Main

The development of complex organs composed of different cell types frequently depends on reciprocal induction events occurring between distinct tissue layers that lie adjacent to one another in the embryo. The pituitary is a well studied case in point. It originates from two ectoderm-derived tissues, with the posterior lobe developing from an evagination of the ventral diencephalon, the infundibulum, and the intermediate and anterior lobes deriving from Rathke's pouch, an invaginating domain in the roof of the oral ectoderm1. In the mature gland, the posterior lobe contains axon terminal projections from two populations of hypothalamic neuroendocrine neurons. The anterior and intermediate lobes contain several different endocrine cell types whose proliferation, hormone synthesis and secretion are regulated by factors secreted from hypothalamic neuroendocrine neurons.

During early development the infundibulum has an inductive role in the formation of the pituitary. For example, the genetic ablation of this domain in Titf1-null mice results in the complete absence of the pituitary gland2. The essential signaling molecules made by the infundibulum are thought to be FGF8 and BMP4, as both are necessary to induce early development of Rathke's pouch. FGF and BMP signals are also required to control the pattern of differentiation of cell types derived from Rathke's pouch3,4.

Even subtle mutations that affect signaling pathways during early organogenesis can have profound effects on subsequent development and specification of mature cell types. These could arise in genes encoding the signaling molecules or their receptors, in transcription factors responsible for their expression or in genes required to specify the interacting tissues. In humans, SOX3, an HMG box protein (for review see ref. 5), is implicated in a syndrome of X-linked hypopituitarism and mental retardation6. In a single family whose males were deficient in growth hormone, a mutation in SOX3 was identified. The consequences of this mutation on the function of the protein are not known. X-chromosome deletions encompassing SOX3 are linked to several syndromes in humans, including mental retardation, but defects in pituitary function have not been reported7,8.

SOX3 is a single-exon gene on the X chromosome in all mammals and is thought to be the gene from which SRY, the Y-linked testis determining gene, evolved9,10. Based on sequence homology, however, SOX3 is more closely related to SOX1 and SOX2 (refs. 5,11). Together they comprise the SOXB1 subfamily and are coexpressed throughout the developing CNS11,12,13. To study the role of Sox3, we targeted null mutations in the gene into XY embryonic stem (ES) cells, but injection of these cells into blastocysts resulted in early lethality of the chimeras due to a gastrulation defect (M. Parsons, C. Wise, S.B., M. Cohen-Tannoudji, K.R., L. Pevny & R.L.-B., unpublished data).

Therefore, to access later functions of SOX3, we initiated experiments using a conditional targeting strategy. In contrast to the chimera experiments in which the targeted ES cells were from an inbred mouse strain (129Sv/Ev), mice with deletions of Sox3 were viable on outbred genetic backgrounds. Sox3-null mice have a range of phenotypes, including craniofacial abnormalities and reduced size and fertility. Because of the association between growth hormone deficiency and SOX3 mutations in humans, we focused our investigations on the role of Sox3 in pituitary development. We show here that Sox3 is required in the presumptive hypothalamus and infundibulum for correct morphogenesis of Rathke's pouch and later for correct function of the hypothalamic-pituitary axis.

Results

Targeted disruption of Sox3

To generate a Sox3 conditional allele, we used homologous recombination in XY ES cells (the 129Sv/Ev line, AB1) to replace the coding region with a construct containing the Sox3 open reading frame flanked by loxP sites (Fig. 1a). We inserted a marker gene encoding green fluorescent protein (GFP)14 downstream to allow expression of GFP driven by Sox3 regulatory sequences on Cre-mediated recombination (Fig. 1a,d).

Figure 1: Conditional targeted inactivation of Sox3.
figure 1

(a) Structure of the mouse Sox3 locus, targeting vector and recombinant alleles. The position of the primers used to genotype the wild-type (P1-P2) and Sox3Δgfp alleles (P3-P4) are indicated by arrowheads. (b) Southern-blot analysis of tail biopsies digested by BglII and hybridized with the 5′ genomic probe showing the expected fragments of 9 kb, 7.5 kb and 5 kb corresponding to the wild-type, Sox3flox.gfp and Sox3Δgfp alleles, respectively (see also a). (c) Genotyping of Sox3 wild-type (XY+/+), heterozygous (XX+/−) and homozygous null (XX−/−) mice by PCR amplification. The amplification product of 323 bp (using primers P1-P2) is specific to the wild-type Sox3 3′ untranslated region, and the product of 712 bp (using primers P3-P4) represents the Sox3Δgfp allele. (d) The live GFP activity domain in a 12.5-d.p.c. heterozygous (XX+/−) embryo was absent in a wild-type (XY+) embryo, but it matched the area of normal Sox3 expression by in situ hybridization (for Sox3). (e) Upper panel: a pair of 2-week-old Sox3-null littermates showing the variability of the phenotype. Bottom panel: a 3-week-old Sox3-null mouse (right) showing craniofacial asymmetry and growth defects, together with a wild-type littermate (left).

