A threshold requirement for Gbx2 levels in hindbrain development Free

The vertebrate neural tube is patterned by a combination of intrinsic factors, such as transcription factors that define positional information, and extrinsic factors that organize local regions of the neuroepithelium. Determining how these factors interact is essential to understanding how the central nervous system (CNS) develops. Early in development, the rostral neural tube is divided along the anteroposterior (AP) axis into a series of brain vesicles: the prosencephalon (forebrain), the mesencephalon (midbrain)and the rhombencephalon (hindbrain) (Wurst and Bally-Cuif, 2001). The hindbrain is transiently segmented during development into seven or eight rhombomeres(Lumsden and Krumlauf, 1996; Wingate, 2001). Rhombomere(r)1 and r2 (metencephalon) give rise to the cerebellum and pons, whereas r3-r8 give rise to the medulla oblongata. Crucial neuronal populations, such as cranial motor nerves, arise within specific rhombomeres(Chandrasekhar, 2004; Cordes, 2001). Rhombomere identity is partially determined by homeobox-containing genes of the Hox family, whose expression respects specific rhombomere borders. The anterior limit of Hox gene expression is r2(Lumsden and Krumlauf, 1996). Thus, the acquisition of AP identity in r1 relies on alternative regulatory genes and inductive signals to confer positional value and cell specification.

Patterning and growth of the midbrain and anterior hindbrain (cerebellum)requires signals originating from the midbrain-hindbrain (MHB) organizing center, which forms at the boundary between the mesencephalon and metencephalon, within an anatomical constriction known as the isthmus(reviewed in Liu and Joyner,2001; Wurst and Bally-Cuif,2001). The organizing activity of the MHB organizer was demonstrated through classic transplantation studies in avian embryos showing that isthmic grafts can induce ectopic midbrain structures in the caudal forebrain and ectopic cerebellar tissue in the posterior hindbrain(Bally-Cuif et al., 1992; Gardner and Barald, 1991; Itasaki et al., 1991; Marin and Puelles, 1995; Martinez et al., 1991).

Wnt1 and Fgf8 are expressed at the MHB boundary and encode signaling molecules that are candidates for isthmic organizing activity(Wurst and Bally-Cuif, 2001). At embryonic day (E) 8.0, Wnt1 is expressed in the entire midbrain,but, by E9.5, Wnt1-positive cells are restricted to a narrow stripe at the most posterior midbrain (Bally-Cuif et al., 1995; Parr et al.,1993; Rowitch and McMahon,1995). Fgf8 expression is activated throughout r1 by E8.5, and is later restricted to a stripe at the anterior hindbrain(Crossley and Martin, 1995). Thus dynamic changes in Wnt1 and Fgf8 expression domains result in their mirror image on the rostral and caudal side of the MHB boundary, respectively. Loss of either Wnt1 (as a conventional gene inactivation) (McMahon and Bradley,1990; Thomas and Capecchi,1990) or Fgf8 (as a hypomorphic allele or an isthmus-specific gene inactivation) (Chi et al., 2003; Meyers et al.,1998) prevents formation of the midbrain and anterior hindbrain,possibly because of aberrant cell death(Chi et al., 2003). When ectopically expressed, Fgf8, but not Wnt1, mimics the isthmic organizer activity (Adams et al.,2000; Crossley et al.,1996; Dickinson et al.,1994; Liu et al.,1999; Martinez et al.,1999; Shamim et al.,1999). Thus FGF8 is sufficient, as well as necessary, for MHB organizing activity.

Given the primary role of FGF8 as the MHB organizer, it is important to determine how isthmic Fgf8 expression is induced and maintained in the proper location. A cascade of transcription factors expressed in the MHB region is required for Fgf8 expression, including Pax2/Pax5and En1/En2 (Liu and Joyner,2001). However, much attention has been focused on two homeobox containing genes, Otx2 and Gbx2, not because they are required for Fgf8 expression per se, but because they control the correct positioning of the isthmic Fgf8 expression domain(Broccoli et al., 1999; Brodski et al., 2003; Millet et al., 1999). Starting at E7.5, Otx2 and Gbx2 expression define two domains within the anterior and posterior neuroectoderm, respectively, whose juxtaposition demarcate the presumptive MHB junction (Ang et al., 1994; Bouillet et al.,1995). This juxtaposition is maintained through mutual antagonism,resulting in a sharp MHB border by E10.5 with Otx2 transcripts found in the forebrain and midbrain, and Gbx2 expression maintained in r1-r3 (Li and Joyner, 2001; Martinez-Barbera et al.,2001). Otx2 is required cell autonomously for development of the midbrain and forebrain (Acampora et al., 1998; Rhinn et al.,1998), whereas Gbx2 is required for the development of r1-r3 (Wassarman et al.,1997). Loss-of-function(Acampora et al., 1997; Puelles et al., 2003; Suda et al., 1997) or gain-of-function (Broccoli et al.,1999; Katahira et al.,2000) genetic manipulation of Otx2 activity results in a rostral or caudal, respectively, shift of Fgf8 expression. Conversely, Gbx2 inactivation or overexpression shifts Fgf8expression caudally or rostrally, respectively(Katahira et al., 2000; Millet et al., 1999; Wassarman et al., 1997). Conditional inactivation studies examining the temporal requirements of Gbx2 in r1 after E9.0 have demonstrated the necessity of its continued expression for the correct establishment of the isthmus and normal cerebellar midline development (Li et al.,2002).

