Genomic Imprinting Variations in the Mouse Type 3 Deiodinase Gene Between Tissues and Brain Regions

The Dio3 gene, which encodes for the type 3 deiodinase (D3), controls thyroid hormone (TH) availability. The lack of D3 in mice results in tissue overexposure to TH and a broad neuroendocrine phenotype. Dio3 is an imprinted gene, preferentially expressed from the paternally inherited allele in the mouse fetus. However, heterozygous mice with paternal inheritance of the inactivating Dio3 mutation exhibit an attenuated phenotype when compared with that of Dio3 null mice. To investigate this milder phenotype, the allelic expression of Dio3 was evaluated in different mouse tissues. Preferential allelic expression of Dio3 from the paternal allele was observed in fetal tissues and neonatal brain regions, whereas the biallelic Dio3 expression occurred in the developing eye, testes, and cerebellum and in the postnatal brain neocortex, which expresses a larger Dio3 mRNA transcript. The newborn hypothalamus manifests the highest degree of Dio3 expression from the paternal allele, compared with other brain regions, and preferential allelic expression of Dio3 in the brain relaxed in late neonatal life. A methylation analysis of two regulatory regions of the Dio3 imprinted domain revealed modest but significant differences between tissues, but these did not consistently correlate with the observed patterns of Dio3 allelic expression. Deletion of the Dio3 gene and promoter did not result in significant changes in the tissue-specific patterns of Dio3 allelic expression. These results suggest the existence of unidentified epigenetic determinants of tissue-specific Dio3 imprinting. The resulting variation in the Dio3 allelic expression between tissues likely explains the phenotypic variation that results from paternal Dio3 haploinsufficiency.

T he type 3 deiodinase (D3) modulates thyroid hormone (TH) action by converting both the prohormone T 4 and the active hormone T 3 into inactive metabolites (1)(2)(3). Its high expression in the pregnant uterus, fetus, and placenta (4,5) and in the developing and adult central nervous system (CNS) (6,7) suggests that D3 is important for the maintenance of appropriate levels of TH in the fetus and the adult. In this regard, we have shown that mice lacking D3 (D3KO or Dio3 Ϫ/Ϫ mice) are overexposed to T 3 during development and subsequently manifest marked deficits in the maturation and function of the thyroid axis, severe growth retardation, and impaired viability and fertility (8,9). These mice also exhibit alterations in cardiovascular (10) and sensory functions (11,12), glucose and insulin homeostasis (13), and brain gene expression patterns (14,15), demonstrating a critical role for D3 in many pathophysiological outcomes.
We and others have demonstrated that Dio3, the gene that codes for D3, is subject to genomic imprinting in the mouse (16,17). Genomic imprinting is an epigenetic phenomenon that affects a small percentage of mammalian genes (18 -20). Imprinted genes are preferentially or exclusively expressed from one of the alleles, depending on the parental origin (18). Although the mechanisms underlying genomic imprinting are not fully understood, a common feature of imprinted loci is the differential methylation (imprint) between parental alleles in key regulatory genomic sequences. These imprints ultimately translate into the allele-specific expression or repression of certain genes within the imprinted domain (18 -20). Genomic imprinting results in a high degree of monoallelic gene expression, and thus poses a higher risk of pathologies, as a random mutation in just one copy of the imprinted gene may lead to deleterious loss or gain of function. Furthermore, alterations in the dosage of imprinted genes due to disrupted genomic imprinting result in abnormalities in development and adult physiology (21,22). In humans, aberrant genomic imprinting is the cause of severe developmental, neurological, and metabolic syndromes (23)(24)(25)(26)(27).
Dio3 belongs to the Dlk1-Dio3 imprinted domain (28), and in the mouse fetus, it is expressed preferentially from the allele inherited from the father (16,17). However, the repression of the maternal Dio3 allele is not complete (17), and in the placenta, considerable expression originates from the maternal allele (29), indicating that Dio3 imprinting in this tissue is significantly relaxed. We have also reported a lesser degree of genomic imprinting in the heads of fetal mice (17). Taken together, these observations suggest that allelic contributions to Dio3 expression vary in a tissue-specific pattern, with resultant functional and phenotypic implications. In the context of the milder gross phenotype that we observe in heterozygous mice with a mutated paternal Dio3 allele when compared with that of null Dio3 mice, herein we use our previously described D3KO mouse and a second novel model of targeted Dio3 disruption to analyze Dio3 allelic expression in various tissues and in regions of the CNS. We show that Dio3 imprinting varies significantly between tissues and developmental stages, especially in the CNS. This suggests that epigenetic information regulates Dio3 expression in a tissue-and/or cell-specific manner and that the resultant alterations in the imprinting pattern of Dio3 effect developmental outcomes.

