Comparative transcriptomics of limb regeneration: Identification of conserved expression changes among three species of Ambystoma

Abstract

Transcriptome studies are revealing the complex gene expression basis of limb regeneration in the primary salamander model – Ambystoma mexicanum (axolotl). To better understand this complexity, there is need to extend analyses to additional salamander species. Using microarray and RNA-Seq, we performed a comparative transcriptomic study using A. mexicanum and two other ambystomatid salamanders: A. andersoni, and A. maculatum. Salamanders were administered forelimb amputations and RNA was isolated and analyzed to identify 405 non-redundant genes that were commonly, differentially expressed 24 h post amputation. Many of the upregulated genes are predicted to function in wound healing and developmental processes, while many of the downregulated genes are typically expressed in muscle. The conserved transcriptional changes identified in this study provide a high-confidence dataset for identifying factors that simultaneous orchestrate wound healing and regeneration processes in response to injury, and more generally for identifying genes that are essential for salamander limb regeneration.

1. Introduction

Few studies have directly compared different salamander species to determine if they regenerate tissue in the same way. In the few studies that have taken a comparative approach, similarities and differences were observed in either regenerative ability or the way regeneration was accomplished [1-13]. For example, it was first proposed that a newt but not an axolotl is capable of lens regeneration after lentectomy [14]. However, later it was shown that axolotls can regenerate lens at an early stage of development but then lose this ability with aging [10]. This and other examples of species and developmental differences [8, 15, 16] suggest a need to pursue more comparative studies. Such studies may identify mechanisms of regeneration that are shared among all salamanders and perhaps all vertebrates. Also, a comparative approach might identify different regeneration strategies among salamanders and thus perhaps more than one way to induce an endogenous regeneration response in non-regenerative organisms.

The power of comparative analyses rest in large part on the phylogenetic relationships of the species that are studied. By definition, comparisons between newts and axolotls, which last shared a common ancestor over 150 million years ago [17], will reveal some deep, mechanistic commonalities and relatively more differences but will not provide perspective about the traits of species within these groups. The axolotl is member of the family Ambystomatidae, which includes approximately 32 species [18] that are widely distributed among North American habitats. Species within this group exhibit considerable variation in colour pattern and life history [19, 20]. For example, the axolotl and other Tiger salamander complex species recently evolved a paedomorphic mode of development from ancestors that typically undergo metamorphosis during ontogeny [21, 22]. All ambystomatids are thought capable of regenerating amputated limbs although the time required to complete limb regeneration varies among some species [13].

Here, we report on a comparative study of the initial, limb regeneration transcriptional response among ambystomatid salamanders. In order to permit more direct comparisons with previously published studies, and limit technical variability in expression estimates for A. mexicanum, we used a standardized Affymetrix microarray [23-32]. Unfortunately, it is cost prohibitive to develop new custom microarrays for species that are related to the axolotl. Thus, we developed a strategy to extend the precision of microarray analysis to non-model species. First, we subjected the same mRNA samples from axolotl limbs to microarray analysis and RNA-Seq. Expression estimates of differentially expressed transcripts identified concordantly by RNA-Seq and microarray show high correlation to estimates produced by quantitative reverse transcription PCR, relative to significant genes identified by any platform alone [33]. Thus, we reasoned that a combined microarray and RNA-Seq approach would produce a high confidence list of A. mexicanum transcripts to query similarities and differences in gene expression across species. Using axolotl transcripts that were identified as differentially expressed using both gene expression methods, we compared these to orthologous, differentially expressed transcripts identified by RNA-Seq for A. andersoni and A. maculatum, close and distant relatives of the axolotl, respectively. We identified 405 non-redundant genes among these species that are commonly, differently expressed after limb amputation. We suggest these genes associate with overlapping wound healing and regeneration processes that are likely conserved among ambystomatids and perhaps other salamander species.

2. Results

2.1. Identification of differentially expressed transcripts during early A. mexicanum limb regeneration

Axolotl larvae were administered amputations (mid-stylopod) and 1 mm of tissue was collected from the distal limb tip at the time of amputation (0hpa) and 24 h post amputation (24hpa). Three replicate tissue samples for each time point were used to isolate three technical RNA replicates for microarray analysis and RNA-Seq. Using the V4 contig assembly from Sal-Site (http://www.ambystoma.org/genome-resources) as a reference, 898,057 axolotl contigs were recovered from de novo assembly of RNA-Seq read pairs, representing a total assembled contig length of 389.47 Mb. The resulting RNA-Seq assembly (V5) has a maximum assembled contig length of 19,732 bp. We note that the V5 is less fragmented compared to V4, as it has a larger N/L50 value (1355/30,340 bp) and fewer assembled contigs than V4 (N/L50 = 798/121,085 bp, assembled contigs = 1,141,056, assembled contig length = 541.48 Mb). Microarray data and the V5 assembly are available from the NCBI GEO database (accession codes GSE116615 and GSE116777 respectively).

