Methyltransferase G9A Regulates Osteogenesis via Twist Gene Repression
Abstract
Here we investigate the role of epigenetic factors in controlling the timing of cranial neural crest cell differentiation. The gene coding for histone H3 lysine 9 methyltransferase G9A was conditionally deleted in neural crest cells with Wnt1-Cre. The majority of homozygous- null animals survived to birth but thereafter failed to thrive. Phenotypic analysis of postnatal animals revealed that the mutants displayed incomplete ossification and 20% shorter jaws as compared to their wild-type littermates. At E13.5, patterns of expression of the osteogenic transcription factor RUNX2 and the mesenchymal transcription factor TWIST are similar in controls and mutants; both overlap in areas of future intramembranous bone formation. At E14.5, the nonosteogenic mesenchyme expressed TWIST, whereas the ossification center had strong RUNX2 and osteopontin expression. In the mutants, TWIST protein was present in the osteogenic mesenchyme, while osteopontin was not expressed until E15.5. In addition, in mutants, small regions of TWIST-positive osteogenic mesenchyme were visible until E15.5. The delay in ossification and reduction in size of the ossification centers were correlated with an earlier decrease in proliferation. We used micromass cultures of the face to investigate the direct effects of G9A inhibition on skeletal differentiation. Addition of a small molecule inhibitor for G9A, BIX-01294, to wild-type cells upregulated Twist genes similar to what was observed in vivo. The inhibitor also caused decreases in several osteogenic markers. Chromatin immunoprecipitation analysis of primary osteogenic mesenchyme from calvaria revealed that Twist1 and Twist2 regulatory regions contain the repressive H3K9me2 marks catalyzed by G9A, which are removed when BIX-01294 is added. Our results establish a role for G9A and H3K9me2 in the regulation of Twist genes and provide novel insights into the significance of epigenetic mechanisms in controlling temporal and tissue-specific gene expression during development.
Introduction
Neural crest cells are described as a fourth germ layer, since they are derived from the ectoderm, but unlike the neural plate, which forms the central nervous system, neural crest cells give rise to nonneural lineages, including ectomesenchymal cells of the head (Le Douarin and Dupin 2016). The cranial neural crest(conditional knockout [CKO]) is useful in the case of epigen- etic modifier enzymes, since germline deletion commonly leads to early lethality. The CKO of the histone deacetylase Hdac8 with Wnt1-Cre resulted in a partially formed skull vault (Haberland et al. 2009). Conditional inactivation of Ezh2 (polycomb repressive complex 2 member that catalyzes theis capable of forming skeletal tissue, including the bones and cartilages of the face, anterior cranial base, and frontal calvaria. The reasons why multipotent trunk neural crest cannot form skeletal tissue may be due to the expression of Hox genes, which supresses skeletogenic capacity (Le Douarin and Dupin 2016). Other restrictions on potency are related to environmen- tal influences, such as the levels of Sonic Hedgehog (Le Douarin et al. 2008). However, since it is clear that the majority of pre- migratory neural crest cells are not restricted to 1 fate, there is room for other mechanisms to regulate lineage restriction. Epigenetic regulation includes DNA methylation, histone acet- ylation and methylation, and microRNA-mediated translational regulation. All these mechanisms control neural crest cell speci- fication and epithelial-mesenchymal transformation and differ- entiation, as reported in several animal models (Hu et al. 2014). In mouse models, cranial neural crest cell derivatives can be targeted by crossing lines expressing Wnt1-Cre with those har- boring a floxed allele (Chai et al. 2000).
