Whole genome microarray analysis of chicken embryo facial prominences
Department of Oral Health Sciences, Life Sciences Institute, University of British Columbia, Vancouver, Canada
Institute of Animal Physiology and Genetics, v.v.i. Academy of Sciences of the Czech Republic, Brno, Czech Republic; Department of Anatomy, Histology and Embryology, University of Veterinary and Pharmaceutical Sciences
The three most frequent congenital defects are caused by abnormal embryonic development of the heart, neural tube, and the face. Of these, neural tube defects are in decline due to folic acid supplementation but the frequency of orofacial clefts has stubbornly remained the same for decades (Kohut and Rusen,2002). Most clefts are mostly nonsyndromic, in other words, they occur in isolation from other abnormalities (Jugessur and Murray,2005). The causes are thought to be multifactorial, including an interaction between genes and the environment (Jugessur and Murray,2005). Clefts begin in embryogenesis and are a result of upper facial prominences that are smaller, have delayed growth or a different morphology. Thus, to understand development of clefts and other more rare facial defects, it is necessary to have a more complete understanding of the genes expressed in the facial prominences, particularly at stages when patterning is still taking place.
The origins of mesenchymal tissues that comprise the face can be traced back to cranial neural crest cells. These multipotent cells arise from the dorsal edges of the neural tube as a result of interactions between the neural plate and surface ectoderm. In the head, they migrate laterally and ventrally as an unbroken sheet to fill the pharyngeal arches, cover the brain and surround the eyes. Neural crest cells arising from the diencephalon to the second rhombomere of the hindbrain are Hox-negative and fill all regions of embryonic face (Creuzet et al.,2005). However, the skeletal pattern derived from these cells and their mesenchymal derivatives is distinct and resident in the neural crest cells (Schneider and Helms,2003). The resultant upper and lower jaw skeletons and the timing of when these patterns arise occurs later, once the neural crest cells have begun to interact with facial ectoderm, endoderm, brain, and eye tissue.
In avian embryos, neural crest cells have ceased migrating by Hamburger and Hamilton stage 14 (Hamburger and Hamilton,1951), when pharyngeal arches begin to form. The next significant event in skeletal development is formation of olfactory placodes at stage 15, which specify the lateral nasal skeleton (Szabo-Rogers et al.,2009). The nasal placodes divide the frontonasal region into a medial frontonasal mass and two lateral nasal prominences. There is little cell movement once neural crest cells are in contact with facial ectoderm such that frontonasal mass cells do not move into the adjacent lateral nasal prominences and postoptic cells (presumptive maxillary cells between the eye and first arch) do not mix with the first arch cells (also known as the mandibular arch; Lee et al.,2001,2004). Budding of the maxillary prominences occurs at stage 18, and these are the last facial prominences to form. Great expansion of the prominences occurs between stage 20 and 28. In addition, the initially separate frontonasal mass and maxillary prominences contact and fuse to form the upper lip. The continuity of mesenchyme between these upper facial prominences is instrumental for the normal development of the upper jaw skeleton. In the absence of contact and fusion, a cleft lip will result.
By stage 18–20, each facial prominence is capable of forming a unique set of skeletal derivatives that can be recognized in isolation from the rest of the head. The frontonasal mass forms the premaxilla and prenasal cartilage as well as other midline skeletal elements (Richman and Tickle,1989). The maxillary prominences form only membranous bones, including the palatine, maxillary, jugal, and quadratojugal bones (Lee et al.,2004). The mandibular prominences form all of the lower jaw, including Meckel's cartilage (Richman and Tickle,1989) while the lateral nasal prominences form the nasal turbinates (MacDonald et al.,2004). Thus, each facial prominence is distinctly patterned and accordingly, is expected to acquire a distinct expression profile.
Expression profiles of facial prominences are predicted to not only include some genes that were on during neural crest cell migration and remain expressed in the mesenchyme, but also a set of genes that are expressed de novo, after neural crest cells have entered the facial prominences. These later expressed genes include those essential for patterning of the mesenchyme such as Distaless related genes Dlx5/6 (Distaless homeobox; Beverdam et al.,2002; Depew et al.,2002) and the upstream signal Edn1 (Endothelin1; Ozeki et al.,2004) or its receptor Endra1 (Ruest et al.,2004). These loss-of-function experiments highlight the fact that genes expressed in the neural crest-derived facial mesenchyme are necessary for large scale jaw patterning. In addition, chicken embryo manipulations also demonstrate that gene expression in postmigratory neural crest cells dramatically changes the identity of facial prominences. Altering the levels of two signaling molecules, retinoic acid and bone morphogenetic protein (BMP), was sufficient to convert the maxilla to a second frontonasal mass (Lee et al.,2001). Hence, by comparing the mesenchymes from the maxillary, mandibular, and frontonasal mass prominences, we might be able to identify additional genes involved in the specification of jaw identity.
Toward this end, several studies have performed expression analysis of facial regions using methods ranging from subtractive hybridization against a nonfacial tissue (liver; Fowles et al., 2003) to microarrays (Brown et al.,2003; Cai et al.,2005; Ivins et al., 2005; Handrigan et al.,2007) and serial analysis of gene expression (SAGE, Cai et al.,2005). Previously, two studies on avian frontonasal mass gene expression were carried out using custom cDNA arrays from older stage embryos when the skeletal condensations are present. These studies on Darwin's finches identified Bone Morphogenetic Protein 4 (BMP4; Abzhanov et al., 2004) and Calmodulin (CAM; Abzhanov et al., 2006) as being differentially expressed in finches with differing beak morphologies. However, despite these efforts, a genome wide approach to characterization of differential expression between normal facial prominences has not been previously attempted (Handrigan et al.,2007). Such studies are essential in improving our understanding of the complex process of normal and abnormal facial development. When normal development is disrupted by genetic or environmental insults in animal models or in humans, it is important to know which genes could be mediating these effects.