A total of 31,886 pairwise alignments with > 98% sequence similarity were identified between 20,036 V3 contigs that were used to design microarray probesets and the V5 RNA-Seq contigs (Fig. 1) (Supplemental File 1). Using limma and a combination of DESeq2/edgeR to analyze the microarray and RNA-Seq datasets respectively, 2360 transcripts were identified as commonly differentially expressed (DE) at 24hpa (or 1522 transcripts under a significance threshold of FDR < 0.05) (Supplemental File 2). We note that this list is redundant with respect to gene names as one or more V5 contigs aligned to the same microarray probeset, and some probesets were designed for the same gene and thus matched the same V5 contig. Fold change estimates between these time points were highly correlated (r = 0.92) (Fig. 2). Overall, 1887 non-redundant probesets corresponded to one or more V5 contigs. The number of up-regulated (N = 1457) and down-regulated (N = 903) genes approximated the number discovered in a previous microarray study that analyzed 10 replicate microarray samples at 24hpa; 1877 of the probesets identified as differentially expressed were also identified in the Voss et al. study [27] (Supplemental File 2). Replication of results across independent studies suggests the initial transcriptional response to injury is precise in the axolotl.

Fig. 1.

(A) Phylogenetic relationships among A. mexicanum, A. andersoni, and A. maculatum. Numbers at nodes of the topology are estimated divergence times in millions of years. (B) The analytical pipeline used to identify 1) 2360 A. mexicanum transcripts that were identified as differentially expressed by microarray analysis and RNA-Seq, and 2) 474 genes that were commonly, differentially expressed among all three species.

Fig. 2.

Comparison of log2 fold change values of 2360 commonly, differentially expressed transcripts identified by microarray analysis and RNA-Seq. Red dots represent upregulated transcripts and blue dots represent downregulated transcripts. Inset plots show highly, differentially expressed transcripts (log2FC > 4 or log2FC < −2). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.2. Comparison of differentially expressed genes among Ambystoma species

To determine if genes identified at 24hpa for axolotl were also differentially expressed in other Ambystoma species, we generated de novo transcriptome assemblies for A. andersoni and A. maculatum. Limb amputations were performed for larval A. andersoni and A. maculatum at 0hpa, and tissues were collected at 0hpa and 24hpa for RNA-seq. The resulting transcriptomes (available from NCBI under accession code GSE116777) were BLAST searched to identify presumptive A. andersoni (N = 7157) and A. maculatum (N = 5065) orthologous transcripts for the technically validated set of 2360 axolotl transcripts (Fig. 1). Of these transcripts, 789 and 1033 transcripts were significantly DE in A. andersoni and A. maculatum respectively, and 485 transcripts exhibited significant DE in both comparisons (Supplemental File 2). These transcripts showed higher correlations when comparing RNA-Seq fold changes (A. andersoni r = 0.88, A. maculatum r = 0.86), to fold changes of all presumptive orthologs for A. andersoni (r = 0.74) and A. maculatum (r = 0.75). Additionally, few A. andersoni (N = 11) and A. maculatum (N = 39) non-redundant transcripts were expressed inversely to corresponding axolotl transcripts. Overall, 474 axolotl transcripts showed the same directional pattern of DE across all three species (321 under a significance threshold of FDR < 0.05). The fold changes of these common regeneration genes (CRGs) were highly, positively correlated (A. andersoni vs. axolotl r = 0.93; A. maculatum vs. axolotl r = 0.89), however the correlation was generally higher for upregulated (axolotl vs A. andersoni r = 0.72, A. mexicanum vs. A. maculatum r = 0.64) than downregulated CRGs (axolotl vs A. andersoni r = 0.35, axolotl vs A. maculatum r = 0.26) (Fig. 3). We note that these lists are redundant with respect to gene names because the validated set of 2360 axolotl transcripts included redundant V5 contigs. The majority of the CRGs (N = 443) were assigned gene names from the V5 contig annotations, however 30 CRGs corresponded to anonymous V5 contigs that may be Ambystoma specific.

Fig. 3.

Pairwise comparison of RNA-Seq log2 fold change of A. andersoni and A. maculatum, to A. mexicanum. Transcripts exhibiting high DE between species are shown.

