Comprehensive expression atlas of fibroblast growth factors and their receptors generated by a novel robotic in situ hybridization platform
Abstract
A recently developed robotic platform termed “Genepaint” can carry out large-scale nonradioactive in situ hybridization (ISH) on tissue sections. We report a series of experiments that validate this novel platform. Signal-to-noise ratio and mRNA detection limits were comparable to traditional ISH procedures, and hybridization was transcript-specific, even in cases in which probes could have hybridized to several transcripts of a multigene family. We established an atlas of expression patterns of fibroblast growth factors (Fgfs) and their receptors (Fgfrs) for the embryonic day 14.5 mouse embryo. This atlas provides a comprehensive overview of previously known as well as novel sites of expression for this important family of signaling molecules. The Fgf/Fgfr atlas was integrated into the transcriptome database (www.genepaint.org), where individual Fgf and Fgfr expression patterns can be interactively viewed at cellular resolution and where sites of expressions can be retrieved using an anatomy-based search. Developmental Dynamics 234:371–386, 2005. © 2005 Wiley-Liss, Inc.
INTRODUCTION
In situ hybridization (ISH) uses labeled RNA or DNA probes to visualize mRNAs in tissues and, thereby, reveals when and where genes are expressed. In particular, in its nonradioactive form, ISH provides gene expression information with single-cell resolution. ISH data are useful in many ways, including identification of cells in which the gene in question may be functioning; and when applied to genetically or otherwise modified organisms, ISH helps to elucidate signaling pathways. In the recent past, systematic efforts to determine gene expression by ISH have been undertaken (http://mamep.molgen.mpg.de, Neidhardt et al., 2000; http://genepaint.org/, Visel et al., 2004; http://brainatlas.org, Boguski and Jones, 2004; http://mahoney.chip.org/mahoney/, Gray et al., 2004). An altogether technically different approach is pursued in the GENSAT project that visualizes gene expression in the central nervous system of the mouse using bacterial artificial chromosome transgenic vectors in which endogenous protein coding sequences have been replaced with a green fluorescent protein reporter (http://www.ncbi.nlm.nih.gov/projects/gensat/, Gong et al., 2003).
In parallel to these experimental expression-mapping projects, Web databases are emerging to access gene expression data published throughout the biomedical literature. For example, Ringwald and colleagues have developed a sector of the Mouse Genome Database that provides search tools to access reproductions of an ever-growing number of published gene expression patterns determined in the mouse embryo (Bult et al., 2004; Hill et al., 2004). The Edinburgh Mouse Atlas Gene Expression Database (EMAGE) offers to the scientific community an opportunity to submit expression data and to map such data into searchable volumetric atlases of mouse embryos (http://genex.hgu.mrc.ac.uk/Emage/; Baldock et al., 2003).
Our laboratory establishes a transcriptome-wide gene expression atlas based on tissue sections of the embryonic day (E) 14.5 mouse embryo (Visel et al., 2004). A recently initiated multi-laboratory project supported by the European Union will extend this effort to ∼20,000 genes within 4 years, and these data will be accessible through genepaint.org and EMAGE. The “Genepaint” technology platform used in this project consists of semiautomated equipment to produce and digitally image ISH data and an Open Access database (www.genepaint.org) housing these data in a searchable manner. In this project, expression is determined with digoxigenin-tagged riboprobes. Although radioactive riboprobes are considered to be the most sensitive way to detect mRNAs, a nonradioactive technique based on dual signal amplification approaches the sensitivity of radioisotopic ISH (reviewed in Speel et al., 1998; Speel, 1999). In the dual amplification strategy, termed “catalyzed reporter deposition” (CARD), hapten-tagged RNA (or DNA) probe is hybridized and subsequently detected in situ with an antihapten antibody conjugated to horseradish peroxidase. The catalytic activity of the peroxidase moiety activates a tyramine–biotin complex so that for each bound antihapten antibody multiple biotinylated tyramine molecules covalently attach to reactive groups in the vicinity of the probe. Thereafter, biotin is detected with an avidin–alkaline phosphatase conjugate, whose enzymatic activity converts nitro blue tetrazolium (NBT) and 5-bromo-4-chloro-3-indolyl-phosphate (BCIP) substrates into insoluble blue–purple indigo and formazan dye precipitates.
Although radioactive ISH can be scaled up to some extent (Lein et al., 2004), automation is easier with nonradioactive ISH as shown for whole-mounts (Plickert et al., 1997; Lein et al., 2004) and with Genepaint for sections (Reymond et al., 2002). In the latter case, sections prepared from frozen specimens are incorporated into temperature-controlled flow-through chambers placed on the platform of a liquid handling robot, with which digoxigenin-tagged riboprobes, solutions for pre- and posthybridization, and CARD are added to the flow-through chambers in a programmable manner (Fig. 1A–C).

Key hardware components of the Genepaint system. A: Custom-made specimen freezing chamber with transparent Plexiglas side walls that are placed into grooves milled into a copper plate (cp). To fasten sidewalls on the top, they each have two stainless steel pegs that fit into defined holes drilled into a top plate (t), which is fastened by two clips (cl) to vertical stainless-steel rods (r). B: Slide in custom-made holder with predefined position for tissue sections (A–D) that were stained blue for better visualization. C: Diagram of flow-through hybridization chamber. Solutions are delivered into the well on the top of the chamber. D: Layout of the working table of a Tecan solvent delivery system. (1) chamber racks housing hybridization chambers, (2) wash station for pipettes, (3) containers for various buffers and other solutions kept at room temperature, (4) receptacle for tubes with riboprobes, (5) temperature-controlled reservoirs for stringency wash solutions. E: Set viewer page for Fgfr1 (see Visel et al., 2004, and www.genepaint.org [“Manual of GenePaint” and “Annotation of Expression” links] for a detailed description of the set viewer window and icons used therein.
The mouse genome contains 22 Fgf genes (Fgf1–Fgf18 and Fgf20–Fgf23; Fgf, fibroblast growth factor) classifiable into seven subfamilies (Ornitz and Itoh, 2001; Itoh and Ornitz, 2004, and references therein). In addition, four Fgf receptor genes are found in vertebrates and all encode membrane-spanning tyrosine kinases that upon ligand binding form dimers and become tyrosine-phosphorylated, triggering signal transduction. Fgfrs are characterized by alternative splicing in their immunoglobulin III domain and receptors encoded by different splice variants have dissimilar ligand specificity (Reuss and Halbach, 2003). A vast body of literature shows that Fgfs/Fgfrs constitute an extremely important family of signaling molecules, and mapping their expression pattern in a standardized manner at specific stages, thus, is informative. Most Fgf genes have been knocked out in mice and phenotypes described (Dono et al., 1998; Ornitz and Itoh, 2001; Reuss and Halbach, 2003), but some phenotypes may not become apparent as a result of redundancy of Fgfs. The decision on which specific mutant mice need to be crossed and analyzed would be helped by knowing which Fgfs are coexpressed in a particular tissue.
Here, we have evaluated sensitivity, signal-to-background ratio, and specificity of the Genepaint procedure. In addition, we have used this method to determine the expression patterns of all Fgfs and Ffg receptors on serial sections through the mid-gestation E14.5 mouse embryo and have compared the resulting expression patterns with published data. We also describe how the Genepaint database allows access to and searches of the Fgf/Fgfr expression patterns using any computer connected to the Internet and equipped with a Web browser.
RESULTS AND DISCUSSION
Sensitivity and Signal-to-Background Ratio
To estimate the number of transcripts per cell that the Genepaint procedure can reliably detect, the number of transcripts of the Down syndrome critical region gene 3 (Dscr3) gene per cell was determined. In the postnatal day 7 (P7) mouse cortex, Dscr3 expression apparent as blue–purple dye precipitates (arrows) is weak and relatively uniform (Fig. 2A). From a freshly collected P7 brain, a 193-mg piece of mostly cortical tissue was dissected, RNA was extracted, and the total number of Dscr3 transcripts was determined by quantitative polymerase chain reaction (PCR; see Experimental Procedures section) using in vitro synthesized Dscr3 RNA as a reference standard. A total of 4.2 × 108 copies of Dscr3 transcripts were isolated from 1.23 × 108 cells averaging ∼3 transcripts per cell. This figure assumes that every cell in the cortex expresses Dscr3, but cortical sections also contained cellular regions devoid of Dscr3 signal (outlined with dashed lines in Fig. 2A). Assuming that the number of expressing and nonexpressing cells is approximately equal (Fig. 2A), sensitivity of the method would be in the range of ∼6 mRNA copies per cell and, hence, is suitable to detect rare transcripts (Braissant and Wahli, 1998). It should be noted that omission of the CARD step results in very weak signal even in the case of a moderately expressed gene (data not shown).