Our study uses a chip in which the whole genome is represented, therefore, differentially expressed genes, be they well characterized or less well annotated, can be detected. Notably, the present study is the first to directly compare different regions of the face and has not only identified general trends in the expression profiles but also identified specific sets of genes that are differentially expressed in specific prominences. The avian embryo was selected due to the fact that it is an amniote and, therefore, more related to humans and mice than fish or amphibians. Similar to fish, large numbers of embryos can be collected at precise stages. Finally, a detailed fate map of the facial prominences has been previously established (Richman and Tickle,1989; Lee et al.,2004; MacDonald et al.,2004). Here, we profile the embryonic face at stage 18 which is after neural crest cells have ceased migrating and before cell differentiation begins in the mesenchyme. Chip-wide analysis shows that out of the three regions of the face surveyed, the one with the most distinct profile is the frontonasal mass. Following validation of a large group of genes with QPCR at two different stages, we present sets of genes that can be used to identify the normal mandibular, maxillary, and frontonasal mass prominences.
RESULTS AND DISCUSSION
Analysis at the Chip and Gene Level Shows Segregation Consistent With Developmental and Evolutionary Origins of the Jaws
The first question we wished to address was whether chips hybridized to cDNA from the same part of the face would cluster together. This would indicate whether our dissections were sufficiently clean to distinguish one part of the face from another. In particular the maxillary region and mandibular prominence are very close to each other at the stage when tissues were dissected. We predicted that frontonasal mass and first arch RNAs would be completely separable, whereas maxillary expression might be a hybrid between the mandibular and maxillary regions. We first filtered out data points that were not flagged as present or marginal in at least two of the chips (PM2 = 18,442 genes of 28,416 predicted genes). Hierarchical clustering was then performed using all of the expression data sorted according to tissue type. No samples formed a distinct branch, separate from their counterparts. Instead clustering analysis confirmed that our first prediction was true. The data clearly demonstrated that the three frontonasal mass samples clustered together and were distinct from the maxillary and mandibular arch samples (Fig. 1A). However, the maxillary and mandibular arch chips were nested, suggesting that many more genes had similar expression profiles in these two regions of the face (Fig. 1A). Because we were careful to leave a margin of tissue between the maxillary and mandibular dissections, the data supports a developmental basis for the similarities.
The reasons for similarity in gene expression profiles between the maxillary and mandibular prominences likely stems from their common neural crest origin. The presumptive maxillary mesenchyme is derived from mesencephalic neural crest cells and these cells mix freely with those entering the first pharyngeal arch (Lumsden et al.,1991). There is no evidence of subpopulations of mesencephalic neural crest cells having unique fates (Noden,1978), except for those near the mid–hindbrain isthmus (Trainor et al.,2002) that contribute to the joint (quadrate).
Our lab and the work of others (Shigetani et al.,2000; Lee et al.,2004) has shown that once the pharyngeal arches form, there is little movement of cells between the first arch and presumptive maxillary prominence. Therefore, commonalities in gene expression reflect earlier developmental history (i.e., neural crest cell origins). Despite the fact that overall, gene expression is quite similar, later on the maxillary and mandibular skeletal patterns are very different, thus expression differences must arise later than stage 18.
The commonality of gene expression profiles in the maxillary and mandibular regions is also related to their phylogenetic origins. The array data fits with the longstanding idea that the upper jaw is as a segment of the first pharyngeal arch (Shigetani et al.,2005; Kimmel and Eberhart,2008). Both examination of the vertebrate fossil record and extant gnathostomes (including sharks) demonstrates that in nonamniotes there was originally a cartilage support for the upper jaw (Depew et al.,2005; Kuratani,2005; Depew and Simpson,2006; Kimmel and Eberhart,2008). This was gradually reduced during evolution such that in amniotes the entire upper jaw is supported by bones and the cartilage is reduced to form the joint. Because embryo development recapitulates phylogeny, our results provide the molecular support for the common evolutionary history of the upper and lower jaws. However, birds are not representative of all vertebrates. It will be interesting in the future to profile expression of other embryos including agnathans to see which gene differences and similarities have been conserved throughout evolution.
The chip-wide analysis showed that the stage18 frontonasal mass is the most distinct of the facial prominences (Fig. 2A,B). This can be traced back to the source of the neural crest cells. The frontonasal mesenchyme is derived from the most anterior population of neural crest cells that originate from the prosencephalon and anterior mesencephalon (Le Lievre,1978; Couly et al.,1993). The migration routes as determined in DiI (1,1′, di-octadecyl-3,3,3′,3′,-tetramethylindo-carbocyanine perchlorate) labeling studies show that cells entering the midline of upper face will pass by the eyes and forebrain (Lumsden et al.,1991). Although both the frontonasal neural crest cells as well as those destined for the mandibular arch are influenced by signals coming from the pharyngeal endoderm such as sonic hedgehog (Brito et al.,2006,2008; Benouaiche et al.,2008), ultimately with expansion of the brain the frontonasal mass is more dependent on signals from the forebrain than the endoderm (Hu and Marcucio,2009) and as such the same signals could also serve to distinguish the gene expression profile of the frontonasal mass. In addition, our results are perhaps biased as a result of the stage chosen for the expression profiling. Exchanges between epithelium and mesenchyme at slightly later stages (i.e., stage 20) have shown that frontonasal mass is fully patterned (Richman and Tickle,1989), and it is likely that by stage 18 many of the gene differences (both higher and lower abundance) are present.