Conditional deletionmethylation of H3K27), was carried out with Wnt1-Cre (Schwarz et al. 2014) and Prrx-Cre (Dudakovic et al. 2015). In Wnt1-Cre CKO embryos, all anterior skull and facial skeletal elements failed to differentiate, whereas the Prrx-Cre embryos had craniosynostosis. The differences are likely due to the expression of Prrx-Cre in mesodermal and neural crest deriva- tives (Logan et al. 2002); therefore, tissue interactions between the dura and the calvaria may have been disrupted.Tri- or dimethylation of histone H3 at lysine 9 (H3K9me) has generally been associated with repressive chromatin states. H3K9me3, catalyzed by the Suv39h1/2 enzymes and Setdb1, is responsible for transcriptional repression in constitutively heterochromatic regions of the genome—for example, at cen- tromeres (Lehnertz et al. 2003), telomeres (Garcia-Cao et al. 2004), or genomic retroviral elements (Martens et al. 2005; Matsui et al. 2010). By contrast, H3K9me2 is catalyzed by G9A/EHMT2 (Tachibana et al. 2001; Tachibana et al. 2002) and its close homolog and heterodimeric partner GLP (Tachibana et al. 2005) and locates to euchromatic regions, where it is often associated with transcriptionally inactive genes (Peters et al. 2003). Importantly, G9A-dependent H3K9me2 appears to be a bona fide example for a tissue-specific chroma- tin mark, as it is organized in large chromosomal domains that increase in number and size during differentiation and occupy regions of tissue-specific gene repression (Wen et al. 2009).G9A has been implicated in the silencing of pluripotency genes in differentiating embryonic stem cells (Feldman et al. 2006), in the efficient induction and repression of cytokine genes during T helper cell differentiation (Lehnertz et al. 2010), and in neuronal gene expression (Roopra et al. 2004; Schaefer et al. 2009; Maze et al. 2010). At the same time, the role of G9A in the regulation of developmental transcription programs dur- ing embryogenesis remains largely unexplored. This is mainly due to the fact that mice carrying a homozygous G9a-null muta- tion exhibit severe overall growth retardation and die between day E8.5 and E12.5 of gestation (Tachibana et al. 2002).Here, we expand the analysis of G9A with respect to its role in regulating neural crest cell fate. Interestingly, whereas the overall facial bone architecture is intact in the absence of G9a, we observe a specific delay in the differentiation and growth of osteogenic progenitors. This indicates that G9a is important for the timing of differentiation and expansion of precursor cells in specific tissues but is otherwise dispensable for the execution of lineage diversification during neural crest cell specification.
The G9A effects on bone formation are mediated by methyla- tion of histones within Twist1 and Twist2 genes, encoding for transcription factors expressed in preosteogenic mesenchyme.The G9afl/fl mice (Lehnertz et al. 2010) were mated with trans- genic mice that express Wnt1-Cre (Rowitch et al. 1998; Chai et al. 2000) to delete G9a primarily in neural crest–derived cells (G9afl/fl,Wnt1-Cre+). Genotyping was carried out as described(Lehnertz et al. 2010). G9afl/+,Wnt1-Cre+, G9afl/+,Wnt1-Cre-, and G9afl/fl,Wnt1-Cre- were used as control embryos. The number of animals used are given in Appendix Table 1. Animal experiments were approved by the University of British Columbia animal care committee (protocol A09-0364), and our study complies with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines for preclinical animal studies.Micro–computed tomography was performed on a SCANCO Medical vivaCT 40 instrument at 3 resolutions (10 µm, p21; 17 µm, p14; or 34 µm, 8 mo), 55 kVp, and 80-µAs exposure (0.5-mm Al filter, calibrated with 1,200 mg of hydroxyapatite/cm3). Picrosirius red/alcian blue staining for histology and alizarin red/alcian blue staining for whole mount preparations was car- ried out as published (Ashique et al. 2002).Endogenous alkaline phosphatase (ALP) activity was detected by BCIP (5-bromo-4-chloro-3-indolyl phosphate p-toluidine salt) plus NBT (nitrotetrazolium blue chloride). Mouse mono- clonal antibodies were used against the following: G9A (1:100; Perseus Proteomics), GLP (1:100; Perseus Proteomics), H3K9me2 (ab1220, 1:100 [Abcam]; 1:50, pan-TWIST [recog- nizes TWIST1 and TWIST2; Abcam]), RUNX2 (1:50; Abnova), osteopontin (1:100; Thermo Fisher Scientific), and Col2a1 (116B3, 1:100; Developmental Studies Hybridoma Bank). The Mouse on Mouse detection kit was used (Vector Laboratories). Two or 3 biological replicates were analyzed for each genotype at each stage (Appendix Table 1).The proportion of BrdU-positive cells in the TWIST- positive osteogenic bone front was determined at E14.5 with a rat monoclonal antibody to BrdU (1:100; Abcam) and an anti- rat Alexa488-conjugated secondary antibody (1:500; Molecular Probes).