We have investigated the effects of reduced Gbx2 levels on development of the isthmus and hindbrain by studying mice homozygous for a Gbx2 hypomorphic allele, Gbx2^neo. These studies reveal different threshold requirements for Gbx2 in refining gene expression and regulating AP patterning within the isthmus and anterior hindbrain.

Mice carrying the Gbx2^neo allele have been described(Wassarman et al., 1997). Gbx2^neo heterozygotes were intercrossed to generate Gbx2^neo homozygotes; Gbx2^neoheterozygous and Gbx2 wild-type siblings served as controls. Mice were maintained on a mixed 129×C57BL/6 background and genotyped as described (Wassarman et al.,1997).

Whole embryo RNA was isolated using RNeasy (Qiagen), incubated with RQ1 RNase^– Dnase (Promega) at 37°C for 15 minutes, followed by ethanol precipitation and resuspension in Rnase^– water(RNase inhibitor, Ambion). RT reactions were performed using oligo-dT primers with Omniscript reverse transcriptase (Qiagen), according to the manufacturer's instructions. cDNA was used as a substrate for PCR amplification. Primers amplifying the hybrid transcript were: f1,5′-GGCAACTTCGACAAAGCCGAGG-3′; and r1,5′-CCAGGCAAATTGTCATCTGAGC-3′. Primers amplifying the neo-resistance (neo-r) gene sequence(Tybulewicz et al., 1991)were: 5′-CCGCTCGAGCGGACTTACAGCGGATCCCCTCA-3′ and 5′-GCTCTAGAGCCTTGCTCCTGCCGAGAAAG-3′.

Quantitative real-time PCR was performed with an ABI Prism 7700 sequence detector (PE, Applied Biosystems). PCR primers 5′-GGCAACTTCGACAAAGCCGAGG-3′ and 5′-CCAGGCAAATTGTCATCTGAGC-3′, and the Taqman probe 5′-CAGGCGTCGCTCGTCGGGGCT-3′ were designed to target the splice junction between Gbx2 exons 1 and 2. The PCR primers 5′-GCATGGCCTTCCGTGTTCCTA-3′ and 5′-CTGCTTCACCACCTTCTTGA-3′, and the TaqMan probe 5′-ACGTGCCGCCTGGAGAAACCTG-3′ targeted a 105-bp fragment of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cDNA. The Taqman probe was labeled with 6-carboxyfluorescein (FAM) and quencher dye 6-carboxytetramethylrhodamine (TAMARA) on its 5′ and 3′ ends,respectively (Keystone Labs). PCR reactions were performed in a final volume of 50 μl and contained 2×Universal PCR Master Mix (Applied Biosystems). The final concentration of PCR primers and Taqman probes was 100 nM. Unknown samples were compared with duplicate samples containing 10⁷ to 10¹ copies of cDNA reverse transcribed from plasmid DNA containing complete Gbx2 or GAPDH ORFs. Cycling conditions were: 2 minutes at 50°C; 10 minutes at 95°C; 40 cycles of 95°C for 15 seconds, 60°C for 1 minute.

2H3 antibody supernatant (Developmental Studies Hybridoma Bank, University of Iowa) was used according to Hogan(Hogan et al., 1994), with slight modifications. Peroxidase deposits were visualized using solutions containing the substrate 4-chloro-1-napthol (Sigma-Aldrich), as previously described (Mark et al., 1993). Stained embryos were incubated in 30% ethanol overnight (4°C), then infiltrated with ethanol-glycerol (1:1) for 2 days. Proliferation was analyzed by examining histone H3 phosphorylation using a rabbit polyclonal antibody against the Ser10 phosphopeptide of histone H3 (Upstate).

Whole-mount RNA in situ hybridization (ISH) was performed as described(Cygan et al., 1997). To detect the hybrid transcript, a 497-bp cDNA fragment of neo-r sequences (Tybulewicz et al.,1991) was cloned into the pBluescript KS(+/–) vector(Stratagene, La Jolla, CA) and used to generate a riboprobe.

