GADD45A Protein Plays an Essential Role in Active DNA Demethylation during Terminal Osteogenic Differentiation of Adipose-derived Mesenchymal Stem Cells*

Methylation and demethylation of DNA are the complementary processes of epigenetic regulation. These two types of regulation influence a diverse array of cellular activities, including the maintenance of pluripotency and self-renewal in embryonic stem cells. It was generally believed that DNA demethylation occurs passively over several cycles of DNA replication and that active DNA demethylation is rare. Recently, evidence for active DNA demethylation has been obtained in several cancer, neuronal, and embryonic stem cell lines. Studies in embryonic stem cell models, however, suggested that active DNA demethylation might be restricted to the early development of progenitor cells. Whether active demethylation is involved in terminal differentiation of adult stem cells is poorly understood. We provide evidence that active DNA demethylation does occur during terminal specification of stem cells in an adipose-derived mesenchymal stem cell-derived osteogenic differentiation model. The medium CpG regions in promoters of the Dlx5, Runx2, Bglap, and Osterix osteogenic lineage-specific genes were demethylated during the increase in gene expression associated with osteogenic differentiation. The growth arrest and DNA damage-inducible protein GADD45A was up-regulated in these processes. Knockdown of GADD45A led to hypermethylation of Dlx5, Runx2, Bglap, and Osterix promoters, followed by suppression of the expression of these genes and interruption of osteogenic differentiation. These results reveal that GADD45A plays an essential role in gene-specific active DNA demethylation during adult stem cell differentiation. They enhance the current knowledge of osteogenic specification and may also lead to a better understanding of the common mechanisms underlying epigenetic regulation in adult stem cell differentiation.

Methylation and demethylation of DNA are the complementary processes of epigenetic regulation. These two types of regulation influence a diverse array of cellular activities, including the maintenance of pluripotency and self-renewal in embryonic stem cells. It was generally believed that DNA demethylation occurs passively over several cycles of DNA replication and that active DNA demethylation is rare. Recently, evidence for active DNA demethylation has been obtained in several cancer, neuronal, and embryonic stem cell lines. Studies in embryonic stem cell models, however, suggested that active DNA demethylation might be restricted to the early development of progenitor cells. Whether active demethylation is involved in terminal differentiation of adult stem cells is poorly understood. We provide evidence that active DNA demethylation does occur during terminal specification of stem cells in an adipose-derived mesenchymal stem cell-derived osteogenic differentiation model. The medium CpG regions in promoters of the Dlx5, Runx2, Bglap, and Osterix osteogenic lineage-specific genes were demethylated during the increase in gene expression associated with osteogenic differentiation. The growth arrest and DNA damageinducible protein GADD45A was up-regulated in these processes. Knockdown of GADD45A led to hypermethylation of Dlx5, Runx2, Bglap, and Osterix promoters, followed by suppression of the expression of these genes and interruption of osteogenic differentiation. These results reveal that GADD45A plays an essential role in gene-specific active DNA demethylation during adult stem cell differentiation. They enhance the current knowledge of osteogenic specification and may also lead to a better understanding of the common mechanisms underlying epigenetic regulation in adult stem cell differentiation.
Methylation of genomic DNA is one of the most important epigenetic mechanisms for gene regulation and is critical for a variety of cellular activities, including X chromosome inactiva-tion, genomic imprinting, chromatin modification, and the silencing of endogenous genes (1). Alterations in DNA methylation are linked to many diseases, including imprinting disorders, autoimmune syndromes, and some cancers (2,3). In humans and mammals, DNA methylation predominantly occurs at CpG dinucleotides that are largely depleted from the genome except at short genomic regions called CpG islands. Methylation is mediated by a number of DNA methyltransferases (DNMT), 3 including maintenance enzyme DNMT1 and de novo methyltransferases DNMT3A/3B (4,5).