Five independent ES cell colonies were correctly targeted, and two of these gave germline transmission from chimeric mice. We verified the presence of the conditional allele, Sox3flox.gfp (flox indicates flanked by loxP sites), by Southern-blot analysis (Fig. 1b). To generate mice with a deletion of Sox3, we bred females carrying Sox3flox.gfp with transgenic males carrying a β-actin–Cre construct conferring widespread Cre expression15. We confirmed the presence of the Cre-recombined allele, Sox3Δgfp, by Southern-blot analysis (Fig. 1b) and by PCR (Fig. 1c). As expected, mutant XY embryos obtained from these matings showed no Sox3 expression by in situ hybridization (data not shown) but expressed GFP in a pattern consistent with that of Sox3 (Fig. 1d)11,12. Despite the complete absence of Sox3, we obtained live-born XY hemizygous null and XX heterozygous mice.

Phenotype of Sox3Δgfp mice

Null mutant males had a variable phenotype. Approximately one-third appeared normal, whereas the most seriously affected mice (Fig. 1e), which had poor growth and general weakness, did not survive to weaning. They also had craniofacial defects, including absence or abnormal shape and position of the pinna and overgrowth of teeth (K.R., unpublished data). Because Sox3 is X-linked, heterozygous females are necessarily mosaic with respect to the mutation. Some had mild craniofacial defects, but they otherwise appeared normal. We bred these females with the less severely affected mutant males to generate homozygous mutant females and to maintain the Sox3Δgfp mutation.

Genotypes of progeny arising from intercrosses between heterozygous mutant females and hemizygous males did not deviate substantially from a normal mendelian distribution when we collected embryos between 8.5 and 15.5 days post coitum (d.p.c.; Fig. 2), but 43% of mutant mice did not survive to weaning (Fig. 2). The sex ratio of the mutant mice did not deviate from normal. Sox3 might have a role in sex determination as a repressor of male development16, but we never observed any evidence for testicular development in homozygous mutant XX embryos at any stage. Therefore, Sox3, at least on this outbred genetic background, is not required for sex determination.

Figure 2
figure 2

Genotypes of offspring obtained from Sox3Δgfp heterozygous (XX+/−) and hemizygous (XY) intercrosses.

Although we did not observe any reduction of fertility in the less severely affected mice, histological examination of testes from the most severely affected males showed considerable abnormalities with some necrotic, empty tubules (Fig. 3a,b), and the ovaries of the most severely affected homozygous females were small, with a high incidence of atretic follicles (Fig. 3c,d). Sox3 is expressed in the adult testis in both humans and mice7,11. The protein was detected in mice in a subset of spermatogonia (Fig. 3e–j), probably type A (Supplementary Fig. 1 online). The absence of SOX3 during this specific phase of germ cell development could explain the abnormal testes phenotype, but we observed no staining for SOX3 in ovaries (data not shown). It is therefore possible that the defects in both testes and ovaries, together with the growth defects seen in these mice, could be explained by pituitary deficiencies. Testes of younger mice (Supplementary Fig. 1 online) were normal, consistent with a pituitary defect17.

Figure 3: Gonadal histology, pituitary hormone levels and growth hormone immunohistochemistry.
figure 3

(a,b) Sections through testes of 2.5-month-old wild-type (a) and Sox3-null males (b). Empty, necrotic tubules were present in the mutant (arrow). (c,d) Sections through ovaries of 2.5-month-old wild-type and mutant females. Some atretic oocytes were present (arrow). (ej) SOX3 immunofluorescence on testis sections of a 4-week-old wild-type mouse. (e) DAPI staining of a tubule. (f) Sox3 staining. (g) Merged image. (h) DAPI staining on a confocal image at high magnification. Sertoli cells are easily recognizable by the presence of distinct nucleoli (arrows). (i) SOX3 staining. (j) Merged image showing the presence of SOX3 in germ cells. (k) Growth hormone radioimmunoassay on pituitary protein extracts from seven wild-type (XY+) and five Sox3-null (XY) 2-month-old littermates. (l) Relationship between concentrations of luteinizing hormone (LH), FSH, growth hormone (GH) and TSH and body weight in five Sox3-null mutant mice. Concentrations on the y axis: luteinizing hormone, 10−7 g per pituitary; growth hormone, 10−6 g per pituitary; TSH, 10−8 g per pituitary; FSH, 10−8 g per pituitary. (m,n) Immunohistochemistry for growth hormone on pituitary sections from 6-week-old wild-type and Sox3 mutant males. The inset in n shows an extra cleft containing cells positive for growth hormone. Ant, anterior lobe; Int, intermediate lobe; Post, posterior lobe.

The pituitary gland is affected in Sox3 mutants

We measured concentrations of pituitary growth hormone, follicle-stimulating hormone (FSH), luteinizing hormone and thyroid-stimulating hormone (TSH) in 2-month-old Sox3 mutant and wild-type male littermates. The Sox3 mutant mice had substantially lower concentrations (about three times lower on average) of pituitary growth hormone than did wild-type littermates (Fig. 3k). Concentrations of luteinizing hormone, FSH and TSH were also lower. These endocrine deficits varied between individual mice and correlated with body weight (Fig. 3l). The range of growth hormone deficits could easily account for the variable post-weaning growth rates in the moderately affected mutant mice18, although the reduction in TSH could also contribute to this. Moreover, the reduction in gonadotrophins, either directly from a pituitary defect or as a secondary consequence of lower TSH levels, could contribute to the gonadal abnormalities.