Animals, genotyping, and tissue harvesting
Animals were kept under a 12-hour light cycle and provided food and water ad libitum. Animals were killed by asphyxiation with CO 2 (adults and weanlings) or by decapitation (neonates and fetuses). Mice of both sexes were used together because no differences in the results obtained were appreciated between males and females. For RNA isolation and D3 enzymatic activity, tissues were dissected, immediately frozen on dry ice, and stored at Ϫ70°C until further use. Brain regions were identified and harvested according to the mouse atlas by Paxinos and Franklin (30). Animal procedures were approved by the Institutional Animal Care and Use Committees of Dartmouth College and the Maine Medical Center Research Institute. Genotyping of the D3KO animals carrying the triple-point mutation was performed by a PCR of genomic DNA from tail snips as previously described (8). Genotyping of mice carrying the novel deletion of the Dio3 gene was performed by Southern analysis (see below) or by a double PCR using a pair of primers that amplify a DNA fragment of the native Dio3 gene and another pair of primers that amplify a DNA fragment specific to the deleted allele (Supplemental Methods).

Mouse embryonic stem (ES) cells and introduction of mice carrying a deletion of Dio3
We (17) described the strategy used to target the Dio3 gene using standard homologous recombination techniques. We used the R1 ES cell line (31), which originated from the 129/Sv mouse strain. Targeted clones were identified by Southern analysis, injected into C57/BL6 blastocysts and reimplanted in CD1 foster mothers. Chimeric males that showed germline transmission were mated to C57/BL6 females to test for the germline transmission of the mutation. Chimeric males were then mated with 129/Sv females to establish the mutant line in a 129/Sv background.

D3 activity and serum levels of TH
D3 enzymatic activities were determined as previously described (8). In brief, tissues were homogenized in a 10-mM Tris-HCl, 0.25 sucrose buffer (pH 7.4). A suitable volume of tissue homogenate was used in the enzymatic reaction to ensure that deiodination do not exceed 40% and was proportional to the amount of protein content. Tissue homogenates were incubated at 37°C for an hour with 2 nM 125 I-labeled T 3 (PerkinElmer) in the presence of 25 mM dithiothreitol. Deiodination was determined based on the percentage of 125 I-3,3Ј-diiodothyronine produced. The latter was determined by measuring the amount of radioactivity associated with the reaction products after separation by paper chromatography as described (32). Total serum thyroid levels were determined using RIA kits following the manufacturer's directions (TKT4 and TKT3; Siemens). Serum TSH was determined by a RIA as previously described (9).

DNA and RNA isolation and Northern and Southern analysis
Total RNA and poly(Aϩ) RNA were isolated from brain tissues using the RNAeasy kit (QIAGEN), including deoxyribonuclease treatment, and the poly(Aϩ) kit from Ambion (Now Life Technologies). Dio3 Northern analysis was performed following standard protocols. In brief, total and poly(Aϩ) RNA samples were electrophoresed in a denaturing 1% agarose gel containing formaldehyde and blotted onto a Nytran membrane (GE Life Sciences). The blots were hybridized at 42°C in buffer containing 50% formamide, washed with 0.1ϫ saline sodium citrate/0.1% sodium dodecyl sulfate at 65°C and autoradiographed for 1-7 days. Probes were labeled with radioactive 32 P-dCTP (MP Biomedicals, Inc) using the oligolabeling kit (Pfizer) and were purified through G-50 columns (Pfizer). A 1.35-kb XhoI restriction fragment comprising the Dio3 coding region and part of the 3Ј-untranslated region was used as a probe.
DNA was isolated from tail snips using a kit from QIAGEN. A Southern analysis was performed using standard protocols on genomic DNA (gDNA) digested with EcoRI. A 0.7-kb SacI/ EcoRI restriction fragment located immediately outside the 5Ј end of the targeting sequencing was used as a probe.