Three biological and technical replicates for each genotype were analyzed and statistical differences determined with t tests.G9A Inhibitor BIX-01294 Treatment of Micromass Cultures and Quantitative Real-time Polymerase Chain Reaction AnalysisMicromass cultures were prepared from E11.5 wild-type embryos (Weston et al. 2000). Mesenchyme from the mediona- sal and mandibular prominences from single embryos was pooled into a cell suspension of 2 × 107 cells/mL. Cultures were treated with 2µM BIX-01294 (Stemgent), which had minimal toxicity in dose-response experiments (0.5 to 5 µM). Following the 6-d culture period, 1 set of cultures was stained with alcian blue, ALP, and hematoxylin. Total culture area (hematoxylin) and ALP-positive area were measured with Adobe Photoshop. A second set of cultures was collected after6 d for RNA expression studies. Each biological replicate (n = 3) consisted of RNA pooled from 3 cultures. The RNeasy kit (Qiagen) was used to extract RNA, and cDNA was transcribed with a high-capacity reverse transcription kit (Applied Biosystems). TaqMan-based chemistry was used (Fast Universal PCR Master Mix; Applied Biosystems). Primers and probe sequences and cycling parameters were as published (Dranse et al. 2011; Appendix Table 2). Ct values were normalized to the expres- sion of Tbp (Tata Binding Protein). Student’s t tests were used to compare relative expression levels in BIX-01294/dimethyl sulfoxide (DMSO)-treated cultures.Primary osteoblasts were isolated from P1 or P2 mouse cal- varia as published (Jonason and O’Keefe 2014). Cells were seeded at 4 × 104 cells/cm2 and maintained in α-MEM + 10% fetal bovine serum (FBS) with β-glycerol phosphate (10 mM) and ascorbate (50 µg/mL). Cells were cultured for 8 d before nuclear DNA was isolated for chromatin immunoprecipitation (ChIP). ChIP was performed with a kit (catalog no. 9003; Cell Signaling Technology). Samples were immunoprecipitated with anti-H3K9me2 (ab1220; Abcam), anti-H3 (4620), and rabbit immunoglobulin G (catalog no. 2729; Cell Signaling Technology). Quantitative real-time polymerase chain reaction (qPCR) was carried out with primers close to the transcrip- tional starting site of Twist1 and Twist2 (Appendix Table 2). Data were expressed as a percentage of input DNA and nor- malized to the signal of glyceraldehyde-3-phophate dehydro-genase (Gapdh) promoter.
