Transcriptome Analysis Identifies 3520 Genes Up-regulated in the Developing Mouse Limb
To identify novel molecules responsible for limb patterning processes and determinants of limb identity, as well as to characterize the expression profiles of genes co-transcribed at high levels during limb development, we generated mouse whole-genome microarray gene expression data for fore- and hind-limbs from five consecutive developmental time points, from E9.5 to E13.5 (), accompanied by data for whole embryos as controls. Fore- and hind-limb samples were compared to whole embryo samples at each time point to determine which transcripts were differentially expressed in the limb. Comprehensively, 6216 unique genes (31.4% of the mouse genome) were differentially expressed in at least one of the comparisons (with 3520 being significantly up-regulated), therefore, unless specified all percentages herein are based on this total count of differentially expressed genes (6216). We found E10.5–E11.5 to have the largest number of significantly up-regulated genes, with 30–36% of the transcripts up-regulated (1840–2263 genes). The least number of differentially expressed genes was found in the E9.5 fore-limb, where only 8% of the genes were up-regulated (), consistent with the very early stage.
Differentially Expressed Genes.
For the top 100 forelimb genes significantly up-regulated at each developmental time point we examined the known molecular function (GO category), description of available knock-out mouse phenotypes (http://www.informatics.jax.org/phenotypes.shtml
) and the documented limb expression in whole mount mouse embryos (http://genome.ucsc.edu/cgi-bin/hgVisiGene
). Most genes were represented in more than one developmental stage, corresponding to a total set of 279 non-redundant transcripts (Table S1
). Thirty genes were up-regulated at all 5 time points, 126 were up-regulated at 4 time points (primarily E10.5–E13.5), 51 were up-regulated at 3 time points, 28 were up-regulated at 2 time points, and 44 transcripts were up-regulated at only one time point, either E9.5 (28 genes) or E13.5 (16 genes).
Among the 279 transcripts significantly up-regulated in at least one developmental stage, we found 21% to be transcription factors, 10% signaling molecules, 9% membrane-associated proteins (including receptors and channels) and 24% of them to be of unknown or novel function. Twenty-eight percent of these transcripts had confirmed musculoskeletal knockout phenotypes (79 genes; highlighted in brown in Table S1
), and a similar percentage (27%; 76 genes) have been previously shown to be expressed in the limb by in situ
hybridization (). These two sets of genes were highly overlapping (67%), and 23 genes with confirmed limb expression patterns corresponded either to previously uncharacterized genes, genes that resulted in no phenotypes (normal mice) when knocked out from the mouse genome, or the available knockout had no reported limb or musculoskeletal defect. Of the genes previously uncharacterized in the limb, 12 were perinatal lethal prior to E11.5 (highlighted in purple in Table S1
), 14 had in situ
expression patterns available and 137 or ~50% were novel, lacking any prior knowledge of their putative limb function (Table S1
; highlighted in grey). In contrast, the available knockout mice for 38 genes had ‘normal’ or no reported limb/musculoskeletal defects described, suggesting that these genes may be functionally redundant and the loss of their proteins is compensated by the presence of another, closely related transcript. Table S1
provides details for all the 279 transcripts. When analyzing the putative gene function off all 279 limb-enriched genes, the largest number of transcription factors (33%) and growth factors (15%) were found enriched at E9.5. In addition, 45 of the total 77 genes with reported limb or musculoskeletal defects in the documented knockout mouse strains were up-regulated at E9.5 suggesting that limb morphogenesis may be driven by very early signaling/transcriptional events that commence immediately after limb bud initiation.
Analysis of the Top 100 Up-regulated Genes.
More than 100 Novel Genes are likely to Contribute to Limb Morphogenesis
Among the 279 transcripts identified as the most highly expressed genes in the murine limb relative to the whole embryo, about half were novel (49% or 137 genes; Table S1
; highlighted in grey). We defined novel genes as genes with no available knockout information and no published reports that would suggest the gene is essential to limb patterning or function. A few of the novel genes, primarily transcription factors, had known embryonic in situs
patterns and those were noted in . We surveyed in situ
expression patterns among 3 groups of transcripts: 1) genes known to function during limb development (Figure S1
), 2) known genes previously uncharacterized in the limb (), and 3) novel genes (Table S1
; highlighted in grey and ), and found that while the majority of genes appeared to be expressed at E10.5 throughout the limb bud (without differentiating between superficial ectodermal staining and mesenchymal deep staining), a large number of genes had restricted and unique limb expression patterns. For example, among the novel genes, C130021l20Rik
was expressed in dorsal ectoderm, Nnat
were primarily expressed in distal mesenchyme, Fibin
was expressed at the base of the limb corresponding to cells migrating from the somites into the limb anlage, while the rest were expressed throughout the limb. The novel genes in are highlighted in grey in Table S1
Enrichment of functionally related genes in groups of co-expressed genes.