In mice carrying the original targeted Gbx2 allele, originally designated Gbx2^flox(Wassarman et al., 1997), the second exon encoding the homeobox was flanked with loxP sites(floxed) so that Cre-mediated recombination produces the null allele, Gbx2^Δhb. Gbx2^Δhb homozygotes die soon after birth and lack a cerebellum (Wassarman et al., 1997). Here, we describe the analysis of mice homozygous for the original targeted allele (prior to Cre-mediated recombination), which we have renamed Gbx2^neo, distinguishing it from a more recently generated Gbx2^flox allele(Li et al., 2002). Gbx2^neo homozygotes, like Gbx2^Δhb homozygotes, die soon after birth without nursing (data not shown). However, a comparison of mid/hindbrains from E17.5 control embryos and embryos homozygous for Gbx2^neo or Gbx2^Δhb(Fig. 1A-C) reveals that only the cerebellar midline (vermis) is absent in Gbx2^neohomozygotes. Thus, the cerebellar phenotype in Gbx2^neohomozygotes is less severe than that in Gbx2^Δhb homozygotes. We found that Gbx2^neo/Δhbtrans-heterozygotes also lacked the entire cerebellum, like Gbx2^Δhb homozygotes (data not shown).

The granule cell is the most abundant neuronal cell type in the developing cerebellum (Altman and Bayer,1997; Wingate,2001). Granule cell precursors are located on each side of the cerebellar midline, where they migrate rostrally over the cerebellar primordium between E11.5 and E13.5(Ben-Arie et al., 1997; Louvi et al., 2003). The external germinal layer (EGL) consists of the vermal EGL (vEGL) that disperses from caudal to rostral and the hemispheric EGL (hEGL), which migrates from lateral to medial where the two meet and fuse(Altman and Bayer, 1997).

To investigate whether normal EGL development occurs in Gbx2^neo homozygotes, we examined Math1expression. Math1 encodes a basic helix-loop-helix transcription factor specifically expressed in granule cell precursors and is required for the genesis of granule cells (Ben-Arie et al., 1997). In E11.5 control embryos, Math1-labeled granule cell precursors outline the two bilateral cerebellar plates (inset in Fig. 1A); however, in Gbx2^neo homozygotes Math1 expression is notably reduced in the cerebellar primordium and is absent in granule cell precursors lining the two adjacent vermal primordia up to the MHB (inset in Fig. 1B).

Histological analysis of E14.5 Gbx2^neo homozygous brains showed that a reduction specifically in the medial region of the developing cerebellum has occurred, including the ventricular cells, the differentiated mantle and the vEGL (Fig. 1D,E). To determine whether a reduction in cell proliferation contributes to this hypoplasia of the medial cerebellar primordium, we examined coronal sections of E14.5 brains by immunohistochemisty for phosphorylated histone H3, which marks mitotic cells(Brenner et al., 2003). In control E14.5 embryos, a uniform layer of mitotically active cells is present across the EGL (Fig. 1F),whereas these cells are markedly decreased specifically throughout the vEGL of Gbx2^neo homozygotes(Fig. 1G). Math1staining at E13.5 revealed a complete absence of midline EGL cells, consistent with this histological analysis (Fig. 1F,G, insets).

Fate-mapping studies in mouse and chick have demonstrated that the cerebellum is derived from r1 (Zervas et al., 2004; Wingate,2001) with the vermis and lateral hemispheres originating in the anterior and posterior portions of r1, respectively(Sgaier et al., 2005). Gbx2^Δhb homozygotes lack r1-r3,hence the lack of cerebellar development in these mutants(Wassarman et al., 1997). Because most of the cerebellum is present in Gbx2^neohomozygotes, most of r1 must also be present. To determine whether other rhombomeres develop in Gbx2^neo homozygotes, we analyzed embryos at early developmental stages by whole-mount RNA in situ hybridization, using probes for Fgf8, Krox20, Cyp26c1 and Hoxa2 mRNA. At early somite stages (ss), Fgf8 is expressed in r1 whereas Krox20 is normally expressed in r3 and r5(Voiculescu et al., 2001; Wilkinson et al., 1989). At E8.0, Cyp26c1, which encodes an enzyme that converts retinoic acid into polar metabolites, is specifically expressed in r2 (where it persists through E9.5) and r4, as well as in lateral mesenchyme flanking r3(Tahayato et al., 2003). Examination of Fgf8 and Krox20 expression at 14 ss in mutant and control littermates demonstrated that r1, r3 and r5 were present in Gbx2^neo homozygotes(Fig. 2A,B). However, the r2 domain defined by the unstained region between r1-specific Fgf8expression and r3-specific Krox20 expression(Fig. 2A) was absent(Fig. 2B). At 6-8 ss, we detected Cyp26c1 expression in two transverse stripes in the neuroepithelium of all control embryos, denoting r2 and r4(Fig. 2C). However, in mutants, Cypc26c1 transcripts were detected only in a single stripe of cells corresponding to r4, suggesting the absence of r2 in Gbx2^neo homozygotes(Fig. 2D). Finally, the anterior limit of Hoxa2 expression, which normally is located at the r1/r2 border (Prince and Lumsden,1994) (Fig. 2E),occurred at the anterior edge of r3 in Gbx2^neo homozygotes(Fig. 2F). Together, these data suggest that Gbx2^neo homozygotes lack r2.