RT-PCR analysis of allelic expression
Total RNA (1 g) was reverse transcribed for 1 hour at 42°C with 1 L of Moloney murine leukemia virus reverse transcriptase (Life Technologies). A reaction without the reverse transcriptase was performed as a negative control to ensure no significant amplification of gDNA in the downstream PCR. The mix was heated at 75°C for 15 minutes to inactivate the reverse transcriptase and diluted 1 to 20 with water. Aliquots of the mixes from a given experiment were pooled before dilution to establish the first point of an internal standard. Three consecutive one to four dilutions of this standard were done to generate three additional standard points. Ten microliters of each of the diluted samples were mixed with 12.5 L of SYBR Select master mix from Life Technologies and 2.5 L of the appropriate genespecific primer mix (3.33 pmol/L each). The mixture was subjected to PCR cycling using a MyiQ single color real-time PCR detection system from Bio-Rad Laboratories. The PCRs were performed in triplicate, and data for each gene and sample were read from the standard curve. Gapdh was used as a control gene, its expression level not varying significantly between experimental groups. The expression levels for each gene are reported in arbitrary units, after correction by Gapdh expression. The sequences of the primers used are shown in Supplemental Methods.

Bisulfite sequencing of the intergenic differentially methylated region (IG-DMR)
The sodium bisulfite sequencing methylation assay was performed as described (33) using 500 ng to 1 g of DNA from 129/Sv mouse tissues of the neonatal age (P1). Briefly, two independent PCR replicates were performed using bisulfitetreated DNA as template. PCR products were gel purified, combined, and cloned using the TopoTA cloning kit (Invitrogen). Clones were analyzed and sequenced at the McGill University and Genome Québec Innovation Centre sequencing platform. Duplicate clones or clones with incomplete C-to-U conversion were excluded from the analysis. Information about the sequence and CpG groups analyzed can be found in Supplemental Methods.

Pyrosequencing
Bisulfite conversion and pyrosequencing was carried out according to Sun et al (34) with the following modifications: gDNA (1 g) was treated using the Imprint DNA modification kit (Sigma) in accordance with the manufacturer's instructions for the two-step conversion and eluted in 20 L. Primer sequences, annealing temperatures, and cycle number are shown in Supplemental Methods. Pyrosequencing was performed on the PSQ HS96 system using PyroGold Q96 SQA reagents (QIAGEN). The degree of methylation at CpG sites (without distinguishing between maternal and paternal alleles) was determined by pyro-Q CpG software (Biotage). Information about the sequence and CpG groups analyzed can be found in Supplemental Methods.

Statistics
We used the GraphPad Prism software for statistical analysis. Unless stated otherwise, differences between more than two experimental groups were determined by ANOVA and Tukey's post hoc test, whereas differences between two groups were determined by the Student's t test. P Ͻ .05 was considered statistically significant.

Results
Paternal inheritance of the mutated Dio3 allele leads to a milder gross phenotype than that in null Dio3 mice Given that Dio3 is preferentially expressed from the paternally inherited allele (16,17), one would predict that heterozygous animals that inherited the mutation from the father (Dio3 mϩ/pϪ) would manifest a phenotype similar to that of a homozygous Dio3 Ϫ/Ϫ mutant. Such is at least partially the case; mice with the Dio3 mϩ/pϪ genotype had a significantly milder phenotype than mice that were completely D3 deficient (8). For example, unlike Dio3 Ϫ/Ϫ mice, adult Dio3 mϩ/pϪ mice do not exhibit significant changes in their serum levels of thyroid hormones, when compared with the wild types (Dio3 ϩ/ϩ ) or with heterozygous animals with maternal inheritance of the mutation (Dio3 mϪ/pϩ mice) (data not shown) (10). However, an increased serum level of T 3 and a partially suppressed serum T 4 level are noted at postnatal day (P) 2 ( Figure 1, A and B). At this age, serum TSH trended lower (68.44 Ϯ 10.09 mU/L in Dio3 mϩ/pϪ mice vs 85.85 Ϯ 11.65 mU/L in Dio3 mϪ/pϩ mice, n ϭ 4 and 5), but the difference was not statistically significant. These thyroid parameters indicate neonatal overexposure to T 3 and partial thyroid axis suppression in Dio3 pϪ/mϩ mice. They are also significantly growth retarded and the weight of their testes is reduced (Figure 1, C and D). These abnormalities indicate that reduced Dio3 dosage has phenotypic consequences. However, these gross phenotypes are significantly milder than those in Dio3 Ϫ/Ϫ mice. Previous observations indicated that P2 Dio3 Ϫ/Ϫ mice feature a 5-fold increase in serum T 3 , an 80% reduction in serum T 4 , and a 30% reduction in weight as young adults (8). Testes size in young Dio3 Ϫ/Ϫ males is also reduced 60%, when compared with Dio3 ϩ/ϩ littermates (unpublished observations, Hernandez, A.). Compared with that of Dio3 Ϫ/Ϫ mice, the milder phenotype of Dio3 mϩ/pϪ mice suggests that Dio3 expression from the maternal allele is partially compensating the lack of expression from the paternal allele in these animals.