Validation of Genes With QPCR and In Situ Hybridization
We had built in three biological replicates into the original experimental design such that we could analyze the variability of the array data. After analysis of variance (ANOVA) analysis, 3,094 genes were identified as significantly different between facial prominences at P value ≤0.05. Tukey's post hoc testing determined there were 1954 genes different between the frontonasal mass and mandibular arch, 1985 differentially expressed genes between the frontonasal mass and maxillary prominence and only 839 genes differentially expressed between the maxillary and mandibular prominences. A heat map of the 89 most highly expressed genes also shows the segregation of genes according to tissue type (Fig. 1B).
We next set out to validate and quantify the relative expression levels using QPCR. To generate smaller lists of genes from the microarray data that were manageable for QPCR, we filtered on fold-change. The total number of genes in the microarray that had greater than a 1.5-fold expression difference between the frontonasal mass and mandibular arch was 466. When the filter was set to three-fold or greater, the number was reduced to 154 genes. Similarly, the total number of genes that were different in the frontonasal mass compared with the maxillary prominence was 431 for 1.5- or 123 for 3-fold. Finally, the number of genes that were different in the mandibular arch and maxillary region was 196 for a 1.5-fold and 58 for a 3-fold difference. Partial lists of genes with their expression differences are provided in Tables 1A–C and full lists of genes with greater than a 1.5-fold difference are listed in Supp. Table S1 (which is available online).
Table 1A. Stage 18 Mandibular Versus Frontonasal Mass, Differentially Expressed Genes in Microarray Successfully Validated With QPCR
Fold change on microarray
QPCR, vs Stage 18 FNM
QPCR, vs Stage 16 FNM
In situ hybridization data
Key: NE, no expression data either at the appropriate stage or for that gene; QPCR, Quantitative PCR; FNM, frontonasal mass; MD, mandibular prominence; MX, maxillary prominence; m, mouse data; c, chicken data.
Validated with multiple primer pairs and fold changes are averaged over the different primers. Individual data for each primer and SD for fold changes are in Table S5. Curated data is restricted to mouse in situs between E9.0-E11.5 and chicken stages 13–20.
In lateral nasal prominence, possibly in medial nasal. No expression in mx or md(m) (Gritli-Linde et al.,2009)
To further increase the probability of being able to validate the results with other methods, we mainly continued analysis of a subset of genes that displayed approximately three-fold or greater difference in gene expression in at least one comparison. Although previously annotated genes were primarily studied, we also included several for which limited information was available, e.g., LOC417741, LOC768624, GGA.2354.1.S1_AT (Table 1A–C; Supp. Table S1). A different set of RNA samples were used for QPCR than those used on the chips (Supp. Table S2). cDNA was prepared from to embryonic stages, stage 16 to 18, in order to search for genes that could vary temporally during facial morphogenesis. A preliminary round of QPCR was performed to determine the best primers for each gene. We subsequently designed multiple primer sets for 28 genes (Supp. Table S3) based on the data from the pilot studies. Repeating the same trend with independent primer pairs provided additional confirmation that the results reflect the true expression patterns. A total of 120 primer pairs were used for these studies (Supp. Table S3). Of the 69 nonhousekeeping genes examined, data from 56 were successful and fit well with the observations on the microarray chips (Supp. Table S4). A total of 13 genes were unsuccessful in QPCR. Nine of the 13 lacked significant expression differences after analysis with ANOVA and ΔΔ Ct (Supp. Tables S4, S5). Some of these genes such as WNT2B are mainly epithelial (Geetha-Loganathan et al.,2009), and their level of expression is lower in relation to mesenchymally expressed genes. The combination of low signal with high variability results in no statistical difference in expression. Others such as PRRX1 have been studied before in the mouse (Beverdam and Meijlink,2001) and are likely highly expressed in all regions of the face at stage 18. Although FGFR3 was higher in the mandibular prominence than the maxillary prominence in the microarray analysis, no significant differences were observed using QPCR. Indeed detailed in situ hybridization studies of FGFR3 by our group (Wilke et al.,1997) were unable to detect expression in any part of the face at stage 15, and it was not until stage 20 when signal became readily apparent in the mandibular prominence. The reason there was no significant difference with QPCR was likely due to very low expression throughout the face at stages 16 and 18.
Three genes had opposite trends in expression to those observed on the microarray chips. We have worked with one of these genes previously, DKK1. The array data showed that the expression of DKK1 in the mandibular and maxillary prominences was higher than in the frontonasal mass, yet the QPCR showed the frontonasal mass was higher than the maxillary prominence. Our previous detailed in situ analysis with radioactive and nonradiolabeled probes for DKK1 agrees completely with the microarray data (Geetha-Loganathan et al.,2009). There is abundant expression in the maxillary and mandibular prominences and much lower expression in the frontonasal mass. Therefore, in cases of disagreement a third method of validation is perhaps warranted. Only a single gene did not give any data in the QPCR reaction (ChEST491f4; Supp. Table S4), and this was because the primers or probe did not hybridize. Overall, we were successful in 81% of the QPCR validations in terms of the close match to the trends observed in the microarray data (Table 1A–C).