ResultsThe G9afl/fl,Wnt1-Cre+ mutants were severely underrepresented in litters collected at weaning (3 wk, 6 of 166 pups, 3.6%). Two- week G9afl/fl,Wnt1-Cre+ mice weighed much less than control lit- termates (42% ± 7% lower weight; n = 3), but the length of the skeleton and size of the body was the same (Appendix Fig. 1). Further analysis of the skulls at 2 and 3 wk revealed skull abnormalities, though no signs of cleft palate, which would have caused postnatal death (Fig. 1C, D). Instead the mutant mice had a large gap between the frontal bones (Fig. 1B, D, F, H; Appendix Fig. 2A, B, same animals as in Fig. 1D, H). There was a single surviving 8-mo-old animal, which also had the same calvarial defect (Appendix Fig. 3A, D). We ruled out a change in cell fate from osteogenic to chondrogenic in the sutures since there was no ectopic cartilage (Fig. 1H; Appendix Fig. 4). Other abnormalities included less dense bone in many of the facial bones (Fig. 1A–H, Appendix Fig. 2A–F), which was not as apparent in the 8-mo-old animal. The examination of the upper jaw revealed a significantly shortened maxilla (22% ± 0.02% shorter maxilla vs. control littermates, n = 3),which may have restricted the airway (Appendix Figs. 2, 3). The smaller jaws combined with a possible airway restriction could have led to the failure to gain weight and early postnatal demise.To assess the expression pattern of G9A and the presence of H3K9me2 in neural crest–derived structures, we analyzed head sections of control and mutant mice, starting from the time when ossification centers are forming (E13.5) up until birth. We found that G9A and its close homologue GLP were uniformly detectable in neural crest–derived and mesoder- mally derived structures of control E14.5 embryos (Fig. 1I, K) and that G9A was qualitatively ablated in neural crest–derived cells (membranous bone and cartilage) of G9afl/fl,Wnt1-Cre embryos (Fig. 1L). Importantly and consistent with other stud- ies (Tachibana et al. 2002; Lehnertz et al. 2010), levels of H3K9me2 and GLP were drastically reduced when G9A was deleted (Fig. 1M, M1, N).We wanted to pinpoint whether there was a defect in mesenchy- mal cell progenitors committing to an osteogenic lineage or whether there were problems with differentiation into osteo- blasts.
ALP staining is a marker of increased phosphatase activ- ity in osteogenic mesenchyme. Indeed, ALP staining is visible in the future site of the dentary bone at E12.5 in control mice (data not shown). Some mutant mice have delayed expression of ALP, but this result varied within a litter (data not shown). By E13.5, all mutant animals were positive for ALP staining in the dentary bone; however, the area was qualitatively smaller (Fig. 2A, D; also see E14.5, Fig. 2G, K). The G9a gene is expressed in chondrocytes, but we did not detect any differences in chon- drocyte specification at E12.5 or E13.5, as shown by COL2A1 immunofluorescence (Appendix Fig. 5A–H).Based on the smaller ossification centers at E13.5 and bone phenotypes postnatally, we wondered whether bone initiation was delayed in the G9afl/fl,Wnt1-Cre embryos. We examined 2 tran- scription factors: RUNX2, which is upregulated during osteo- blast commitment (Komori et al. 1997), and TWIST, which is expressed in neural crest–derived mesenchyme prior to com- mitment to a specific lineage (Miraoui and Marie 2010). We also included a marker for early nonmineralized bone matrix (osteoid), osteopontin (OPN; secreted phosphoprotein 1). In E13.5 mutants, RUNX2-positive cells overlapped the ALP staining (Fig. 2B, E), and expression in mutants was similar to that of controls, though not as widely distributed. TWIST expression was largely complementary to that of RUNX2, and this correlates with uncommitted mesenchymal cells that are not programmed to be osteogenic (Fig. 2C, F). At E14.5, RUNX2 was expressed in a complementary fashion to ALP and TWIST in control embryos (Fig. 2G, I). The first expression ofOPN was seen in the dentary bone (Fig. 2J), and this region lay within the RUNX2 expression domain. However, in E14.5 mutants, RUNX2 and TWIST expression overlapped at the edges of the RUNX2 domain in adjacent sections (Fig. 2L, M). Unexpectedly, OPN expression was absent in the mutant ani- mals (Fig. 2N). By E15.5, expression of TWIST had mostly downregulated except in the odontogenic mesenchyme (Fig.3A, B); however, in the mutants, TWIST was still present at the edges of the bone or in the osteogenic front (Fig. 3E). RUNX2 expression was present throughout the intramembranous bone (data not shown).
