Systematic analysis of palatal transcriptome to identify cleft palate genes within TGFβ3-knockout mice alleles: RNA-Seq analysis of TGFβ3 Mice

Analysis of RNA-seq data

As outlined in the experimental workflow of our study (Figure 1), RNA samples were extracted from mouse palatal tissues of TGFβ3 knockout mice (homozygous (−/−), heterozygous (+/−), and wildtype (+/+)) at three developmental stages (E14.5, E15.5, and E16.5); libraries were prepared, converted to cDNA, and sequenced by using the Illumina HiSeq2000 next generation sequencer. The bioinformatics analysis of RNA-Seq data was processed using the Tuxedo protocol [31], which includes TopHat, Cufflinks and CummeRbund packages from R. The CP genes were identified from OMIM and MGI databases and downstream effects were analyzed using IPA.

Our data clearly demonstrate that TGFβ3 wildtype (+/+) palates from E15.5 expressed increased TGFβ3 protein, double the expression of heterozygous (+/−) and as expected the homozygous (−/−) palates were completely devoid of any TGFβ3 expression (Additional file 1). Extracted RNA samples were evaluated by RNA integrity numbers (RIN) to monitor quality and degradation with BioAnalyzer (Agilent Technologies, Palo Alto, CA). We confirmed that all of the palatal RNA samples had an RIN ≥ 8.4, indicating the high quality and low degradation of our experimental samples (Additional file 2). The RNA libraries were prepared for cDNA conversion and the samples were sequenced by Illumina techniques using HiSeq 2000. We obtained total reads from the RNA-Seq samples ranging from 14 million to 96 million (Additional file 3). Among all samples, only 4 samples have less than 40 million reads. We did not apply sequence manipulations on either 3^′ or 5^′ ends because the RNA-Seq sequencing quality was high across all samples based on our FastQC data quality control analysis. The relative abundance of transcripts was normalized to the total read number and shown as fragments per kilobase per million (FPKM) and mapped via Cufflinks.

In order to evaluate the reproducibility of our experimental samples and procedures, two independent biological and technical replicates for each gestational age and each allele resulting in 36 (n = 9 × 2 × 2) paired-end samples were assayed [32]. Using the scatter plots generated by CummeRbund, biological and technical replicates of our samples were validated based on the comparison of their read numbers per transcript. We detected a high correlation among both the biological and technical replicates of each gestational age and allele (Figure 2) (r = 0.98, Pearson’s correlation; p < 0.001). The replicates of E14.5 of TGFβ3−/− mice are presented in Figure 2. Additional scatter plots for all of the other samples with their replicates are shown in Additional file 4. Because the expression values (FPKM) of certain CP related genes differed significantly among the gestational ages, we transformed the FPKM value of each gene into log₁₀ (FPKM + 1), rounded up to 2 decimal places, and used transformed values for further analysis.

Overall transcriptome expression profile

Identification of CP related genes and their expression patterns

On the basis of our differential expression profile data between genotypes and gestational ages (Table 1), we decided to group and analyze the potential CP genes based on their expression patterns throughout different stages of palatogenesis. We categorized the 322 CP genes identified above into 9 possible expression patterns based on the comparison of their expression values (upregulated, downregulated, or unchanged) across the 3 different gestational age groups analyzed by RNA-seq (E14.5, E15.5, and E16.5) (Figure 3). The patterns (p) represent the change in expression value of the transcript first from E14.5 to E15.5, and second from E15.5 to E16.5, which is explained in detail in Figure 3. The genes in each expression pattern group are listed and common genes among them are diagrammed in Additional file 9.