If r2 is indeed absent in Gbx2^neo homozygotes, then r2-derived cranial nerves should not develop properly in these mutants. Motoneuron cell bodies of the trigeminal (nV) and facial (nVII) cranial nerves arise in r2/r3 and r4/r5, respectively(Chandrasekhar, 2004; Cordes, 2001). To determine whether development of the r2-derived nV occurs in Gbx2^neohomozygotes, we analyzed Krox20 expression in E10.5 embryos. Consistent with earlier findings (Maro et al., 2004; Voiculescu et al.,2001), we observed Krox20 expression in the neural crest-derived boundary cap cells of nV and nVII cranial nerves of E10.5 littermate controls, but trigeminal boundary cap cells were absent in Gbx2^neo homozygotes(Fig. 2G,H), consistent with the absence of r2.

The ganglion of nV is composed of sensory neurons derived from neural crest cells and placodal ectoderm, which populate the proximal and distal regions of the ganglion, respectively (Baker and Bronner-Fraser, 2001; Dunty et al., 2002). To determine whether development of these sensory neuron populations is impaired in Gbx2^neo homozygotes, we analyzed cranial nerves and ganglia at E10.5 by whole-mount immunohistochemistry. Neurofilament staining indicates that neural crest-derived sensory neurons forming the proximal ganglion of nV are absent in Gbx2^neo homozygotes, causing a discontinuity between nV and the hindbrain (compare Fig. 2I with 2J). The distal ganglion, which is placodal in origin, develops normally. We also observed that the mandibular branch of nV is absent in the first branchial arch in mutants (Fig. 2I,J). This is noteworthy because muscles in this location are required for proper jaw movements in suckling(Chandrasekhar, 2004; Cordes, 2001). Hence, an inability to suckle may account for the death of Gbx2^neohomozygotes.

Together, these data demonstrate that Gbx2^neohomozygotes lack r2 and its derivatives, as well as the cerebellar vermis. As these phenotypes are less extreme than that found in Gbx2^Δhb homozygotes (which lack the entire cerebellum and r1-r3), we conclude that Gbx2^neobehaves like a hypomorphic allele.

The selectable marker in the gene-targeting construct that generated the Gbx2^neo allele was a standard neo-resistance cassette(neo-r) inserted into the Gbx2 intron(Fig. 3A)(Wassarman et al., 1997). In other loci, a similarly embedded neo-r can result in altered splicing of RNA transcripts due to the use of cryptic splice sites within the neo-r cassette, possibly causing a reduction in wild-type mRNA levels(reviewed by Lewandoski,2001). To determine if this occurs in the Gbx2^neo allele, we performed RT-PCR using RNA from E10.5 wild-type and Gbx2^neo homozygous embryos. Using primers located in Gbx2 exons 1 and 2(Fig. 3A), we amplified a 732-bp cDNA product derived from E10.5 Gbx2^neo homozygous mutant RNA that was absent in wild-type cDNA(Fig. 3B). Sequence analysis showed this cDNA contained 497 bp of neo-r sequence, spliced between the normal Gbx2 exon 1 and 2 sequences(Fig. 3A). Thus, cryptic splice-acceptor and -donor signals within neo-r interfere with normal Gbx2 splicing, resulting in a reduction in wild-type Gbx2mRNA (Fig. 3B). As the number of bases in the neo-r segment is not a multiple of three, a frameshift occurs upstream of the homeobox, and presumably the hybrid mRNA encodes a non-functional protein.

Cre-mediated inactivation of Gbx2 starting at 8 ss and concluding at 15 ss (E9.0) causes a vermis deletion similar to that observed in Gbx2^neo homozygotes(Li et al., 2002). To determine whether Gbx2 mRNA levels were reduced prior to this time point in Gbx2^neo homozygotes, we performed whole-mount ISH at early head-fold stages (prior to formation of the first somite) and at 12-13 ss. A qualitative reduction in Gbx2 mRNA levels was detected(Fig. 3C). To quantify this reduction in Gbx2 transcripts, we performed TaqMan real-time quantitative PCR using total RNA from wild-type and mutant embryos at 6-8 ss(∼E8.5) and at 24 ss (∼E10.0). Gbx2 mRNA levels in mutants are reduced to 6-10% of that found in wild-type embryos at all stages examined(Fig. 3D), presumably because of the aberrant splicing caused by neo-r sequences. To determine whether this splicing defect occurs throughout normal Gbx2 expression domains during embryogenesis, we performed whole-mount ISH analysis using probes specific for Gbx2 mRNA sequences or for the neo-r sequences of the hybrid transcript. Because the neo-r cassette is inserted in the opposite orientation relative to the Gbx2transcription unit, this `neo-r' probe does not hybridize to neo-r mRNA driven from the PGK promoter in the selection cassette,and thus is specific to the hybrid neo-Gbx2 message.