Sensitivity and signal-to-background ratio of Genepaint procedures. A: Dscr3 is broadly expressed in postnatal day (P) 7 cerebral cortex. Red arrows point at medium-sized dye precipitates (medium expression) and green arrows indicate small precipitates reflecting weak expression (for a discussion of the relationship between strength of expression and size of dye precipitates, see Carson et al., 2005). Dashes circumscribe regions that do not express Dscr3. B,C: Ear2 riboprobe hybridized to embryonic day (E) 14.5 kidney of wild-type (B) and Ear2−/− mutant (C); any stain seen in C represents background. D,E: Trhde riboprobe hybridized to wild-type (D) and Trhd−/− mutant (E) adult hippocampus. F,G: Trhde riboprobe hybridized to wild-type (F) and Trhd−/− mutant (G) adult neocortex. Note the negligible staining in mutant tissues (E,G). CA, cornu amonis; DG, dendate gyrus; I, II, III, cortical layers I, II, and III.
Weak gene expression appears as small precipitates (green arrows in Fig. 2A) overlaying diffuse staining of tissue such as seen in Figure 2C,E,G. We interpret such a diffuse staining as nonspecific background staining and not as gene transcription. To gain support for this conclusion, we have analyzed the expression of the nuclear orphan receptor gene Ear2 and of thyrotropin-releasing hormone degrading ectoenzyme (Trhde) in wild-type and mutant tissue. Northern blots showed that Ear2 (Warnecke et al., 2005) and Trhde mutants (K. Bauer and L. Geffers, personal communication) are devoid of the corresponding transcripts. Thus, staining detected in null mutant tissue would represent background. Ear2 transcripts are detected in developing kidney (Fig. 2B), but in mutant kidney, virtually no signal is seen (Fig. 2C). In the case of Trhde, strong signal is seen in all cells of the cornu ammonis (Fig. 2D) with the nearby cell-dense dentate gyrus exhibiting only a weak, diffuse staining similar to that seen in Trhde−/− tissue (Fig. 2E). Individual neurons in the retrosplenial granular cortex of wild-type mice strongly express Trhde (Fig. 2F), but mutant cortex exhibits only a slight staining in this layer (Fig. 2G). Thus, Genepaint procedures are obviously capable to effectively discriminate between specific signal and background staining and are very sensitive.
Probe Specificity Issues
Ideally, probes are designed in such a way that they share little or no sequence homology with other genes. For various reasons, including availability of characterized clones, incomplete knowledge about the transcriptome, the abundance of gene families and the requirement for reasonably long probes (600–1,000 nucleotides), such ideal probes cannot always be made. This finding raises the question of how much sequence homology is tolerable for Genepaint procedures. Fgfrs derive from an ancestral gene (Itoh and Ornitz, 2004) and, therefore, show considerable sequence homology (Table 1), which offers the opportunity to test the issue of cross-hybridization of probes. Homology of the four Fgfr probes used in the present work to the four Fgfr transcripts is shown in Table 1. For example, the sequence of the Fgfr1 probe shares homology regions with all three other receptors (Table 1). To assess specificity of our Fgfr probes, we identified structures in the embryo that express only one type of receptor. A subset of neurons in the dorsal root ganglia expresses only Fgfr1 (Fig. 3A–D). Fgfr2 is the only receptor gene expressed in thymus (Fig. 3E–H), Fgfr3 the sole receptor gene expressed in proliferating chondrocytes in the core region of ribs (Fig. 3I–L), and Fgfr4 the only receptor gene expressed in intercostal muscles attached to the sternum (Fig. 3M–P). If for example Fgfr1 probe were to cross-hybridize with Fgfr2 mRNA, then one would see a signal in Figure 3E, which is not the case. Note that this is so despite that the sequence homology between Fgfr1 probe and Fgfr2 mRNA is in the range of 75% in a region of several hundred nucleotides (top data row in Table 1). Fgfr4 is the sole Fgfr expressed in intercostal muscle fibers (Fig. 3M–P). Again, Fgfr1, -2, and -3 probes have significant sequence homology with the Fgfr4 mRNA (last column in Table 1), yet no hybridization of any of these three probes to intercostals muscle is detected (Fig. 3M–O). Analogous arguments can be developed for the remaining examples shown in Figure 3, leading to the conclusion that even an ∼75% sequence homology over as many as 500 nucleotides still produces a specific signal. With the level of specificity delivered by the Genepaint procedures, it should be possible, therefore, to design specific probes on a transcriptome-wide scale and trust the resulting ISH expression patterns even in the case of gene families. The obvious caveat is that, in case of large disparities in expression levels of related genes, a probe designed for a rare transcript might hybridize to cells expressing the more abundant transcript.
Receptor type (GenBank accession no. of reference sequence) | |||||||
---|---|---|---|---|---|---|---|
Fgfr1 IIIb (AF176552) | Fgfr1 IIIc (U22324) | Fgfr2 IIIb (NM_201601) | Fgfr2 IIIc (NM_010207) | Fgfr3 IIIb (none)a | Fgfr3 IIIc (M81342) | Fgfr4 (NM_008011) | |
Fgfr riboprobe | |||||||
Fgfr1 | 495/501 (98%)b | 783/783 (100%) | 328/444 (73%) | 390/510 (76%) | 170/222 (76%) | 294/390 (75%) | 108/138 (78%) |
156/156 (100%) | 44/54 (81%) | ||||||
Fgfr2 | 255/338 (75%) | 390/510 (76%) | 455/464 (98%) | 715/715 (100%) | 218/297 (73%) | 356/486 (73%) | 207/282 (73%) |
107/107 (100%) | |||||||
Fgfr3 | no homology | no homology | 70/91 (76%) | 70/91 (76%) | 807/812 (99%) | 807/812 (99%) | 149/199 (74%) |
Fgfr4 | 109/138 (78%) | 108/138 (78%) | 208/282 (73%) | 207/282 (73%) | 315/415 (75%) | 315/415 (75%) | 762/762 (100%) |
44/54 (81%) | 44/54 (81%) | 66/83 (79%) |
- a No full-length sequence of mouse Fgfr3-IIIb is available in GenBank. BLAST search was performed using the predicted mRNA sequence, in which nucleotides 1139–1280 (IIIc domain) of M81342 were replaced by nucleotides 781–928 (IIIb domain) of the truncated Fgfr3-IIIb Ensembl transcript ENSMUST00000067171.
- b Two entries reflect two regions of homology.

A–P: Evaluation of cross-hybridization with Fgfr riboprobes with sequence homology to some or all Fgfr transcripts (see Table 1 for percentage homology). The four panels in each column were hybridized with the probe listed in the header. To assess probe specificity, tissue that hybridizes with one and only one of the four receptors was selected from embryo sections. A–D: Dorsal root ganglia express Fgfr1 in A, but dorsal root ganglia in B–D (outlined in red) are devoid of signal. E–H: Thymus expresses only Fgfr2 (F), because sections hybridized with probes of any of the other three receptors do not reveal in the thymus, which is outlined in red in E, G, and H. K: The proliferating chondrocytes in the core region of ribs express only Fgfr3. P: Intercostal muscle fibers near the sternum express only Fgfr4. For details, see text. D, dorsal root ganglion; H, heart; L, lung; MF, intercostals muscle fiber; R, rib; T, thymus.
Accessing the Expression Data on Genepaint.org
Because genepaint.org/ currently contains ∼3 TB of data, one needs adequate tools to access such data. Even the limited number of Fgf and Fgfr gene expression patterns determined at a single developmental stage represents a wealth of information that cannot be done justice by a limited number of figures in a publication. Thus, genepaint.org/ permits access to the images collected with the help of diverse queries. The chief features of genepaint.org/ were described previously (Visel et al., 2004) and a user manual is also found in the “quick guide” accessed by means of the genepaint.org/ home page. Expression patterns for all expressed Fgfs and Fgfrs in the E14.5 mouse embryo are represented by a set of 20 to 24 sagittal serial sections spaced approximately 150 μm apart. In the case where expression is not detected, only a near-midline section is shown (Fgf16, -22, and -23). Each of these sets is characterized by a “Genepaint set ID” (Table 2). The expression pattern of, e.g., Fgfr1 is accessed from the home page by typing into the top left blank field either this ID (FG34), the EntrezGene gene symbol (Fgfr1) or the GenBank accession number. Clicking the “GO” button opens a new window on whose right half (“Results” heading) is a line representing the data set for Fgfr1. By clicking the “set viewer” link, a new window opens that consists of four quadrants (Fig. 1E). Top left is an “info box” with self-explanatory information and links to probe sequence, EntrezGene and GenBank. Underneath the info box resides the “annotation table,” describing in which tissues Fgfr1 is expressed. Top right is the “image navigator” with a low-resolution image of the first section of the Fgfr1 data set. One can browse through all images by either using the forward or reverse arrows underneath the image or alternatively by clicking any of the image links listed in the “image directory” located underneath the image navigator. The first image of the Fgfr1 set is named “Embryo_EX215_1_1A”, the second image is “Embryo_EX215_1_1B”, and the last image is “Embryo_EX215_1_6D”. In the text below where we discuss expression patterns, only the last two terms of the image name (1A, 1B, 6D, and so on) are used to refer to the particular section. The low-resolution images appearing in the image navigator quadrant can be enlarged and zoomed-in using one of the three viewer buttons (“HTML”, “Applet”, “Plugin”). The HTML and applet viewers will not require any user intervention, but when using the plug-in viewer, the first time, it will either self-install or can be down-loaded from a Web site listed on the last page of the Genepaint quick guide (currently no plug-in viewer is available for Apple's OS10). The use of any of these viewers is intuitive, but instructions are found in the quick guide. If images are produced with either the HTML or the applet viewer they can be printed using browser-provided printing features.