2.3. Assessment of anonymous CRGs

We next attempted to resolve the identity of the 30 anonymous CRGs using BLAST to search new axolotl transcriptomic resources [34-36] (Supplemental File 3). In total, we identified gene names for 27 CRGs. The remaining anonymous CRGs were compared to the NCBI NR database using BLASTx, but no identities were resolved. Anonymous CRGs were then assessed for coding potential using the Coding Potential Analysis Tool [37] and Coding Potential Calculator 2 [38]. One CRG exhibited ambiguous RNA coding potential (Ax_TR487141_c2_g8, avg. Pr. = 33.5%) and two CRGs exhibited high, non-coding RNA potential (Ax_TR491465_c9_g1, avg. Pr. = 5%; Ax_TR501104_c0_g1, avg. Pr. = 7%). Both non-coding transcripts are > 700 bp, thus implicating them as long non-coding RNAs. Anonymous CRGs were also assessed for conserved protein domain (CD-Search), protein family (Pfam), noncoding RNA family (Rfam), and micro-RNA hits, however no significant hits were identified (E < 1.0).

2.4. Gene ontology annotation of CRGs

Exempting genes that did not return a significant match to a proteincoding sequence, 405 non-redundant gene names were identified for the CRGs (Supplemental File 4). The resulting CRG list and PANTHER [39] were used to identify significantly enriched Gene Ontology (GO) terms. The 169 upregulated genes significantly enriched three Cellular Component (Plasma Membrane, Extracellular Space, and Specific Granule) and two Molecular Function (Protein Binding and TNF-Activated Receptor Activity) GO terms (Table 1); these terms suggest a bias for proteins that locate to plasma membranes and extracellular spaces, and that functionally bind to other proteins. Also, enrichment of the Specific Granule term implicates genes (arg1, cnn2, cyba, cybb, cxcl1, lyz, olfm4, tnfrsfb1, tollip) that encode proteins associated with secretory vesicles of neutrophils and other leukocytes. A much greater number of Biological Process (BP) terms (N = 26) were identified as enriched, implicating cellular responses that are regulated during wound healing and tissue development. For example, 112 of the upregulated genes redundantly enriched Regulation of Signal Transduction, Regulation of Developmental Process, Immune System Process, and Programmed Cell Death. Other significantly enriched BP terms included Regulation of Cell Migration, Regulated Exocytosis, Cytokine-Mediated Signaling Pathway, Inflammatory Response, Vascular Development, Response to Reactive Oxygen Species, and Keratinocyte Differentiation. We note that a group of 27 “highly connected” genes (aif1, areg, arg1, cdkn1a, ctgf, cyr61, dusp, egr1, ereg, fos, foxo1, hspa5, klf2, left, lgals9, ptgs2, rhob, thbs1, tnfrsf1b, tnfrsf1a, tp53inp1, sdc1, serpinf2, spi1, tnfrsf21, ubc, xbp1) each redundantly enriched 9 or more BP terms and collectively were represented among all 26 significant BP term gene lists. The 93 downregulated genes with gene names enriched 23 GO terms (Table 2), the majority of which were associated with muscle tissue (e.g. Muscle Contraction, Z Disc, Sarcolemma, and Muscle Myosin Complex. Calcium Ion Binding, Basement Membrane, Extracellular Matrix Structural), and several cellular metabolism (Energy Derivation by Oxidation, Electron Transport Chain, ATP Metabolic Process, Regulation of ATPase Activity) GO terms were also identified as enriched by the downregulated genes. Declining transcript abundances among these broad categories may largely reflect epidermal and ECM remodeling, and muscle tissue loss by histolysis at 24hpa. Matrix metalloproteinases (mmpl, 3, 19, 28), which are known to encode proteins that breakdown tissue and extracellular matrix components, were highly expressed at 24hpa. Tissue histolysis, and perhaps also dedifferentiation, would be expected to decrease the number of differentiated muscle cells and the number of transcripts these cells contribute to the total RNA pool. As noted above, genes that are specific to muscle cells (e.g. cox7a1, myh6, myh3, mybpc3, myl1, myl2, myl3, tpm1) showed a decline in expression as did genes that encode proteins that function in cellular metabolism, including many genes that locate to mitochondria and function in the electron transport chain (cox4i2, did, ndufb9, ndufc2, ndufs7, ndufv2, slc25a12, uqcrc2, uqcrfs1). Expression profiles from Voss et al. [27] show that these muscle and metabolic gene have highly correlated patterns of expression throughout limb regeneration. However, whereas muscle-specific transcripts decline to approximately null levels, the group of metabolic genes declined to moderate expression levels throughout regeneration (Supplemental File 5). These two patterns are shown by comparing the expression of cox7a1 and cox4i2; both of these genes function in the electron transport chain but the former is specifically expressed in muscle [34] (Fig. 4). The moderate and incomplete decrease of cellular metabolism genes is consistent with their loss from a proportion of cells that die or that undergo reprogramming, but not all cells that were sampled from the limb. Interestingly, we observed higher expression of MMPs in A. maculatum than in the paedomorphic A. mexicanum and A. andersoni, and also a greater decline of muscle and metabolic genes (Fig. 5). These results reveal two aspects of the conserved transcriptional response of ambystomatid salamanders to limb amputation: 1) Up-regulation of genes that are typical of wound-healing responses and developmental processes, and 2) Down-regulation of genes that are typically expressed in muscle.