A Gene Profile for the Frontonasal Mass
Comparison of the mandibular and maxillary prominences to the frontonasal mass. This revealed a large number of transcription factors characteristic of the frontonasal mass. These included ALX1, ASCL1, EYA2, FOXG1, SIX3, SOX10, SOX8, SP8, and TBX22 (Table 1A; Fig. 2). Although ZIC1 was detected on the microarray, this gene was not validated as being differentially expressed in frontonasal mass versus the mandibular prominence with QPCR. However, ZIC1 did become significant when compared with the maxillary prominence (Table 1C). SOX10 was significantly different in the QPCR analysis between the frontonasal mass and mandible but is eclipsed by the much larger relative expression in the maxillary prominences (Fig. 2; Table 1B). Several of the same transcription factors that were identified in the mandibular comparison to the frontonasal mass were also present in the maxillary comparison. These included ASCL1, SIX3, SOX8, SP8, and TBX22 (Table 1C; Fig. 2, and Supp. Fig. S2). Thus combinations of this group of transcription factors appear to identify the frontonasal mass. Other transcription factors are notably lower in the frontonasal mass than other parts of the face and it is possible that these genes must remain low in order for identity to be established. These include DLX1, LHX8, MAB21L1, MAB21L2, PITX2, EVI1, and TWIST2.
Several of the differentially expressed frontonasal mass genes identified herein have previously been studied with in situ hybridization, including TBX22 (Haenig et al.,2002; Handrigan et al.,2007), SOX8 (Bell et al.,2000), SOX10 (Barembaum and Bronner-Fraser,2007), FOXG1 (Geisha ID BF1Qin), ALX1 (Geisha ID ALX1.53Z.A1), and SP8 (Szabo-Rogers et al.,2009). Because facial expression was not always the focus of these studies, we examined several of these genes in more detail. Expression of TBX22 is very light at stage 16 and this increases at stage 18 (Higashihori et al., in press). Thus, the relatively higher level of TBX22 in the stage 18 frontonasal mass compared with the mandibular prominence (Table 1A; Fig. 2); this is consistent with our in situ data. Maxillary expression of TBX22 is not visible until stage 20, so QPCR data that shows much higher expression in the frontonasal mass at stage 18 and 16 is identical to the in situ patterns (Table 1C). At all stages, TBX22 expression is restricted to the mesenchyme. ALX1, although expressed nicely in the limbs, could not be detected in the face in our hands. Nonetheless, by stage 24 expression can be detected in frontonasal mass in the Geisha database. QPCR analysis showed that, compared with the stage 18 mandibular prominence, the same stage frontonasal mass has significantly higher ALX1 expression and this was confirmed with three primer sets (approximately six-fold higher; Fig. 2). ALX1 is also expressed in the maxillary prominence, however, and as such did not result in statistically significant differences between these two prominences in the microarray or QPCR analysis. SOX8 is also highly expressed in the frontonasal mass at stage 16 and 18 (Fig. 2). Others have shown strong expression in the frontonasal mass of the stage 24 embryos and have confirmed this signal was in the mesenchyme (Bell et al.,2000). Our microarray and QPCR data show that SOX8 is also expressed at younger stages, specifically in the frontonasal mass.
Genes with other functions were also confirmed with QPCR to be more abundant in the frontonasal mass than the mandibular prominence including POSTN, OGN, CYP26A1, NR2E1, and SCARA5 (Table 1A; Supp. Fig. S1). When compared with the maxillary prominence, we found that SCARA5 was again detected as being highly expressed in the frontonasal mass (14-fold greater, Table 1A; Supp. Fig. S1); therefore, it will be interesting in the future to pursue the expression of this gene in more detail. A gene that came up only in the maxillary comparisons is TMEM16A, a member of a very large gene family of transmembrane proteins (Rock and Harfe,2008). There was much higher expression in the frontonasal mass (10-fold). Although there is no chicken expression data, a recent mouse study showed high levels of expression in the upper face at E10.5 (Gritli-Linde et al.,2009). It is uncertain whether the medial nasal prominences (equivalent to the frontonasal mass) are expressing the gene, but it is clear that there was no expression in the maxillary or mandibular prominences. Thus, the chicken appears to mirror the expression patterns observed in the E10.5 mouse embryo (Table 1C). Although the germline knockout of Tmem16a had no craniofacial phenotype (Rock and Harfe,2008), this may be explained by the large number of paralogues, several of which have overlapping patterns in the face (Gritli-Linde et al.,2009).
Markers of the Nasal Pit Identified as Part of the Frontonasal Mass Genes
Because our dissections included the nasal pits, we isolated a group of transcription factors that mark this early olfactory epithelium. SP8, EYA2, and SIX3, were some of the most highly expressed genes in the frontonasal mass (Fig. 2, Supp. Fig. S2A,B) and all have in common high levels of expression in the nasal pit. Quantitative analysis of SP8 expression revealed a greater than 300-fold difference between the frontonasal mass at stage 16 and the mandibular prominence (Supp. Fig. S2A,B). The reasons for this stage specificity become apparent in the in situ hybridization samples. SP8 is expressed in the epithelium of the frontonasal mass and subsequently becomes restricted to the nasal pits (Supp. Fig. S2C–E′). This temporal and spatial pattern very much resembles that of fibroblast growth factor-8 (FGF8), a growth factor that is in the SP8 pathway (Kawakami et al.,2004). Mice with targeted deletion of Sp8 have a very severe forebrain phenotype due to failure of closure of the anterior neuropore (Bell et al.,2003). However, because it is very likely that the facial abnormalities are secondary to abnormal development of the telencephalon, a conditional targeting experiment is necessary to really determine the requirement for SP8 in facial development. The prediction is that because SP8 lies in the same pathway as FGF8 (Kawakami et al.,2004), similar truncations of the facial skeleton would result (Trumpp et al.,1999).