In particular, the transcript corresponding to the novel gene C130021I20Rik
, which was one of the most highly enriched transcript [classified among the top 100 genes at all developmental stages examined (E9.5 to E13.5)], was found to have a highly restricted expression pattern, primarily in the dorsal mesenchyme, highly similar to LIM homeodomain transcription factor 1 beta (Lmx1b
(). Further examination of the genomic locus of this transcript, revealed that in the mouse, its transcription start site is physically located 682 base pairs upstream of Lmx1B
, and is transcribed on the opposite strand (). Using nucleotide blast we compared the Lmx1B
1.2 kb mRNA (NM_010725) to the C130021I20Rik
mRNA (AK147796) and found no significant sequence homology between the transcripts. We also did a detailed in situ
analysis of these two genes from E10.5 to E12.5 () and found them to have highly similar expression patterns in the limb. The close proximity of these genes, the lack of sequence homology shared by the transcripts and the divergent transcriptional direction of these two genes suggest that they are transcribed from a bidirectional promoter and are likely to share cis
-regulatory elements that may be required for their shared limb-specific expression patterns. LMX1B
mutations in humans cause an autosomal dominant inherited disease called nail-patella syndrome (NPS), which is characterized by abnormalities of the arms and legs as well as kidney disease and glaucoma 
. Expression of Lmx1b
has been described as dorso-mesenchymal and this gene has been shown to be critical for specification of dorsal limb cell fates and consequently dorso-ventral patterning of limbs 
also appears to be restricted to the dorsal mesenchyme (), suggesting that this new transcript may functionally cooperate with Lmx1b
to play a central role in fate determination and/or cell differentiation in the dorsal limb.
C130021I20Rik is a Novel Gene Co-transcribed from the Lmx1b Promoter, a well Known Limb Gene.
Limits of Transcriptome Analysis: Missing Functional Genes with Low Limb Expression
To identify transcripts with elevated levels of expression in the limb we initially compared each limb sample to its corresponding whole embryo sample. In this comparison, genes that are highly expressed in the limb but at relatively low levels in all other parts of the embryo will emerge as ‘up-regulated’ in the limb. Even if a gene is expressed at high levels in other parts of the embryo, it will still be identified as ‘up-regulated’ in the limb as long as its overall RNA concentration (total RNA per sample volume) remains lower in the whole embryo as compared to the limb sample. In contrast, a gene that is ‘ubiquitously’ expressed will have similar RNA concentrations and hence will not be identified as ‘up-regulated in the limb. Thus, one potential limitation of this analysis stems from comparing the limb expression to the whole embryo expression; while powerful at extracting genes that are robustly expressed in the limb in relationship to other tissues, it falsely eliminates transcripts that are expressed in the limb at very low levels, or play important roles in non-limb tissues and are thus highly expressed in other parts of the embryo as well.
To estimate the rate of false negatives we examined how many genes known to cause limb defects are up-regulated in our comparisons. We extracted from the Mouse Genome Informatics Database (http://www.informatics.jax.org/
) all the genes (674 genes) annotated with abnormal limb morphology (MP:0002109) as well as categories directly related to this one in the Phenotype Ontology hierarchy. Of these 674 genes, only 449 were represented on the array and 186 (41%) were significantly up-regulated in the limb in at least one comparison. Next we examined whether the genes that were not significantly up-regulated had different expression levels in the limb and in the whole embryo as compared with the genes up-regulated in the limb. We found that the genes that were not up-regulated in limb were expressed at significantly lower levels in the limb (P-value
, Wilcoxon rank sum test), but at approximately the same levels in the whole embryo. Furthermore, these genes were found to be expressed at lower levels in the limb as compared with all the genes on the array (P-value 2.9×10−6
, Wilcoxon rank sum test), but not in the whole embryo. This analysis suggests that it is the particularly low level of expression of these genes in the limb rather than the differences in their average level of expression in the whole embryo that reduces our power to detect them in the context of this analysis (Figure S2
We also examined 200 genes corresponding to the BMP, WNT, TGF-beta, hedgehog, FGF, and ROR signaling pathways, homeobox transcription factors along with other known signaling molecules and transcription factors linked to these pathways. We found 40% of these genes to be up-regulated in the limb in at least one comparison (82/200; Table S2
). Twenty-nine percent (58) of these transcripts were down-regulated in the limb, compared to whole embryos, at least at one time point, and unchanged at all other time points, suggesting that a different part of the embryo had much greater expression of these transcripts than the expression observed in the limb. An additional 6 transcripts displayed a mixture of up-regulated and down-regulated incidences at specific time points, mostly in the Hox
transcription factor category (4/6 genes). Most of the hedgehog signaling pathway genes were underrepresented among the up-regulated genes (Gli2
; 1/7 was enriched in the limb), primarily because of their wide tissue distribution that masked their transcriptional enrichment in the limb.