Pituitary sections from wild-type mice immunostained for growth hormone showed clearly demarcated posterior, intermediate and anterior lobes, with heavy staining for growth hormone–producing somatotrophs (Fig. 3m,n). Except at the margins, the anterior lobe was separated from intermediate lobe cells by a single pituitary cleft, the remnant of the lumen of Rathke's pouch (Fig. 3m). In the Sox3 mutants (Fig. 3n), the anterior lobe was much smaller and the distinction between anterior and intermediate pituitary tissue was disrupted on one side by the presence of a second Rathke's cleft structure delineated by typical columnar lining cells. Between these clefts, an aberrant region contained growth hormone–positive anterior lobe cells mingling with intermediate lobe cells.

We then examined pituitary development in the Sox3 mutants. A sagittal section through the brains of wild-type embryos at 11.5 d.p.c. showed the typical structure of Rathke's pouch overlain by the infundibulum and the more anterior presumptive hypothalamus (Fig. 4a). In contrast, the pouch was expanded and bifurcated in mutant littermates (Fig. 4b). The evagination of the infundibulum was much less pronounced, and the presumptive hypothalamus appeared thinner and shorter in mutant embryos. All six mutant embryos sectioned and analyzed in this way had these defects to variable extents. Coronal sections from mutant mice at 12.5 d.p.c. showed that the floor plate of the diencephalon was markedly expanded relative to the typical V shape seen in wild-type embryos in the region overlying the distended bifurcated Rathke's pouch (Fig. 4c,d). Sections from mice at 14.5 d.p.c. (Fig. 4e,f) and 16.5 d.p.c. (Fig. 4g,h) showed the presence of extra Rathke's lumens, presumably originating from the bifurcated pouch, in mutant mice.

Figure 4: Abnormal morphogenesis of the hypothalamus, pituitary and midline CNS structures in Sox3 mutant mice.
figure 4

(a,b) Sagittal sections through the brains of 11.5-d.p.c. wild-type (a) and Sox3-null embryos (b). In the mutant, Rathke's pouch was dorsally expanded and bifurcated and the evagination of the infundibulum was less pronounced (arrow). (c,d) Coronal sections through the brains of 12.5-d.p.c. wild-type (c) and Sox3-null embryos (d). The morphology of the ventral hypothalamus and infundibulum (arrow) was abnormal in the mutant. (e,f) Sagittal sections though the brains of 14.5-d.p.c. wild-type (e) and Sox3-null embryos (f). An extra Rathke's pouch lumen was present in the mutant (arrows). (g,h) Sagittal sections though the brains of 16.5-d.p.c. wild-type (g) and Sox3-null embryos (h), confirming the persistence of an extra Rathke's pouch lumen in the mutant (arrows). (i,j) Sagittal sections through the heads of newborn wild-type (i) and Sox3-null (j) littermates. An extra lumen and cleft in Rathke's pouch was present in the mutant (arrow). (k,l) Transverse section through the brains of 3-week-old wild-type and Sox3 mutant littermates, showing dysgenesis of the corpus callosum in the mutant (arrows). Rp, Rathke's pouch; Inf, infundibulum; Hyp, presumptive hypothalamus; Cca, corpus callosum; Dhc, dorsal hippocampal commissure; Icf, intercerebral fissure. Scale bars: a (for aj), 0.1 mm; k (for k,l), 0.5 mm.

All three pituitary lobe tissues were present in mutant mice at birth, but extra Rathke's clefts and lining cells were evident (Fig. 4i,j). We also observed additional midline CNS abnormalities in the most severely affected mice, including dysgenesis of the corpus callosum, failure of the dorsal hippocampal commissure to cross the midline and apparent continuity of the intercerebral fissure with the third ventricle (Fig. 4k,l).

Altered expression of Fgf8 and Bmp4

Sox3 was not expressed in Rathke's pouch (Fig. 5a) but was present throughout the CNS and, notably, at high levels in the ventral diencephalon, including the infundibulum (Fig. 5a). The observed defects in the Rathke's pouch are probably a secondary consequence of the absence of Sox3 in this domain. We then looked at the expression of Bmp4 and Fgf8 in the infundibulum, two genes encoding secreted factors required for the development of Rathke's pouch3,4.