SIX3 similarly has very strong expression in the olfactory placodes in the chicken (Bovolenta et al.,1998). The QPCR data confirms that this expression continues until at least stage 18. Similar to SIX3, EYA2 is expressed in the nasal pit of stage 15 embryos and the maxillary region (postoptic mesenchyme) in stage 14 embryos (Ishihara et al.,2008). No data on older embryos are presented in this study; however, our QPCR results show that EYA2 is also expressed strongly at stage 18 (Fig. 2). Of interest, the stage 16 embryos also had higher levels of EYA2 in the maxillary prominence than the mandibular prominence (Fig. 2), mirroring the in published in situ results (Ishihara et al.,2008).
The expression of CASH1/ASCL1 is similar in some respects to SP8, EYA2, and SIX3. This gene was described in the brain previously by us and others (Franco del Amo et al.,1993; Jasoni et al.,1994; Casarosa et al.,1999; Hornbruch et al.,2005); however, the facial expression had not been noted. In comparison to both the mandibular and maxillary prominence, ASCL1 had significantly higher expression in the frontonasal mass (Table 1A,C). These results were validated in QPCR analysis where the stage 18 frontonasal mass had higher levels of transcript than the stage 18 mandibular prominence (Fig. 3A). Our in situ results show that, at stage 15, expression is very high in the optic recess (Fig. 3B,B′) but not in the adjacent frontonasal mesenchyme or epithelium. This may be why we could not detect a significantly different expression level between the stage 16 frontonasal mass and stage 16 mandibular or maxillary prominences with QPCR. At stage 18, there is strong expression in the invaginating nasal pits and in the ventral telencephalon (Fig. 3C,C′), fitting with the QPCR results. At stage 25, expression is limited to the cranial, olfactory epithelium of the nasal slit (Fig. 3D,D′). There is no mesenchymal expression for ASCL1 at any stage we have examined. Our expression data fit well with the olfactory expression described in the mouse and the phenotypes of the mouse knockout (Guillemot and Joyner,1993; Guillemot et al.,1993). Despite such a long and well-developed connection between Ascl1 and olfaction in the mouse, ours is the first study to show ASCL1 expression is conserved in the chicken olfactory system.
Genes Enriched in the Maxillary Prominence and Trigeminal Ganglion
The stage 16 embryo has a well-developed trigeminal ganglion very close the ectoderm of the lateral head. It happens that our dissection of the maxillary prominence included parts of this ganglion (Fig. S5). Several genes in our maxillary gene list were expressed in the ganglion including SOX10 (Table 1B; Fig. 2). The expression in the maxilla was 50- to 70-fold higher than the mandibular prominence. The expression data from in situ hybridization shows that SOX10 is initially expressed in migratory neural crest (Richman et al.,2002; McKeown et al.,2005) and then contributes to the crest-derived parts of the cranial sensory ganglia (Bondurand et al.,1998; Bennetts et al.,2007). Therefore, the microarray, QPCR and in situ data for SOX10 are consistent.
TAGLN3, also known as Neuronal protein 25 or 22 (NP25, NP22) was not previously shown to be expressed in the head. TAGLN3 was isolated originally from rat brain (Ren et al.,1994) and recent functional studies have shown that TAGLN3 encourages neurite outgrowth from chicken dorsal root ganglia (Pape et al.,2008). In the same study, the expression in the chicken spinal chord, dorsal root, and sympathetic ganglia is reported (Pape et al.,2008). We extend these studies to the craniofacial region. QPCR analysis of TAGLN3 found that the maxillary prominence at both stage 16 and 18 had more abundant expression than the mandibular prominence (Table 1B). There was also some expression in the frontonasal mass at stage 16 (Table 1A). Using a probe raised to a different region of the gene than the previous study by Pape et al. (2008), we mapped expression from stage 11 to stage 28. At stage 11, we found punctuate expression in the ventral mesencephalon (Fig. 4B,B′) and even numbered rhombomeres (Fig. 4B′). Expression was consistent with newly differentiating neurons as neurons in even-numbered rhombomeres differentiate slightly before those in odd-numbered rhombomeres (Eickholt et al.,2001). Their mediolateral position in r4 also suggests this (i.e., located close to pial as opposed to ventricular surface). At this stage, no expression was detected outside the central nervous system. Slightly later (i.e., beginning at stage 15), expression was also localized to several of cranial sensory ganglia including the trigeminal ganglion (V), facial ganglion (VII), acoustic ganglion (VIII), glosso-pharyngeal ganglion (IX), and vagal ganglion (X; Fig. 4C,C′, and data not shown). Interestingly, expression was mainly confined to the neural crest-derived proximal portion of each cranial ganglia rather than the distal placodal-derived region and as in the hindbrain is coincident with the onset of neuronal differentiation (D'Amico-Martel and Noden,1983). We also found strong expression beginning at stage 15 in the dorsal root and sympathetic ganglia (data not shown) also neural crest-derived. Expression persisted to stage 26, the latest stage examined and included staining in the dorsal half of the spinal cord along the entire body axis (Fig. 4D,D′). The mediolateral position of TAGLN3-expressing cells in the spinal cord suggests this gene is transiently expressed as cells exit the cell cycle, which is also the conclusion reached in the study on NP25 (Pape et al.,2008). Although at stages 15–20 there was very little signal for TAGLN3, in the face, by stage 22, it was possible to detect expression in the presumptive olfactory epithelium (Fig. 4E). At stage 26, additional domains of TAGLN3 expression can be seen in the medial edges of the maxillary prominences (Fig. 4F). As such, the in situ data agree with the microarray data in several respects. First, there is no detectable expression of TAGLN3 in the mandibular prominence at any stage examined. Second, there is higher expression in the maxillary prominence than the mandibular arch, likely because the maxillary area includes the trigeminal ganglion. Thus, in this case, detailed in situ hybridization analysis did fit with the QPCR data on microdissected regions of the face but also added structural detail of the ganglion expression that could not otherwise have been appreciated.