Finally, we examined the expression values for genes transcribed at low, moderate and high levels in the limb. Genes on the arrays had (log2 transformed) expression values ranging from 2.4 to 14.6, with a median of 6.1. Using the average expression values for Lrp5
, genes known to have important roles in the limb, but expressed at low levels (Lrp5
; 4.2–5.6 expression value range) or in a highly restricted cluster of cells (Shh
; 4.7–8.8 expression value range), we identified 19 genes with low expression values (<5) at all 5 developmental time points. This list included 4 genes known to have limb defects as heterozygous or homozygous null (Tbx6
). These analyses suggest that up to 60% of genes of limb genes expressed at very low levels may be missed by this method (Table S2
; genes highlighted in yellow). However, additional biological validation will be needed to confirm the true rate of false negatives.
Shared Expression Profiles Reveal Gene Specific Functions in Limb Development
Genes significantly up-regulated in the forelimb clustered into 31 groups according to their expression profiles. Twenty of these groups were comprised primarily of the up-regulated expression profiles of 3213 transcripts or 91% of the total genes found to be up-regulated in at least one pair-wise comparison with the corresponding whole embryo control. These 20 clusters were further grouped into four major categories: early genes, peak genes, late genes and oscillating genes (; genes corresponding to each cluster are listed in Table S3
). The expression of 247 transcripts was elevated at E9.5 and at least one other time point, as part of clusters 1–4 comprising the ‘early gene’ category, where only 44 transcripts were enriched at all examined time points ().
Temporal Expression Pattern Analysis in the Developing Forelimb.
A second category of genes, defined as ‘peak genes’, were distinctly up-regulated either at a single time point or within a narrow range of time points (). The largest cluster consisted of transcripts up-regulated at E10.5 and E11.5 (491). A smaller cluster encompassed genes enriched between E10.5 and E12.5 (401) and 151 transcripts were significantly up-regulated at E11.5 and E12.5. Genes that peaked at a single time point were classified as ‘stage-specific’ genes and are discussed in the next section. A third category of genes, termed ‘late genes’ turned on past E9.5 and their up-regulated expression persisted up to E13.5, and possibly beyond. A large number of transcripts (475) were found significantly up-regulated from E10.5 to E13.5, while smaller cohorts were enriched from E11.5 (73) or E12.5 (77) and beyond.
The last category of genes, ‘oscillating genes
’ is represented by clusters of genes that display patterns of oscillating up-regulation in the limb, relative to whole embryo. A small number of genes (151) were distributed among these 5 clusters, with the smallest clusters representing genes that are up-regulated only at E9.5 and E11.5 (22), or at E10.5 and E13.5 (20). To determine if these genes truly have an oscillating behavior or whether their profiles are misleadingly derived because of fluctuating expression levels in the whole embryo we normalized the log2 expression value of each gene so that the mean and standard deviation across all the arrays are 0 and 1 respectively and plotted the limb and whole embryo expression in Figure S3
. For most whole embryo samples we found the log2 expression to be negative, corresponding to low level of expression in the whole embryo. In most cases we found that, indeed, the limb expression oscillates, but the oscillating behavior was due either to variability in the limb expression only, or both in the limb and whole embryo (Figure S3
). For example, in the case of genes that are “on” in the forelimb at E10.5 and E12.5 (51 genes), we actually have two effects at work: these genes are particularly highly expressed in the limb at these time points, but they are also slightly less expressed in the whole embryo as well, magnifying this effect. Similar observations were made for genes “on” in the fore-limb at E10.5 and E13.5: their expression was slightly decreased in the forelimb at E11.5 and E12.5, but also slightly increased in the whole embryo. Although the observed changes in expression are obviously not restricted to the limb, there is a limb-specific trend in their expression and this trend in the limbs is contrary to that in the whole embryo.