Figure 5: Altered domains of expression of Fgf8 and Bmp4 during early development of the pituitary in Sox3 mutants.
figure 5

(a) Sagittal section of a 10.5-d.p.c. wild-type embryo hybridized to Sox3. (b,c) Sagittal sections of 10.5-d.p.c. wild-type (b) and Sox3-null (c) embryos hybridized to Bmp4 (purple staining) and Isl1 (blue staining). In the mutant (c), the Bmp4 expression domain was expanded ventrally and fused to the more ventral Isl1 domain (compare double-headed arrows in b and c). (df) Sagittal sections of 10.5-d.p.c. wild-type (d) and Sox3-null (e,f) embryos hybridized to Fgf8. Fgf8 expression in the infundibulum (arrows) was expanded ventrally in the mutants (e,f). The extent of this extra Fgf8 expression domain correlated negatively with infundibular evagination (compare e and f). (g,h) Sagittal sections of 10.5-d.p.c. wild-type (g) and Sox3-null (h) embryos hybridized to Hesx1, showing an expansion of its expression domain (arrow) in the mutant Rathke's pouch (h). (i,j) Sagittal sections of 11.5-d.p.c. wild-type (i) and Sox3-null (j) embryos hybridized to Hesx1. Hesx1 was expressed in the bifurcated pouch in the mutant (j). (k,l) Sagittal sections of 11.5-d.p.c. wild-type (k) and Sox3-null (l) embryos hybridized to Bmp4. At this stage, the expression of Bmp4 in the mutant was more restricted dorsally than at 10.5 d.p.c. (m,n) Sagittal sections of 11.5-d.p.c. wild-type (m) and Sox3-null (n) embryos hybridized to Fgf8. Like Bmp4, Fgf8 expression in the mutant (n) was more dorsally restricted in the abnormal infundibulum. Rp, Rathke's pouch; Inf, infundibulum; Hyp, presumptive hypothalamus. Scale bars: a,b, (for bn), 0.1 mm.

We examined embryos at 10.5 d.p.c., 24 h before any morphological defect became apparent in Rathke's pouch in the Sox3 mutants. In the wild-type control (Fig. 5d), Fgf8 expression was limited to the infundibulum overlying the dorsal edge of Rathke's pouch. In contrast, Sox3 mutants showed Fgf8 expression in an expanded ventral territory (Fig. 5e,f). The extent of this territory varied among the mutant embryos and correlated negatively with infundibular evagination: the greater the infundibular evagination, the less the ventral expansion of Fgf8 expression (compare Fig. 5e and f). The expansion of Fgf8 expression could reflect the aberrant morphology of the infundibulum (Fig. 3) or a more general defect in the ventral diencephalon. We therefore decided to compare the expression of another infundibular marker, Bmp4, with that of a more ventral, presumptive hypothalamic one, Isl1 (ref. 19) using double in situ hybridization. In sections from control mice (Fig. 5b), as for Fgf8, expression of Bmp4 was restricted to the infundibulum above Rathke's pouch, whereas Isl1 was present more ventrally, overlying the oral ectoderm. Mutant mice showed an expansion of the Bmp4 expression domain, such that it was in direct contact with the Isl1 expression territory (Fig. 5c). This result suggests that the expansion of Bmp4 and Fgf8 could be the result of ectopic expression. This expansion was only transient, because the expression of the two molecules was restricted more dorsally in Sox3-null embryos at 11.5 d.p.c., by which time the morphological defects were obvious (Fig. 5k–n).

This transient expansion of the domain responsible for inducing Rathke's pouch would be expected to have molecular consequences in the responding tissue. We therefore looked at the expression of the transcription factor Hesx1, whose expression is restricted to Rathke's pouch at 10.5 d.p.c. (ref. 20; Fig. 5g). In mutant mice (Fig. 5h), the Hesx1 expression domain was expanded anteriorly. After separation from the oral ectoderm at 11.5 d.p.c., Hesx1 was expressed throughout Rathke's pouch in wild-type mice (Fig. 5i) and in the bifurcated pouch in mutant mice (Fig. 5j).

Mosaic analysis in XX Sox3Δgfp heterozygous embryos

Despite the widespread expression of Sox3 in the CNS, the main defects observed in mutant embryos were restricted to specific territories. Are these territories more sensitive to subtle modifications that did not cause obvious defects elsewhere, or does this phenomenon reflect a particular requirement for SOX3 in specific regions? Because Sox3 is X-linked and only one copy of the X chromosome is active in somatic cells20, XX embryos heterozygous with respect to Sox3Δgfp should have mosaic expression of the wild-type and mutated alleles. Such mosaicism should be reflected in the activity of the GFP reporter in territories where Sox3 is normally expressed. Any nonrandom distribution of GFP-positive cells would indicate differences in properties between cells expressing Sox3 and those not expressing Sox3.

We first assessed the distribution of GFP immunofluorescence in Sox3-null embryos at 10.5 d.p.c. (Fig. 6a,c). As expected, all cells in the CNS were GFP positive and the infundibulum and hypothalamic region had an intense signal (Fig. 6b). In heterozygous (XX+/−) embryos (Fig. 6d,f), we observed a bright, seemingly homogenous GFP-positive region (Fig. 6e) in this territory, whereas the rest of the CNS had the speckled pattern expected from a mosaic X-inactivated allele (Fig. 6e). Examination of higher magnification, confocal sections confirmed the ubiquitous expression of GFP in the Sox3-null embryos (Fig. 6g,l), both in the infundibulum and hypothalamic region (Fig. 6g–i) and in the forebrain (Fig. 6j–l). They also showed that GFP-positive, Sox3-negative cells were the main, if not only, cells present in the infundibulum and hypothalamic region of heterozygous embryos (n = 8; Fig. 6m–r). Examination of heterozygotes at later stages showed that this GFP-positive domain persisted, such that by 15.5 d.p.c. it was present in the hypothalamus and infundibulum (Supplementary Fig. 2 online).