Considering the differential expression of CP genes throughout palatogenesis, we generated individual heatmaps according to the expression patterns of CP related genes and grouped them into corresponding genotypes (Figure 4). In TGFΒ3−/−, the majority of transcripts (n = 294; 92%) were placed in p1, which denotes unchanged expression of genes throughout gestation. Similarly, in TGFΒ3+/− samples, the majority of transcripts were in p1 (n = 293; 91%). Surprisingly, the list of genes in p1 from TGFΒ3−/− and TGFΒ3+/− were highly comparable (92%) (Additional file 9). The genes clustered in p1 unique to TGFΒ3−/−, but not in TGFΒ3+/−, were Chrng, Foxc2, H19, Kcnj13, Lhx8, Meox2, Shh, and Six3 (Table 3). We concluded that these 8 genes demand additional attention due to their unchanged level of expression during crucial stages of palatogenesis.

Finally, in WT samples, the transcripts were more equally dispersed into p1, p2 and p4 as 29%, 32%, and 23% respectively. These results suggest that in normal mice, the majority of CP transcripts are required to be upregulated between E14.5 and E15.5 (denoted as p2 and p4), which corresponds to the most crucial stages of palatogenesis- the adhesion and the fusion stages [34]. Some of these upregulated genes were eventually downregulated (p4) or remained at the same expression levels (p2), which suggest that they may contribute to the final phase of palate development.

Downstream pathway analysis of CP genes

We used the IPA software to determine how the expression pattern-clustered CP genes interconnected with each other to facilitate cellular biofunctions and signaling. According to the core analysis of all CP genes (n = 322) using the IPA, the most significant top 5 cellular functions were cellular development, gene expression, cellular growth and proliferation, cell signaling and cell morphology (Table 4). Although the numbers of molecules were higher in some cellular functions, the ranking was based on p-values calculated by IPA according to relevance of datasets with IKB. Differentially expressed molecules clustered into GO of these cellular biofunctions are tabulated in Additional file 10. The pattern-oriented distribution of the top cellular biofunctions (n = 8) in relation to the TGFβ pathway are illustrated with a color coded matrix with the darker color representing higher relevance between cellular functions and uploaded CP gene pattern datasets (Figure 5).

The CP genes unique to p1 of TGFΒ3−/− (n = 8) may have necessary functions for the complete development of palatal shelves. Therefore, we analyzed the interaction of these genes with each other, as well as with key biofunctions and TGFβ signaling using IPA (Figure 6). We found that 5 of these genes (Shh, Foxc2, Six3, Meox2, and H19) are closely related to each other, with Tnf (tumor necrosis factor) and Ccnd1 (Cyclin d1) being their intersecting molecules. These genes were mostly associated with cell morphology and cell death biofunctions, and were indirectly associated with TGFβ signaling through Pias4 (protein inhibitor of activated STAT, 4).

According to IPA core analysis, there were 19 molecules associated with TGFβ signaling [Ratio: 19/89 (21%)] among all of the CP genes (n = 322). TGFβ signaling molecules along the pathway and their cellular locations are illustrated in Figure 7. TGFβ signaling transcripts expressed in both TGFΒ3−/− and TGFΒ3+/− were (n = 18): Acvr1, Bmp2, Bmp4, Bmp7, Crebbp, Grb2, Inhba, Inhbb, Kras, Map2k1, Map2k2, Nkx2-5, Runx2, Smad3, TGFβ2, TGFβ3, TGFβr1, TGFβr2 (Table 3). All of these molecules followed the p1 expression pattern, which represents a uniform level of expression throughout all time points. However, within WT transcripts, TGFβ-associated molecules were almost equally distributed among p1, p2, p4, and p8. All represented TGFβ pathway molecules (n = 19) were grouped into genotypes according to their expression patterns (Table 5). When the significantly altered CP genes were overlaid with TGFβ signaling, all of these genes followed the Smad-dependent pathway (Figure 7).

