Integration site of the GFP transgene
We have previously reported that the GFP transgene used in this screen has integrated as an approximately 11 copy array on chromosome 1 [
10]. We were keen to further characterize the integration site. PCR using primers specific to the 3' end of the transgene construct in combination with degenerate random tagging primers (genome walking) revealed that the transgene had integrated into chromosome 1 at 47,747,486 bp (UCSC web browser, July 2007 assembly). This site of integration is neither centromeric nor telomeric, and so presumably the silencing is related to the multicopy nature of the transgene array [
13,
14]. The integration site does not disrupt any annotated genes, and is approximately 1 Mb from the closest annotated transcript.
The identification of MommeD7-D10
We have now screened an additional 400 G1 offspring and recovered four more
Mommes,
MommeD7-D10 (Figure , Table ). The fluorescence activated cell sorting (FACS)-based screening is carried out on a drop of blood taken at weaning, using a gate set to exclude 99.9% of autofluorescing cells. Under these conditions, wild-type mice homozygous for the transgene express GFP in 55% of erythrocytes.
MommeD7 is a suppressor of variegation, that is, the percentage of cells expressing the transgene was significantly higher than it was in wild-type individuals (Table ).
MommeD8,
D9 and
D10 are enhancers of variegation, that is, the percentages of expressing cells were significantly lower than they were in wild-type individuals (Table ). The mean fluorescence level in expressing cells also changed. We have reported previously that as the percentage of expressing cells increases, the mean fluorescence of the expressing cells increases [
10]. We presume that since the mice are homozygous for the GFP transgene, this is mainly due to an increase in the proportion of expressing cells with two active GFP alleles. However, in the case of
MommeD7 the level was more than double that seen in the wild-type littermates. We hypothesized that this was likely to be the consequence of an increase in the percentage of reticulocytes in the peripheral blood of this mutant, as mature red cells, with no ability to produce new proteins, would have, on average, less GFP than reticulocytes (see below). In the cases of
MommeD8,
D9 and
D10 the mean fluorescence levels were slightly lower than that seen in the wild-type littermates, consistent with a presumed reduction in the proportion of cells with two active GFP alleles.
| Table 1Quantitative analysis of GFP expression following heterozygous intercrosses |
For each
MommeD, the heritability of the mutation has been tested and confirmed over at least 5 generations by using at least 30 litters. During the breeding of each mutant line, the expression profiles remained constant, consistent with the idea that we were following a single mutation in each case. The frequency with which we found these mutations, 1 in 100 G1 offspring, was similar to our previous results [
10].
Homozygous lethality
Following heterozygous intercrosses, the proportion of expression types observed in the offspring at weaning was consistent with a semidominant homozygous lethal mutation in the cases of MommeD7 and MommeD9, since only two GFP expression profiles were observed (Figure , Tables and ) and there was a significant litter size reduction in both cases (Figure ). In the cases of MommeD8 and MommeD10, three expression profiles were observed, suggesting viability of some homozygotes (Figure , Tables and ) and in the case of MommeD10 this was later confirmed by genotyping for the point mutation. In both cases, fewer individuals with the severe phenotypes were observed than predicted by Mendelian inheritance, suggesting reduced viability of the homozygotes (Table ). This conclusion is supported by significant litter size reductions in both cases (Figure ). There is also a suggestion of some heterozygous death in the case of MommeD8 and, to a lesser extent, MommeD10 but this is not statistically significant.
| Table 2Genotype ratios of offspring at weaning following heterozygous intercrosses |
Homozygous lethality occurs at different stages of development
Litter size reductions following heterozygous intercrosses have already been reported for
MommeD1-D6 at weaning [
10], but the timing of the losses has only been reported for
MommeD1,
D2 and
D4.
MommeD1 is homozygous lethal in females only, with death occurring around mid-gestation [
10,
11].
MommeD2 and
MommeD4 are homozygous lethal at 8.5 days post-coitus (dpc) and 17.5 dpc, respectively [
10,
12]. Here we describe the timing of the losses for
MommeD3,
D5,
D6,
D7,
D8,
D9 and
D10.