Table 1.

Gene ontology terms enriched by upregulated CRGs.

GO Biological Process	#	P-Value
Negative regulation of cellular process (GO:0048523)	78	2.90E-02
Regulation of signal transduction (GO:0009966)	64	4.41E-04
Regulation of developmental process (GO:0050793)	54	1.12E-03
Immune system process (GO:0002376)	46	1.77E-02
Regulation of apoptotic process (GO:0042981)	39	3.46E-03
Programmed cell death (GO:0012501)	32	4.44E-04
Response to drug (GO:0042493)	30	1.19E-03
Positive regulation of cell proliferation (GO:0008284)	26	4.61E-03
Positive regulation of protein phosphorylation (GO:0001932)	26	3.66E-02
Regulation of cell migration (GO:0030334)	26	2.96E-03
Tube development (GO:0035295)	26	6.02E-03
Positive regulation of cell death (GO:0010942)	24	5.78E-03
Regulated exocytosis (GO:0045055)	23	4.86E-02
Cytokine-mediated signaling pathway (GO:0019221)	22	1.18E-03
Vasculature development (GO:0001944)	19	2.11E-02
Inflammatory response (GO:0006954)	18	1.70E-04
Response to toxic substance (GO:0009636)	18	3.80E-02
Response to antibiotic (GO:0046677)	15	3.33E-02
Cellular response to nutrient levels (GO:0031669)	14	2.90E-03
Response to lipopolysaccharide (GO:0032496)	14	1.04E-02
Negative regulation of cytokine production (GO:0001818)	12	2.92E-02
Response to reactive oxygen species (GO:0000302)	12	2.73E-02
Keratinocyte differentiation (GO:0030216)	10	8.51E-03
Response to glucocorticoid (GO:0051384)	10	4.20E-02
Actin filament bundle organization (GO:0061572)	8	5.95E-03
Cellular response to interleukin-4 (GO:0071353)	6	8.78E-03
GO Molecular Function	#	P-Value
Protein binding (GO:0005515)	152	1.23E-02
Tumor necrosis factor-activated receptor activity (GO:0005031)	4	6.58E-03
GO Cellular Component	#	P-Value
Plasma membrane (GO:0005886)	73	1.75E-02
Extracellular space (GO:0005615)	57	5.01E-03
Specific granule (GO:0042581)	10	1.61E-02

Table 2.

Gene Ontology terms enriched by downregulated CRGs.

Down-Regulated Genes
GO Biological Process	#	P-Value
Muscle filament sliding (GO:0030049)	17	3.60E-15
Cardiac muscle contraction (GO:0060048)	16	1.93E-10
ATP metabolic process (GO:0046034)	15	3.47E-03
Energy derivation by oxidation of organic cmpds (GO:0015980)	15	2.58E-02
Electron transport chain (GO:0022900)	13	2.59E-02
Sarcomere organization (GO:0045214)	11	3.59E-07
Regulation of striated muscle contraction (GO:0006942)	10	2.38E-03
Regulation of ATPase activity (GO:0043462)	10	2.74E-03
Cardiac muscle cell development (GO:0055013)	8	4.10E-03
Skeletal muscle contraction (GO:0003009)	7	1.87E-03
Ventricular cardiac muscle tissue morphogenesis (GO:0055010)	7	3.53E-02
GO Molecular Function	#	P-Value
Calcium ion binding (GO:0005509)	26	8.27E-04
Actin filament binding (GO:0051015)	14
		1.58E-03
Structural constituent of muscle (GO:0008307)	10	7.83E-06
Extracellular matrix structural constituent (GO:0005201)	8	1.47E-02
Myosin heavy chain binding (GO:0032036)	4	1.14E-02
Troponin T binding (GO:0031014)	4	2.31E-03
GO Cellular Component	#	P-Value
Mitochondrial membrane part (GO:0044455)	13	2.90E-02
Sarcolemma (GO:0042383)	11	2.60E-03
Z disc (GO:0030018)	11	5.44E-03
Basement membrane (GO:0005604)	9	1.06E-02
Muscle myosin complex (GO:0005859)	6	5.63E-03
Troponin complex (GO:0005861)	6	5.68E-06
Dystrophin-associated glycoprotein complex (GO:0016010)	5	1.52E-02

Fig. 4.

Expression difference between muscle-specific cox7a1 and the more ubiquitously expressed cox4a1. Data are from Voss et al. [27].

Fig. 5.