As we are interested in specification of the maxillary prominence, we focused on several transcription factors that had relatively higher expression in the presumptive maxillary region (Table 1B,C). MAB21L2 is one of two closely related genes whose cellular compartment is the nucleus but whose function is to act as a cofactor for other transcription factors such as SMADs (Baldessari et al.,2004). The expression published for the mouse is somewhat higher in the maxillary and mandibular prominences at E9.5 and E10 (Mariani et al.,1999; Wong et al.,1999). We found that, in the chicken embryo at the stage equivalent to that used in the microarray, there was approximately equal expression in the maxillary and mandibular prominences and no expression in the frontonasal mass (Fig. 5B,C). The QPCR and microarray data also show that the maxillary and mandibular prominence have much higher expression than the frontonasal mass (Table 1A,C; Fig. 5A). We extended the in situ analysis to older stages and found that MAB21L2 is down-regulated in the mandibular prominence and remains very strong in the maxillary prominence. The signal is restricted entirely to the mesenchyme (Fig. 5H). Thus, although not a specific marker of the maxillary prominence at early stages, this changes with ontogeny. Other areas of strong expression include the limb buds. Interestingly, there is a clearing of transcripts directly under the apical ectodermal ridge in the mesenchyme. The knockout mouse does appear to have malformations of the face from gross morphology but these have not been characterized any further (Yamada et al.,2003). It is possible that MAB21L1 is compensating for the loss of MAB21L2 and that more severe facial and limb phenotypes will be produced in double homozygous embryos.
MEIS2 is a transcription factor best known for its roles in proximodistal limb development (Capdevila et al.,1999; Mercader et al.,2000; Salsi et al.,2008). Facial expression data are not available for this gene. Our microarray data show four-fold greater expression in the maxillary region than in the frontonasal mass but no significant difference when compared with the mandibular prominence (Fig. S3A). The in situ data at stage 18 shown higher signal in the maxillary prominence than other regions (Fig. S3B). There was also light expression in the mesenchyme beneath the nasal placodes (Fig. S3B′). At stage 20, there is expression in the maxillary prominences (Fig. S3C) and epithelium of the frontonasal mass and mandibular prominences (Fig. S3C′). At stage 24, expression intensifies in the edge of the frontonasal mass, lateral nasal, and is especially prominent in the cranial maxillary prominence (Fig. S3D, D′). BMP4 is expressed in a strikingly similar pattern to MEIS2 in all regions of the face (Ashique et al.,2002), and the potential for gene interactions is, therefore, possible. There is relatively lower expression in the mandibular prominence at these later stages. Thus, the microarray data fit well with the in situ data over many developmental stages. The function of Meis2 in craniofacial development is not known, in part, because at the time of writing, there is no Meis2 knockout.
A third transcription factor that is highly expressed in the maxillary prominence is ID1 (inhibitor of DNA binding 1, dominant-negative helix–loop–helix protein). ID1 is a direct target of BMP signaling (Miyazono and Miyazawa,2002), and BMP levels are important for specifying the skeletal pattern of the face. There is a close correspondence between the expression of ID1 in the anterior maxillary prominence, medial mandibular prominence, and caudal edge of the frontonasal mass with expression of BMP4. The QPCR analysis with two different primers showed that the stage 18 mandibular prominence has higher expression than the stage 18 frontonasal mass (four-fold; Table 1A). Statistically significant differences between the maxillary prominence and the other regions of the face were not detected, however (Supp. Fig. S4A). This was explained when we analyzed the in situ hybridization signal. We found that, as reported by others (Kee and Bronner-Fraser,2001a,b), ID1 was strongly expressed in both the maxillary and mandibular prominences at stage 17 and 20 and at low levels in the caudal edge of the frontonasal mass (Fig. S4B–C′). It will be interesting to test whether ID1 mediates the effects of BMPs in the face. Thus far, knockouts of Id1 have not resulted in early embryo phenotypes, but because there are four family members with overlapping expression patterns (Gray et al.,2004), functional redundancy is possible.
Genes That Characterize the Mandibular Prominences
The genes with higher expression in the mandibular prominences when compared with the frontonasal mass included many that are known to be characteristically mandibular in the chicken and mouse. These genes included DLX1, HAND2, LHX8, MSX2, PITX2, and TWIST2 (Table 1A). Three of the genes found to be differentially expressed in the mandibular and maxillary prominences were BETA3, HAND2, and MSX2. Of these, the gene than stands out as being the most highly expressed in the mandibular prominence is HAND2, which is several hundred-fold more abundant than in the maxillary prominence or frontonasal mass. Genetic targeting causes early death of the embryo before formation of the head skeleton. However, other targeting experiments in which the first branchial arch enhancer of Hand2 was selectively deleted showed major reductions in the mandible (Yanagisawa et al.,2003). Hand2 is in the pathway that controls mandibular identity (Depew et al.,2002; Ruest et al.,2004), but no changes in identity were observed in the enhancer knockout. Studies that investigate genetic interactions between Hand2 and Dlx5/6 or Ednra (Endothelin receptor A) may yet reveal a contributing role for Hand2 in specifying identity. Our expression data from the chicken are consistent with a major role in controlling mandibular patterning.