Dio3 is preferentially expressed from one allele in mouse ES cells
We have previously reported the targeting of the Dio3 gene in mouse ES cells to generate a mouse deficient in the type 3 deiodinase (17). We have used Dio3 targeted and nontargeted ES cell clones from that experiment to analyze Dio3 expression. Targeted clones are heterozygous for a Dio3 mutation that introduces an XbaI restriction site in the selenocysteine codon, thus rendering the enzyme inactive (35). We prepared gDNA and cDNA from Dio3-targeted clones and used PCR to amplify a DNA fragment containing the inserted Dio3 mutation. XbaI digestion of these fragments ( Figure 2A) revealed partial digestion when the fragment had a gDNA origin, whereas almost complete digestion occurs when the fragment was amplified from cDNA generated by reverse transcription of total RNA. Sequencing of the mutation site in these fragments demonstrated that mutated DNA accounts for approximately half of the gDNA but for a much higher percentage of the RNA-derived cDNA ( Figure 2B). Similar results were obtained for all five Dio3-targeted clones. In addition, D3 activity in these targeted clones was less than 20% of the average D3 activity measured in nontargeted clones ( Figure 2C). These results indicate not only that Dio3 is preferentially expressed from one of the alleles in mouse ES cells but also that the preferentially expressed allele is targeted more frequently in the homologous recombination event.

Dio3 allelic expression in the fetal head
Previous data suggested that allelic contribution to Dio3 expression may be only slightly biased toward the paternal allele in the fetal head (17). We have analyzed the allelic contribution to Dio3 expression in the fetal eye, brain, and the rest of craniofacial structures. We used animals generated by reciprocal matings of Dio3 ϩ/Ϫ and Dio3 ϩ/ϩ mice so that we could track the parental origin of the mutated allele in the offspring and thus the allele that is contributing to Dio3 expression (8). Fetal eye D3 activities in heterozygous mice that inherited the mutation from the father (Dio3 mϩ/pϪ) were similar to that in heterozygous mice that inherited the mutation from the mother (Dio3 mϪ/pϩ), and both were approximately 50% of D3 activity measured in wild-type (Dio3 ϩ/ϩ ) mice ( Figure 3A). D3 activity in the brain of Dio3 mϪ/pϩ mice was 20% lower than in Dio3 ϩ/ϩ mice, whereas it was significantly higher than in Dio3 mϩ/pϪ mice, in which D3 activity was reduced 55% compared with that in wildtype mice. In the tissues comprising the rest of the head and craniofacial structures, D3 activity in Dio3 mϪ/pϩ mice was not significantly reduced compared with that of wild-type animals, but a 75% reduction was observed in Dio3 mϩ/pϪ mice ( Figure 3A), indicating strong preferential expression from the paternal allele in this tissue.
Dio3 mRNA expression from the mutated allele was evaluated by real time quantitative PCR using primers specific for the mutated Dio3 allele. Mutated Dio3 mRNA was expressed at similar levels in the eyes of Dio3 mϪ/pϩ and Dio3 mϩ/pϪ fetuses ( Figure 3B). In the brain, the expression of the mutated Dio3 mRNA was reduced in Dio3 mϩ/pϪ fetuses compared with that in Dio3 mϪ/pϩ fetuses. In the rest of the fetal head, this reduction was more marked, indicating a stronger preferential Dio3 expression from the paternal allele. These activity measurements indicate a strong correlation between Dio3 mRNA and protein levels and confirm the previously defined imprinting pattern based on mRNA expression levels.  Because the rodent brain expresses Dio3 transcripts that are larger than the 2.2-kb transcript (7) that is most abundant in placenta and fetal tissues, we also performed a Northern analysis of mRNA from fetal tissues. These samples were extracted from mice in which the neomycin cassette used for targeting in ES cells was still present in the 3Ј-untranslated region of the Dio3 mRNA (17). As a result, the Dio3 mRNA species carrying the mutation is approximately 2 kb larger than the native Dio3 mRNA. As expected, Dio3 mRNA transcripts above 4 kb in size are detected in tissues from Dio3 ϩ/Ϫ animals. In the fetal eye and nonneural craniofacial structures of the wild-type mice, the typical 2.2-kb transcript is essentially the only one detected (Figure 3C), whereas a larger, similarly abundant, 2.7-kb Dio3 transcript is also observed in the wild-type fetal brain ( Figure 2C, lanes 1, 4, and 7). The abundance of the mutated 4-kb and the native 2.2-kb Dio3 transcript is similar in the eye of Dio3 mϩ/pϪ and Dio3 mϪ/pϩ animals ( Figure 3C, lanes 2 and 3), indicating biallelic Dio3 expression in this tissue. However, in the nonneural head samples, the mutated transcript was much more abundant than the native transcript in Dio3 mϩ/pϪ mice, whereas the native Dio3 transcript was much more abundant than the mutated one in Dio3 mϪ/pϩ mice ( Figure 3C, lanes 8 and 9), demonstrating strong preferential expression from the paternal allele. Similarly, in the fetal brain Figure 3C, (lanes 5 and 6), transcripts that originated in the paternal allele are also more abundant, although the pattern is not as marked as in the tissues contained in the rest of the head. Overall, these results indicate no imprinting of Dio3 in the fetal eye and marked preferential Dio3 expression from the paternal allele in the craniofacial tissue and, to a lesser degree, in the fetal brain.