Following intercrosses between MommeD3-/+ (genotypes determined by GFP fluorescence and progeny testing), dissections at 14.5 dpc suggested that death of homozygotes had already occurred (Table ). This was confirmed following a FVB/C57 F1 MommeD3-/+ intercross, where embryos could be genotyped using microsatellite markers across the linked interval (Table ). These data suggest MommeD3-/- mice were dying at or before 14.5 dpc. Similar results were obtained for MommeD5 at 14.5 dpc (Table ). Once the MommeD5 point mutation had been found (see below), these crosses were repeated and dissections were performed at 10.5 dpc. Again, a significantly higher than expected proportion of developmentally delayed embryos were detected (Table ). These embryos were genotyped and found to be MommeD5-/- in all cases, indicating developmental arrest at around 8-9 dpc. Results obtained for MommeD6 (genotypes determined by GFP fluorescence and progeny testing) were similar (Table ), suggesting MommeD6-/- embryos arrest around 8-9 dpc.
| Table 3Embryo dissections to determine time of death of homozygotes |
Following MommeD7-/+ intercrosses (genotypes determined by GFP fluorescence and progeny testing), a small but significant increase in abnormal embryos was detected at 14.5 dpc (Table ). This increase is not enough to account for all expected MommeD7-/- mice. At 17.5 dpc, approximately one-quarter of the embryos were pale (Table , Figure ), suggesting a red cell defect in the homozygotes. Homozygous MommeD7 mutants were never seen at weaning (Table ), and preliminary observations suggest that they die in the first few days after birth. Further analysis of adult heterozygous individuals revealed severe splenomegaly (Figure ) and a marked increase in reticulocytes in peripheral blood (Figure ).
We hypothesized that this increase in reticulocytes was responsible for the larger than expected increase in the average fluorescence level of the GFP transgene in expressing cells observed in this line (Table ). We performed FACS analysis on whole blood after staining for reticulocytes with propidium iodide. As expected, MommeD7-/+ mice had a threefold increase in the percentage of reticulocytes compared to MommeD7+/+ mice (Figure ), and the percentage of GFP fluorescence in both MommeD7-/+ and MommeD7+/+ was higher in reticulocytes than mature red cells (Figure ). Although this is only significant for MommeD7-/+ (p < 0.005), the trend is there for MommeD7+/+ mice (p = 0.07). This is consistent with the idea that a change in the erythroid cell populations contributes to the dramatic increase in the average fluorescence level of the GFP transgene in MommeD7-/+ mice.
Some MommeD8-/- mice (classified by their GFP expression profile and progeny testing) were viable at weaning but they were rare (Figure , Table ). Following MommeD8-/+ intercrosses, dissections at 14.5 dpc showed no increase in the number of abnormal or resorbed embryos (Table ). Litter size at birth was not significantly different from that seen in wild-type litters (data not shown), suggesting that the death of most MommeD8-/- individuals occurred after birth and before weaning. The only obvious phenotypic abnormality seen in MommeD8-/- mice that survived to weaning was reduced size. MommeD8 homozygotes were significantly smaller (6.60 g ± 0.25 standard error of the mean (SEM)) than their wild-type (8.54 g ± 0.33 SEM, p < 0.001) and heterozygous (8.65 g ± 0.29 SEM, p < 0.0001) littermates.
Dissections following MommeD9-/+ (determined by GFP fluorescence and progeny testing) intercrosses revealed a pattern similar to that seen for MommeD5 and MommeD6, suggesting MommeD9-/- embryos arrest before 9.5 dpc (Table ). In the case of MommeD10 the data suggest that death of homozygotes occurred after birth (Table ), and preliminary data collected at P7 indicated death in the first week of life (data not shown). Some MommeD10-/- individuals survived to weaning but they were extremely rare. This was confirmed by genotyping once the point mutation was identified.
So in all ten MommeDs produced so far, homozygosity for the mutation is associated with embryonic or perinatal lethality (Tables and ).
| Table 4Summary of MommeD screen for epigenetic modifiers |
Abnormal phenotypes associated with heterozygosity for MommeD7-D10
Extensive phenotyping of the heterozygous
MommeD mutant lines has not been carried out. However, in some cases heterozygous effects were obvious, for example, the haematopoietic defect in
MommeD7-/+ mice described above. We have also noticed some litter size reduction during the breeding of these strains. The data for the breeding of
MommeD7,
D8,
D9 and
D10 are shown in Figure . Following crosses between heterozygous males and wild-type females in the FVB strain, we found significant litter size reductions in the cases of
MommeD9 and
MommeD10, but not in the cases of
MommeD7 and
MommeD8. A breakdown of the offspring by sex and genotype revealed that for
MommeD9 and
MommeD10, the litter size reduction was associated with a deviation from Mendelian patterns of inheritance (
p < 0.05 in both cases) and a reduction in the number of mutants (Figure ). These two cases of transmission ratio distortion have not been investigated further but they do suggest that heterozygosity for the
MommeD mutations is associated with some level of disadvantage. There also appears to be a skewed sex ratio in the wild-type offspring of
MommeD9 sires, suggesting the phenotype of the father can affect his wild-type offspring. While we have not characterized this in any more detail, the idea of a paternal effect is not new. We have previously published examples of paternal effects resulting from haploinsufficiency of modifiers of epigenetic gene silencing in the mouse [
12].