Differentially expressed metabolic (orange background) and muscle genes (green background) among three ambystomatid salamander species. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3. Discussion

A relatively large number of studies have characterized global patterns of gene expression during axolotl limb regeneration [23-27, 31, 32, 34, 36, 40-43]. In a comprehensive microarray study of axolotl forelimb regeneration [27], a study that examined 10 replicate samples for 20 post-amputation time points, > 1000 genes were identified as significantly differentially expressed by 24hpa. This punctuated phase of transcription included a diversity of genes, and not unexpectedly, genes that are typically associated with early wound healing events. With less replication in this study and because transcription is punctuated, dynamic, and variable within narrow windows of time [27, 32], only a portion of this initial transcriptional pulse was sampled. However, the genes that were identified were technically verified between microarray and RNA-Seq methodologies, and further shown to be highly overlapping with previous studies. Thus, the resulting list of 24hpa genes provide robust biomarkers for elucidating early transcriptional responses in future studies of the axolotl.

After identifying robust gene expression changes in the axolotl, we searched for similar changes in A. maculatum and A. andersoni, which diverged from the axolotl approximately 21.47 and 4.27 MYA, respectively [44]. All three species regenerate limbs and histological changes observed during A. maculatum limb regeneration mirror those in A. mexicanum [45]. Thus, we anticipated that some genes would be expressed similarly among these species, even against the backdrop of species-specific developmental variations and lineage specific evolution, and the difficulty of correctly determining gene orthology relationships among non-model species [46]. The list of conserved expression changes reported among these three ambystomatid species is conservative, representing a robust portion of the conserved gene expression response to amputation injury.

Our axolotl-centered strategy yielded a high-confidence set of commonly-expressed genes, but largely left unresolved the identity of transcripts that are differentially expressed among species. Given caveats noted above when performing comparative transcriptomics, analysis of a single interval of time with limited replication has the potential to yield species-specific gene lists with many false positives. We did observe a few genes in A. andersoni and A. maculatum that were expressed in opposite directions from their presumptive axolotl orthologs (A. andersoni N = 11, A. maculatum N = 39). The smaller number of oppositely expressed genes between A. andersoni and the axolotl supports the idea that gene expression divergence correlates with phylogenetic distance [47]. We also identified genes that showed significant expression in two species but not all three. Of note, sall4, hmga2, and ngr1 were upregulated in A. maculatum and the axolotl, but were not significant in A. andersoni. Several gene expression studies have found sall4 and hmga2 to be significantly upregulation at 24hpa in axolotl, and sall4 is expressed during Xenopus limb regeneration [48]. Also, nrg1 is essential for nerve dependent axolotl limb regeneration [49]. These genes, as well as others in the datasets we report, may be shown in future studies to be expressed differently among salamander species.

Limb amputation causes tissue damage, bleeding, and necrosis. These events release damage-associated molecules and induce stress signals that initiate wound healing responses [50]. By 24hpa, the amputation site is covered by epithelial cells that migrated during the first 6–8 hpa from margins surrounding the wound [51]. Beneath this wound epithelium (WE) are extruded plasma and blood cells from damaged vasculature, as well as cellular and extracellular matrix debris [23]. Proper formation of the wound epithelium (WE) is crucial for host defense, tissue remodeling, and subsequent blastema formation [52]. Consistent with this injury scenario, we identified multiple genes that either annotate to the Keratinocyte Differentiation GO term or that are predicted from the literature to function in epidermal functions (cdh2, cnn3, cdknla, col4a1, ereg, dsg2, krt5, krt12, krt17, lamb2, lgr4, lef1, mycn, pde4d, orai1, rgma, sdc1, ski, sostdc1, tgm1, tmem79, tollip, thbs1, umod). For example, thbs1 has been shown to be highly upregulated at 24hpa, specifically in basal cells of the axolotl WE [53]. At 24hpa, the WE is a few cell layers thick and lacks a basement membrane. Thus, basal epithelial cells are capable of secreting molecules that affect underlying stump tissues. It is known, for example, that the WE becomes enriched with matrix metalloproteinases soon after injury [54] and this is coincident with the migration of keratinocytes from the wound margin, as well as histolysis of underlying tissues, ECM, and cellular debris. Thus, an early function of the wound epithelium is tissue remodeling, through the synthesis of metalloproteinases and regeneration permissive ECM molecules [52]. Genes that encode MMPs (mmp1, mmp3, mmp19, mmp28) and timp1, a regulator of MMP activity, were upregulated at 24hpa, as were genes that generate extracellular matrix components (has1, tnc). Also, the upregulated genes significantly enriched the Vesicle Exocytosis GO term; this might be associated with WE secretion, although the genes that enriched this GO term may also be expressed by infiltrating leukocytes that release cytokines and reactive molecules (see below). These results implicate the WE transcriptional program as a highly conserved feature of ambystomatid limb regeneration.