Genes with extracellular functions that appear to be promising as mandibular patterning genes include CHODL. This is an integral protein in the cell membrane that has a long extracellular domain and, therefore, could be involved in cell–cell interactions, cell movements, or cell aggregation. Very little expression data exist for this gene (Weng et al.,2003), and certainly this is the first report of expression at an early stage of facial development.
Stage-Specific Differences in Expression Detected With QPCR
Important differences in the plasticity of facial prominences to respond to chemical manipulations exist between stage 16 and stage 18 (Lee et al.,2001; and Butchová, unpublished data). The null hypothesis is that most gene expression differences would follow trends similar to those found in stage 18 faces. We found that the majority of genes (56) did not change between stage 18 and stage 16 within a single facial prominence, supporting the null hypothesis (Table 2). However, a group of 13 genes did show expression differences in the mandibular or frontonasal mass. Interestingly, the majority of these showed down-regulation as the embryo matured. For example, MEIS2 and ID1 are four- and six-fold more abundant, respectively, at stage 16 than at stage 18. This trend is visible in the whole-mount in situs we have performed for MEIS2 (Fig. S3). There is relatively more expression in the epithelium and mesenchyme at stage 15 (Supp. Fig. S3B,B′) and mainly epithelial expression at stage 20 (Supp. Fig. S3C,C′). For ID1, the differences were not visible in whole-mount in situ hybridization (Supp. Fig. S4). The diffuse signal for this gene makes such a qualitative assessment difficult.
Table 2. Differences Between Stage 16 and 18 Facial Prominences Detected With QPCR
Of interest, there were no instances of differences in maxillary prominence gene expression between stage 16 and 18, despite this being the part of the face with the greatest responsiveness to identity change (Table 2). It is likely that expanding the microarray analysis to other stages would reveal more stage specific changes in gene expression. Such an analysis has been carried out for the first arch from 4, 5, and 6 week human embryos (Cai et al.,2005). Here, they used two methods of gene profiling, an earlier version of the human genome chip (HG-U95Av2) and SAGE (Cai et al.,2005). We estimate that the fourth week human gestation is approximately equivalent to stage 16 and the 5th week to stage 18 used in the present study. An overview of the differential expression in the first pharyngeal arch was presented, but the caveat was that it was necessary to use mouse tissues to validate a small subset of 14 genes. One gene that we also encountered was PITX2. In the mouse QPCR experiment, Pitx2 was 1.9-fold more abundant in the first pharyngeal arch of E10.5 embryos compared with E9.5 embryos. We did not find significant differences between stage 16 and 18 mandibular prominences, although as we have discussed there were differences in the frontonasal mass. The authors did the bulk of their comparisons between the SAGE and microarray data. Although 97 genes were considered as stage specific with SAGE, only 2 of these were also identified with the microarray analysis. Thus, these two methods are giving different sets of genes that then need yet further validation. The one gene that was present in the first pharyngeal arch and confirmed to be up-regulated at 5 weeks by both SAGE and microarray was c-fos induced growth factor or vascular endothelial growth factor D, (FIGF/VEGFD). We also detected the same gene as differentially expressed in the chicken embryo among the facial prominences; however, the fold-change was less than 1.5, so it was not studied further here. This gene has subsequently been knocked out in two different labs, and there are no craniofacial defects reported (Baldwin et al.,2005; Lijnen et al.,2009). The original study by Cai et al. (2005) did not compare expression between the facial prominences, although the RNA was collected. Thus, perhaps in the future, a more in depth comparison between the gene profiles of the human and chicken facial prominences will be possible.
We used the chicken genome microarray as a way to obtain an unbiased and complete picture of gene expression in specific regions of the developing chicken face. Although we only surveyed one stage of development, we subsequently extended the information about expression patterns using QPCR and in situ hybridization. We found that, by selecting an early stage with respect to facial development, we have identified genes with restricted expression patterns across many stages of development.
One of the main aims of our study was to identify genes that could be used as markers of the different facial prominences. There are an increasing number of studies where genetic or molecular manipulations have resulted in dramatic homeotic transformations of the jaws (Miller et al.,2000; Lee et al.,2001; Kimmel et al.,2003,2007; Ruest et al.,2004; Sato et al.,2008). One difficulty in determining whether the molecular landscape of the transformed region is converted to the new identity is that it is difficult to find genes that are restricted to different regions of the face. We have succeeded in describing sets of genes that together can be used to identify regions of the face. Moreover, the gene lists will be a valuable resource for future studies, particularly for those where a mutation in a gene not previously connected to craniofacial development, causes a facial phenotype either in humans or in animal models.
Fertilized White Leghorn chicken eggs obtained from the University of Alberta were incubated at 38°C. Embryos were collected at stage 16 or at stage 18 (Hamburger and Hamilton,1951). These time points were selected as, at stage 16, neural crest cells have finished migration but still may carry with them some of the gene expression patterns present in migratory neural crest. Stage 18 is just when the maxillary prominences can be discerned and identity is fixed as determined by grafting experiments. Tissue from three regions of the face, the maxillary prominence, frontonasal mass, and mandibular arch, was isolated by microdissection (Fig. S5). Dissected tissues from approximately 20 embryos were pooled to form a biological replicate (Table S2). Each sample was processed independently using the RNAeasy mini kit (Qiagen). There were a total of six replicates for each region of the face at stage 18. Half of the samples were used to generate the RNA for microarray chips (3 facial areas × 3 biological replicates = 9 microarray chips). The remaining nine replicates were used for QPCR validation. At stage 16, there were two replicates of the frontonasal mass and three each of the maxilla and mandible for a total of eight samples and all of these were used for QPCR analysis.