In contrast to DNA methylation, less is known about DNA demethylation. It was generally believed that DNA demethylation occurred mainly by passive processes during DNA replication (6), although the occurrence of active DNA demethylation has long been elusive (7). Recently, active DNA demethylation was demonstrated in several cells, such as human embryonic kidney cells, breast cancer cells, T cells, and neurons (8 -12), and several functions of active demethylation have been described (8 -10, 13, 14). For example, active DNA demethylation occurs at the genome-wide levels during early embryonic development in humans and several animal models, such as zebrafish (Danio rerio), Xenopus laevis, and mouse (15)(16)(17). Paternal DNA demethylation occurs after fertilization and during the development of primordial germ cells (18 -21). Active DNA demethylation also occurs at the interleukin-2 locus (8) and is critical for cyclical DNA demethylation at estrogen receptor ␣ targets (11). In embryonic stem cell (ESC) models, it has shown that active DNA demethylation usually takes place in progenitor cell differentiation but rarely occurs in terminal cell specification (22)(23)(24), suggesting that active DNA demethylation is restricted to the early development of progenitor cells (25). Whether active DNA demethylation occurs in the terminal differentiation of adult stem cells has yet to be determined.
Although active DNA demethylation occurs during various cellular processes, the mechanism underlying this process remains a matter of dispute (26). A recent study revealed that the growth arrest and DNA damage-inducible ␣ (GADD45A) protein was critical for active DNA demethylation in X. laevis oocytes (16), suggesting GADD45A might be a central regula-tor of active DNA demethylation during other processes. The GADD45 family consists of three members, GADD45A, GADD45B, and GADD45G, that are highly conserved and are key mediators of cellular stress responses (27). Among these family members, the GADD45A has received the most attention. It is a small (18 kDa) p53-regulated histone-fold protein known to participate in the regulation of cell proliferation, DNA repair, cell cycle, and apoptosis (28,29). Consistent with these diverse functions, GADD45A interacts with a multitude of proteins, including PCNA, MAP3K4, XPG, p21, Cdk1, and histones (28, 30 -32). Several animal models were established to study the role of GADD45A in active DNA demethylation, including X. laevis, D. rerio, and mouse. As in X. laevis, GADD45A was found to promote active DNA demethylation in zebrafish zygote development (15). However, not all studies support the involvement of GADD45A in active DNA demethylation. Several studies failed to substantiate the role of GADD45A in demethylation of DNA in mice (33,34). For example, neither global nor locus-specific methylation was increased in Gadd45a ؊/؊ mice (34). Recently, GADD45A was shown to be recruited to the rDNA promoter and to promote demethylation and rRNA transcription in HEK293T cells (12). However, the rDNA promoter is a pol I promoter recognized by RNA polymerase I rather than polymerase II. Therefore, the question of whether GADD45A participates in active DNA demethylation at a gene-specific pol II promoter and promotes mRNA transcription in mammals is still unanswered.
In this study, the role of GADD45A in adult stem cell differentiation was investigated. The adipose-derived mesenchymal stem cell (ADSC)-derived osteogenic specification model, one of the most widely used adult stem cell differentiation models, was employed for this purpose (35). The results clearly showed that active DNA demethylation occurs at osteogene-specific gene promoters during terminal adult stem cell differentiation and that GADD45A plays an essential role in this process.

EXPERIMENTAL PROCEDURES
Isolation and Culture of ADSCs-Six-week-old male ICR mice obtained from the Laboratory Animal Unit of Zhejiang Academy of Medical Sciences (Hangzhou, China) were used in the experiments. All animal experiments were performed in accordance with legal regulations, including approval by a local ethics committee. The ADSCs were isolated as described previously (36). Briefly, the epididymal fat pads were excised and washed extensively with sterile phosphate-buffered saline (PBS) to remove contaminating debris. Tissues blocks were then cut into small pieces and incubated with 0.075% type I collagenase (Sigma) in PBS for 60 min at 37°C with vigorous agitation. After neutralization of the collagenase, cells released from adipose fragments were filtered and washed three times with Dulbecco's modified Eagle's medium: Nutrient Mixture F-12 (DMEM/F-12, Hyclone). Cells were seeded in T-175 flasks (Greiner Bio-One GmbH, Germany) at 5 ϫ 10 6 cells/ml in DMEM/F-12 supplemented with 10% fetal bovine serum (FBS, Hyclone, MD), 100 units/ml penicillin (Invitrogen), and 100 g/ml streptomycin (Invitrogen). After 24 h, nonadherent cells and debris were removed. Cells were harvested by 0.25% trypsin (Sigma) at 80% confluence. For most of the experiments, ADSCs at the 2nd to 4th passages were used for experiments. Media were changed twice a week during culture.