Figure 6: GFP expression from Sox3Δgfp in XY hemizygous and XX heterozygous embryos and pituitary defects in heterozygotes.
figure 6

(af) Confocal imaging of GFP immunofluorescence on sagittal sections of 10.5-d.p.c. Sox3-null (ac) and heterozygous (df) embryos. (gr) High-magnification confocal imaging of two regions of the same sections. (gi,mo) Images through the infundibulum and hypothalamic region (corresponding to the white arrow in e) of Sox3-null (gi) and heterozygous (mo) embryos. (jl,pr) Images through the forebrain (region corresponding to the white arrowhead in e) of Sox3-null (jl) and heterozygous (pr) embryos. In the Sox3-null embryos, the infundibulum, hypothalamic and forebrain regions were uniformly stained for GFP (compare h and k), whereas in Sox3 heterozygotes, GFP expression was mosaic in the forebrain region but uniformly bright in the infundibulum and hypothalamic region (compare q and n). (s) Sagittal section through the brain of an 11.5-d.p.c. heterozygous embryo. Rathke's pouch was expanded and bifurcated (arrow). (t) Transverse section through the pituitary gland of a 3-month-old heterozygous female showing the presence of extra clefts (arrows). Scale bars, 0.1 mm.

The presence of the GFP-positive patch at 10.5 d.p.c. confirmed that there is a specific requirement for Sox3 in the infundibulum and hypothalamic region. Furthermore, if aberrant signaling from this region is responsible for the morphological abnormalities of Rathke's pouch in null embryos, then XX+/− embryos should also have a bifurcated Rathke's pouch, as signaling from this coherent patch of Sox3-negative cells should be similarly abnormal. Histological analysis of embryos at 11.5 d.p.c. confirmed the presence of a bifurcated Rathke's pouch in three of four heterozygous embryos. Moreover, three of four adult XX+/− mice had abnormal pituitary morphology (Fig. 6s,t), similar to that seen in null mice, suggesting that the patch of Sox3-negative, GFP-positive cells was responsible for inducing the dysmorphic Rathke's pouch.

The nonrandom distribution of GFP-positive cells could be due to altered rates of cell proliferation. We therefore used 5-bromodeoxyuridine (BrdU) to label dividing cells at 10.5 d.p.c. Very few cells in the GFP-positive patch in heterozygotes showed BrdU staining compared with cells outside this region (Fig. 7). This was not seen in wild-type controls. We counted the number of BrdU+ cells in two restricted areas, in the GFP-positive patch and adjacent to this in the CNS, in heterozygous embryos and in equivalent regions in controls. In the wild-type controls (n = 5), the distribution of BrdU+ cells was homogeneous across the infundibulum and hypothalamic region with 47.6% of BrdU+ cells located in the region corresponding to the patch, whereas in heterozygous embryos (n = 4), only 23.6% of the BrdU+ cells were located in the GFP-positive patch (mean ± s.e. for angular transformation: wild-type, 43.6 ± 1.6%; heterozygotes, 29.1 ± 3.43%; P < 0.005). The percentage of BrdU+ cells outside the patch in mutant embryos was similar to that in controls, indicating that proliferation was reduced by a factor of 3 within the GFP-positive patch. By TUNEL assay, we observed no differences in the cell death patterns in the GFP-positive domain in heterozygous and wild-type embryos (Supplementary Fig. 3 online). These results confirm that within this domain, cells lacking SOX3 have properties different from those with SOX3 and indicate that the protein is necessary to maintain normal rates of proliferation in this population of neuroepithelial precursors.

Figure 7: Proliferation in the infundibulum and hypothalamic region of XX heterozygous Sox3Δgfp embryos.
figure 7

(ad) BrdU staining and GFP expression in the infundibulum and hypothalamic region of a 10.5-d.p.c. heterozygous (XX+/−) Sox3Δgfp embryo. In the GFP-positive patch (bracket), few cells showed BrdU staining compared with cells outside this region. (eh) BrdU staining in a corresponding domain (bracket) of the infundibulum and hypothalamic region of a wild-type embryo showed a homogenous BrdU staining pattern.