Mapping
We have mapped the mutations in all ten cases to relatively small regions of the genome (Table ). The mapping of
MommeD1-D6 has been documented [
10]. Here we report the mapping of
MommeD7-10.
MommeD7 maps to a 0.25 Mb region on chromosome 7 between markers D7Mit220 and rs13479441 (using 134 phenotypically mutant and 135 phenotypically wild-type mice). This region contains 10 genes.
MommeD8 maps to a 4 Mb region on chromosome 4 between markers rs6337795 and D4Mit279 (using 234 phenotypically mutant and 177 phenotypically wild-type mice). This region contains 54 genes.
MommeD9 maps to a 3 Mb region on chromosome 7 between markers rs31712695 and rs6193818 (using 103 phenotypically mutant and 127 phenotypically wild-type mice). This region contains 80 genes.
MommeD10 maps to a 4 Mb region between markers D5Mit165 and rs13478547 on chromosome 5. Twenty-four phenotypically homozygous and 312 phenotypically non-homozygous (heterozygous and wild-type mice combined) were used (see Materials and methods). These data show that each of the ten
MommeD mutations maps to a unique region of the genome.
MommeD5 has a mutation in Histone deacetylase 1
MommeD5 was localized to a 1.42 Mb region on chromosome 4 flanked by the markers rs27486641 and rs27541967 [
10] (Table ). This interval contains 46 genes and
Hdac1 was chosen as the best candidate because of its known role in chromatin modification. Sequencing of all exons, including exon/intron boundaries, from two heterozygous and two wild-type individuals revealed a 7 bp deletion (GAAGCCA) in exon 13 in the mutants (Figure ). This mutation was subsequently verified in over 100 mice classified as mutants by GFP expression profiling. The chance of a second point mutation occurring in a functional region in a linked interval of this size is extremely small. Using the estimated mutation rate following ENU mutagenesis of 1 in 1.82 Mb [
15,
16], the probability of such an event can be calculated [
15,
17]. This website takes into account the amount of coding sequence in a given linked interval. In this case, the probability of a second point mutation occurring in the coding region is
p < 0.0005.
The mutation produces a frameshift, resulting in changes to the last 12 amino acids, and an additional 25 amino acids. Protein modeling predictions based on human HDAC8, for which the crystal structure has been solved [
18,
19], suggest that the carboxyl terminus of Hdac1 is relatively unstructured and so the mutation is unlikely to affect the stability of the protein (J Matthews, personal communication). An antibody that recognizes the carboxyl terminus of Hdac1 showed a 50% reduction of binding in 10.5 dpc
MommeD5-/+ embryos, and negligible binding in
MommeD5-/- embryos (Figure ), confirming that this region of the protein has been altered in the
MommeD5 line. An antibody that recognizes the amino terminus of Hdac1 showed that the levels of the protein are not altered between mutant and wild-type mice (Figure ). Lysine 476 at the carboxyl terminus has been shown to be sumoylated and important for enzymatic function of the wild-type protein [
20] and the absence of this amino acid in the
MommeD5 mutant protein is likely to impair function. A knockout of
Hdac1 has been reported and the homozygous embryos died around 9.5 dpc [
21], similar to the time of death observed in
MommeD5-/- embryos (Figure ). Together, these results argue that the mutation in
Hdac1 is causative of the
MommeD5 phenotype. Consistent with this, the level of Hdac2 increased in both
MommeD5-/+ and
MommeD5-/- embryos, as predicted from the reports of compensatory upregulation of Hdac2 in embryonic stem cells null for Hdac1 [
21]. Indeed, this upregulation may explain why
MommeD5 was identified as an enhancer, rather than a suppressor, of variegation. Loss of histone deacetylase function is generally associated with transcriptional activation, but exceptions to this have been reported and the upregulation of Hdac2 could explain these results [
22].