One goal of using cluster analysis for such comprehensive expression datasets is to detect co-expressed genes, potentially revealing gene networks likely to regulate functionally distinct developmental processes such as chondrogenesis, osteoblastogenesis or myogenesis that progress concomitantly (). While the limb has long been an important model system for examining the molecular mechanisms of tissue patterning during development, its tissue heterogeneity, with many different cell types and signaling pathways intersecting to build complex multifunctional structures, poses a great challenge for the identification of functionally related genes. In we have outlined the events, developmental landmarks, and few molecular markers currently known for the formation of skeletal elements, muscle, nerves and skin between E9.5 and E14.5. To determine whether we can identify new gene candidates likely to contribute to one, or more of these pathways, we examined all clusters with more than 40 transcripts (15 clusters) for the presence of enriched functional categories (enrichment ≥2 fold with p-values adjusted for multiple testing ≤0.05) using the available Gene Ontology (GO) annotation for each transcript. ~50% of the clusters examined (8/15) were found to be statistically enriched in genes that share a molecular function or a biological process ().
Differentiation Events Corresponding to Stages of Limb Development and Functional Category Enrichments that Describe some of these Events.
Clusters 1, 2, 5, and 6 were found to be enriched in transcripts involved in limb and skeletal morphogenesis, while clusters 1, 2, 5 and 12 were enriched in genes associated with transcriptional regulatory functions. The transcripts in cluster 1 (44 transcripts) were up-regulated in the limb at all examined time points; 45% of them had documented limb and/or skeletal defects (20/44) and 40% corresponded to transcription factors (18/44), a 12-fold enrichment for DNA binding transcription factor activity (; S3).
We plotted the relative expression levels for the genes corresponding to osteoblast differentiation, epithelial differentiation, anterior-porterior patterning, neuronal differentiation, digit morphogenesis, epithelial-mesenchymal transition, cartilage development, embryonic skeletal morphogenesis, and limb morphogenesis enriched categories in and examined their expression dynamics vis-à-vis what is known about the morphological events during this developmental time course (). For most of these clusters we found a tight correlation between their expression pattern and the function they are known to participate in. For example anterior-posterior pattering is an early event, therefore most of the genes associated with this function had the highest expression level at E9.5, and continued to decline beyond this time point (). In contrast, digit morphogenesis occurs beyond E10.5, and most transcripts associated with this function were not expressed at E9.5, but were sharply up-regulated at E10.5 and beyond (). The most surprising observation was for genes associated with osteoblast differentiation (). Osteoblasts themselves are not ‘present’ in the limb until late E12.5, early E13.5, yet, these genes seemed to be consistently expressed at high levels during most time points, suggesting that these genes either participate in multiple independent events during limb morphogenesis, or that osteoblast specification and differentiation may initiate much earlier than anticipated and new cell-type specific markers may be needed to discriminate pre-osteoblasts from mature osteoblasts. While functional correlation of genes comprising these clusters can only be achieved through experimentation and validation, these new relationships can assist prioritization in future functional characterization of limb genes and conclusively link the genes with the biological process they mediate during musculoskeletal morphogenesis.
Stage Specific Limb Development is Regulated by Redundant Mechanisms
With the aim of identifying the genes that determine limb patterning at particular developmental stages, we first established differences among cohorts of up-regulated genes between all pairs of successive time points. Overall, less than 27% of genes were differentially expressed between consecutive time points, with most changes ranging between 3–10% of the total 6216 differentially expressed transcripts were limb specific. We found the most dramatic up-regulation of gene expression to occur between E9.5 and E10.5 fore-limb comparisons, where ~8 times more transcripts or 27% of total transcripts were more highly expressed at E10.5 forelimb than at E9.5. In the hindlimb, a similar transition was observed between E11.5 and E12.5 where the E11.5 hindlimb had 6 times more genes expressed than the E12.5 hindlimb. Since hindlimb development is delayed approximately by half a day in relationship to fore-limb development 
, we compared the overlap between E10.5 fore-limb specific genes (1675) and E11.5 hind-limb specific genes (1193). We found 607 transcripts or 50% of genes to be shared among these two cohorts, suggesting that E10.5 in forelimb and E11.5 in hindlimb may be proceeding through similar morphological processes and may represent key transitional time points during limb development, with the greatest number of up-regulated transcripts during the E9.5 to E13.5 developmental window.