The hybrid mRNA was detected in the MHB region at 10 ss and at E10.5, as well as in other CNS regions, such as the forebrain and prospective spinal cord (Fig. 4B,D,G,H). Furthermore, the hybrid mRNA was also detected in non-neuroectodermal Gbx2 expression domains, such as the branchial arches, otic vesicles and limb buds (Fig. 4C,D). Thus, a careful comparison of Gbx2 expression in wild-type embryos(Fig. 4A,C,F) with expression of the hybrid transcript in either Gbx2^neo heterozygotes(Fig. 4D,G) or Gbx2^neo homozygotes(Fig. 4B,H) demonstrates that the expression pattern of the hybrid transcript recapitulates that of Gbx2. One hypothesis we considered to explain the vermal deletion in Gbx2^neo homozygotes is that Gbx2 mRNA levels are more reduced in the primordium of the vermis than in the lateral cerebellar regions (Sgaier et al., 2005). However, we detected the hybrid transcript throughout the entire cerebellar primordium of both E10.5 Gbx2^neo heterozygotes and homozygotes, albeit with stronger staining detected in the latter(Fig. 4G,H). Because the neo-r cassette interferes with splicing throughout the cerebellar primordium but only affects midline development, these results suggest that the vermal primordium is more sensitive to a reduction in Gbx2transcript levels.

In total, our phenotypic, genetic and molecular analyses described above demonstrate that Gbx2^neo is a hypomorphic allele. Moreover, we demonstrate that anterior hindbrain abnormalities in Gbx2^neo homozygotes are due to reduced levels of wild-type Gbx2 gene expression caused by aberrant splicing defects that are attributable to the neo-r sequences.

Previous studies have shown that mutual antagonism between anterior hindbrain cells expressing Gbx2 and posterior midbrain cells expressing Otx2 is essential in refining gene expression within the isthmus (Liu and Joyner, 2001; Wurst and Bally-Cuif, 2001). We performed whole-mount ISH analyses to determine the effect of reduced Gbx2 levels on Fgf8, Wnt1 and other crucial genes expressed in the embryonic mid/hindbrain region shown to be essential in anterior hindbrain development (Rhinn and Brand,2001; Wurst and Bally-Cuif,2001). At early ss (9-10), we observed a broad caudal expansion of the expression domains of Fgf8, Wnt1 and Otx2 in Gbx2^neo homozygotes. The extension of the caudal Otx2 border results in its overlap with that the Gbx2expression domain (compare Fig. 4A with Fig. 5A,B). Fgf8, which is normally detected in a band of Gbx2-expressing cells located in the rostral end of the prospective hindbrain (Fig. 5C), is extended caudally and spans the entire r1 region in Gbx2^neo homozygotes(Fig. 5D). The Wnt1expression domain is extended into anterior rhombomeres where it is normally absent so that there is continuous expression from the MHB and through the mesencephalon (Fig. 5E,F).

At E10.5, the earlier broad Fgf8 and Wnt1 expression domains are normally reduced to two adjacent narrow transverse rings at the mid/hindbrain boundary (Fig. 6A,K). Importantly, Fgf8 expression is restricted to the isthmus, and, hence, is an isthmic marker(Mason et al., 2000; Crossley and Martin, 1995). In addition, other genes shown through gene inactivation studies to control cerebellar development, such as Fgf17, En1, En2 and Otx2,are expressed either at the mid/hindbrain boundary or in the posterior midbrain and anterior hindbrain domains in normal embryos(Rhinn and Brand, 2001; Wurst and Bally-Cuif, 2001)(Fig. 6C,G,E,I). Otx2and Wnt1 are misexpressed caudally, but instead of the broad expansion that occurs at earlier stages, the ectopic expansion of these markers is restricted to the dorsal midline and patches within the r1 alar plate, with Wnt1 displaying somewhat more extensive misexpression(Fig. 6J,L). En1 and En2 expression are abnormally extended through the dorsal midline of the isthmus and along the anterior region of r1(Fig. 6F,H). Likewise, the normal expression domains of Fgf8 and Fgf17 at the isthmus extend caudally through the dorsal midline of the isthmus and alar plate of r1, and mediolaterally into the anterior rhombic lip(Fig. 6B,D). Strikingly,mutants viewed laterally and hybridized for Fgf8 expression appear to have a `double isthmus' (Fig. 6A,B, insets).