Gene name | Accession no. | Template size | Forward primer | Reverse primer | Genepaint ID |
---|---|---|---|---|---|
Fgf1 | NM_010197 | 947 | T7_AGGAGCAAGGAGACAGGATG | SP6_ATGCAGTACCCCTGGAGTTG | FG50 |
Fgf2 | NM_008006 | 426 | T7_AGCGGCATCACCTCGCTTCC | SP6_TATGGCCTTCTGTCCAGGTC | FG39 |
Fgf3 | NM_008007 | 681 | T7_TGCTGCTCAGCTTGCTGGAAC | SP6_TTGGAGTGGCCCTGGTAGAC | FG24 |
Fgf4 | NM_010202 | 379 | T7_GACTACCTGCTGGGCCTCAA | SP6_TACCTTCATGGTAGGCGACA | FG44 |
Fgf5 | M30643 | 737 | T7_TCTTCTGCAGCCACCTGATC | SP6_TCTGTACTTCACTGGGCTGG | FG26 |
Fgf6 | M92416 | 873 | T7_AGCTGGAGAGATTTCGGGTG | SP6_AATCCTGCTGACTCGACAAG | FG46 |
Fgf7 | NM_008008 | 524 | T7_ATGGATACTGACACGATCC | SP6_TCTTCCCTTTGACAGGAATC | FG58 |
Fgf8 | Z46883 | 981 | T7_TGTTGCACTTGCT | SP6_TGCGGCTGTAGAGCG | MH358 |
Fgf9 | AH008071 | 766 | T7_ACCTCGCCTAGTGTCTCCTG | SP6_AAGAACCCACCGCATGAAAG | FG52 |
Fgf10 | D89080 | 556 | T7_CTGTTGCTGCTTCTTGTTGC | SP6_GGAGGAAGTGAGCAGAGGTG | FG47, MH359 |
Fgf11 | NM_010198 | 562 | T7_TAGCCTGATCCGACAGAAGC | SP6_GCTGCCTTGGTCTTCTTGAC | FG25 |
Fgf12 | AK018774 | 934 | T7_ACGCTCAAGGAAAAGTTCGG | SP6_AATGCAAGAAGCCTGTGCTC | FG53 |
Fgf13 | AF020737 | 753 | T7_TCGCTCATCCGGCAAAAGAG | SP6_AGGTTCTGTTATAGAGCCCTCG | FG21 |
Fgf14 | NM_010201 | 715 | T7_TCGCCAGCGGCTTGATCCG | SP6_TTGTTGACTGGTTTGCCTCC | FG54 |
Fgf15 | AK017829 | 988 | T7_AGACGATTGCCATCAAGGAC | SP6_ATGGGACAGAGACAAGCTCC | FG49 |
Fgf16 | AB049219 | 577 | T7_TCTTTGCCTCCTTGGACTGG | SP6_ATGGAGGGCAACTTAGAAGG | FG42 |
Fgf17 | NM_008004 | 567 | T7_ACCAGTACGTGAGGGACCAG | SP6_ACTAAGGCCTCCCTGACTAC | FG51 |
Fgf18 | AF075291 | 895 | T7_TCCGCCTGCACTTGCCTGTG | SP6_TGGTTTCTCGCAGTTTCCTC | FG55 |
Fgf20 | AB049218 | 306 | T7_ATCCTGGAATTCATCAGTGTG | SP6_TTCTGGGTCTACTGGTCTTGG | FG40 |
Fgf21 | NM_020013 | 390 | T7_TCAAGTCCGGCAGAGGTACC | SP6_TGGAGCAGGCCTGGCATGGG | FG41 |
Fgf22 | AK008922 | 684 | T7_AGATGTGCACCACCAGGCTG | SP6_ACCAGGCAGTGGGCCTGTAG | FG32 |
Fgf23 | AB037889 | 635 | T7_ACCTGCCTTAGACTCCTGGTG | SP6_TCCTCTGCGCTCGGCAGCTC | FG43 |
Fgfr1 | U22324 | 783 | T7_TAGCTCCCTACTGGACATCC | SP6_TCTGGCTATGGAAGTCGCTC | FG34 |
Fgfr2 | NM_010207 | 715 | T7_ACCGAGAAGATGGAGAAGCG | SP6_AGATGACTGTCACCACCATG | FG35 |
Fgfr3 | NM_008010 | 809 | T7_TCTCCACAGAGGCGTTCTCC | SP6_TGTGTATGTCTGCCGGATGC | FG36 |
Fgfr4 | NM_008011 | 762 | T7_AGCACCCTACTGGACACACC | SP6_TAGTGGCCACGGATGACTTG | FG37 |
Often biologists are interested in identifying genes that are expressed in a certain tissue. To access this information, genepaint.org/ data provides a “structure selection tool”. For an increasing fraction of data sets of genepaint.org/, the pattern and strength of expression is annotated for 96 structures. Therefore, the users can query for genes that are expressed in a particular structure or in a combination of structures by entering into the structure selection tool (advanced search page) one or several structures and the desired type of expression patterns (regional, scattered, ubiquitous, strong, medium, weak; Carson et al., 2005).
Genepaint.org/ also contains an anatomical atlas of an E14.5 NMRI strain mouse embryo consisting of Nissl-stained serial sections prepared identically to those used for ISH. The 96 structures annotated for gene expression are indicated in the atlas by red pointers. The atlas, accessed from the set viewer page by clicking a dedicated button above the image directory, can be browsed through so that a particular ISH section and the closest atlas section can be viewed side-by-side.
Expression Patterns of Fgfr Genes
Fgfr1 (Genepaint set ID: FG34).
The defining feature of Fgfr1 expression in the central nervous system (CNS) is the presence of high levels of transcript throughout the ventricular zone (4D), including that of the spinal cord (4B). Additionally, several brain regions contain scattered Fgfr1-positive cells including striatum (2A), brainstem, pons (4D), and spinal cord (4B). High expression is seen in the hippocampal anlage, and this expression is continued into the cortical plate albeit at a weaker level (4D). In addition to expression in the CNS, Fgfr1 is expressed in many other tissues, mostly of mesodermal and mesenchymal origin.
Fgfr2 (Genepaint set ID: FG35).
Fgfr2 is strongly expressed in the ventricular zone but expression strength varies locally (3D, 4A, 5D). The choroid plexus strongly expresses this gene (4A, 5A), but unlike Fgfr1, no signal is seen in the cortical plate and expression in the hippocampal anlage is near background. Numerous scattered neuronal precursors expressing Fgfr2 are detected in the brainstem, pons, and midbrain (5A). The spinal cord is characterized by strong expression in the roof (4A) but also in neurons elsewhere (3D). Fgfr2 is expressed in the pineal gland primordium, the pituitary (4B), and in various regions of the anterior hypothalamus (4D), i.e., regions involved in diverse neuroendocrine functions. Numerous tissues outside the CNS express Fgfr2 such as kidney, adrenal (3B), pancreas (3C), heart valves (3C), lung (3B), gut (3D), and the axial (3A), and appendicular (1B) skeletal systems.
Fgfr3 (Genepaint set ID: FG36).
Fgfr3 expression is detected in the ventricular zone with gaps similar to those seen for Fgfr2 (3B). Transcripts are detected in numerous scattered neurons of the pons and medulla, midbrain (3B,C), and the striatum (1C). Expression in spinal cord is intense in the ventricular zone and in many cells surrounding this region (4A). Neither the choroid plexus (3B, 3C, 4A), the hippocampal anlage, nor the cortical plate (3B) express this gene. The hypothalamus contains a few scattered Fgfr3-positive neurons (3A). Outside the CNS, Fgfr3 transcripts are detected in endocardium (3D), skeletal primordium (3D, 1B), lung epithelium and mesenchyme (2B), lens (1A), pancreas (3C), intestinal epithelium (3A), and weakly in kidney (5B).
Fgfr4 (Genepaint set ID: FG37).
The defining feature of expression of this gene is strong signal in striated skeletal muscles (5B). Within the CNS Fgfr4 is expressed in the ventricular zone, but expression is less widespread than for any of the three other receptors. Rhombic lip, area postrema, cerebellum (4B), and neuronal precursors presumably exiting the neuroepithelium lining the anterior edge of the fourth ventricle express Fgfr4 (5A). There is a rostrocaudal expression gradient of Fgfr4 in the ventricular zone of the neocortex (4D). The spinal cord contains numerous Fgfr4-positive small, scattered cells, but the level of expression is less pronounced and less regional than observed for any of the three other receptors (4A). Outside the CNS, Fgfr4 is expressed very prominently in skeletal muscles (deLapeyriere et al., 1993) and also in lung, liver, kidney, intestine, pancreas, and cartilage (3B).