Dio3 allelic expression in the perinatal brain
Based on D3 activity data, brain Dio3 follows a similar pattern of allelic expression in late gestation. At embryonic day (E) 18.5, Dio3 expression is reduced in the forebrain and hindbrain of Dio3 mϪ/pϩ animals compared with that of Dio3 ϩ/ϩ mice ( Figure 4A), but it is reduced significantly further in Dio3 mϩ/ pϪ, indicating preferential expression from the paternal allele. At P1, a marked imprinting pattern of expression is observed in the thalamus and hypothalamus, with normal and very reduced Dio3 expression in Dio3 mϪ/pϩ and Dio3 mϩ/pϪ animals, respectively ( Figure 4B). No imprinted pattern of expression is noticed in the cerebellum, in which Dio3 expression seems equally biallelic. A relaxed pattern of imprinting is observed in the cerebral cortex and the midbrain and pons. In this case, significant preferential expression from the paternal allele is noted, but it is not sufficient to achieve Dio3 expression levels comparable with those in wild-type mice ( Figure 4B). Lack of Dio3 imprinting is noted in certain nonneural tissues, such as the newborn testes ( Figure 4B).

Dio3 allelic expression in the late postnatal brain
Northern analysis of P7 brain mRNA reveals that the larger 2.7-kb Dio3 transcript is the only one detected in the neocortex ( Figure 5A, left panel) and does not show significant preferential allelic expression; mutated (and larger because it contains the neomycin cassette) and the wild-type version of this RNA species appear similarly abundant in Dio3 mϪ/pϩ and Dio3 mϩ/pϪ mice. Quantification of the mutated Dio3 mRNA species reveals similar levels of expression in heterozygous mice, irrespective of the parental origin of the mutated allele ( Figure 5B, left panel). In a P7 specimen comprising the thalamus, hippocampus, and striatum, both 2.2-and 2.7-kb Dio3 transcripts are detected, the latter being more abundant (Figure 5A, right panel). In this tissue, the pattern of allelic expression in Dio3 mϪ/pϩ and Dio3 mϩ/pϪ mice still shows significant preferential expression from the paternally inherited allele ( Figure 5, A and B, right panel).
At P21, Dio3 ϩ/ϩ mice exhibit wide variation in Dio3 expression levels across brain regions, D3 activity being very high in the hypothalamus and in the cortex, striatum, and hippocampus; low in the cerebellum, midbrain, and pons and in the corpus colliculus; and intermediate in the amount in the olfactory bulb and thalamus ( Figure 5C). At this age, thalamic and cerebellar Dio3 shows no imprinting pattern of expression, D3 activity being comparable in Dio3 mϪ/pϩ and Dio3 pϪ/mϩ mice and significantly lower than in Dio3 ϩ/ϩ mice ( Figure 5D). The rest of the brain regions analyzed still exhibit preferential expression from the paternal allele, although the relative contribution of this allele to overall Dio3 expression appears reduced compared with previous developmental stages. Thus, at this stage, expression from the paternal allele is sufficient to achieve normal levels of Dio3 expression in the midbrain and pons, corpus colliculus, and the cortex, striatum, and hippocampus, whereas the maternal allele can achieve more than 50% of the normal levels of expression when the paternal allele is mutated ( Figure 5D).