Isthmic FGF signaling is thought to control a negative-feedback loop by inducing members of the Sprouty (Spry) gene family(Minowada et al., 1999), which act as negative regulators of the FGF-RAS-ERK pathway(Suzuki-Hirano et al., 2005). We examined the expression of the four Spry genes in E10.5 normal and Gbx2^neo homozygous embryos. Consistent with previous reports, we found that Spry1 and Spry2 are expressed in the posterior midbrain and anterior hindbrain, and that Spry3 expression was absent (Minowada et al.,1999) (Fig. 6M,data not shown). Surprisingly, we also detected Spry4 expression in midbrain and hindbrain strips flanking the isthmus in control embryos(Fig. 6O). The expression domains of both Spry1 and Spry4, but not Spry2,mimic the abnormal Fgf8 expression pattern in Gbx2^neo homozygotes, by extending caudally and mediolaterally into the anterior region of r1(Fig. 6N,P).

These data indicate that expression domains, encoding key regulatory molecules, such as OTX2 and WNT1, are expanded caudally at early somite stages in Gbx2 hypomorphic mutants, in a manner similar to that occurring in Gbx2^Δhb homozygotes(Millet et al., 1999; Wassarman et al., 1997). However, at E10.5, the ectopic expression of these genes more resembles that occurring in conditional mutants in which Cre-mediated recombination inactivates Gbx2 by E9.0 (Li et al., 2002). Likewise, the caudal expansion of robust Fgf8and Fgf17 expression, along with that of the FGF target genes Spry1 and Spry4, is reminiscent of the altered Fgf8domain in conditional mutants (Li et al.,2002).

The extension of the isthmic expression domains of Fgf8, Fgf17 and the FGF targets Spry1 and Spry4 into the anterior dorsal region of r1 suggest that this region may be converted into an isthmus-like tissue in Gbx2^neo homozygotes. To investigate this further, we examined the status of the upper rhombic lip in Gbx2^neo homozygotes by using Math1 as an indicator. Math1 is expressed in the dorsal midline of r1 and rhombic lip precursors at E10.5 (Fig. 7C). In Gbx2^neo homozygotes, Math1expression is absent in the anterior rhombic lip (compare Fig. 7D,F with 7C,E) where Fgf8 is abnormally expressed (Fig. 7B).

Cellular proliferation in the isthmus occurs at lower levels relative to the surrounding neuroepithelium (Fig. 7G) (Altman and Bayer,1997; Li et al.,2002; Trokovic et al.,2005). To determine whether this characteristic of the isthmus is also extended into the anterior region of r1, we examined saggital sections of E10.5 embryos immunostained with an antibody against a phosphorylated form of histone H3 (H3P) (Brenner et al.,2003) (Fig. 7G,H). We observed that the normal decrease in mitotically active cells at the isthmus was abnormally extended through the anterior region of r1 in Gbx2^neo homozygotes(Fig. 7H). We also examined the expression of the D-type cyclins (D1, D2 and D3), which may underlie the regulation of this altered proliferation, as these factors are key regulators of G1 exit, and their developmental expression patterns during neurulation and midgestion are well characterized(Lundberg and Weinberg, 1999; Ross et al., 1996; Wianny et al., 1998). We found that in control E10.5 embryos a robust increase of cyclin D2 expression occurs in r1 (Fig. 7I,K), in agreement with previously published data (Ross et al., 1996). Additionally, cyclin D2 expression is reduced in the isthmus of control E10.5 embryos (Fig. 7I,K). In Gbx2^neo homozygotes, cyclin D2 expression is diminished throughout r1 and thus the isthmic region of reduced expression is expanded caudally into the anterior region of r1(Fig. 7J,L). These data are consistent with the abnormal caudal expansion of the medial isthmus into r1 in Gbx2^neo homozygotes as a zone of reduced proliferation.

Here, we present genetic and molecular data demonstrating that Gbx2^neo is a hypomorphic allele, and, by analyzing Gbx2^neo homozygotes, reveal Gbx2 dosage requirements for the development of different hindbrain regions. In contrast to Gbx2 null homozygotes, which lack r1-r3, Gbx2^neo homozygotes have a less severe phenotype in that they lack only the cerebellar vermis and r2. A probable molecular mechanism causing the hypomorphic phenotype is the presence of the neo-r cassette in the intron of Gbx2^neo. neo-r interferes with normal Gbx2 splicing, resulting in a hybrid transcript lacking homebox sequences, and causes a reduction in the level of Gbx2transcripts to 6-10% of wild type. Thus, Gbx2 levels drop below a threshold required for normal development of the anterior region of r1 (the vermis primordium) and r2. However, Gbx2 levels below this threshold are apparently sufficient for normal development of the posterior r1 region(the primordium of the cerebellar lateral hemispheres) and r3.