Published data describing the expression patterns of Fgfrs are generally consistent with our findings. At E14.5, Fgfr1 and Fgfr2 are expressed in the ventricular zone of the brain and spinal cord and Fgfr2 transcripts are high in the choroid plexus (Orr-Urtreger et al., 1991; Reid and Ferretti, 2003). Expression of Fgfr3 at E14.5 has been reported in the ventricular zone of the spinal cord as well as in neurons that have exited this epithelium (Peters et al., 1993). These authors also noted expression in the ventricular zone of the fore- and hindbrain and in neurons that appear to have migrated from the ventricular zone. Findings concerning detection of Fgfr4 in the CNS are contradictory (Stark et al., 1991; Ozawa et al., 1996). In agreement with the present study, Ozawa et al. (1996) detected Fgfr4 transcripts in the ventricular zone of the forebrain but Hasegawa et al. (2004) do not report such expression. In part, the contradictory results may arise because the various studies use different probes. This study and that of Ozawa et al. used probes that correspond to the coding region, whereas Stark et al. used a probe in the 3′ UTR.
Expression Pattern of Fgf Ligand Genes
Fgf1 (Genepaint set ID: FG50).
Expression of this ligand in the CNS is detected in the ventral part of the spinal cord (3C, D), in the nuclei of the cranial nerves, in the superior cervical ganglion (5A), and the dorsal root ganglia (3A). Strong expression signal is seen in the facial nucleus (3A), the motor trigeminal nucleus (2C), the oculomotor nucleus (3C), the abducens nucleus (3C), the ambiguous nucleus (2D), and the superior paraolivary nucleus (4D). These data are fully consistent with autoradiographic studies by (Alam et al., 1996). A striking and novel feature is the high level of expression of Fgf1 in the neurohypophysis (4A). Outside the nervous system, Fgf1 transcripts are readily found in the lung epithelium (3C; Bellusci et al., 1997), in the submandibular salivary gland (2D; Bellusci et al., 1997; Hoffman et al., 2002), in facial cartilage (3C), and the developing ear (2D).
Fgf2 (Genepaint set ID: FG39).
Expression levels of Fgf2 in the CNS are very low, but at high magnification, tiny blue dye precipitates are detected throughout the brain (e.g., 4C). Even outside the brain expression is weak and possibly ubiquitous. Fgf2 transcripts are clearly recognizable in the long bones of the developing limb (1B) and less clearly in the facial and axial cartilage (3C). Dirks et al. (2001) have carried out Northern blot analyses of E13.5 mouse tissues and consistent with our observation find Fgf1 expression in limbs and facial structures and to a much lesser extent in the brain. Of note, Fgf2 expression is not detected in a microarray analysis of whole E14.5 embryos (see Fig. 4 below).

Expression of Fgfs and Fgfrs determined by microarrays. A: Signal intensity (absolute signal) for each Fgf (black) and Fgfr (red) gene is plotted against the detection P value of the probe set that had the lowest P value. Threshold P values (dashed lines) for determination of the absolute call were P < 0.05 (present), 0.05 ≤ P ≤ 0.065 (marginal), and P > 0.065 (absent). B–K: Fgfs with conflicting microarray and in situ hybridization results (Table 3) are usually expressed in a highly regional manner. Fgf2, weak expression in developing radius and ulna; Fgf3, expression in cerebellar anlage; Fgf4 and Fgf20, expression in the enamel knot of developing teeth; Fgf5, expression in the bed nucleus; Fgf8, expression in the developing kidney; Fgf10, expression in the developing inner ear; Fgf17, expression at the midbrain–hindbrain boundary; Fgf21, expression in tongue epithelium.
Gene name | Affymetrix probe set | Signal A | p-value A | Signal B | p-value B | Array result | ISH expression pattern | Consistency | |||
---|---|---|---|---|---|---|---|---|---|---|---|
Absolute call | Regional | Regional and widespread | Widespread | Not detected | |||||||
Fgf1 | 1423136_at | 21.0 | 0.00586 | 19.3 | 0.03760 | present | medium | yes | |||
Fgf2 | 1449826_a_at | 1.1 | 0.53394 | 4.8 | 0.30371 | absent | low | no | |||
Fgf3 | 1422923_at | 16.2 | 0.30371 | 19.1 | 0.27417 | absent | high | no | |||
Fgf4 | 1450282_at | 27.8 | 0.08057 | 41.7 | 0.11157 | absent | high | no | |||
Fgf5 | 1426186_a_at | 7.6 | 0.56763 | 9.2 | 0.50000 | absent | medium | no | |||
Fgf6 | 1427582_at | 54.5 | 0.05615 | 47.0 | 0.08057 | marginal* | medium | yes | |||
Fgf7 | 1422243_at | 37.6 | 0.00586 | 46.3 | 0.00122 | present | medium | yes | |||
Fgf8 | 1451882_a_at | 0.8 | 0.88843 | 1.0 | 0.88843 | absent | high | no | |||
Fgf9 | 1420795_at | 46.6 | 0.05615 | 50.6 | 0.09522 | marginal* | medium | yes | |||
Fgf10 | 1420690_at | 24.0 | 0.30371 | 16.1 | 0.33447 | absent | high | no | |||
Fgf11 | 1439959_at | 35.1 | 0.03760 | 42.7 | 0.08057 | present* | medium | yes | |||
Fgf12 | 1451693_a_at | 43.6 | 0.00073 | 37.1 | 0.00073 | present | medium | yes | |||
Fgf13 | 1418497_at | 550.1 | 0.00024 | 495.2 | 0.00024 | present | high | yes | |||
Fgf14 | 1435747_at | 10.8 | 0.00122 | 18.7 | 0.00024 | present | medium | yes | |||
Fgf15 | 1418376_at | 36.8 | 0.01856 | 23.8 | 0.19458 | present* | high | yes | |||
Fgf16 | 1420806_at | 6.5 | 0.29614 | 5.9 | 0.30371 | absent | ND | yes | |||
Fgf17 | 1421523_at | 17.6 | 0.01856 | 26.2 | 0.01074 | present | high | yes | |||
Fgf18 | 1449545_at | 84.4 | 0.01856 | 83.9 | 0.03760 | present | high | yes | |||
Fgf20 | 1421677_at | 1.5 | 0.75391 | 2.2 | 0.80542 | absent | high | no | |||
Fgf21 | 1422916_at | 3.8 | 0.60107 | 7.8 | 0.39893 | absent | high | no | |||
Fgf22 | 1460296_a_at | 5.4 | 0.56763 | 18.1 | 0.33447 | absent | ND | yes | |||
Fgf23 | 1422176_at | 2.4 | 0.30371 | 4.8 | 0.24609 | absent | ND | yes | |||
Fgfr1 | 1424050_s_at | 298.7 | 0.00024 | 315.8 | 0.00073 | present | high | yes | |||
Fgfr2 | 1420847_a_at | 63.9 | 0.00024 | 66.9 | 0.00073 | present | high | yes | |||
Fgfr3 | 1421841_at | 239.2 | 0.00293 | 195.3 | 0.00586 | present | high | yes | |||
Fgfr4 | 1418596_at | 96.7 | 0.00122 | 119.8 | 0.00195 | present | high | yes |
- * Absolute call based on that replicate with the lower p-value.
- ND, not detected.
Fgf3 (Genepaint set ID: FG24).
Fgf3 is regionally expressed at E14.5 with strong signal seen in the cerebellar anlage (1D, 4C; Wilkinson et al., 1989), in the septal primordium (4C), and a rostral tegmental nucleus (4C). A few scattered, weakly expressing cells are detected in the brain stem (2B). Retina (1A; Wilkinson et al., 1989), the mesenchymal part of the developing teeth (3B, 5B; Wilkinson et al., 1989), the inner ear (5C; Wilkinson et al., 1989), and the thymus (3B) are other prominent sites of Fgf3 expression.
Fgf4 (Genepaint set ID: FG44).
This gene is not expressed in the CNS, but the suitability of the riboprobe used is validated by the presence of a strong signal in enamel knot cells during the cap stage of tooth development (2D and 5B) as had been described previously (Niswander and Martin, 1992; Kettunen and Thesleff, 1998).
Fgf5 (Genepaint set ID: FG26).
In the brain, this gene is significantly expressed in the bed nucleus of the stria terminalis (3A) and in the pons in a scattered group of cells (3D). Weak and ubiquitous expression is seen in the lateral part of the midbrain (5B) and in the dorsal aspect of the cerebellum (5C). Although expression in the cerebellum is inconspicuous at E14.5, by E18.5, a marked up-regulation of expression has taken place and signal is detected in the posterior folia of the cerebellum (data not shown). Fgf5 expression is also observed in the dorsal root ganglia (4B) and the acoustic ganglia (4D) but not in other cranial ganglia. Outside the nervous system, Fgf5 transcripts are detected in mastication and extrinsic ocular muscles (1B,C), intercostal muscles best visible near the sternum (4B), muscles surrounding the pharynx (3B), a subpopulation of limb muscles (5C), and a thin layer of muscular tissue surrounding liver (3B), kidney, gut, and bladder (3B). Our data are consistent with and extend an earlier study (Haub and Goldfarb, 1991).