Global transcriptome of TGFβ3 alleles

Palatogenesis is a complex process involving multiple steps, including: palatal shelf growth, elevation, adhesion and fusion (Additional file 11). It requires the orchestration of several key cellular functions (such as cellular proliferation, differentiation, movement, communication, and death) regulated by numerous genes and signaling pathways. It has been well documented that disruption of structural and regulatory genes involved in any stage of palatogenesis, which occurs between E11.5 and E16.5 in mice, contributes to the formation of cleft palate [34]. On the basis of our studies and others [5, 6, 35], the stages from E14.5 to E16.5 have been identified as the most critical phases. During these stages, the palatal shelves adhere to each other with a seam and the seam begins to disintegrate in order to complete fusion, resulting in a confluent palate. Therefore, we anticipated that the majority of the genes contributing to physiological palatogenesis would be progressively regulated during these phases. In this study, we took advantage of high-throughput sequencing technology, RNA-Seq, to quantitatively analyze the expression of transcripts in mouse embryonic palatal tissues at three developmental stages.

The integrity and quality of the isolated RNA are the primary factors contributing to successful RNA-Seq studies [36], therefore we ensured that our palatal RNA samples were of high quality based on their RIN (>8.4). In our study, we used the “paired end” approach for RNA-Seq, wherein a single 100bp-long read is sequenced at both ends, which allows the tracking of alternate splice junctions, as well as further analysis of SNPs, insertions and deletions; thus providing more accurate and informative sequencing results from our samples [37, 38]. We submitted the raw reads of RNA-Seq data (fastq files) to SRA and the assembly data to the Transcriptome Shotgun Assembly Database.

Our global transcriptome analyses of TGFβ3 alleles showed that significant numbers of transcripts (~6000 genes) were differentially expressed between E14.5 and E15.5, particularly within the WT fetuses, indicating functional roles of these genes during this time period of palatogenesis (Table 1). Interestingly, we found lower numbers of transcriptome alterations between the same time points within the TGFβ3−/− mice (n = 2770), which can be explained by the consequential suppression of gene clusters due to inactivation of the TGFβ3 gene. However it was surprising to observe that the lowest number of transcriptome changes were in TGFβ3+/− mice, n = 1993, which are phenotypically identical to WT mice [25]. This may suggest that only this number of gene alterations is sufficient to enhance the palatal growth and fusion.

Furthermore, in normal mice (WT) the majority of transcripts were upregulated during the early transition of palatogenesis (E14.5 - E15.5) and downregulated during the later transitions (E15.5 - E16.5), suggesting that these are critical stages of palatogenesis requiring extensive gene regulation. This pattern of upregulated-downregulated expression suggests that particular CP genes may follow the same expression pattern to stimulate disintegration of the palatal seam, resulting in a confluent palate.

Understanding the expression of CP genes

The OMIM and MGI databases provide up-to-date and adequately classified information about genetic diseases in humans and mice, respectively; therefore, the identification of CP related genes from either species was uncomplicated. However the merger of the information from these databases required further computer analysis to identify common and individual CP genes between the two species (Additional file 7). To our knowledge, our list of CP genes (Table 2) covers the most comprehensive number of genes (n = 322), which cause CP when mutated in either species alone or both species [39, 40].

It was expected that TGFβ3 allelic mice follow 9 different expression pattern combinations during the E14.5-E15.5 and the E15.5-E16.5 transitions (Figure 3). Interestingly, none of the genes followed pattern 9, which represents the continuous downregulation of the transcript throughout all palatal stages. We observed that the majority of the CP genes were equally clustered into p2, p4, and p8 in WT samples (WT), representing normal palate development (Figure 4). Conversely, in both the TGFΒ3−/− and TGFΒ3+/− samples, almost all CP transcripts (91%) were accumulated in p1, and demonstrated high similarity between datasets (92%) (Additional file 9). This suggests that expression regulation of CP genes in both of these allelic mice was either delayed or abandoned.

Unique TGFΒ3−/− genes following the p1 pattern

Although the pattern of transcriptional expression between TGFΒ3−/− and TGFΒ3+/− are unexpectedly similar, phenotypic expression of CP occurs only in TGFΒ3−/− and not in TGFΒ3+/− pups [24, 25]. This suggests that the unique p1 genes of TGFΒ3−/− (n = 8; Chrng, Foxc2, H19, Kcnj13, Lhx8, Meox2, Shh, and Six3) may play crucial roles in the completion of palatogenesis. Therefore, we analyzed their specific functions within the cell, their roles in palatogenesis, and their interactions with the TGFβ signaling pathway using IPA and GeneCards [41].