Microarray Chip Processing and Analysis
RNA quality was examined on a Bioanalyzer 2100 (Agilent). Total RNA was sent to London Regional Genomic Centre (University of Western Ontario), where hybridization and scanning of array chips were performed using the manufacturer's recommended protocol. Chicken Affymetrix microarray chips containing probe sets interrogating approximately 28,000 full-length genes and expressed sequence tag clusters from the UniGene database were used (Chicken GeneChip). Microarray datasets have been deposited in the Gene Expression Omnibus, record number GSE18477. Data were analyzed using Genespring GX version 7.3 (Agilent Technologies). Normalization was performed as follows: Values below 0.01 were set to 0.01. Each measurement was divided by the 50th percentile of all measurements for that sample. Each gene was divided by the median of its measurements for all samples. If the median of the raw values was below 10 then each measurement for that gene was divided by 10 if the numerator was above 10; otherwise the measurement was discarded. To create gene lists, three filters were used in the following order: (1) present or marginal expression level in at least two samples (PM2, present or marginal in 2 samples); (2) one-way ANOVA (P = 0.05); and (3) fold-change of expression values. Lists of genes were compared between samples, and mainly those with greater than three-fold differences were selected for further validation. Other chip-wide analyses included hierarchical clustering performed on PM2 data. The gene descriptions in Genespring were manually updated by the authors by first importing the more complete Affymetrix annotations into the Genespring annotation file and then uploading the merged file. Individual unknown genes were then blasted to identify any that had homology to known genes. These annotations were added to the individual genes in the Genespring annotation file.
Validation With High Throughput Quantitative PCR
To increase the number of genes that could be validated with Q-PCR, we used a microfluidics-based array (BioMark 48 × 48 dynamic array, Fluidigm), which greatly decreased the sample volume required and reduced the number of pipetting steps. Furthermore, the use of a universal probe library (Roche Applied Sciences) in combination with custom primers afforded us greater economy when studying the chicken genome. Probes and primer sequences were selected using Roche's Probe Finder assay design software.
Three housekeeping genes were tested, β-Actin (ACTB1), eukaryotic translation elongation factor 1 alpha 1 (EEF1A1), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Of the three housekeeping genes on the array, we found that the gene with the least variability was ACTB1. The threshold number of cycles required to detect ACTB1 was 19.84 ± 1.16 averaged across all the samples. We had previously used ACTB1 as a reference gene for QPCR and found that it was consistent across experimental conditions and replicates (Higashihori et al.,2008). The cDNA samples were preamplified using Applied Biosystem's TaqMan PreAmp Master Mix. Preamplified cDNA was used for Q-PCR in the BioMark™ Real-Time PCR System. The cycling program used consisted of 95°C for 10 min, 40 cycles for 95°C for 15 sec, 70°C for 5 sec, and 60°C for 1 min. Data were analyzed using the BioMark Gene Expression Data Analysis software to obtain Ct values. Three technical replicates were performed for each sample (see Supp. Methods for more details).
Analysis of QPCR data was carried out by first averaging the three technical replicates for each of the 17 samples and then normalizing each mean value for the gene of interest to the mean ACTB1 value for the same sample (Gene of Interest/ACTB1). One-way ANOVA was performed on the ratios followed by Tukey's post-hoc testing (Statistica) to determine which samples were statistically different from each other. A P value of < 0.05 was considered significant. To determine relative fold-changes we calculated the ΔΔCt relative to the lowest expressing sample. The ABI (Applied Biosystems) Bulletin #2 was used as a guideline for making the ΔΔCt calculations. The formula 2ˆ-(Mean Gene of Interest – Mean ACTB1) was used to determine relative abundance of a gene in relation to other samples. The fold-changes and standard deviations for each possible comparison are given in Supp. Table S5. All but one primer set gave useable data (Supp. Table S4).
Validation With Whole-Mount In Situ Hybridizations
In situ hybridization was performed for validation of selected genes from the microarray datasets. Whole-mount in situ hybridization was performed on chicken embryos as previously described (Song et al.,2004) using an Intavis in situ robot (Hölle and Hüttner AG). The 920-bp fragment of SP8 (bp 1776 to 2696) and the 780-bp fragment of TAGLN3 (bp 154 to 980). The following individuals generously supplied these cDNAs: MASH1 (CASH1), T. Reh; ID1, M. Bronner-Fraser; MAB21L2, P. Antin, and MEIS2, D. Schulte.
The authors thank M. Underhill for getting us started on this microarray study and for invaluable help with Genespring GX analysis. We also thank C. Wicking for her insightful comments on the manuscript and David Carter at the London Regional Genomics Center for performing the microarray hybridizations. J.M.R. is funded by the Canadian Institutes of Health Research. M.B. is funded by the Grant Agency of Czech Republic. W.P.K. is supported by the Harvard Catalyst, Clinical and Translational Science Award. C.L. is supported by grants from the Natural Sciences and Engineering Research Council of Canada. S.N. and P.G.-L. are supported by MSFHR postdoctoral fellowships.