Fgf6 (Genepaint set ID: FG46).
Fgf6 is not expressed in the nervous system but transcripts are prominent in skeletal muscles throughout the embryo (2A), including those of the limbs (1B), and also in muscles of the tongue (5A), eye (1C), and in mastication muscles (1C). This pattern is consistent with previous reports (deLapeyriere et al., 1993; Han and Martin, 1993).
Fgf7 (Genepaint set ID: FG58).
Weak but localized expression of Fgf7 is observed in the ventricular zone of the lateral most part of the lateral ventricle of the forebrain (1B), an expression site that had been seen previously (Mason et al., 1994). Numerous sites outside the CNS express Fgf7, including mesenchyme-derived tissues distributed throughout the embryo especially visible in face (3B). Fgf7 transcripts are also present in muscles (4B; Mason et al., 1994; Finch et al., 1995; Bellusci et al., 1997). 4A also reveals a striated expression pattern in tongue muscle.
Fgf8 (Genepaint set ID: MH358)
Fgf8 shows strong regional expression throughout the brain. Specifically, expression is detected in the anlage of the pineal gland (3C), the region of the lamina terminalis neuroepithelium (3B to D), in a line of cells along the anterodorsal thalamus (3A, D), in the optic chiasm (2B, 4A), the midbrain/hindbrain boundary (3A, 4B, C), the midline neuroepithelium lining the floor of the third ventricle (3C, D), and the posterior pituitary (3C). Outside the CNS, Fgf8 is expressed in the inner ear (1B, 5D), in kidney (1D) and in the nasal passage (2B). These data are fully consistent with published data (Crossley and Martin, 1995; Lin et al., 2002), except that, at E14.5, Fgf8 is not expressed in the telencephalic roof plate (3B,C), which it is at E12.5 (Crossley and Martin, 1995; Lin et al., 2002).
Fgf9 (Genepaint set ID: FG52).
In the CNS, Fgf9 expression was observed in the postmitotic neurons of the olfactory bulb (4D), the trigeminal nucleus (5A), and the motor column of the caudal-most region of the spinal cord (4A). Outside the nervous system expression is seen, e.g., in limbs (1B, 6A), the developing whiskers (6B), tongue (3C) and appendicular muscles (6C), heart (4B), cartilaginous, dental and epithelial structures of the lower and upper jaw (4C,D), retina (6C), intestinal epithelium (4B), and very prominently in the semicircular canals of the developing ear (6A–C). These expression sites are fully in agreement with those described previously (Kettunen and Thesleff, 1998; Colvin et al., 1999).
Fgf10 (Genepaint set ID: FG47, MH359).
Fgf10 expression in the brain is restricted to cranial nerve nuclei such as the trigeminal nucleus (2C), the superior paraolivary nucleus (2D), ambiguous nucleus (2D) and facial nucleus (3A), the posterior part of the pituitary anlage (3D) and associated hypothalamic neuroepithelium, and the spinal cord (3B). Expression in the pituitary is also reported in the rat embryo of the stage corresponding to E14.5 of mouse (Yamasaki et al., 1996). Outside the nervous system, strong expression of Fgf10 was observed in lung mesenchyme (3B; Colvin et al., 2001), the developing ear (2B; Pirvola et al., 2000), periocular mesenchyme (2D), lacrimal gland tissue (2C; Makarenkova et al., 2000), the mesenchyme of the submandibular gland (2C; Hoffman et al., 2002), and the dental papillae of lower and upper incisor teeth (3C; Harada et al., 2002). Other positive tissues included kidney (3A), intestine (3D), the tongue (2D), tissue around the thymus (3D), the rostral part of the sternum (3D), the perichondrium of the distal-most limb skeletal elements, and in dermal mesenchyme limb of the zeugopods (1B).
Fgf11 (Genepaint set ID: FG25).
At E11, Fgf11 is prominently expressed in the neuroepithelium (Smallwood et al., 1996), but by E14.5, only weak Fgf11 expression is seen the intermediate zone throughout the telencephalon (3A) and in scattered cells of the inferior colliculus of the midbrain (3A) as well as in the ventral aspect of the spinal cord (3B). Outside the CNS, no convincing expression can be identified with the possible exception of lung (2C) and salivary gland (2C). Using reverse transcriptase-PCR (RT-PCR), Dichmann et al. (2003) reported Fgf11 expression in pancreas, which we were not able to detect in our ISH data.
Fgf12 (Genepaint set ID: FG53).
Olfactory postmitotic neurons and olfactory epithelium, brainstem, pons, and tegmentum (3C), and pretectum (3A,B) express Fgf12 and so do hypothalamus, septum (4B,C), and spinal cord (4C). Peripheral nervous system structures positive for Fgf12 signal are the dorsal root ganglia (5C) and a subset of cranial ganglia, i.e., the trigeminal- and vestibulo-cochlear ganglia (3B). Transcripts of Fgf12 are detected in the retina (1C), and in the caudal-most axial skeleton (3C). In all the above tissue, Fgf12 transcription is clearly detectable, but can be considered mostly weak. By using autoradiography (Hartung et al., 1997), we found a very similar expression pattern in the CNS.
Fgf13 (Genepaint set ID: FG21).
The Fgf13 gene exhibits significant expression throughout the ventricular zone of the telencephalon (2B). Moreover, the marginal zone of the neocortex contains individual Fgf13 expressing cells (5D). Such individual Fgf13-positive cells are also found in tegmentum, inferior and superior colliculi, in pons, medulla, and spinal cord (3B). Noteworthy is the presence of transcripts in the area giving rise to the substantia nigra (4B, ventral tegmental area) and in the locus coeruleus (2C). Cranial nuclei and cranial (2B) and dorsal root ganglia strongly express Fgf13 (2C). Outside the CNS, transcripts are found in the oral epithelium, the kidney (2B), and in the parasympathetic neurons of the intestine (3D) and stomach (6B). Hartung et al. (1997) reported identical results in their autoradiographic study of Fgf13 expression. Microarray data suggest that the level of expression of Fgf13 in whole E14.5 embryos is approximately one order of magnitude higher than that of Fgf12 (Fig. 4A), a result that is consistent with our ISH data.
Fgf14 (Genepaint set ID: FG54).
Fgf14 is expressed in the amygdala (1C), striatum (1C), hypothalamus, preoptic area (4B), and septum (4C). Weaker expression can be seen in the thalamus (4C), and scattered/regional expression of Fgf14 is detected in pons, medulla, pretectum, and spinal cord (4B). The thymus strongly expresses Fgf14 (4B). These expression sites are identical to those seen in the autoradiography-based study of Wang et al. (2000).
Fgf15 (Genepaint set ID: FG49).
Expression is restricted to the CNS with the exception of individual cells in the olfactory epithelium (4C) and the retina (1A; Thut et al., 2001). Within the CNS, expression is found in the lateral ventricle in the amygdaline region (1C), the septal area (3C and 4C), the ventricular zone of the olfactory bulb, and of the midbrain (4C), the preoptic area (4C), thalamus (3C, 4B), the lamina terminalis region (4B), and in spinal cord (4B). Cerebellum and single cells in the brainstem and the ventral premamillary nucleus also strongly express Fgf15 (5A). Noteworthy are scattered cells in the ventral (5A) and dorsal thalamus (4C). Our results are in agreement with those reported by McWhirter et al. (1997) and Gimeno et al. (2002).
Fgf16 (Genepaint set ID: FG42).
No expression of Fgf16 was seen at E14.5.
Fgf17 (Genepaint set ID: FG51).
At mid-gestation Fgf17 expression is highly restricted to three regions: the midbrain/hindbrain junction (4B), lamina terminalis region/subfornical organ (4B) and in a very restricted way in the suprachiasmatic region (3B, 4C). Xu et al. (2000) report identical sites of expression in the CNS but in addition also find Fgf17 transcripts in the major arteries and the periosteum and chondrocytes of dorsal costal bone.
Fgf18 (Genepaint set ID: FG55).
Fgf18 expression in the CNS is complex. Fgf18 transcripts are observed in the marginal zone (4B; Hasegawa et al., 2004, interpret the Fgf18-expressing cells in the neocortex as the cortical plate), the neuroepithelium of the lamina terminalis region (4A), the midline neuroepithelium lining the floor of the third ventricle (4A), the midbrain/hindbrain boundary (3C), certain nuclei in the pons and brainstem (5C, 4C), very strongly in a cluster of cells in the tegmentum (4B), and in the optic chiasm (4C). Spinal cord tissue reveals Fgf18 transcripts in its ventral region (4C). In the peripheral nervous system, Fgf18 transcripts are detected in the dorsal root ganglia (3B), cranial ganglia (2C), and in a subpopulation of neurons of the olfactory neuroepithelium (3A). Outside the nervous system, Fgf18 transcripts are prominently detected in joints (1A, B), in the perichondrium of long bones of limbs (5D), in the axial skeleton including ribs (4B, C), sternum (4B), and intervertebral discs (3B); the osteogenic mesenchyme around the brain and in the developing upper and lower jaws (3C). Kidney and adrenal are surrounded by Fgf18-positive tissue (5A), and lung mesenchyme surrounding the airway epithelium (5A) also expresses this gene. Many of these sites of Fgf18 expression have been reported previously (Hu et al., 1998; Liu et al., 2002; Ohbayashi et al., 2002; Usui et al., 2004).