Sox3 expression persists in the postnatal hypothalamus

Although heterozygotes had morphological abnormalities in Rathke's pouch (Figs. 3n and 6t), there were no phenotypic defects with respect to growth or fertility, and concentrations of pituitary growth hormone in 2-month-old heterozygotes were similar to those in wild-type littermates (40.9 ± 4.1 versus 39.1 ± 4.8 μg growth hormone per pituitary, respectively; n = 7 per group). This could be because the defects in Rathke's pouch were less severe in the heterozygotes or because there were additional defects in the null mice. Production of pituitary growth hormone is controlled by hypothalamic trophic factors, but these effects are mostly manifested postnatally21,22. Therefore, to assess whether the morphological defects compromised pituitary function before hypothalamic input, we measured growth hormone levels at 18.5 d.p.c. and found these to be identical (3.5 ± 0.04 μg) in pituitary extracts from mutants (n = 5) and wild-type littermates (n = 9). Therefore, there must be additional defects in the null mice, probably at the level of hypothalamic control.

We therefore carried out histological examination of the mature hypothalamus. In sections from wild-type mice (Fig. 8a), the cellular densities corresponding to several nuclei (e.g., paraventricular or arcuate (ARC)) were visible. These were recognizable in the Sox3 mutant mice (Fig. 8b) with no obvious abnormalities, although the general cellular density was slightly reduced in medial regions of the hypothalamus. We also examined the hypothalamic ARC–median eminence region for GFP expression postnatally. We observed abundant GFP-positive cells throughout the ARC, including the ventrolateral zones, in the Sox3 mutants (Fig. 8c,d). At higher magnification, GFP could be seen in cell bodies and in axon projections throughout the median eminence (Fig. 8d). GFP-positive cells also delineated the ependymal lining of the third ventricle. We observed similar expression in wild-type brains by in situ hybridization for Sox3 (data not shown).

Figure 8: Histological analysis and GFP immunofluorescence of the hypothalamus in the postnatal brain of Sox3 mutants.
figure 8

(a,b) Transverse sections through the hypothalamus of postnatal day 5 (P5) wild-type (a) and Sox3-null (b) male littermates. The general cellular density was reduced in the mutant. (c,d) Detection of GFP expression by immunofluorescence on a transverse section through the hypothalamus of a P3 Sox3-null mouse. DAPI image (c) and GFP immunofluorescence (d) were present throughout the ARC. Confocal high-magnification images of the median eminence (ME; insets in c and d) showed GFP staining throughout the median eminence. 3rdV, third ventricle.

Discussion

Mice with a deletion of Sox3 have pleiotropic phenotypes, including craniofacial abnormalities, midline CNS defects and hypopituitarism. We show that Sox3 activity is required in the ventral diencephalon for morphogenesis and differentiation of the infundibulum and hypothalamus and to correctly induce the morphogenesis of Rathke's pouch. Although pituitary development is abnormal in Sox3-null mice, this may not in itself be sufficient to explain the hypopituitarism, which is probably due to additional defects at the level of hypothalamic control. We observed a slight reduction in the number of cells in the hypothalamus, but the lack of SOX3 could also compromise the function of specific neurons in the ARC, in which its expression is normally maintained, without having much effect on their survival. The ventral ARC–median eminence region also selectively maintains Sox3Δgfp expression. As the primary role of this structure is to convey signals from the hypothalamus to the pituitary, a Sox3-dependent deficit in this region could make a continuing contribution to the dysfunction of an abnormal pituitary.

There is wide variation in the severity of phenotypes in Sox3-null mice, for which there are at least two explanations. First, there is variation due to their mixed genetic background. Fewer mice were born as we backcrossed the Δgfp mutation onto the 129sv/ev strain, suggesting that some were lost at embryonic stages (data not shown). This is consistent with our previous studies in which Sox3-null 129sv/ev XY ES cells generated chimeras that died from gastrulation defects (M. Parsons, C. Wise, S.B., M. Cohen-Tannoudji, K.R., L. Pevny & R.L.-B., unpublished data).

Second, SOX3 has a largely indirect effect on pituitary morphogenesis: development of Rathke's pouch is disrupted despite Sox3 not being expressed in the pouch. Thus, the primary defect in the Sox3Δgfp mutants must lie in the inducing tissues of the ventral diencephalon, as shown by the flattened morphology and expanded domains of Bmp4 and Fgf8 expression. The defects in Rathke's pouch are almost certainly due to overinduction, as they are similar to, though less severe than, those observed when FGF8 was expressed ectopically in Rathke's pouch4. Other inducing molecules may also be involved. Supporting the overinduction explanation, the expression domain of the homeobox gene Hesx1 in this domain was expanded at the same time. Similar Rathke's pouch phenotypes have been reported in mice with a deletion of Hesx1, and mutations of the homolog in humans are responsible for some cases of septo-optic dysplasia23. But Hesx1 is also expressed earlier, during the formation of the prospective forebrain, and so its absence from the CNS may be indirectly responsible for the malformed Rathke's pouch, as shown for Sox3.

An independent Sox3 mutation in mice results in a similarly variable phenotype24, which included defects in growth, craniofacial development and fertility, similar to our results described here. But the authors did not report any pituitary abnormalities, focusing instead on the gonadal abnormalities. With respect to the latter, they concluded that SOX3 was expressed in Sertoli cells, and we interpret their data to be consistent with ours, which shows that the protein is found in spermatogonia.