Chrng (Cholinergic receptor, nicotinic, gamma) is a transmembrane glycoprotein and has been shown to play a role in neuromuscular organogenesis and ligand binding [42]. Mutation of Chrng causes the developmental disorder multiple pterygium syndrome in humans, exhibited by isolated CP, short stature, vertebral (spine) defects, joint contractures, and webbing of the neck, armpit, elbow, and knee [42, 43]. Pterygium is also associated with Irf6 and p63 mutations in humans [44], and both Irf6 and p63 are known to be closely associated with TGFβ signaling. Nevertheless, according to current literature, we could not find any relationships between Chrng and other molecules of the TGFΒ3−/− unique p1 list (Figure 6), or the TGFβ pathway.

Foxc2 (Forkhead box C2) is a member of the forkhead box (FOX) family of transcription factors and has been shown to be functional during development of mesenchymal tissues [45]. Mutations in Foxc2 are responsible for the hereditary lymphedema-distichiasis syndrome with CP observed in some patients [46]. In addition, Foxc2 is also involved in cancer metastases. In particular, expression of Foxc2 is induced when epithelial cells undergo epithelial-mesenchymal transition (EMT) [47]. Recently, Lindley at al. (2010) showed that Foxc2 may play a role in EMT maintenance when human mammary epithelial cells are treated with TGFβ [45]. (The Foxc2 knockout mouse is not available, according to KOMP). Overall, these relationships suggest that Foxc2 could be a vital element of palatal confluency through facilitation of EMT. Furthermore, failure of proper Foxc2 transcript regulation may result in CP in TGFβ3−/− fetus.

H19 is an imprinted maternally expressed transcript and expresses a long noncoding RNA. It has been shown that an enhancer deletion affects the expression of both H19 and Insulin-like growth factor 2 (Igf2), thus the Igf2 and the H19 genes are proposed to utilize a set of common enhancers [48, 49]. When Thomas et al. (1997) produced transgenic mice that express the Igf2 gene under the control of the H19 enhancers, a large fraction of homozygous mice developed CP [44]. In our network analysis (Figure 6), we found that H19 is indirectly related with TGFβ signaling through TNF, which suggests that it may play a crucial role during palatogenesis under regulation of TGFβ.

The Kcnj13 gene encodes a member of the inwardly rectifying potassium channel family of proteins and regulates ion transmembrane transport and mutations in Kcnj1 are associated with snowflake vitreoretinal degeneration [50]. There were no significant relationships found between Kcnj12 and other molecules of the TGFΒ3−/− unique p1 list (Figure 6), or the TGFβ pathway. However, this does not rule out its potential regulatory interaction with TGFβ signaling, and therefore demands additional studies to determine its role during palate development.

Lhx8 is a member of the LIM homeobox family of transcription factor proteins, which are involved in patterning and differentiation of various tissue types. Lhx8 is expressed in the mesenchyme of the mouse palatal structures throughout their development [11]. Zhao et al. (1999) showed that in Lhx8 homozygous mutant embryos, the bilateral primordial palatal shelves formed and elevated normally, but they often failed to make contact and to fuse properly, resulting in a cleft secondary palate [51]. Inoue M et al. (2006) proposed that the L3/Lhx8 gene contributes to epithelial mesenchymal interactions during facial morphogenesis and that Fgf-8b and Tgfβ3 were responsible for Lhx8 expression in the maxillary process [52]. Our RNA-Seq study of TGFβ3 −/− fetuses further confirmed the requirement of upregulated expression of Lhx8 throughout palatal morphogenesis.