Many of the upregulated genes also enriched immunological GO terms, which is consistent with the presence of neutrophils, granulocytes, and macrophage populations at 24hpa in regenerating axolotl limbs [55]. These genes included biomarkers of hemostasis (f2rl2), wound closure (eppk, krt17), tissue histolysis (mmp1, mmp3, mmp19, mmp28, timp1), inflammation (ptgs2), and the production and delivery of host-defense molecules by immunological cells at sites of injury (cdc42ep2, cyba, cybb, hvcn1, mpo, rab11a, zyx). Leukocyte-associated genes were also upregulated, including biomarkers of platelets, eosinophils, B-cells, neutrophils, macrophages and T-cells (cdkn1a, cxcl1, ereg, foxo1, irs2, lgals9, map3k7, mpo, oraI, ubc). Diverse leukocytes in the post-injury environment, coupled with a secretory WE, would be expected to generate and release a repertoire of chemokines. Indeed, 22 upregulated genes (cdkn1a, cnn2, cxcl1, egr1, ereg, fos, foxo1, irs2, gpr35, mmp1, mmp3, ptgs2, sdc1, socs1, timp1, tnfrsf1b, tnfrsf1a, tnfrsf21, tnfrsf6b, tollip, ubc, ywhaz) enriched the Cytokine-Mediated Signaling Pathway GO term. Cytokine signaling activates diverse biological processes; this may explain why many of the same upregulated genes enriched a diversity of GO terms associated with immune responses, cell migration, intracellular signaling, tissue development, proliferation, and cell death. For example, reactive oxygen species function in pathogen defense and provide cues that are essential for regeneration in invertebrate and vertebrate models [56-59]. In Love et al. [57], knockdown of cyba, a gene that encodes a protein required for NADPH-dependent ROS production, inhibited Xenopus tail regeneration. Consistent with a similar requirement in Ambystoma, cyba and cybb were both upregulated at 24hpa, as well as other genes that function in ROS signaling (areg arg1, dusp1, fos, foxo1, hba1, klf2, mmp3, mpo, plk3, rhob, sdc1). Future studies of the highly-connected, upregulated genes identified in this study may identify transcriptional regulators that simultaneously mediate wound-healing responses and limb regeneration processes. Potential mediators of these events maybe found among genes in the Ambystoma CRG list that encode epigenetic, transcription and signaling factors (e.g. areg, cbfb, ctgf, cyr61, dkk2, dusp1, dusp5, dusp7, egr1, elf3, ereg, fos, foxo1, hdac7, hmga2, klf2, klf3, klf10, lef1, marcks, mas1, plekhg3/5, plekho2, plek, spil, ski, smad7, sostdc1, sox7, sp7).

A characteristic, early response to limb amputation is muscle tissue remodeling in the limb stump [60, 61]. Multinucleated muscle fibers fragment and give rise to mono-nucleated cells that act as muscle progenitor cells in adult newts but not axolotls [8]. Previous gene expression studies have documented significant reductions of muscle specific transcripts during limb regeneration [27, 32, 40, 41]. We similarly found that downregulated CRGs significantly enriched muscle tissue GOs. Additionally, we discovered that genes associated with various cellular metabolic processes were also down-regulated. Interestingly, when we examined the expression profiles of these metabolic genes from Voss et al. [27], we observed that almost all (22 of 28) showed the same temporal pattern of change across the first 28 days of regeneration. Thus, as was previously observed for muscle-associated genes, these metabolic genes exhibited a highly correlated pattern of expression. These patterns suggest that genes are regulated in a tissue-specific manner and/or are members of a common gene expression network. We note an important difference in the pattern of decline previously documented for muscle-specific transcripts in comparison to the metabolic genes report here. By 10dpa, muscle-specific transcripts decrease to levels that suggest absence of muscle tissue in samples collected for RNA isolation and expression [27]. This suggests a requirement for muscle tissue depletion in the area that the blastema forms; perhaps, until muscle tissue is depleted, progenitor cell recruitment and blastemal outgrowth is inhibited. In contrast to this pattern, transcript abundances for metabolic genes declined by approximately 2–1 fold to a moderate expression level that was sustained throughout limb regeneration. The moderate and not absolute decrease in transcripts of metabolic genes may be explained by their broader expression among tissues. While these transcripts are lost from cells that undergo reprogramming and/or that are lost to cell death, surviving cells continue to express these transcripts. If this inference is correct, it would suggest a connection among metabolic reprograming, tissue histolysis, and perhaps, activation of progenitor cells, with these events regulated in concert by 24hpa. Indeed, stem cell studies have shown a link between metabolic reprogramming and the promotion of stem-cell like properties [62].

Although we infer a conserved mechanism of muscle histolysis for Ambystoma, we note that muscle-specific and metabolic genes were downregulated to a greater extent in the metamorphic A. maculatum compared to the other two paedomorphic species. Metamorphosis is the ancestral developmental mode in Ambystoma and it is characterized by dramatic tissue remodeling events that are mediated by MMPs in amphibians [63]. During metamorphosis, the composition and structure of limb muscle is altered before individuals necessarily leave ephemeral aquatic larval habitats and use their limbs for terrestrial locomotion [64]. We hypothesize that the metamorphic life history presents a strong, stabilizing selective pressure for robust and rapid tissue remodeling in response to limb injuries that occur during the larval stage [65]. The limbs of paedomorphic species that have completely aquatic life histories are important for locomotion and balance in water, however limb regeneration is not time-constrained by metamorphosis.