Selectable marker cassettes, such as neo-r, are required to select targeted ES-cell clones for use in generating mouse lines carrying altered alleles. Placement of these cassettes in intronic, 5′ or 3′ UTR sequences often cause a partial loss-of-function allele (reviewed by Lewandoski, 2001), which can cause a novel hypomorphic phenotype or a threshold effect as shown here. Although the mechanism of gene interference is not usually determined, in those cases in which such studies were carried out, it has been determined that cryptic splice sites in the neo-r cassette obstruct normal gene splicing (Carmeliet et al.,1996; Jacks et al.,1994; Levin and Meisler,2004; Meyers et al.,1998; Nagy et al.,1998; Wang et al.,1999; Xu et al.,2001). In two different loci where both the targeted gene and neo-r were transcribed from the same DNA strand, identical cryptic splice sites were active (Meyers et al.,1998; Nagy et al.,1998). However, in Gbx2^neo and three other loci in which the direction of neo-r transcription opposes that of the targeted gene, different cryptic sites are active(Carmeliet et al., 1996; Jacks et al., 1994; Levin and Meisler, 2004). Thus, how an intronic neo-r cassette affects splicing is locus specific. In our case, we could not `rescue' the hypomorphic allele via Cre or FLP-mediated removal of neo-r because the targeted allele was not designed for this option (Wassarman et al., 1997). Thus, it remains a formal possibility that the loxP sites in Gbx2^neo cause the observed reduction in wild-type Gbx2 mRNA levels. This is very uncommon(Lewandoski, 2001) and is unlikely in this case, especially considering that the placement of loxP sites in Gbx2^neo is very similar to that in Gbx2^flox, which is genetically identical to a wild-type allele (Li et al., 2002).

This work furthers our understanding of the growing complexity concerning the role of Gbx2 in regulating rhombomere development. Previously, Gbx2 has been shown to be essential for the development of r1-r3 and their derivatives (Wassarman et al.,1997). More recently, it has been appreciated that r4-r6, although present, do not display normal gene expression in Gbx2 null homozygotes (Li et al., 2005; Lin et al., 2005). The role of Gbx2 in r3 development is to repress the caudal expansion of Otx2 into this region, as r3 is restored in Otx2^–/–; Gbx2 ^–/–double mutants (Li and Joyner,2001). Cre-mediated deletion of Gbx2 in the anterior hindbrain demonstrated that, with the exception of the cerebellar midline, Gbx2 is not required after E9.0 for r1 development(Li et al., 2002). In this study, we demonstrate that 6-10% of normal Gbx2 levels is sufficient to support development of the cerebellum and r3, but not development of the cerebellar midline or r2, demonstrating a lower threshold requirement for Gbx2 in these regions. Our data are consistent with a dual role for Gbx2 in anterior hindbrain development. First, as a cell-autonomous factor acting to maintain specific rhombomere development (r1-r3), with the most stringent requirement for Gbx2 function in r2. Second, as a regulator of the expression of both secreted and intrinsic factors, such as Fgf8 and Math1, which are involved in patterning and cell specification within the anterior hindbrain(Ben-Arie et al., 1997; Martinez et al., 1999).

A number of studies have shown that Gbx2 expression in the anterior region of r1 (metencephalon) determines the posterior limit of mesencephalic Otx2 expression at the MHB(Li and Joyner, 2001; Li et al., 2002; Martinez-Barbera et al., 2001; Wassarman et al., 1997). Moreover, mutual antagonism between anterior hindbrain cells expressing Gbx2 and posterior midbrain cells expressing Otx2 is essential in refining gene expression within the isthmus(Li and Joyner, 2001; Wassarman et al., 1997). Here,we show that at early somite stages, the MHB markers Otx2, Fgf8 and Wnt1 are expanded caudally in Gbx2^neohomozygotes, much as they are in Gbx2 null homozygotes(Millet et al., 1999). Yet,unlike mice lacking Gbx2(Wassarman et al., 1997), a cerebellum (lacking the midline) forms, indicating that although 6-10% of normal Gbx2 gene product cannot maintain normal expression of these markers, it is sufficient to maintain normal lateral cerebellar development despite these altered gene expression domains. At E10.5, when these expression patterns are normally refined with sharp borders, the expression of many MHB markers in Gbx2 hypomorphic mutants differs from that in both control embryos and Gbx2 null homozygotes, and more resembles mutants generated by the inactivation of Gbx2 by E9.0 by Cre-mediated recombination, which also results in deletion of the vermis(Li et al., 2002). In particular, the expression of Fgf8, Fgf17 and the FGF targets Spry1 and Spry4, instead of displaying the reduced and diffuse pattern occurring in the absence of Gbx2, is strongly expressed and expanded from the isthmus along the anterior alar plate of r1. Our data indicate that in Gbx2 hypormorphic mutants (and possibly in conditional mutants when Gbx2 is inactivated at E9.0) Gbx2levels are high enough for r1 formation, but are too low to maintain a proper MHB border. We propose this results in a change in specification of the r1 anterior region to isthmus-like tissue. Thus, this region undergoes a physical reduction, as does the normal isthmus(Zervas et al., 2004; Palmgren, 1921), and, as it is normally the primordium of the cerebellar midline(Sgaier et al., 2005), the vermis fails to form.