Identification of Isolated ADSCs-The isolated ADSCs were identified by their antigen expression profiles. Briefly, cells were harvested by trypsinization and incubated with the following rat to mouse monoclonal antibodies: FITC-conjugated CD11b and CD45 and PE-conjugated CD44, CD73, and CD90 (all from Caltag Laboratories, San Diego). Cell fluorescence signals were measured using a FACScan flow cytometer (BD Biosciences) equipped with an argon laser, with emission at 488 nm. At least 10,000 events were collected. Data were analyzed with Cell-Quest software (BD Biosciences).
Osteogenic Induction of ADSCs-To test for osteogenic induction, ADSCs at over 90% confluence were incubated in a differentiation medium for 2 weeks, with medium changes every 3 days. The differentiation medium was composed of DMEM/F-12 supplemented with 10% FBS, 10 Ϫ8 M dexamethasone (Calbiochem), 50 M ascorbic acid 2-phosphate trisodium salt (Fluka Chemie GmbH, Switzerland), and 10 mM ␤-glycerophosphate disodium (Sigma). To detect the effect of DNA methylation on osteogenic gene expression, 5 M of the DNMT inhibitor 5-aza-2-deoxycytidine (Sigma) was added to differentiation medium and used to treat 50% confluent ADSCs for 48 h.
Quantification of mRNA by Real Time PCR-After osteogenic induction for 6 or 9 days, total RNA was isolated using TRIzol reagent (Invitrogen) according to the manufacturer's instructions and quantified by UV spectroscopy. One microgram of total RNA was reverse-transcribed to complementary DNA using the SuperScript III first-strand synthesis system (Invitrogen) with oligo(dT) primers. Real time PCR was performed with Platinum SYBR Green qPCR SuperMix-UDG with ROX (Invitrogen) according to the manufacturer's instructions. The PCR thermocycle profile was stepped at 94°C for 2 min, followed by 40 cycles at 94°C for 20 s, 60°C for 20 s, and 68°C for 20 s. Signals were detected with a Realplex5 real time PCR system instrument (Eppendorf, Germany). The gene-specific primers for Gapdh, Dlx5, Hoxa10, Runx2, Osterix, Col1a1, Bglap, Gadd45a, Gadd45b, and Gadd45g are listed in supplemental Table 1. All transcript levels were normalized to that of Gapdh.
Bisulfite Sequencing-Bisulfite conversion was performed as described previously (37). Briefly, total genomic DNA was isolated from ADSCs using a DNeasy tissue kit (Qiagen). The genomic DNA (2 g) was denatured in a volume of 50 l of freshly prepared NaOH (final concentration 0.3 M) for 30 min at 42°C. After denaturation, 30 l of freshly prepared hydroquinone (10 mM) and 510 l of sodium bisulfite (3.6 M, pH 5.0) were added and incubated at 50°C for 16 h. The modified DNA was purified using the DNeasy spin column (Qiagen) and then desulfonated by adding NaOH to a final concentration of 0.3 M for 15 min at 37°C. Samples were neutralized by adding 33 l of ammonium acetate (10 M, pH 7.0), followed by ethanol precipitation and resuspension of DNA in 40 l of 1 mM Tris-Cl (pH 8.0). Nested primers (supplemental Table 1) were used for PCR, and the products were gel-purified using the QIAEX II gel extraction kit (Qiagen). Transcripts were then cloned into pUCm-T vector (Sangon, China) and transformed into Escherichia coli strain DH5␣. The DNA samples from at least six positive clones per original set of cells were sequenced. The osteogenic gene promoter sequences were analyzed at UCSC Genome Informatics browser database, and the patterns of methylation were evaluated using DNAMAN (Lynnon Corp., Canada).