We note in closing that some CRGs may be specific to salamanders or have regeneration specific functions that are not represented among annotated GOs. For example, prod1 was significantly upregulated among all three ambystomatid species. This gene is essential for newt limb development and regeneration [66, 67], and perhaps other salamander species [68]. As another example, rrad is expressed in response to limb amputation in newts, specifically in the nuclei of muscle cells at the site of injury [69]. This expression pattern suggests a role for rrad in muscle fiber dedifferentiation and the generation of muscle progenitor cells. A somewhat similar result was found in a mouse model of skeletal muscle regeneration [70]; rrad expression increased in response to chemical injury of gastrocnemius muscle and expression was localized to myogenic progenitor cells. Expression of rrad in mammals is thought to inhibit myoblast differentiation via the blocking of L-type Ca^{2 +} channels [71], a mechanism that may affect a time-delayed enhancement of cell proliferation [70]. This is consistent with salamander limb regeneration; progenitor cells do not undergo cell proliferation during the early wound healing phase [52]. However, an in vitro study of neonatal rat cardiomyoctes showed that rrad inhibits p38-MAPK signaling and initiates apoptosis [72], suggesting a mechanism which is consistent with muscle histolysis in salamanders. Studies are needed to examine the function of rrad in axolotls, which regenerate muscle tissue from satellite cells in the same manner as mammals, but of course to a much higher degree.

4. Conclusion

In this study we compared changes in gene expression during early limb regeneration among three Ambystoma salamanders. We identified differentially expressed genes that showed concordant expression changes as estimated by microarray and RNA-Seq platforms. The majority of these genes were also found to be significantly differentially expressed in a previously study [27]. We then used this set of A. mexicanum genes to compare their expression against presumptive orthologs from Ambystoma andersoni and Ambystoma maculatum, close and distant relatives, respectively. The new transcriptomic resources developed for A. andersoni and A. maculatum enabled us to identify differentially expressed orthologs that comprise a conserved transcriptional response initiated post-injury in ambystomatid salamanders.

5. Materials and methods

5.1. Animal care, tissue collection, and RNA isolation

Ethical animal procedures performed in this study were approved by the University of Kentucky IACUC committee (protocol 00907 L2005). Six larval A. mexicanum (~4–5 cm) were obtained from the Ambystoma Genetic Stock Center. Axolotls were maintained at 17–18C in 40% Holtfreter’s solution. Six larval A. andersoni (~4–5 cm) were generated from a single cross using two adults that were collected from Lake Zacapu under permit from the Mexican government. Six larval A. maculatum (~3–4 cm) were reared from an egg clutch collected from Stanton, KY under permit from the State of Kentucky. The larvae from the three species had fully formed fore and hind limbs and the A. maculatum did not show signs of metamorphosis. Animals were anesthetized in 0.001% benzocaine and limbs were amputate mid-stylopod with a sterile razor blade. Exactly 1 mm of distal limb tissue was collected for all species immediately after amputation (0dpa) and 24 h after amputation (24hpa). Three replicates samples were collected for each time point, with each replicate formed by pooling tissue from 5 different larvae. Total RNA was isolated according to the protocol in Voss et al. [27]. mRNA was extracted using the TruSeq RNA Library Prep Kit (Illumina).

5.2. Microaway and RNA-Seq data preparation

The six A. mexicanum mRNA samples were examined using both microarray and RNA-Seq methodologies. The six A. andersoni and A. maculatum mRNA samples were examined using only RNA-Seq. Microarray analysis was performed by the University of Kentucky Microarray Core using a custom Affymetrix GeneChip with 20,080 probe sets [73]. Microarray data were normalized using Robust Multichip Averaging (RMA) [74] as implemented by the Affymetrix Expression Console. HudsonAlpha Institute for Biotechnology generated the RNA-Seq data by preparing cDNA libraries and sequencing approximately 50,000,000,100 bp read pairs per sample (HiSeq, Illumina). De novo assembly of RNA-Seq isotigs was accomplished using Trinity’s three-stage protocol (Inchworm, Chrysalis, and Butterfly). The previous V4 assembly was incorporated during de novo assembly using the −long_reads option in Trinity. RSEM was used for count estimation with default parameters [75]. Expected counts were extracted from the RSEM output and used for data processing and statistical analyses.