The Meox2 gene (mesenchyme homeobox 2) encodes a member of a subfamily of non-clustered, diverged, antennapedia-like homeobox-containing genes [53, 54]. Jin et al. (2006) found that Meox2 is highly expressed in the posterior region of the developing palate, and the reduction of Meox2 gene levels increased the susceptibility of mice to cleft palate through a novel post-fusion mechanism [53]. The posterior cleft in E15.5 Meox2 −/− mice contained no epithelial cells, therefore they proposed that Meox2 may function to strengthen the fusing zone of palatal shelves. The lack of strong fusion of epithelial cells during stable expression of Meox2 may explain why the palatal shelves of TGFβ3−/− mice attach to each other, but fail to fuse [25]. Recently, Valcourt et al. (2007) showed that ectopic Meox2 suppressed epithelial cell proliferation in cooperation with Tgfβ1, and mediated induction of the cell cycle inhibitor gene p21 [54]. Furthermore, they showed that Meox2 failed to promote EMT and partially blocked Tgfβ1-induced EMT [54]. Considering the requirement of EMT during palatal fusion, Meox2 expression needs to be downregulated at the later stages of palatogenesis, as observed in both WT and TGFβ3+/− mice in our study.

Sonic hedgehog, Shh, is a member of the hedgehog family of secreted proteins and has been implicated as the key inductive morphogen in patterning of vertebrates during organogenesis; particularly the ventral neural tube, the anterior-posterior limb axis, and the ventral somites [55]. Shh signaling plays essential roles in craniofacial development by regulating a number of transcription factors and signaling interactions that take place between the epithelium and mesenchyme during normal palatogenesis [56]. Lan et al. (2009) showed that Shh is predominantly expressed in palatal epithelia and signals directly to the palatal mesenchyme to regulate palatal mesenchyme cell proliferation through maintenance of cyclin D1 and D2 (Ccnd1 and Ccnd2) expression [56]. Recently, Sasaki et al. (2007) showed that Shh expression in the palates of Tgfβ3-null mice was reduced throughout E12.5–E15.5, and thereby proposed that Shh may be involved in TGFβ3 regulation of normal palatal fusion [57]. Our results further support that upregulated expression of Shh is necessary for the successful completion of palatogenesis, as observed in the TGFβ3+/− and WT mice, but not in TGFβ3−/−.

SIX homeobox 3 (Six3) is a transcription factor crucial to embryonic development by providing the necessary instructions for the formation of the forebrain and eye development [58]. Similar to Shh [59], mutations in the homeodomain of the human Six3 gene cause holoprosencephaly and are associated with midline facial cleft - tessier cleft 14 [60]. In mice, it has been shown that Six3 protein increases expression of Ccnd1 protein [61], which is directly related to TGFβ signaling (Figure 6). Our results indicate that Six3 expression has to be upregulated after E14.5 for the precise progression of palate development. Therefore, the interaction of Six3 with TGFβ3 and their mechanism of activation deserve further analysis using biological techniques.

Cellular biofunctions and TGFβ pathway molecules

The expression pattern-clustered CP genes interconnect each other to facilitate cellular biofunctions and canonical pathways. The relevance of these groups of gene datasets was detected using IPA core analysis (Figure 5).

Our primary objective in this study was to analyze the transcriptomal differences among the alleles of TGFβ3 knockout mice. Thus, it was critical to find the relationship of CP genes with the TGFβ signaling pathway, which encompasses 89 molecules (Figure 7). TGFβ3 controls the fusion of palatal shelves by EMT and apoptosis [5, 14, 62]. Furthermore, TGFβ1 and TGFβ2 regulate mesenchymal cell proliferation and extracellular matrix synthesis in the palate [63]. Although not all of the molecules of the TGFβ signaling pathway are associated with CP genes, it is noteworthy that all of the related CP genes followed the Smad-dependent pathway. This may indicate that TGFβ3 regulates the CP genes through Smad-signaling, with ancillary involvement of ERK, PI3K, JNK, MAPK, etc. Thus, the promoter regions of these CP genes should be analyzed for Smad-binding elements using ChIP-PCR or ChIP-Seq. The majority of these CP genes followed the p8 expression pattern in WT mice, which represents the unchanged-downregulated expression pattern. This suggests that TGFβ molecules are required during E14.5 and E15.5, but are less necessary during E16.5, which correlates with our previous findings demonstrating the sequential function of TGFβ3 throughout the degradation of the palatal seam [5].