Methylation of the IG-DMR and maternally expressed gene 3 (Meg3) intron in different tissues
To begin to address the mechanisms of the differences in Dio3 imprinting between tissues, we analyzed the methylation pattern of the IG-DMR ( Figure 6A). This region regulates imprinting in the Dlk1-Dio3 imprinted domain and is usually methylated in the paternal allele in the mouse fetus (36), in which it appears to play a role in the repression of the maternal copy of Dio3 (28,36). First, the methylation status of 33 CpGs corresponding to the IG-DMR (37) was assessed by bisulfite sequencing of DNA isolated from the cerebellum and hypothalamus/ thalamus of P1 mice. We could not distinguish between maternal and paternal alleles in these samples, but we observed two types of clones from both tissues, either hypomethylated or hypermethylated ( Figure 6B). Overall levels of methylation were comparable in both tissues despite the observation that allelic expression of Dio3 is different in those tissues, biallelic in the cerebellum and preferentially paternal in the hypothalamus. This suggests that loss of Dio3 imprinting in the cerebellum is not due to changes in the methylation status of the IG-DMR.
Pyosequencing analysis of the methylation status of seven CpG (CG5 to CG12) of the IG-DMR indicated approximately 50% methylation in the P2 cerebellum, hypothalamus, and testis and in the fetal eye and face (Figure 7, left panel). Modest but significant differences in the degree of methylation were noticed at P2 between craniofacial structures and the testis. Using the same technique and the same tissues, we analyzed the methylation status of a region at the 5Ј-end of the Meg3 gene (or gene trap locus 2, Gtl2), including the first intron. This region is differentially methylated between alleles and contains a putative binding site for the zinc finger protein CCCTC binding factor (CTCF), which plays a role in the regulation of imprinting (28) ( Figure  6A). Overall methylation was in the range of 30% for all tissues (Figure 7, right panel). Again, modest but significant differences in methylation level were observed, with fetal craniofacial structures exhibiting significantly lower methylation than the neonatal testes and cerebellum.
In summary, the changes in methylation profiles are associated with differences in Dio3 allelic expression in some tissues, suggesting an impact on the regulation of Dio3 genomic imprinting. But this is not the case for all tissues, an observation that points to other tissue-specific mechanisms influencing the epigenetic regulation of Dio3.

Deletion of the Dio3 gene and promoter does not change imprinting at the Dio3 locus
We have generated by homologous recombination in mouse ES cells an additional mouse model in which an approximately 4.5-kb genomic fragment between endogenous NotI and HindIII restriction sites has been deleted in the Dio3 locus and replaced by the neomycin cassette used for clone selection. The deleted fragment contains the Dio3 exon, 1.8 kb of 3Ј-flanking region and approximately 500 bp of Ј-flanking region including the Dio3 promoter ( Figure 8A). A Southern blot analysis of EcoRIdigested gDNA isolated from tail snips of wild-type mice and mice heterozygous and homozygous for the deletion demonstrated the germline transmission of the deletion ( Figure 8B). A Northern blot analysis of RNA isolated from a pregnant uterus and fetal heads demonstrated the absence of Dio3 mRNA in mice homozygous for the deletion (Dio3 Del Ϫ/Ϫ) ( Figure 8C). Using this animal model, we analyzed mRNA expression of neomycin and Dio3 in various tissues of heterozygous animals with different parental inheritance of the deletion. In the fetal liver, brain, and body of the fetus, neomycin mRNA levels were much higher in animals with paternal inheritance of the deletion (Dio3 Del mϩ/pϪ) than in heterozygous animals that inherited the deletion from the mother (Dio3 Del mϪ/pϩ) ( Figure 8D). Because neomycin mRNA is expressed only from the deleted Dio3 allele, these results indicate preferential expression from the paternal allele in these tissues.
In contrast, the contribution to neomycin mRNA levels was not significantly different between alleles in the neonatal testes and the adult brain ( Figure 8D). Essentially the opposite pattern is observed for Dio3 mRNA levels. Fetal tissue Dio3 mRNA levels were much lower in animals with paternal inheritance of the deletion (Dio3 Del mϩ/pϪ) than in heterozygous animals that inherited the deletion from the mother (Dio3 Del mϪ/pϩ) ( Figure  8E). Again, contributions to the Dio3 mRNA levels in the adult brain were similar between alleles ( Figure 8E). These tissue-specific imprinting patterns of neomycin and Dio3 mRNAs in this animal model recapitulate the results obtained in the Dio3 mutation mouse model, in which the Dio3 gene and promoter are intact. These results demonstrate that, at least in the tissues studied, the Dio3 pro- moter is not necessary for conferring tissue specificity to the allelic expression at the Dio3 locus. This is consistent with studies indicating that imprinting control is mediated by differentially methylated regions such as the IG-DMR at the Dlk1-Meg3 part of the cluster. It also shows that the normal imprinting pattern can be imposed on the regulation of the phosphoglycerate kinase promoter driving the neomycin gene; therefore, specific characteristics of the Dio3 promoter do not confer sensitivity to imprinting.