5.3. Assignment of orthology between V5 and other datasets

The program BLAST [76] was used to assign orthology between RNA-Seq contigs and contiguous sequences that were used to design microarray probesets. Comparable sequences were identified by performing BLASTn searches and selecting pairwise alignments with ≥ 98% sequence similarity. To identify interspecies orthologs, the reference transcriptomes of A. andersoni and A. maculatum were queried against the A. mexicanum transcriptome database using BLASTn and BLASTx searches respectively. Transcript alignments were retained between A. mexicanum and A. andersoni if sequence similarity was at least 95%. A lower threshold (80% sequence similarity) was used to identify orthologs between A. mexicanum and the more distantly related A. maculatum. Additionally, an e-value cutoff of < 1E-09 was used in all BLAST searches.

5.4. Differential gene expression analysis

RMA normalized values for microarray and expected count values for all RNA-Seq datasets were filtered prior to identification of differentially expressed transcripts. Microarray platforms do not precisely detect low intensity values; thus it is useful to remove these probes to lessen the effect of multiple testing in downstream analyses [77, 78]. Probes with an average expression value greater than the first quartile of all values were considered. Filtered probes were then used for differential gene expression analysis using limma [79]. RNA-Seq expected count values were transformed to counts per million (CPM) or fragments per million (FPM), using edgeR [80] and DESeq2 [81] respectively; transcripts with a CPM/FPM value > 0 in all samples were considered.

Limma was run on all filtered probes to identify probesets that had significant differential expression (FDR < 0.1, moderated t-test) while directionality was determined based on log2 base fold change (upregulated: log2FC > 0; downregulated: log2FC < 0). DESeq2 and edgeR packages were used on filtered RNA-Seq transcripts to identify significant transcripts (FDR < 0.1) with similar log2FC requirements for directionality. DESeq2 identified significant transcripts using the likelihood ratio test after normalization based on dividing each samples’ counts by the estimated size factor of that sample. edgeR transformed the expected counts using TMM (trimmed means of M values) normalization and identified significant transcripts based on a general linear model likelihood ratio test. Comparisons between time points were done as 0hpa against 24hpa samples for both microarray and RNA-Seq analyses. All analyses were done using R [82]. Complete linkage, hierarchal clustering of DE gene expression estimates identified one A. mexicanum RNA-Seq sample (SRV.0043) as an outlier and this sample was removed from the RNA-Seq analysis of DE genes.

5.5. Anonymous CRG characterization

To characterize anonymous CRGs, A. mexicanum sequences were queried against the Am_3.4 transcriptome, available on https://axolotl-omics.org/blastn [35] and the A. mexicanum transcriptome v1.0, available on http://www.cruzomics.net:8888/ [36], using BLASTn and default settings. Additionally, local BLAST searches were performed using the tissue mapped transcriptome described in [34] using BLASTn (version 2.2.17, E-value < 1E-10 and percent identity ≥ 98%). CRG sequences were also compared to the NCBI non-redundant (NR) database [83] to identify sequence hits, using BLASTx (E < 1E-5 and default parameters). Once matches were identified, the unknown CRG was annotated based on the sequence with the best bit score and longest alignment length. For CRGs possessing no gene identifier, the coding potential was calculated using the Coding-Potential Assessment Tool (CPAT) [37] and Coding Potential Calculator 2 (CPC2) [38]. CPAT and CPC2 were used in conjunction to discriminate between potential coding (average Prob. ≥ 40%), non-coding (average Prob. ≥ 10%), and ambiguous RNAs (10% < average Prob. < 40%). CPAT was utilized to calculate coding potential using RNA features (ORF length, ORF coverage, Fickett TESTCODE score, and hexamer usage bias) from human (hg19) and zebrafish (Zv9danRer7) genome training sets; CPC2 was calculating using similar RNA features (Fickett TESTCODE score, ORF length and integrity, and isoelectric point) from human (hg38), zebrafish (danRer7), and Xenopus tropicalis (xendTro3) genome training sets. Cutoffs for coding assignments followed CPAT recommendations. Additionally, unknown transcripts were queried against the NCBI Open Reading Frame (ORF-Finder) [84] and Conserved Domain (CD-Search) [85] databases to identify the longest open reading frame and conserved domains respectively. Rfam [86] and Pfam [87] online tools were utilized to identify conserved families as well. To identify micro-RNA (miRNA) matches, predicted non-coding RNA transcript sequences from all species were queried against the online miRBase [88] database for significant hits to mature miRNA sequences (BLASTn, E: < 1).

5.6. PANTHER annotations

Gene ontology annotations were identified using PANTHER [39]. The statistical overrepresentation test (adjusted p-value < .05, Binomial test with Bonferroni correction) was utilized to separately query upregulated and downregulated CRGs against the complete Biological Process, Molecular Function, and Cellular Component Gene Ontology (GO) lists. The most specific GO subclass was reported and not more inclusive parent terms. The reference list consisted of all gene IDs from the axolotl Affymetrix microarray.