MommeD10 has a mutation in Baz1b
MommeD10 was localized to a 4 Mb region on chromosome 5 flanked by the markers D5Mit165 and rs32250347 (Table ). Interestingly, this interval encompasses the region syntenic with the Williams Beuren syndrome (WBS) critical region in humans. WBS, also known as Williams syndrome, is an autosomal dominant disorder affecting approximately 1 in 10,000 individuals. The classic presentation of WBS includes a characteristic craniofacial dysmorphology (elfin face), supravalvular aortic stenosis, multiple peripheral pulmonary arterial stenoses, statural deficiency, infantile hypocalcaemia and a distinct cognitive profile with mild mental retardation. The linked interval for
MommeD10 contains 52 genes and
Baz1b was chosen as the best candidate because it contains a bromodomain (a domain commonly associated with chromatin remodelers) and has recently been shown to form at least two chromatin remodeling complexes and associate with replication foci and promoters [
23-
25]. Sequencing of all exons, including exon/intron boundaries, from two homozygous, one heterozygous and one wild-type individual revealed one point mutation (T → G transversion) in exon 7 in the mutants (Figure ). This mutation was subsequently verified in over 100 mice classified as mutants by GFP expression profile. The mutation results in a non-conservative amino acid change, L733R, in a highly conserved region of the protein (Figure ). Western blot analysis showed reduced levels of Baz1b protein in both embryonic and adult
MommeD10-/- tissue, with
MommeD10-/+ tissue showing intermediate levels (Figure and data not shown), suggesting that the mutant protein is much less stable than its wild-type counterpart. Quantitative real-time PCR performed on cDNA from 14.5 dpc embryos showed all three genotypes have similar levels of mRNA (Figure ).
Effects of MommeD5 and MommeD10 on DNA methylation at the transgene locus
Transgene silencing can be associated with changes in both DNA methylation [
26,
27] and chromatin accessibility [
28]. This particular transgene promoter consists of a GC-rich segment of the human alpha-globin promoter, which we were unable to analyze by bisulfite sequencing because the cloned bisulfite converted fragment was refractory to growth in bacteria. The transgene also contains the HS-40 enhancer, which is known to regulate the locus in humans [
29]. We analyzed the methylation state at this region by bisulfite sequencing. As predicted from the variegated nature of the transgene expression, the methylation pattern differed from clone to clone in all cases (data not shown). The percentage of methylated CpGs in the HS-40 element was approximately 55% (averaged across all clones) in spleen from 4-week-old wild-type FVB/NJ mice (Figure ). Samples from
MommeD5-/+,
MommeD10-/+, and
MommeD10-/- mice showed similar levels of CpG methylation (52%, 47%, 59% respectively; Figure ). Mice heterozygous for a null allele of
Dnmt3b, which showed an increase in expression of the GFP transgene from 37 ± 3% in the wild-type mice to 55.5 ± 2.5% in the
Dnmt3b+/- mice (in both cases mice were hemizygous for the transgene; see Materials and methods), showed a decrease in CpG methylation at the HS-40 element (31%; Figure ) compared to that seen in the wild-type C57BL/6J mouse strain (60%; Figure ). These data suggest that
MommeD5 and
MommeD10 mutants alter the expression of the transgene by changing the chromatin state rather than by altering DNA methylation levels. This is consistent with the fact that both genes encode proteins involved in histone modification and chromatin remodeling [
21,
23-
25,
30-
33]. Modifiers identified in this screen include DNA methyltransferases, chromatin remodelers and genes involved in histone modification, all of which have wide-ranging effects in the genome, making it difficult to unravel direct and indirect effects at any particular locus.
Craniofacial analysis of MommeD10 mice
Surviving MommeD10 homozygotes were significantly smaller than littermates at weaning (Student's t-test, p < 0.0001; Figure ). A similar size differential was evident in utero at 18.5dpc (Student's t-test, p < 0.01), indicating that this is not simply due to poor post-natal feeding (Figure ). MommeD10 homozygotes also appeared to have widened, bulbous foreheads and shortened snouts (Figure ). To examine the craniofacial phenotype more accurately, three heads from 4-week-old male mice of each genotype (MommeD10-/-, MommeD10-/+ and MommeD10+/+) were subjected to micro-computed tomography. Heads from one 4-week-old female of each genotype were also examined at this level. They followed the same trend as males. Visual inspection of the three-dimensional reconstructions confirmed the original observation that homozygote's skulls were more bulbous and showed a flattening of the nasal bone and upward curvature of the nasal tip (Figure ).