Next, we turned to the analysis of strictly stage-specific genes, i.e., genes that are exclusively up-regulated at one time point (; stage specific genes). E11.5 had the largest number of stage-specific
transcripts (405 or 7%), E12.5 has the least (159 or 3%), while all other time points had an average of 192 genes (3%). Twenty-three of these transcripts were also found among the top 100 overexpressed genes in the fore-limb, and are highlighted in Table S1
by 1 or 3 asterisks. We examined the known molecular function, description of available knock-out mouse phenotypes and the documented limb expression in whole mount mouse embryos for these 1121 genes (Table S4
). E9.5 enriched transcripts had the largest number of confirmed musculoskeletal null phenotypes (30/184 or 16%), while E13.5 had the lowest (9/183 or 5%). The reverse relationship was observed for genes without any reported function in the limb, where 82% of the E13.5 enriched transcripts (150/183) had no available knockout, nor was the gene ever linked to any embryonic limb developmental function; in contrast 57% of the E9.5 enriched genes present in this category (106/184) ().
Analysis of Stage Specific Gene Expression.
The in situ
expression pattern survey of these 1121 stage specific genes revealed that with the exception of E9.5 and E10.5 genes where 18.5% and 11.5% were confirmed to be expressed in the limb, all other time points had less than 7% of genes with reported limb expression data. In we show the spatial expression pattern for representative stage specific genes from each time point (marked by an asterisk in Table S4
). While a majority of these transcripts appear to be expressed throughout the limb bud (without differentiating between superficial ectodermal staining and mesenchymal deep staining), a large number of these genes also exhibit restricted and unique limb expression patterns. Interestingly, even transcripts where the available knockout mouse has not been reported to have a musculoskeletal defect show highly restricted limb expression patterns. Such genes include Lmo1, Lmo3, Irx2
transcription factors (). While Lmo1
knockouts have no obvious phenotypes, the Lmo1/Lmo3
double knockout dies shortly after birth due to neural tube defects. In contrast Lmo2
knockouts die embryonically before E10.5, suggesting that some or all of the 4 Lmo
family members have redundant and overlapping functions, possibly including aspects of limb development 
Several other gene families were overrepresented among these stage specific enriched genes: 19 transcription factors belonged to the Kruppel associated box (KRAB-containing zinc finger) transcription factor family of transcriptional repressors (in descending order of expression: E9.5 enriched Zfp771, Zfp775; E10.5 enriched Zfp763, Zfp418, Zfp68; E11.5 enriched Zfp90, Zfp783, Zfp13, Zfp637; E12.5 enriched Zfp57, Zfp354b, Zfp110, Zfp324, Zfp715; and E13.5 enriched Zfp760, Zfp826, Zfp280d, Zfp207, Zfp788); 11 genes belonged to the Tmem family of transmembrane-like proteins (E9.5 enriched Tmem59l, Tmem115, Tmem54, Tmem178, Tmem101; E10.5 Tmem177, Tmem2, Tmem120a ; E11.5 Tmem138; E12.5 Tmem208, Tmem11); 4 genes are metallopeptidases as part of Adam and Adamts familes (E9.5 enriched Adamts18 and Adamts1; E10.5 Adam10; E11.5 Adamts7 and E13.5 Adamts6); and 3 genes belonged to the SWI/SNF related, matrix associated, actin dependent chromatin regulators or Smarc genes (E9.5 SmarcD2; E11.5 SmarcD3 and SmarcE1).