The three Soxb1 genes are expressed in neuroepithelial progenitor and stem cells from the earliest stages. Some direct evidence that they are involved in neuroectoderm induction and maintenance has come from studies in fruit flies, frogs and chicks25,26, suggesting that mutations in Soxb1 genes in mice will probably affect the formation of the CNS.

The lack of SOX3 led to dysgenesis of the corpus callosum, but the infundibulum and prospective hypothalamus were present, although they were thinner and flatter. Because SOX1 and SOX2 continued to be expressed, any effects were likely to be subtle. We therefore examined XX Sox3Δgfp heterozygotes, taking advantage of X inactivation to compare the properties of cells expressing the wild-type Sox3 allele with those of cells expressing Sox3Δgfp within the same embryo.

We expected that cells expressing the mutated allele might be competed out in areas where SOX3 function is crucial, especially if cell proliferation was compromised. But we observed a patch of Sox3-negative, GFP-positive cells specifically in the region of the infundibulum and presumptive hypothalamus. Overproliferation could explain an apparent coherent patch, but in fact these cells had a marked reduction in proliferation, consistent with the association of Soxb1 gene activity in dividing neuroepithelial cells. Alternatively, cells expressing Sox3 could preferentially differentiate and migrate away or the lack of Sox3 could alter the adhesive properties of cells in this region. We looked at some neuronal differentiation markers (N-cadherin, Hes5 and Mash1) but observed no obvious precocious differentiation of cells throughout the region (data not shown). The remaining explanation is that the cells in the GFP-positive patch have properties that promote their separation from surrounding cells expressing Sox3.

The reduction in proliferation provides a simple explanation for the failure of the infundibulum to fully evaginate and for the subsequent flattening and stretching of the region that also includes the prospective hypothalamus. This may account in part for the enlarged domains of Fgf8 and Bmp4 expression and could explain the presence of a shorter interval between the apparently expanded infundibulum and the Isl1-positive presumptive hypothalamus. Although we used Isl1 expression as a landmark, however, it is conceivable that the lack of SOX3 led to an expansion of its normal territory. There could be a more general alteration in rostral-caudal patterning in this region.

The abundance of Sox3 expression in a group of ventral hypothalamic cells and the median eminence postnatally is notable, as these structures are implicated in the control of pituitary function. The distribution of GFP-positive cells does not suggest restriction to any single type of neuroendocrine cell. But the ventrolateral distribution of GFP-positive cells overlaps that of growth hormone–releasing hormone cells, whose product directly regulates secretion of pituitary growth hormone27. Compromise of the number, connectivity or activity of growth hormone–releasing hormone cells in the absence of SOX3 postnatally would cause functional growth hormone deficiency. GFP-positive cells were scattered in the hypothalamus of heterozygotes, unlike the clustering pattern seen in the posterior pituitary, and so the cells expressing SOX3 are probably sufficient to achieve normal pituitary function.

The hypopituitary phenotype of the Sox3-null mice resembles that described for humans with an 11-mer polyalanine tract expansion in SOX3. Some of these individuals also have craniofacial abnormalities that could relate to those in the mutant mice6. Another group of individuals with X-linked hypopituitarism have duplications of the chromosomic region including SOX3, and their phenotype is probably due to overexpression of the gene28. As these individuals do not have craniofacial defects, the polyalanine tract expansion is probably a loss-of-function mutation. The wide phenotypic variation of Sox3 mutant mice is also evident in both sets of affected humans. Presumably, as in the mice, this variation is due to a combination of genetic background effects and the indirect nature of the final phenotype.

Our results have several other implications for hypopituitarism in humans. If both gain- and loss-of-function mutations can lead to hypopituitarism, then pituitary development must be very sensitive to SOX3 dosage. SOX3 should therefore be considered a good candidate, not only for association with X-linked inheritance, but also for sporadic cases of hypopituitarism. As these are more prevalent in boys than girls29, SOX3 may be a good candidate for unexplained hypopituitarism in boys. Our studies in mice also highlight the potential involvement of hypothalamic abnormalities in Sox3 mutants, in addition to their pituitary dysmorphogenesis. This may be difficult to detect with magnetic resonance imaging in individuals with SOX3 mutations, although some of these individuals have an ectopic posterior pituitary (P.Q.T., unpublished data; M. Dattani, personal communication). Because the severity of the phenotype is so variable in mice, the expected phenotype for humans with SOX3 mutations could be quite broad, ranging from mild growth hormone deficiency to severe panhypopituitarism with CNS midline abnormalities. Examination of such individuals might also identify other CNS midline defects, as noted in the mutant mice. Moreover, we would predict that apparently healthy mothers of affected XY offspring are likely to have abnormal Rathke's pouch derivatives, even if this is not reflected in overt hormone deficiencies.

Methods

Targeting constructs, ES cell transfection and germline transmission.