Consistent with this interpretation are the alterations we observed in Math1 and cyclin D2 expression. Math1 is normally expressed with an anterior border at the isthmus and, subsequently, in the EGL of the developing cerebellum (Akazawa et al.,1995; Ben-Arie et al.,1997). The anterior border of Math1 expression is shifted to the posterior r1 region in Gbx2^neo homozygotes. In control E10.5 embryos, we detect robust cyclin D2 expression in both anterior r1 and the midbrain, with a stripe of reduced expression in the isthmus, thus confirming the immunohistological analysis(Ross et al., 1996). In Gbx2^neo homozygotes, this strip of reduced expression is expanded caudally into the anterior portion of r1. As development proceeds,cyclin D2 is expressed in proliferative granule cell precursors in r1(Ciemerych et al., 2002; Ross et al., 1996; Wianny et al., 1998), and mice lacking cyclin D2 have fewer granule cells within the cerebellar EGL(Huard et al., 1999). Consistent with this observation, our histological and immunohistochemical data show a reduction in vEGL cells in Gbx2^neo homozygotes at E14.5.

The reduced cyclin D2 expression that occurs normally at the isthmus may contribute to a decrease in cellular proliferation that occurs in this region relative to the surrounding neuroepithelium(Altman and Bayer, 1997; Li et al., 2002; Trokovic et al., 2005). Consistent with our observation that the anterior region of r1 has been converted to isthmic-like tissue in Gbx2^neo homozygotes,the isthmic domain of reduced proliferation is also expanded caudally and thus is the cellular basis of the vermis deletion. This is consistent with a model for the origin of the vermis that postulates that an expansion of the anterior region of r1 by differential proliferation generates the cerebellar midline(Sgaier et al., 2005; Sidman and Rakic, 1982).

It is curious that we observed a caudal expansion of the expression domains of both Fgf8 and Fgf17 given that a deletion of the vermis is also observed in mutants where isthmic FGF signaling is reduced or obstructed, such as in Fgf17^–/–;Fgf8^+/– mutants (Liu et al., 2003), or in embryos in which MHB-specific Fgfr1has been inactivated by Cre activity(Trokovic et al., 2003). Similarly, the cerebellar midline is absent and Fgf8 expression is diminished in mutants that ectopically express Otx2 throughout the MHB (Broccoli et al., 1999). However, Li et al. (Li et al.,2002) also observed a vermis deletion correlating with an expansion of the Fgf8 expression domain that resembles, but is not identical to, that reported here. Thus, either a reduction or an increase in isthmic FGF signaling can be correlated with vermis deletion.

These conflicting data can be reconciled in two different models. In one model, the normal isthmic FGF signal is necessary for vermis development,possibly by maintaining proliferation within the adjacent vermis primordium(anterior r1). When this primordium also expresses Fgf8 [as in Gbx2^neo homozygotes or conditional Gbx2 mutants(Li et al., 2002)], FGF targets encoding antagonists of the FGF signaling pathway are upregulated,causing an inhibition of a subset of FGF target genes(Storm et al., 2003) necessary for vermis development. In support of this idea, we observed a caudal expansion of the FGF antagonists Spry1 and Spry4 into the anterior region of r1. However, when we assayed, by immunohistochemistry, the FGF-mediated activation of the RAS-extracellular signal-regulated kinase (ERK)pathway (Corson et al., 2003),we observed no diminution of signal (data not shown), suggesting that overall FGF signaling was not diminished.

We postulate in our second model that a minimal FGF signal is required for formation of the vermis, thus accounting for mutants where reduced FGF8 signaling occurs. Independent of this FGF signal, if the amount of Gbx2 gene product drops below a certain threshold, it is not sufficient to confer proper specification of the anterior region of r1, which acquires then isthmus-like characteristics, and therefore fails to form the vermis. Consequently, the expression domain of genes like Fgf8 is expanded caudally as an isthmic marker but is not the direct cause of the loss of the vermis. This Gbx2 threshold requirement exists after E9.0, as the Cre-mediated deletion of Gbx2 after this stage also causes a vermis deletion (Li et al.,2002). The testing of these two models awaits the development of a Cre mouse that can be used to inactivate a conditional Fgf8 allele(Meyers et al., 1998)specifically in the anterior region of r1, thus restoring a relatively normal Fgf8 pattern in a homozygous Gbx2^neo background,which will rescue vermis formation only in the first model.

This research was supported by the Intramural Research Program of the NIH,National Cancer Institute, Center for Cancer Research. We would like to thank Cindy Elder, Dr John Julias, Barbara Kasprzak, Jennifer Matta and Catherine Wilson for excellent technical assistance. We also thank Dr Chuck Sherr(cyclin D2) and Dr M. Petkovich (Cyp26c1) for providing probes. Finally we thank Dr Gail Martin, in whose lab this work began.