC. For comparison purposes, the expression level of each gene in the st53 d1PA RNA was arbitrarily set to “1” after normalization.


We thank the Indiana University Center for Regenerative Biology and Medicine for its support and the Center for Medical Genomics for its support and assistance with microarray analysis and interpretation.