Fgf20 (Genepaint set ID: FG40).
Distinct expression of Fgf20 was found in the developing teeth (5B,C) and possibly also in a subgroup of cells in the cochlea (4C, 5B). The Fgf20 probe exhibits 80% identity with the Fgf9 mRNA over 184 nucleotides, which might cause cross-hybridization. A strong site of Fgf9 expression is the motor trigeminal nucleus (FG28, 5A), a structure that does not show any signal when sections are hybridized with Fgf20 riboprobe. To our knowledge, Fgf20 has not been described at a stage comparable to ours.
Fgf21 (Genepaint set ID: FG41).
Strong expression in cells of the tongue epithelium and the dorsal epithelium of the oral cavity is observed (3C–4A). Notably, expression is confined to tissue in the midsagittal plane. To our knowledge, no published data for this developmental stage currently exists.
Fgf22 (Genepaint set ID: FG32).
No expression is seen at E14.5.
Fgf23 (Genepaint set ID: FG43).
No expression is seen at E14.5.
Complex Expression Patterns of Fgfs and Fgfrs at E14.5
Availability of ISH data for multiple genes in a set of standard sections makes it straightforward to compare expression patterns of genes that belong to a signaling pathway or to a gene family. In the case of the brain-specific expression of Fgfs and their receptors, several general findings emerge. All four receptors are expressed to a greater or lesser extent in the neuroepithelium. In the case of Fgfr1, expression encompasses the entire ventricular zone, while in the case of the other receptor genes, only a part of the ventricular zone exhibits expression. In addition to their characteristic expression in the proliferating cells of the ventricular zone, all four receptors are expressed in numerous neurons scattered throughout the brain. Specifically, such cells are found in the striatum (Fgfr1 to 3), in the hypothalamus (Fgfr2 and 3), in midbrain (Fgfr1 to 3), and in pons and medulla (all four Fgfrs). To determine whether any of these neurons express multiple receptors or whether expressing cells are merely in spatial proximity will require double-labeling experiments.
With regard to Fgf expression in the brain, the following themes emerge. First, all Fgfs, with the exception of Fgf16, -22, and -23, are expressed, and overlapping expression of multiple Fgfs is quite common. Fgf1, -10, and -13, for example, are coexpressed in several cranial nerve nuclei. Fgf11, -12, -13, -15, and -18 are expressed in parts or throughout in the ventricular zone and Fgf3, -15, and to a lesser extent also Fgf5 are all expressed in the cerebellum. Second, several secreted Fgfs (7, 8, 15, 17, 18) and nonsecreted Fgfs (11, 12, 13) are expressed in cells located in proximity to the ventricles. For example, Fgf7 and -15 are expressed in the lateral ventricle of the forebrain, and Fgf8 and Fgf17 transcripts are found in lamina terminalis and the midline neuroepithelium lining the floor of the third ventricle. In the case of the secreted Fgfs, this finding opens up the possibility that they are delivered into the ventricles from which they could diffuse and bind to Fgfrs that are abundantly expressed throughout the ventricular neuroepithelium. In the case of the nonsecreted Fgfs (11 to 13), receptor activation could be achieved by means of an autocrine mechanism. Thus, Fgf13 is prominently expressed nearly throughout the ventricular zone of the telencephalon. In keeping with the theme of short-range signalling, one also observes overlapping and nearby expression of receptors and ligands in the deeper regions of the brain such as in the striatum, where solitary Fgfr3-expressing cells are embedded in a mass of cells expressing Fgf14. This synopsis makes it clear that, in the brain, similar to other organs, the components of the Fgf signaling system exhibit complex, partly overlapping expression patterns. The genepaint.org/ database provides a wealth of data required to identify expression overlaps and, thus, will be very useful in the design of appropriate mouse genetic experiments such as the generation of double and triple mutants and the design of tissue-specific knockouts. Needless to say, in the context of such experiments, the genepaint.org/ data may be extended to other developmental stages using our validated templates (Table 2).
Consistency of ISH and Microarray Expression Data
An ever-growing number of studies illustrate the power of analyzing gene expression by microarrays in which a large number of genes are analyzed for expression in parallel and where quantification of expression is standardized (Barrett et al., 2005). This finding raises the question of how expression based on a microarray analysis compares with ISH results. To begin to address this question, we have compared Fgf and Fgfr expression results obtained by ISH with a microarray analysis of whole E14.5 mouse embryos. Figure 4A depicts a scatter plot in which the intensity of expression is plotted against the detection P value for Fgfs and their receptors expressed in the E14.5 mouse embryo. As judged by the detection P values, all four Fgfrs and nine Fgfs score as present, two Fgfs are marginally present and 11 are absent. The signal intensities for those transcripts that are present range between 19 (Fgf14) and 550 (Fgf13), thus encompassing a rather wide range. Table 3 juxtaposes the signal intensities determined by array analysis with the expression patterns as revealed by ISH. Array results are represented with the signal intensities determined on two replicate arrays, P values and the absolute call of each gene. ISH expression patterns are classified as “regional” (gene is predominantly expressed in a very restricted manner, e.g., Fgf8), “regional and widespread” (gene is expressed in numerous tissues but also regionally, e.g., Fgf13), and “not detected” (e.g., Fgf22). In addition, the strength of expression was classified as “low,” “medium,” or “high” (see Carson et al., 2005 for definition of expression strength). For 18 of 26 cases, there is a match between the array and ISH data. In the eight instances where the results differ (Fgf2 to -5, Fgf8, Fgf10, Fgf20, and Fgf21), the ISH analysis reveals that expression is highly regional, with Figure 4B illustrating such localized expression. For example, Fgf4 transcripts are only found in the enamel knot of the developing teeth, Fgf20 transcripts in developing teeth and ear, and Fgf21 transcripts are exclusively detected in the tongue epithelium. It should be noted, however, that ISH identifies Fgf17 expression as highly regional (mid-/hindbrain junction, lamina terminalis, subfornical organ, and suprachiasmatic region), and the array results classify this gene as present. Thus, genes expressed in regionally restricted manner may but not necessarily do escape detection by arrays. Taken together, despite some shortcomings in the case of regionally expressed genes, the comparison of the two types of data suggests that a microarray analysis, even when based upon RNA isolated from a whole embryo, provides useful initial information for prioritization of a genome-wide ISH analysis.
Conclusions
Similar to genome sequences, comprehensive and interactive atlases of gene expression are valuable resources for biology. Although several such atlases now exist, most of them have limited capability to access primary data and/or apply elaborate search strategies that require extensive structural annotation. This study examines the technology on which the data of genepaint.org/ are based. We show that sensitivity, signal-to-background ratio, and probe specificity of the semiautomated procedure used here are such that reliable data can be produced. This finding is not only apparent from a battery of experimental tests but also emerges from the fact that, in their majority, the expression patterns of Fgf and Fgfr genes determined in this study are fully consistent with previously published results. A practical conclusion from our limited initial comparison of ISH and microarray data is that the prioritization of which genes should be examined first in a transcriptome-scale ISH can be based on microarray data. The caveat is that some genes with a highly restricted expression pattern would be missed, a deficiency that could be reduced with arrays of single embryonic organs, tissues, or even cells (Tietjen et al., 2003).
Genepaint produces large amounts of information that cannot be published in a conventional way but is accessed, viewed, and mined through Web-based tools provided by genepaint.org/. We intentionally refrained from showing specific expression patterns for Fgfs and Fgfrs but instead point the reader to pertinent images on genepaint.org. This interactivity requires some extra effort, but the obvious reward of making primary data accessible in a searchable format is that an intelligent access to the data opens up opportunity for further research. For example, clustering algorithms can be developed that allow genes with similar complex expression patterns to be identified in silico. The structure selection tool of genepaint. org is a first step in this direction. As thousands of gene expression patterns eventually populate gene expression databases, a complex “molecular map” of localized gene expression will emerge that will advance our understanding of gene function, gene interaction, gene regulation, and in conjunction with other “-omic” data, will lead to testable hypotheses (Ge et al., 2003).