Overall, these sequencing and bioinformatics analysis studies are used as gene identification tools for future experiments to further understand the detailed mechanism of palate development. The 8 putative CP causing genes and Smad-dependent TGFβ signaling molecules will be analyzed in detail using gene and protein expression studies, as well transcription factor and promoter binding assays.

Mouse breeding and genotyping

The generation and genotyping of TGFβ3−/− mice have been described previously [25]. TGFβ3 heterozygous (+/−) C57BL/6J male and female mice were provided by Tom Doetschman (BIO5 Institute, University of Arizona, AZ). Mice were housed in accredited animal facilities at the University of Nebraska Medical Center (UNMC). All procedures were approved by the IACUC (Institutional Animal Care and Use Committee) of UNMC. Homozygous (−/−) embryos were obtained by mating TGFβ3^+/− male and female mice. The morning that the vaginal plug was pulled was considered embryonic day (E) 0.5. Tissue samples were individually labeled and embryonic tails were used for genotyping by PCR. Genomic DNA was prepared using the Wizard Genomic DNA Purification Kit (Promega – A1120). The primers used for TGFβ3 were as follows:

TGFβ3 Forward 5^′ TGGGAGTCATGGCTGTAACT 3^′

TGFβ3 Reverse 5^′ CACTCACACTGGCAAGTAGT 3^′

pMC1-Neo 5^′ GCCGAGAAAGTATCCATCAT 3^′

These primers amplified 600 bp and 400 bp fragments for the mutated and WT alleles, respectively. PCR conditions were 30 cycles of 95°C for 30 seconds, 57°C for 45 seconds, 72°C for 1 minute, followed by one cycle of 72°C for 10 minutes.

Typically, 8 fetuses were obtained from each pregnancy in a Mendelian-fashion 3 allelic-genotype (1:2:1) as 2 TGFβ3−/−, 4 TGFβ3+/−, and 2 WT. Palates were extracted from fetuses as described earlier [2] at 3 time points (E14.5, E15.5, and E16.5) as duplicates and the tissues were stored in RNAlater RNA Stabilization Reagent (QIAGEN, CA) to preserve transcript integrity. In order to evaluate the reproducibility of our experimental samples and procedures, two independent biological and technical replicates were assayed for each gestational age and each allele resulting in 36 (n = 3 × 3 × 2 × 2) samples for sequencing.

Protein, RNA extraction, construction of small RNA libraries, and RNA-Seq

Total proteins were extracted from the palatal shelves collected from embryonic (E) 15.5 and western blot was performed as previously done in our lab and described in [9, 10] using antibody against TGFβ3 (R & D Systems, MN). We extracted the RNA using the RNeasy Kit (QIAGEN, CA) and evaluated the RNA for purity and concentration by ultraviolet spectroscopy (NanoDrop, Wilmington, DE). RNA integrity numbers (RIN) [64, 65], based on an algorithm for judging the integrity of RNA samples, were evaluated using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Libraries were prepared using the Illumina mRNA-Seq Sample Preparation Kit (San Diego, CA) according to the manufacturer’s instructions. Briefly, poly(A) + RNA was recovered from 1 μg of total RNA using two rounds of isolation with oligo-dT-coated Sera-Mag magnetic beads. The recovered poly (A) + RNA was then chemically fragmented. RNA fragments were converted to cDNA using SuperScript II and random primers. The second strands were synthesized using RNaseH and DNA Pol I. The ends of the cDNA were repaired using T4 DNA polymerase, T4 polynucleotide kinase, and Klenow DNA polymerase. A single adenosine was added to the 3^′ end using Klenow fragment (3^′ to 5^′ exo minus). Adaptors were attached to both ends of the cDNA using T4 DNA ligase. RNA fragments were extracted from a 2% low range ultra agarose sizing gel. The fragments were amplified by 15 cycles of PCR using Phusion DNA polymerase. Libraries were validated with Agilent Bioanalyzer (Palo Alto, CA). Libraries were diluted to 10 pM and applied to an Illumina flow cell using the Illumina Cluster Station. Sequencing was performed on Illumina HiSeq 2000 according to the manufacturer’s instructions [66, 67].