Discussion
In mice, the absence of a functional D3 enzyme leads to systemic overexposure to T 3 during development (8) and to excessive TH action in certain adult tissues, such as the CNS (14). Given the pleiotropic actions of TH, this overexposure leads to a number of abnormalities. Because Dio3 is expressed from the paternal allele during fetal life (16,17), it is reasonable to expect that heterozygous animals inheriting the Dio3 mutation from their fathers (Dio3 mϩ/pϪ) would feature similar abnormalities. We show herein that this is at least partly the case because Dio3 mϩ/pϪ mice exhibit elevated serum T 3 and sup-pressed serum T 4 as neonates, growth retardation, and small testes. These findings are also observed in the Dio3 null mouse, but their severity is significantly reduced in the Dio3 mϩ/pϪ mutant. Furthermore, viability and fertility of Dio3 mϩ/pϪ mice is significantly improved when compared with Dio3 Ϫ/Ϫ mice, and their serum TH levels appear normal in adult life (10). The more moderate phenotype observed in Dio3 mϩ/pϪ mice compared with Dio3 Ϫ/Ϫ mice is likely due to considerable Dio3 expression from the maternal allele in certain tissues.
We demonstrate that mouse ES cell clones targeted for Dio3 inactivation exhibit a high degree of monoallelic Dio3 expression at the mRNA and protein levels, suggesting that strong genomic imprinting of Dio3 is present around the blastocyst stage of early development. This is not surprising because most imprinted genes, including those in the Dlk1-Dio3 imprinted domain, manifest such patterned expression in ES cell lines and at early developmental stages (38). It is remarkable that all five Dio3 targeted clones exhibited markedly deficient levels of Dio3 expression, suggesting a significant bias in the homologous recombination event, which favored the preferentially expressed, presumably paternal allele. This suggests that the expression and associated alterations in chromatin facilitate homologous recombination. The result is also consistent with observations that suggest the existence of epigenetic boundaries that are associated with a differential timing for DNA replication at imprinted loci (39). If this is also the case for the Dio3 locus, it is then possible that gene targeting is facilitated when the preferentially expressed allele is replicated.
Although marked preferential Dio3 expression from the paternal allele is found in ES cells and the fetus as a whole, it does not necessarily apply to all tissues and cell types. Many imprinted genes exhibit tissue-specific patterns of allelic expression (40,41). We show that in the fetal eye, both alleles contribute equally to Dio3 expression. The absence of Dio3 imprinting in this tissue suggests that deficient expression from the paternal Dio3 allele only is less likely to have an impact on the retinal and vision phenotype that is observed in Dio3 null mice (42). The absence of Dio3 imprinting is also observed in the newborn testes. In this regard, the decrease in testis size in Dio3 pϪ/mϩ mice is much less severe than in Dio3 Ϫ/Ϫ mice (unpublished results, Hernandez, A.) and may be due to the still significant neonatal exposure to elevated circulating levels of T 3 in these animals. Other fetal tissues (eg, nonneural craniofacial tissue) show a high degree of preferential expression from the paternal allele, but it is the CNS that displays a wide regional variation in the relative contributions from each allele to overall Dio3 expression. Except for the cerebellum, which exhibits equal Dio3 expression from both alleles, most regions of the perinatal and late neonatal brain exhibit preferential Dio3 expression from the paternal allele. However, Dio3 genomic imprinting appears relaxed in some brain regions because the expression from the paternal allele alone is not sufficient to achieve normal Dio3 expression levels, and the expression from the maternal allele is close to 50% of wild-type values in some cases. Only the perinatal hypothalamus exhibits an allelic expression pattern that is consistent with robust genomic imprinting of Dio3, but even in this brain region, the monoallelic pattern of expression is relaxed by weaning age. The overall relaxation of genomic imprinting of Dio3 by weaning age in most brain regions is consistent with the reported late neonatal loss of imprinting of Dlk1 in neurogenic structures of the brain (43) and the in-creased plasticity of brain imprinting for several genes in the cluster (37). Thus, it is possible that common mechanisms of imprinting loss or relaxation exist for the Dlk1-Dio3 imprinted domain in the CNS and other tissues.