EXPERIMENTAL PROCEDURES
Collection of Tissues and Sectioning
E14.5 embryos (NMRI or C57BL/6J) were collected in ice-cold phophate buffered saline, blotted dry with a filter paper, immersed in ice-cold OCT 4583 (Tissue-Tek) for ∼5 min, and transferred into an OCT-filled, custom-made freezing chamber (Fig. 1A). The embryo was oriented and the chamber placed on top of a metal block immersed in a mixture of dry ice and ethanol (−70°C) and left there until freezing OCT reached the embryo. Thereafter, the chamber was kept at room temperature for a few minutes and then transferred to a −20°C freezer. A few hours later, the chamber was disassembled and the block was stored in a sealed plastic bag at −80°C. Before sectioning, the OCT block was equilibrated at −20°C for up to a week and frozen sections were cut in a Leica CM3050S cryostat with the chamber and object holder temperatures set so that the temperature in the proximity of the specimen was approximately −10°C as measured with a thermometer placed near the specimen. Sections (20 or 25 μm thick) were collected (Fig. 1B) from eye-to-eye on Superfrost slides that were subsequently placed into metal slide racks (Leica). Slide racks were stored in Zip-lock bags overnight (−20°C) in the presence of silica gel. Thereafter, the unthawed sections were fixed at room temperature in 4% paraformaldehyde for 15 to 30 min, acetylated twice for 5 min each in fresh 0.25% v/v acetic anhydride in 0.1 M triethanolamine (pH 8.0), and dehydrated in the Leica autostainer XL in an ethanol series ending with 100% ethanol. Air-dried sections were stored at −80°C in sealed slide boxes containing desiccant. Sections were collected as follows: The slide holder frame (Fig. 1B) defines four areas (A–D). The first section cut was placed to position A of the first slide, the second section to position A of the second slide, and so forth. The sixth section went to position A on the sixth slide, whereas the seventh section was placed to position B of the first slide etc. This resulted in six sets of slides, each consisting of five or six slides.
DNA Template Synthesis
Templates for Fgfs, Fgfrs, and Ear2 were generated from a cDNA cocktail derived by reverse transcription of mRNA from E10.5 and E14.5 mouse embryos and P7 and P56 mouse brains. In the PCR, specific forward and reverse primers (Table 2) extended at their 5′ end with either T7 or SP6 promoter sequences were used (T7, GCGTAATACGACTCACTATAGGG; SP6, GCGATTTAGGTGACACTATAG; the three underlined extra bases facilitate RNA polymerase binding). For example, the Ear2 gene specific primers were as follows: Ear2 forward, T7_5′-ACCGAGTATGTGCGTGCCCAG-3′; Ear2 reverse, Sp6_5′-TGGTGCCCTCCGTGGACCATG-3′. Temperature gradient PCR (55 to 65 °C) was carried out with six cDNA aliquots, which contained 2 μl of forward and reverse primers (5 pmol/μl, MWG Biotech), 2 μl of cDNA, 2 μl of dNTPs (2 mM, Roche), 10× PCR buffer (Qiagen), 5× Q enhancer (Qiagen), and 0.1 μl Taq polymerase (5 U/μl, Qiagen), and water to a volume of 10 μl. PCR conditions were as follows: 2 min initial template denaturation at 94°C, 35 cycles with 20 seconds denaturation (94°C), 20 seconds primer annealing (55–65°C), 1 min per kilobase of elongation (72°C), and a final elongation cycle for 9 min at 72°C. The resulting product was analyzed and purified on an agarose gel, and the desired band was extracted using the Qiagen gel extraction kit. Initial PCR products were re-amplified in a reaction consisting of 10 μl each of the gene-specific forward and reverse primer (5 pmol/μl), 20 to 50 ng of PCR product, 10 μl of dNTPs (2mM), 10× PCR buffer, 5× Q enhancer, and 0.5 μl Taq polymerase, and water to a total volume to 100 μl. PCR conditions were as described above, except that primer annealing temperature was that as determined to be optimal in the gradient PCR above. Templates were purified using QIAquick PCR purification spin columns (Qiagen) and aliquots (∼200 ng) of the purified DNA (∼10 μg) were analyzed on an agarose gel. PCR products were sequence verified. Trhde template (accession no. NM 146241, nucleotide positions 1382–1865) was PCR amplified from a subcloned fragment with T7 and SP6 primers as described above.
RNA Probe Synthesis
In vitro transcription was carried out in a cocktail of 2 μl of rATP, rCTP, rGTP (10 mM, Roche), 1.3 μl of rUTP (10 mM), 0.7 μl of digoxigenin-UTP (10 mM, Roche), 2 μl of dithiothreitol (0.75 M), 1 μl of ribonuclease inhibitor (40 U/μl MBI Fermentas), 1 μg of DNA template, 0.5 μl of RNA polymerase (T7: 50 U/μl, SP6: 20 U/μl, both New England Biolabs), 5× transcription buffer and DEPC water to a final volume of 20 μl. The reaction mix was incubated at 37°C for at least 150 min followed by a 20-min incubation with a DNase I/MgCl2 mix to remove the DNA template (1.6 μl of 300 mM MgCl2, 2 μl of DNase I [10 U/μl, Roche], 16.4 μl of DEPC water). RNA was ethanol precipitated, re-dissolved in 25 μl of DEPC water while vigorously shaking at room temperature. Two microliters of the RNA were analyzed on a 1% agarose gel, and 2 μl were used to determine concentration in a photometer. Probes were diluted with DEPC-treated water to 150 ng/μl and stored as 20-μl aliquots at −80°C. Before use in ISH, riboprobes were diluted in hyb-mix (Ambion, catalog no. B8807G) to a final concentration of 150–400 ng/ml.
Quantitative Real-Time PCR
To determine the mRNA copy number per cell of a weak to moderate expressed gene (Dscr3), mRNA was isolated by two sequential incubations from 193 mg (wet weight) of P7 mouse forebrain using Dynabeads (Oligo (dT)25, Dynal Biotech). A Dscr3 RNA standard was synthesized using the Ambion Competitor Construction 1356 kit. Tissue-derived mRNA and Dscr3 RNA standard were reverse transcribed in parallel reactions using a Dscr3-specific reverse primer (ACTGCTTAATGAACCTAAG). From both types of cDNAs, serial dilutions were made and subjected to Q-PCR in an iCyler iQ (Bio-Rad) with SYBR green detection. The reaction volume was 50 μl containing 25 μl of SYBR Green Supermix (Bio-Rad), 1 μl of each internal primer (Dscr3 forward, 5′-AACTGCGCTATCACGCAGC-3′; Dscr3 reverse, 5′-TCCTGTTGCATTCCAGACG-3′, 10 pmol/μl), 2 μl of a serial dilution of brain cDNA or Dscr3 standard and 21 μl of water. The threshold cycles of the serially diluted Dscr3 standards were determined and semilogarithmically plotted against the copy number of the RNA standard, yielding a standard curve from which the Dscr3 transcript copy number present in the brain extract was interpolated. To determine Dscr3 transcript copy number per cell, the number of cells in the brain piece was estimated through DNA content measurements as described in Sambrook and Russell (2001).
Microarrays
E14.5 embryos were dissected from timed-pregnant C57BL/6J mothers and homogenized in liquid nitrogen. Total RNA was isolated from two different aliquots of frozen embryo powder with RNAzol B. Duplicate cRNA synthesis, hybridization to two Mouse Genome 430 2.0 arrays (Affymetrix), and initial data analysis were carried out by the Affymetrix Service Provider at the German Resource Center for Genome Research (RZPD, Berlin). For Fgf and Fgfr genes that were represented by more than one probe set on the 430 2.0 array, that probe set with the lowest P value was selected for further analysis (Table 3). Absolute calls for each gene were determined based on the P values (present, P < 0.05; marginal, 0.05 ≤ P ≤ 0.065; absent, P > 0.065). For 22 of 26 genes, the absolute calls (absent, marginal, present) were identical between the two replicate arrays. For those four genes that had different absolute calls between the two replicates (Fgf6, Fgf9, Fgf11, Fgf15), the result with the lower P value was used for further analysis. For genes with present or marginal absolute calls, the correlation coefficient between the signal intensities of the replicates was > 0.99.
Genepaint Hardware
The Genepaint instrument constructed for automated ISH consists of flow-through hybridization chambers (Fig. 1C), which are inserted into temperature-controlled chamber racks located on a Tecan Genesis solvent delivery robot (Carson et al., 2002). The layout of the platform used for the present study is shown in Figure 1D. Appendix 1 details the ISH protocol automatically executed by a script that controls pipette movements, buffer deliveries, riboprobe addition, and duration and temperature of all reactions. After cover-slipping, slides are scanned in a microscope equipped with a custom-made stage and controlled by custom-made software (Carson et al., 2002). Images are collected with a CCD camera in brightfield or interference contrast with a 10× objective (NA 0.40). The resulting TIFF images with a resolution of 1.6 μm/pixel are stitched together and deposited on the genepaint.org/ database along with metadata such as specimen, gene name, and probe sequence.