Unlike at E14.5, OPN was expressed in the ossification centers of controls and mutants (Fig. 3C, F). A delay in the onset of OPN expression was also observed in the frontal bone at E16.5 (Fig. 3G–L). Interestingly, there were differences in the advancement of the osteogenic fronts of the frontal bones at E14.5 to E16.5 (Fig. 3G, J; Appendix Fig. 6A– F). In control mice, the frontal bone had reached the dorsal sur- face of the forebrain (Fig. 3G), whereas in mutants, the bone front was located at the lateral eminence of the forebrain (Fig. 3J). RUNX2 expression was similarly less advanced toward themidline in mutants (Fig. 3H, K). No OPN expression was observed in the mutant frontal bone, even though the dentary had strong expression by this time (Fig. 3I, L; data not shown). The immunofluorescence and size differences of the differenti- ating intramembranous bone suggest that there may be growth inhibition of the osteogenic mesenchyme.One possible explanation for the bone defects was a decrease in proliferation of osteogenic mesenchymal cells or increase in apoptosis. There was no change in apoptosis (data not shown), but there was a significant decrease in percentage of BrdU- labeled cells in G9a CKO embryos (Fig. 4A–D; n = 3, P < 0.05). BrdU incorporation in cells of the Meckel’s cartilage was equiv- alent between controls (Fig. 4C; 9.5% ± 0.4%) and mutants (Fig. 4D; 8.7% ± 0.9%). Thus, in addition to a delay in differentiation, a decrease in the proliferation of osteogenic progenitors may contribute to the reduced bone mass in G9afl/fl,Wnt1-Cre+ mice.Methyltransferase Activity of G9A Is Required for Cartilage and Bone Formation In VitroTo discern how G9A-dependent transcriptional regulation was responsible for the observed phenotype, we investigated the markers for osteoblasts and chondrocytes in primary cultures of facial mesenchyme.
We employed a small molecule antagonist of heterodimers of G9A/GLP, BIX-01294 (Kubicek et al. 2007; Chang et al. 2009). BIX-01294 inhibits enzymatic activity medi- ated by G9A/GLP heterodimers. The proportion of ALP-positive matrix was not decreased in BIX-01294-treated micromass cul- tures (Fig. 5A, C); however, the number of cartilage nodules was significantly reduced (Fig. 5B, D; n = 12 ± 7 for BIX-01294 and n = 116 ± 10 for controls). The expression of Runx2, Sp7, Spp1, Bsp, Bglap, Alpl, Ocn, and Opn were significantly decreased in BIX-01294-treated samples as compared with controls (Fig. 5E). In contrast, levels for uncommitted mesenchyme Twist1 and Twist2 were significantly upregulated in BIX-treatedcultures, while Sox9 and cartilage matrix genes Acan (aggrecan) and Col2a1 (collagen alpha-1[II] chain) were significantly decreased (Fig. 5E). One possible explanation for the discrep- ancy between in vivo and in vitro cartilage results is that the drug BIX-01294 inhibits G9A/GLP activity, whereas the mutants could still have activity from homodimers of GLP leading to normal cartilage differentiation. It appears that the loss of meth- ylation is indirectly leading to a reduction of osteogenic geneexpression, possibly mediated by the increase in Twist1 andTwist2 expression.The prolonged expression of TWIST protein in G9afl/fl,Wnt1-Cre+ mice and Twist1/2 upregulation in BIX-01294-treated micro- mass cultures suggest that Twist genes are directly regulated by G9A. In addition, the timing and complementary expression of RUNX2, OPN, and TWIST in our immunofluorescence stud- ies suggested that TWIST is a negative regulator of RUNX2 expression. The loss of methylation on the promoter of Twist in G9a mutants could have led to prolonged expression and, thus, the repression of osteogenesis in a temporally specific manner. Therefore, we analyzed the methylation status of Twist genes in response to BIX-01294 with ChIP assays. More than 107 cells are needed to do a ChIP; as such, it is difficult to obtain enough cells from mouse embryonic facial mesenchyme. Moreover, as we have seen, there are temporal differences in the timing of osteogenesis in different regions of the jaws and skull. Pooling all the cells from E14.5 to E16.5 embryos would obscure methylation status changes. Thus, we used osteoblasts derived from P1 or P2 calvaria. Expression of RNA markers for bone differentiation typically spikes between 6 and 8 d in primary osteoblast cultures (Jonason and O’Keefe 2014), so we reasoned that Twist expression would normally be methyl- ated and therefore repressed around this time. Indeed, the pres- ence of the H3K9me2 mark was confirmed in control cells at 8 d (Fig. 5F). H3K9me2 was significantly reduced at both the Twist1 and Twist2 loci in BIX-01294-treated osteoblasts as compared with DMSO-treated osteoblasts (Fig. 5F). Given these in vitro data, we conclude that loss of G9A activity in the mutants leads to histone demethylation and subsequent pro- longed activation of Twist genes, transiently repressing Runx2 and the osteogenic program.