Different epigenetic mechanisms regulating the expression of the larger, approximately 2.7-kb Dio3 mRNA transcript, expressed in the mouse brain, may also explain the relaxation of Dio3 imprinting in this tissue. In the fetal brain, this transcript is as abundant as the better characterized 2.2-kb transcript but becomes by far the most abundant by P7, being the only Dio3 transcript detected in the P7 neocortex by Northern analysis. In the untreated adult rodent brain, this transcript is also the most abundant (7). Indirect evidence from the studies of DIO3 in human cell lines (44) and preliminary observations in our laboratory suggests that the larger Dio3 mRNA contains additional sequence in the 5Ј-untranslated region of the mRNA, with transcription thus being directed by an unidentified upstream promoter. It is thus possible, as observed with other imprinted genes (41), that this alternative promoter may not be regulated by the same epigenetic mechanisms. At a more specific cell and tissue level, it is possible that other imprinting patterns of Dio3 expression exist. In this regard, Sittig et al (45) reported modest preferential Dio3 expression from the maternal allele in the hippocampus of a rat model of developmental alcohol exposure, whereas Correa et al (46) reported preferential Dio3 expression from the maternal allele in pancreatic ␤-cells.
The varied tissue and developmental patterns of Dio3 allelic expression suggest the existence of tissue-specific mechanisms of epigenetic regulation of Dio3 expression. An analysis of the methylation status of two regions in the imprinted domain with known regulatory roles on allelic Dio3 expression (36), the IG-DMR and the Meg3 intron, reveal subtle differences in the degree of methylation between some tissues. Although statistically significant, these differences are rather small and may not explain the allelic expression observed in a given tissue. For instance, Dio3 is not imprinted in the neonatal testes and cerebellum. These tissues show increased Meg3 intron methylation compared with fetal craniofacial tissue, in which Dio3 is imprinted. In this case, the difference in methylation is associated with a difference in the ratio of allelic expression. However, Meg3 intron methylation in the newborn hypothalamus, in which Dio3 imprinting is robust, is similar to that in testes and cerebellum, which show no Dio3 imprinting. It is thus uncertain whether these methylation differences influence the observed tissue specific pattern of allelic Dio3 expression.
Studies on a new mouse model in which the Dio3 gene and its characterized promoter are deleted and substituted by the neomycin gene driven by a different putative promoter reveal no significant role for the endogenous Dio3 promoter in mediating tissue-specific Dio3 imprinting. At least for the tissues analyzed in this mouse model, the patterns of neomycin and Dio3 allelic expression resemble those observed for the intact and mutated allele in mice carrying the Dio3 mutation. This demonstrates that sequences in the endogenous Dio3 promoter are not relevant in mediating the tissue specificity of Dio3 allelic expression. Although transcription of the antisense Dio3os gene may be affected in mice carrying this deletion, our results indicate that this mouse model can be a useful tool for imprinting studies and for other types of investigations that benefit from the absence of the Dio3 mRNA and protein.
Overall, our observations indicate that genomic imprinting of Dio3 varies significantly between tissues, developmental stages, and Dio3 transcripts. This applies in particular to the brain, an organ in which Dio3 is prominently expressed throughout life. It is possible that the extent of control conferred by the IG-DMR may differ between one tissue and another; because Dio3 is the most distant from the IG-DMR end of the cluster, it may be more prone to distance effects, resulting in the loss of Dio3 imprinting in some contexts. This likely explains the less severe gross phenotype of heterozygous mice with inactivation of the paternal Dio3 allele (as compared with null mice) and suggests that certain abnormalities of the Dio3 null mice will be more likely to occur than others when imprinting of Dio3, and thus Dio3 dosage, is disrupted. In that context, it is likely that phenotypes that originate in tissues with high Dio3 expression and a marked pattern of monoallelic expression will be the most susceptible to abnormal epigenetic regulation of Dio3. Future studies will address this hypothesis.