The targeting vector was based on a 7.5-kb EcoRI-BglII fragment of genomic Sox3 (Fig. 1a). We inserted a loxP site at the XhoI site, downstream of the TATA box, creating an additional XmnI site in the locus. We cloned the GFP coding sequence, a gift from M. Zernicka-Goetz14 (Wellcome Trust/Cancer UK Gurdon Institute of Cancer and Developmental Biology, Cambridge, UK) downstream of a loxP site and upstream of an SV40 polyA site. We placed the pgk-neor cassette between two direct FRT sequences in the FRT2 vector30. We then cloned this sequence downstream of the GFP coding sequence in the loxP-GFP construct, in opposite transcriptional orientation, creating the final selection cassette. We introduced this cassette into the SpeI site of the previously modified genomic fragment, downstream of the Sox3 polyA sequence. Both loxP sites are in the same orientation.

We electroporated the targeting vector into AB1 XY ES cells31, derived from the black agouti 129sv/ev mouse strain. We screened 500 G418-resistant clones by Southern blotting. We identified correctly targeted ES cell clones by the presence of BglII fragments of 7.5 kb instead of 9 kb (Fig. 1b), by the presence of BglII-XmnI fragments of 3.5 kb instead of 4 kb and by the presence of BamHI fragments of 14 kb instead of 11 kb (data not shown). Five ES clones carried the Sox3neo floxed allele. The sequences of the primers used for genotyping are available on request.

Generation and analysis of chimeric embryos.

We injected targeted ES cells into 3.5-d.p.c. C57BL/6 (B6) blastocysts and transferred them into 2.5-d.p.c. CBA × C57BL/10 pseudopregnant mothers. We obtained high-percentage chimeras (70–100% agouti pigmentation) for all the Sox3flox.gfp clones. Male chimeras were bred to MF1 females, and their agouti pups were genotyped to confirm germline transmission of the targeted allele (Fig. 1). We confirmed germline transmission of the targeted allele in offspring from two targeted ES cell clones.

Mice.

We maintained the Sox3 conditional mutation on MF1 (random bred) and 129SvEv inbred backgrounds (NIMR). The β-actin–Cre line15 was maintained on a MF1 background and was used to obtain the Sox3 deleted allele (Sox3Δgfp) in embryos and mice. We obtained fertile hemizygous male (XY) and heterozygous females (XX+/−) directly in this way, on a largely MF1 outbred background, and then bred them to derive embryos and mice carrying the Sox3Δgfp allele. The selection cassette was present in the mice used in this study. We carried out all mouse experiments using protocols approved under the UK Animal (scientific procedures) Act.

Histology, in situ hybridization and immunofluorescence.

For histology, we fixed embryos in Bouin's, dehydrated them through graded ethanol series and embedded them in paraffin. We stained sections with hematoxylin and eosin as described32. We carried out whole-mount in situ hybridization of embryos as described33. We carried out single and double in situ hybridization on frozen sections as described34, using Sox3 (ref. 35), Bmp4 (ref. 36) Fgf8 (ref. 37) and Isl1 (ref. 19) probes. For frozen sections, we fixed the embryos in 4% paraformaldehyde for up to 2 h at 4 °C, cryoprotected them in 20% sucrose and embedded them in OCT (BDH). We analyzed immunofluorescence on 12-μm frozen sections as described38 using a 1:1,000 dilution of an antibody to GFP (Molecular Probes) and a 1:500 dilution of a goat antibody to rabbit conjugated to Alexa-488 or Alexa-594 (Molecular Probes). Confocal images were obtained using a Leica TCS SP microscope and TCSNT software.

Radioimmunoassay and growth hormone immunohistochemistry.

We homogenized anterior pituitary glands in phosphate-buffered saline and assayed them for growth hormone, luteinizing hormone, FSH and TSH by specific radioimmunoassays using NHPP reagents provided by A.F. Parlow (National Hormone & Peptide Program, Torrance, California, USA). For growth hormone immunohistochemistry, we used a 1:2,000 dilution of a monkey antibody to rat growth hormone serum from the NHPP as previously described39.

BrdU incorporation.

We injected pregnant females intraperitoneally with 100 μg of BrdU (Sigma) per g of body weight and killed them after 1 h. We processed embryos as described above for immunofluorescence. After 30 min post-fixation in 4% paraformaldehyde, we incubated sections for 1 min in 0.05% pepsin (Sigma) in 0.01 N HCl at 37 °C. BrdU labeling was exposed by a 30-min treatment with 2 N HCl at 37 °C, neutralized by three washes in 0.1 M Borate, pH 8.5. We then incubated sections overnight at 4 °C with a 1:100 dilution of a fluorescein-conjugated monoclonal antibody to BrdU (Boehringer Mannheim). Under the conditions used for the BrdU staining, GFP fluorescence was undetectable. We counted BrdU+ cells by selecting a square that fit the narrowest part of the region to be examined and counting BrdU+ cells inside this square. We did this five times per section in each of the two areas to be examined, inside a region corresponding to the GFP-positive patch and directly above it. We counted one section for each embryo and calculated the percentage of proliferating cells in the GFP-positive patch. After angular transformation of the percentages40, we estimated the significance of the difference in proliferation between the two genotypes using Student's t-test.

Note: Supplementary information is available on the Nature Genetics website.