Discussion
G9A appears to be required in a temporally restricted manner, following lineage specification, during the expansion and dif- ferentiation of osteoprogenitor cells. Although G9A and H3K9me2 are expressed in all neural crest–derived head mesen- chyme as well as in mesodermal components, the osteogenic mesenchyme appears to be the main target tissue. Importantly, deletion has no obvious impact on the specification, migration, or generation of most cranial neural crest–derived lineages. Deletion of G9a in Neural Crest Cells Reveals a Specific Epigenetic Regulation of the Timing of Twist Gene Expression During Intramembranous Bone Formation To our knowledge, our study represents the first report of the requirement of a histone methyltransferase being involved in the timing of intramembranous bone differentiation. Indeed, we saw approximately a 24-h delay in the mandibular and fron- tal bone differentiation in the mutants, as shown by OPN stain- ing, which was accompanied by prolonged expression of TWIST. Additionally, the ChIP analysis showed that the TWIST promoter is methylated; thus, its expression could be repressed by G9A. Supporting these ideas, we showed that chemical antagonism of G9A catalytic activity leads to upregu- lation of Twist1 and Twist2 RNA. The CKO of Twist1 with Wnt1-cre blocks intramembranous ossification (Zhang et al. 2012; Bildsoe et al. 2013). The transcriptional repression of Runx2 by TWIST has been described (Bialek et al. 2004). These authors showed that Twist and Runx2 are initially coexpressed and that it is necessary for TWIST to be downregulated before osteogenesis can begin. We have now added to this story by showing that epigenetic regulation of TWIST is the mecha- nism for derepression of RUNX2 at the correct time and place.
Conclusion
The novel result discovered here is that H3K9 dimethylation is a mechanism used to control the ossification of intramembra- nous bones in the head. Indeed, such a precise epigenetic mechanism is likely applicable to all bones, including those derived from paraxial mesoderm. Our model is that by main- taining TWIST expression in G9a mutants, the number of pro- liferating osteoprogenitor cells is reduced at the critical time when bone size is being determined. From the single 8-mo ani- mal (probably a hypomorph), we can see that this transient block can be overcome in time such that most of the bones are fully formed except for the calvaria. There may also be clinical relevance to our findings. Haploinsufficiency of TWIST1 (Paznekas et al. 1998) causes craniosynostosis, perhaps mimicking the condition of hyper- methylation. In addition, the condition of cleidocranial dysos- tosis causes hypoplastic bones and delayed ossification of the fontanels (Online Mendelian Inheritance in Man: 600211). Most patients have loss-of-function mutations in RUNX2. There is a promising recent animal study BIX 01294 demonstrating that epigenetic modifier enzymes are suitable therapeutic targets. By antagonizing histone deacetylases, it is possible to rescue the open sutures in Runx2 heterozygous mice (Bae et al. 2017). Thus, future therapies that target methylation could also be valuable in treating developmental abnormalities in ossification.