Mapping and quantification of transcripts

RNA-Seq datasets were analyzed by following the recently published RNA-Seq protocol [31]. The 36 samples (paired-end reads, including biological replicates and technical replicates; n = 3 × 3 × 2 × 2) were individually mapped onto the mouse genome (mm9, build name NCBIM37) by using TopHat v1.4.1 [68–70]. Preprocessing steps were not applied to the raw data (fastq) files because the sequencing quality was very high in both 3^′ and 5^′ ends according to the FastQC reports [71]. The relative abundance of reads was normalized to the total read number and shown as fragments per kilobase per million mapped (FPKM) [71]. After the alignment, Cufflinks v1.3.0 [72–74] was used to estimate the abundances of transcripts. Cuffdiff from Cufflinks repository was used to perform pairwise comparisons between different genotypes and gestational groups. The dispersion between each biological replicate was also tested by Cuffdiff, as well as the dispersion between each technical replicate. CummeRbund (http://compbio.mit.edu/cummeRbund/) software [31] provides functions for creating commonly used expression plots, such as: volcano, scatter and box plots, using processed data from Cuffdiff. We used CummeRbund to generate scatter plots to use for quality assessments of both biological and technical replicates.

We used the VennDiagram [75], an R package that enables the automated generation of highly-customizable, high-resolution Venn diagrams, to illustrate the number of transcripts in each experimental group which passed the significance and fold change (FC ≥ 2.0) filtrations through Cufflinks. Pattern-oriented listings of CP genes were illustrated with GeneVenn, a web application for comparing gene lists (Additional file 5).

Identification of cleft palate genes

We used the OMIM (Online Mendelian Inheritance in Man) database to retrieve 280 genes and loci involved in genetic diseases associated with CP syndromes in humans. We used information from developmental biology studies demonstrating CP in knockout mouse models and the Mouse Genome Informatics (MGI) website with the mammalian phenotype browser, the international database resource for the laboratory mice, to retrieve 128 genes (out of 223 genotypes) causing CP in mice. The pattern-oriented heatmaps of CP genes were illustrated by using R pheatmap package

Biological interpretation and pathway analysis

The effects of TGFβ alleles on CP genes during palatogenesis were examined in the context of differentially expressed human and mouse genes using Ingenuity Pathway Analysis (IPA), (Ingenuity Systems, CA), a web-delivered application used to discover, visualize, and explore relevant networks. After performing statistical analysis and filtering of the RNA-Seq data, the Entrez IDs and FPKM values of each group were uploaded to IPA as a dataset. Each identifier was mapped to its corresponding gene object in the IKB. Interactions were then queried between these datasets and all other gene objects stored within IPA to generate a set of direct interaction networks that were overlapped. The significant genes were categorized, compared to functional categories in the IPA database, and ranked according to their p-values. P-values less than 0.05 indicate a statistically significant, nonrandom association between a set of significant genes and a set of all genes related to a given function in the IKB [76]. Through the assessment of differentially expressed genes, cellular processes that were most significantly altered in our dataset were: gene expression, cellular development, cellular growth and proliferation, cell signaling, cell morphology, cell death and survival, cellular movement, and cell cycle. Based on these IPA analyses, the cellular biofunction categories mentioned above and the TGFβ pathway were filtered among other cellular processes and canonical pathways; then ranked according to their p-values and represented as a color coded matrix. Datasets associated with the TGFβ pathway were overlaid using the IPA Pathway Designer.