Twenty cranial landmarks and nine mandibular landmarks were located on each skull using approximately 70 micron resolution datasets and inter-landmark measurements were compared (Figure and Additional data file 1). Statistical analyses were carried out using the data collected from males only. Homozygote skulls were significantly different to wild type (Student's
t-test, length:height ratio,
p < 0.01; width:height ratio,
p < 0.01; length:width ratio,
p < 0.05), confirming the bulbous appearance of the skulls on the reconstructed images. Much of this difference could be attributed to reduction of the parietal and nasal bones (both > 12.5% shorter in homozygotes compared to an overall mean length and width reduction of approximately 9%). The reduced parietal bone length and the reduction and upward angulation of the nasal bones in these mice (Figure ) are reminiscent of the decrease in the volume of the parieto-occipital region and the appearance of the nose in WBS patients [
34,
35]. Heterozygotes also had a decreased cranium width:height ratio (Student's
t-test,
p < 0.05) and decreased length:height ratio (Student's
t-test,
p < 0.05) compared to wild-type skulls. Of note, heterozygotes showed a reduction in palatine bone width of similar magnitude to that seen in homozygotes, suggesting a greater sensitivity of some parts of the skull to decreased Baz1b protein levels. Measurements of the lower jaw revealed relative mandibular hypoplasia in homozygotes that was most pronounced in the posterior region (approximately 20% reduction), encompassing the condyle, angle and coronoid processes (Figure and Additional data file 1). The posterior aspects of the mandibles of heterozygotes were also reduced in size when compared to wild-type mandibles, albeit to a lesser degree than in the homozygotes.
Expression of Baz1b during mouse embryogenesis
It has previously been shown that
Baz1b is expressed in the mouse embryo from around 9.5 dpc and whole mount
in situ at 11.5 dpc showed expression in limb buds, tail and brain [
24]. We have gone on to characterize the expression of
Baz1b in more detail, and show that at 8.25 dpc
Baz1b is expressed in the headfolds, the caudal tail bud region and the presumptive hindbrain in a pattern reminiscent of rhombomere staining (Figure ). From 9.5 dpc expression is evident in the somites and in the forelimb bud as it emerges from the lateral plate mesoderm (Figure ). Diffuse mesenchymal expression in both the forelimb and hindlimb continues until 12.5 dpc when it is restricted to the interdigital mesenchyme (data not shown).
In the facial primordia, Baz1b is expressed in branchial arch 1 as it first emerges (Figure ), and continues later in development in both the maxillary and mandibular components of branchial arch 1 and branchial arch 2 (Figure ). Expression in the branchial arches is primarily mesenchymal, and is enriched in the rostral distal margin of the mandible, and the caudal distal margin of branchial arch 2 (Figure ). Baz1b is also expressed in the frontonasal process (Figure ) and later in the mesenchyme of both the medial and lateral nasal prominences as they elevate to surround the nasal pits (Figure ). Baz1b is expressed strongly in all the major facial primordia from early in embryogenesis.
A possible role for Baz1b in Williams syndrome
Overall, the skull shape in mutant animals is reminiscent of the head shape seen in WBS, including a small upturned nose with flat nasal bridge, micrognathia (or mandibular hypoplasia), malocclusion, bi-temporal narrowing and prominent forehead [
34]. WBS is known to be associated with a hemizygous deletion of approximately 28 genes in humans, but which of these genes are responsible for the craniofacial phenotype remains controversial. People with atypical deletions, and varying degrees of craniofacial abnormalities, implicate both proximal and distal ends of the deletion, suggesting that more than one gene is responsible [
36-
41]. Tassabehji and colleagues [
42] reported craniofacial defects in a transgenic (c-
myc; Tgf-α) mouse line that harbored a chromosomal translocation fortuitously disrupting the
Gtf2ird1 gene, the human orthologue of which resides in the WBS critical interval [
43]. Mice homozygous for this transgene produced little
Gtf2ird1 mRNA and exhibited craniofacial dysmorphism, suggesting a role for
Gtf2ird1 in the craniofacial phenotype. Mice hemizygous for the transgene were indistinguishable from wild-type animals. Disruption of
Gtf2ird1 in this mouse was associated with a 40 kb deletion at the site of integration, the addition of 5-10 tandem copies of a c-
myc transgene, and translocation of the entire segment to chromosome 6 [
43], providing opportunity for disruption to the expression of other genes, such as
Baz1b, in the region. A targeted knockout of
Gtf2ird1, produced by others, failed to find craniofacial dysmorphology or dental abnormalities [
44]. We checked the sequence and expression of
Gtf2ird1 in
MommeD10 mutants and found no abnormalities (data not shown). The chance of a second point mutation occurring in a coding region in this linked interval is extremely small (
p = 0.0008, based on a mutation rate of 1 in 1.82 Mb) [
15-
17].
Our studies show that the chromatin remodeler Baz1b is expressed strongly in cranial neural crest-derived mesenchyme that drives facial morphogenesis and that homozygosity for a missense mutation in Baz1b produces an array of craniofacial features that are similar to those that characterize the typical WBS face. Significantly, some craniofacial features are also apparent in heterozygotes. These results suggest that reduction in Baz1b protein levels contributes to the elfin facies characteristic of WBS and that WBS is, at least in part, a chromatin-remodeling factor disease.
1
This article has been 