Preparation of Gadd45a, Gadd45b, and Gadd45g shRNAs Lentiviruses-The short hairpin RNA (shRNA) oligonucleotides designed for this experiment contained sense strands of mouse Gadd45a, Gadd45b, and Gadd45g nucleotide sequences, followed by short spacers (TTCAAGAGA), the reverse complement of the sense strands, and five thymidines as RNA polymerase III transcriptional stop signal. The oligonucleotides used in generating the Gadd45a, Gadd45b, and Gadd45g shRNA (10,38), as well as the scrambled shRNA (as the negative controls), are shown in supplemental Table 2. These oligonucleotides were annealed and ligated into the BglII and HindIII sites of pSUPER.retro.puro vector (Oligoengine, WA). The oligonucleotides and H1 promoter were then subcloned into lentiviral PLB plasmid (Addgene plasmid 11619). All the plasmids were confirmed by sequence analysis. Replication-defective lentiviruses were produced by co-transfecting lentiviral vector (for Gadd45a, Gadd45b, or Gadd45g knockdown), packaging vectors pCMV-dR8.2 dvpr (Addgene plasmid 8455), and pCMV-VSVG (Addgene plasmid 8454) into 293T cells using Lipofectamine 2000 (Invitrogen). After transfection, three samples of viruses were harvested at 48, 60, and 72 h and filtered through a Millex-HV 0.45-m PVDF filter (Millipore, Ireland). The lentiviral particles were concentrated by centrifugation at 26,000 rpm for 90 min. Pellets were left for resuspension overnight at 4°C in PBS. After gentle pipetting, aliquots were stored at Ϫ80°C. Infection of ADSCs was performed by 6 h of exposure to dilutions of viral supernatant in the presence of Polybrene (5 g/ml). The efficiency of transfection was evaluated by determining the percentage of green fluorescent protein (GFP)-positive cells under a fluorescent microscope. Differentiation of infected ADSCs was then induced for 14 days to evaluate osteogenesis.
Alkaline Phosphatase and Calcium Mineralization Assays-Alkaline phosphatase (ALP) activity was detected by histochemical staining after osteogenic induction for 14 days. Modified Gomori staining was performed as reported previously (39). For calcium mineralization assays, cells were fixed in 4% paraformaldehyde after 14 days of culture and stained with 0.1% Alizarin red S (Sigma) for 20 min. Mineralization was quantified by measurement of the average photo density of calcium mineralization with a pathology image analysis system (Mike Audi Image Analysis Inc., China). Measurements were carried out in duplicate, and each experiment was repeated at least three times.
Western Blot Analysis-ADSC cultures were washed twice with ice-cold phosphate-buffered saline (PBS) and then lysed in 40 ml of lysis buffer (Promega, WI) and 1 ml of proteinase inhibitor mixture (Sigma). The total protein concentration was measured using a Bradford assay containing Coomassie Plus protein reagent (Bio-Rad) according to the manufacturer's instructions. Equivalent amounts of total cell lysate per lane were separated by SDS-PAGE using 12% polyacrylamide gels. Proteins were electroblotted to PVDF membranes (Millipore, MA). The membranes were then blocked and incubated in anti-GADD45A antibody (rabbit polyclonal; 1:1,000; Millipore, MA) overnight at 4°C. HRP-conjugated goat anti-rabbit IgGs (1:1,000; Millipore, MA) were used as secondary antibodies for detection. The membranes were incubated with enhanced chemiluminescence detection reagents (Bioind, Israel) according to the manufacturer's instructions and detected using the Versa Doc imaging system (Bio-Rad).
Immunofluorescence Staining-Immunofluorescence staining was performed as described previously (21). Briefly, the cells grown on coverslips were washed three times with PBS, fixed using 4% paraformaldehyde, and then permeabilized for 15 min with PBS containing 0.5% Triton X-100. Nonspecific binding sites were blocked with 10% goat serum in PBS containing 0.05% Tween 20 for 1 h at room temperature. The cells were then incubated overnight at 4°C with rabbit anti-GADD45A antibodies (Millipore) at 1:500 dilution, washed, and incubated with Alexa Fluor 555-conjugated goat anti-rabbit IgG antibodies at 1:1000 dilution (Invitrogen). Their nuclei were visualized using 1 M DAPI (Sigma) and examined under a Zeiss fluorescent microscope (Carl Zeiss Axiovert 40 CFL).
Chromatin Immunoprecipitation (ChIP) Assay-The ChIP assay was performed as described previously (25). Briefly, 2 ϫ 10 7 cells were cross-linked with 1% formaldehyde for 15 min at room temperature. The reaction was stopped through the addition of 125 mM glycine. The cells were collected, lysed, and sonicated to obtain chromatin fragments with ϳ200 -800-bp length. The chromatin aliquots were precleared with protein A-Sepharose coated with sonicated salmon sperm DNA and bovine serum albumin. The samples were subjected to immunoprecipitation using anti-GADD45A antibodies for 2 h at 4°C; rabbit IgG was used as the control. Immune complexes were mixed with precoated protein A-Sepharose and incubated overnight via constant rotation at 4°C. The immunoprecipitated DNA was further reverse cross-linked, purified, and subjected to quantitative PCR analysis using SYBR Green reagent (Invitrogen). The primer sequences are listed in supplemental Table 2.