Acknowledgements
We thank Drs. Lars Geffers and Dr. Marei Warnecke for cDNAs, Sarah Herzog, Frauke Grabbe, Barbara Fischer, Ana Martinez-Hernandez, Polina Spies, and Markus Uhr for technical assistance, and Dr. Nigel Pringle for providing the initial information on the sequences of murine Fgfs and their receptors. We thank Dr. Moritz Rossner for critical reading of the manuscript. A.V. was supported by a Boehringer Ingelheim Fonds scholarship. G.E. is a recipient of a Burroughs Wellcome Award in Functional Genomics. This paper is dedicated to the memory of Reiner Pisalla whose enthusiasm for laboratory automation was essential for the development of robotic in situ hybridization in our laboratory.
APPENDIX 1
A script (available upon request) is executed by the Tecan Gemini software and manages a multichannel solvent delivery system used for the controlled addition of reagents to the hybridization chambers as described in the protocol below. This protocol lists reagent composition, sequence of addition, incubation times, and solution volumes. The protocol is optimized for in situ hybridization (ISH) of embryonic day (E) 14.5 embryos but, with minor adjustments, can be used for other types of specimens. Solutions are prepared at different stages of the process. Most are made once per week, but a few are made on the day that they are used and, hence, are referred to as either “day 1” or “day 2” solutions. Altogether, the ISH process lasts approximately 24 hr. On day 1, the hybridization chambers are assembled, followed by prehybridization, overnight hybridization with probes, and stringency washes. Day 2 encompasses blocking steps, hapten detection, and signal amplification (CARD). All solutions required up to and including the overnight hybridization were made with DEPC-treated ultra-pure water (e.g., Milli-Q) in sterile plastic vials or glassware baked at 180°C. Solutions required for subsequent steps were made with regular ultra-pure water. Pipetting volumes were 350 μl unless noted otherwise, and all solutions contained 0.05% Tween-20 except for methanol/0.6%hydrogen peroxide solution and the hybridization buffer. Because reactions occur in a narrow 80-μm-thick hybridization chamber, several solutions have to be degassed, which is achieved by a combination of preheating and vacuum application. This process prevents the formation of bubbles in the hybridization chambers.
Day 1: Prehybridization
- 1
Slides were thawed while they were sealed inside slide boxes at 37°C for 30 min. Thereafter, they were assembled into flow-through hybridization chambers (Fig. 1C) and placed into the chamber racks located on the Tecan platform (Fig. 1D). The process was started by incubating five times for 5 min with methanol containing 0.6% hydrogen peroxide (peroxide is added to methanol just before use).
- 2
Incubate seven times for 5 min with phosphate-buffered saline (PBS).
- 3
Incubate twice for 5 min with 0.2 N hydrochloric acid.
- 4
Incubate four times for 5 min with PBS.
- 5
Incubate once for 5 min with 400 μl of proteinase K buffer (50 mM Tris, 5 mM ethylenediaminetetraacetic acid [EDTA], pH 8.0).
- 6
Incubate twice for 10 min with 2 μg/ml proteinase K (Roche, 3115887) in proteinase K buffer. Proteinase K is freshly added to the buffer as step 5 commences, mixed, and placed onto the Tecan platform. Different developmental stages and tissues require adjustment of proteinase K concentration.
- 7
Incubate seven times for 5 min with PBS.
- 8
Incubate twice for 10 min with 4% paraformaldehyde (PFA; EMS, 19210).
- 9
Incubate seven times for 5 min with PBS.
- 10
Incubate with hybridization buffer. Hybridization buffer (Ambion, B8807G) is preheated at 65°C in an incubator oven for ∼10 min. After cooling down to room temperature, add dithiothreitol (Sigma) to 1.5mg/ml.
- 11
A second aliquot of hybridization buffer is added, and incubation is continued at room temperature for 15 more min, after which the temperature of the chamber rack is increased to 64°C.
Overnight: Hybridization Step (64°C)
- 1
RNA probe (150–400 ng/ml final concentration) is dissolved in hybridization buffer prepared as described under 10 above and was added to each hybridization chamber by the script. After 2.5 hr, a fresh aliquot of probe is pipetted by the script and hybridization process is continued for 3 hr.
Overnight: Stringency Washes (62°C)
- 1
Incubate five times for 5 min with 5× standard saline citrate (SSC; 0.15 M NaCl, 0.015 M sodium citrate, pH 7.0).
- 2
Incubate five times for 10 min with 2× SSC containing 50% formamide.
- 3
Incubated five times for 12 min with 1× SSC containing 50% formamide.
- 4
Incubated four times for 8 min with 0.1× SSC. During the last cycle, the script ramps-down the chamber rack temperature to 25°C.
Overnight and Day 2: Immunohistochemical Detection of Riboprobe
- 1
Incubate four times for 5 min with NTE (500 mM NaCl, 10 mM Tris, 5 mM EDTA, pH 7.6).
- 2
Incubate six times for 5 min in 20 mM iodoacetamide (Sigma) in NTE. This solution is made freshly on day 1, degassed for 30 min in a vacuum oven at room temperature and placed onto the platform.
- 3
Incubate four times for 5 min with NTE.
- 4
Incubate two times at room temperature for 5 min each with TN (100 mM Tris, 150 mM NaCl, pH 7.6).
- 5
Incubate six times for 5 min in sheep serum in TN. On day 1, mix 4% (w/v) heat-inactivated sheep-serum (Gibco, 16070-096) with TN, vacuum filter through a 0.45-μm HV Durapore membrane (Millipore), degas in a vacuum chamber, and place on platform.
- 6
Incubate four times for 5 min with TN.
- 7
Incubate two times at room temperature for 10 min with TNB blocking buffer (100 mM Tris, 150 mM NaCl, 0.5% (w/v) blocking reagent [Perkin-Elmer Lifesciences, FP1012] pH 7.6). This solution is prepared on day 1 as follows. An Erlenmeyer flask containing the TNB mixture is placed on a magnetic stirrer set to 60°C, which is located in a 65°C incubator oven. Stir for 2 hr until solution is translucent. Continue stirring for 1 hr at room temperature, and then degas for 30 min in a vacuum oven.
- 8
Incubate two times for 5 min with TN solution.
- 9
Incubate two times for 5 min with maleate wash buffer (MWB; 100 mM maleate, 150 mM NaCl, pH 7.5). This solution is prepared on day 1.
- 10
Incubate two times for 10 min with a day 1-prepared MWB containing 1% (w/v) blocking reagent (1% BR; Roche, 1096176). An Erlenmeyer flask containing the 1% BR mixture is placed on a magnetic stirrer set to 60°C that is located in a 65°C incubator oven. Stir for 2 hr until solution is translucent. Continue stirring for 1 hr at room temperature and filter. Then degas for 30 min in a vacuum oven at room temperature.
- 11
Incubate two times for 5 min with MWB.
- 12
Incubate two times for 5 min with TN solution.
- 13
Incubate three times for 5 min with a day 1-prepared TMN solution (0.1 M Tris, 0.05 M MgCl2, 0.1 M NaCl, pH 9.5). Degas for 30 min in a vacuum oven at RT and filter.
- 14
Incubate four times for 5 min with TN solution.
- 15
Incubate four times for 10 min each with TNB blocking buffer (see step 7).
- 16
Incubate two times for 30 min with a day 1-prepared anti-digoxigenin-POD (Roche, 1207733) diluted to a final concentration of 0.2925 U/ml in degassed TNB.
- 17
Incubate six times for 5 min with TN.
- 18
Incubate for 20 min with 300 μl tyramide–biotin diluted 1:50 with TSA amplification diluent (Perkin-Elmer Lifesciences, FP1120). This solution is prepared on day 2, warmed to 37°C for 5 min, and is used immediately.
- 19
Incubate six times for 5 min with MWB.
- 20
Incubate two times for 30 min with a day 2-prepared neutravidin–alkaline phosphatase (N-AP) conjugate (Pierce, 31002) diluted to a final concentration of 2.85 μg/ml in 1% BR. The MWB with 1% (w/v) blocking reagent is degassed for 30 min in a vacuum oven at room temperature, filtered, and N-AP is added.
- 21
Incubate six times for 5 min with MWB.
- 22
Incubate four times for 5 min with TN.
- 23
Incubate two times for 5 min with TMN containing levamisol (0.5 mg/ml, Sigma) prepared as described under step 13.
- 24
Incubate three times for 10 min in chromogenic reagent that is prepared as follows. Degas and filter TMN and add 5-bromo-4-chloro-3-indolyl-phosphate (BCIP; 0.15 mg/ml, Roche, 1383221), nitro blue tetrazolium (NBT; 0.4 mg/ml, Roche, 1383213) and, while gently stirring, add levamisol (0.5 mg/ml). The terminal incubation step can be extended depending on signal strength.
- 25
Incubate four times for 5 min with 400 μl of system liquid water (provided by the Tecan robot).
- 26
Incubate for 5 min with NTE.
- 27
Incubate for 20 min with 200 μl of 4% PFA.
- 28
Incubate four times for 5 min with 400 μl of system liquid water.
- 29
Disassemble hybridization chambers while they are immersed in dust-free water, air-dry slides overnight, and coverslip them with aqueous medium (Hydro-Matrix, Micro-Tech-Lab, Graz, Austria). Squeezes out bubbles and remove excess medium with a moist tissue paper.