Zinc finger proteins containing the KRAB motif represent the largest single family of transcription factors, estimated to make up ~30% of all annotated zinc finger transcription factors in the human genome (290/799) 
, therefore it is no surprise that ~2% of all the stage-specific enriched genes or 12% of the total transcription factors belong to this category. What is unexpected, however, is that only 2 of the 11 genes identified have been previously studied; Zfp110
knockout has been described to be perinatal lethal by E12.5, and Zfp826
has documented skeletal and craniofacial defects 
. Among the 91 transcription factors in this category, 51 have available knockouts, 46 (90%) of which have been described to exhibit various musculoskeletal/limb defects or are perinatal lethal. The remaining 5 KOs are either normal or do not have any reported limb defects (Table S4
). This high percentage of transcription factors with skeletal defects suggests that most of the other 40 novel transcription factors in this category are likely to participate in important events during limb morphogenesis, particularly with a likely role during the narrow developmental time point when their transcription is relatively up-regulated.
Disintegrin and metalloproteinases (ADAM) and ADAMs with thrombospondin motifs (ADAMTS) are two subfamilies of metalloproteinases closely related to the matrix metalloproteinases (MMPs). Some members of these subfamilies are associated with various physiological and pathological processes involved in embryonic development, angiogenesis, coagulation, and arthritis 
. Nineteen distinct ADAMTS genes have been identified in humans, 4 (21%) of which are enriched in the limb. Their substrates include procollagen, hyalectans, decorin, fibromodulin and cartilage oligomeric matrix protein, hence based on their previously characterized roles these 4 Adamts
genes enriched during limb development are likely to contribute to joint and cartilage formation. Adamts1
knockout has a described adipose tissue defect, while the function of all the other 3 (Adamts 6, 7
) have not yet been examined in knockout mice. ADAM10, the other metalloproteinase found to be up-regulated in the limb at E10.5, causes embryonic lethality by E9.5 due to failure of the cardiovascular and nervous system failures 
Hindlimb Identity is likely achieved by Inhibition of Forelimb-Specific Genes
The analysis of differentially expressed genes is a powerful approach for elucidating the genetic mechanisms underlying the morphological and evolutionary diversity of serially homologous structures within the same organism (hand vs. foot) or among different species (hand vs. wing). Most genes known to be involved in limb patterning processes have been shown to confirm highly similar expression patterns in both the fore- and the hind-limb, however they dictate the formation of skeletal elements that results in distinctly unique structures such as hands and fingers in the forelimb and feet and toes in the hindlimb. Despite these dramatically different phenotypic skeletal patterning outputs, to date only a few genes have been determined to regulate limb-type identity. These include the forelimb-restricted Tbx5
and the hindlimb-restricted Tbx4
transcription factors 
. It has been hypothesized that additional key regulators of limb identity exist. This hypothesis is based on several observations: Pitx1
are all transcription factors however their target genes are not known, nor are the upstream transcriptional regulators that confirm their limb specificity. In addition, gain- and loss-of-function mutations in these genes result in partial limb-type transformation phenotypes suggesting that additional key molecules participate in limb-type specificity.
To identify new genes and formulate new hypotheses about initial patterning control and molecular pathways that dictate limb-specific identity we compared fore- and hind-limb expression at each time point (E10.5–E13.5, ). Comprehensively, 855 transcripts were found to be up-regulated specifically in the fore- and 511 in the hind-limb, at least in one developmental time point. 230 transcripts were enriched in the forelimb at E9.5, 155 at E10.5, 245 at E11.5, 298 at E12.5 and 105 at E13.5. In the hindlimb, we observed 278 enriched transcripts at E10.5, 151 at E11.5, 36 at E12.5 and 114 at E13.5 ().
Identifying Genes that Contribute to Limb Identity.
show genes up-regulated 2-fold or more in fore- (76) or hind-limb (11), relative to each other, in at least one time point. Furthermore, we examined the known molecular function, description of available knock-out mouse phenotypes and the documented limb expression pattern in whole mount mouse embryos for these genes and found 37% of them to have confirmed limb and skeletal phenotypes (fore- 29/76; hind-limb 6/11) in available knockout mice (Table S5
). In addition, all genes that have been previously shown to be differentially expressed between fore- and hind-limbs, including Pitx1
, and several Hox
genes were also found to be differentially expressed at least at one sampled time point (; red arrows). The identification of these genes provided a positive control, and these differences were confirmed by whole mount in situ
hybridization (). Among the genes for which we could evaluate a protein function description based on GO classification confidence level and available experimental data, we found 25% (19/76) of fore- and 45% (5/11) of hind-limb genes to be known or putative transcription factors, suggesting that limb identity may be driven by a divergent transcriptional program, that uses different cohorts of transcription factors to instruct limb identity.
Expression Patterns of Fore- and Hind-limb Specific Genes.