Statistical Analysis-All data are presented as the mean value Ϯ S.D. of each group. Statistical differences between group means were evaluated using the Student's t test, and p Ͻ0.05 was accepted as statistically significant.

Identification of ADSCs and Osteogenic Differentiation-
Freshly isolated cells from epididymal fat pads mainly consisted of erythrocytes and nonadherent cells. After 24 h, nonadherent cells were removed by changing the culture medium. The ADSCs were subcultured until passage 4 when they had grown to 80% confluence. The isolated cells were identified by immunophenotypic and functional analysis, respectively. Immunophenotypic analysis by flow cytometry showed that the majority of cells after multiple passages were negative for CD11b and CD45, which are hematopoietic lineage markers, although positive for CD44, CD73, and CD90, which are known as the cellsurface antigens of mesenchymal stem cells (Fig. 1A). The osteogenic differentiation potency was examined for functional evaluation. The ADSCs were cultured in an osteogenic inducing medium for 14 days before characterization. By histochem-ical staining by ALP and Alizarin red S for calcium mineralization, it was shown that the majority of cells exhibited a high potency for osteogenic differentiation (Fig. 1, B-D). These results clearly demonstrated that the subcultured cells were ADSCs.
Gene Expression during Osteogenic Differentiation-To reveal the expression patterns of osteogenic genes during differentiation and to determine which genes were most appropriate for subsequent demethylation analysis, six osteogenic lineage genes (Dlx5, Hoxa10, Runx2, Osterix, Bglap, and Col1) were selectively assayed by qRT-PCR after 3, 6, and 9 days in osteogenic differentiation medium. As shown in Fig. 2, there were no noticeable changes in expression of any of the examined genes on day 3. Most osteogenic genes, however, were significantly up-regulated by day 6. A greater than 200% increase was observed in Dlx5, Runx2, Bglap, and Col1 transcripts on day 6, whereas Osterix increased 79-fold. Osteogenic gene expression was still elevated on day 9 except for Dlx5 and Col1. This induction pattern is typical of osteogenic specification, thereby validating the in vitro model of osteogenic differentiation of ADSCs. Indeed, this is an attractive model for osteogenic differentiation because the ADSCs are easy to prepare and maintain and replicate many aspects of the in situ gene expression pattern during differentiation. Furthermore, all the genes ana-   Runx2, Col1a1, Osterix, Hoxa10, and Bglap expression was assayed by qRT-PCR on days 3, 6, and 9 at early stage of osteogenic differentiation. Each gene expression pattern was normalized to Gapdh. Compared with undifferentiated ADSCs, there were no noticeable changes on day 3; however, the expression of most osteogenic genes increased significantly on day 6. A Ͼ200% increase was seen in Dlx5, Runx2, Bglap, and Col1 transcripts on day 6, and Osterix increased 79-fold. Osteogenic gene expression was still elevated on day 9 except Dlx5 and Col1. *, p Ͻ 0.05; **, p Ͻ 0.01. lyzed were suitable for further studies of DNA demethylation as all the expressions were increased during osteogenic differentiation.
DNA Hypomethylation Was Associated with Osteogenic Differentiation-5-AzadC is a typical inhibitor of DNA methyltransferase widely used to produce DNA hypomethylation. Our previous study revealed that 5-AzadC promoted osteogenic gene expression by repression of DNA methylation (40). To determine whether DNA hypomethylation was associated with osteogenic differentiation of ADSCs, the expression of Dlx5, Hoxa10, Runx2, Osterix, Bglap, and Col1 was examined under differentiation conditions during DNMT inhibition. As expected, 5-AzadC promoted the expression of the osteogenic genes, Dlx5, Runx2, Osterix, and Bglap. For example, a Ͼ300% increase in the expression of Osterix, Dlx5, and Bglap was observed in 5-AzadC-pretreated cells on days 6 and 9 (Fig. 3). In contrast, neither Hoax10 nor Col1 expression was enhanced by 5-AzadC, suggesting that only Osterix, Dlx5, Runx2, and Bglap are regulated by demethylation during differentiation. An actual role for active DNA demethylation was then examined.
DNA Demethylation Was Associated with Osteogenic Gene Expression-Previous high throughput experiments demonstrated that down-regulation of gene expression by DNA methylation partly depended on CpG density and that promoters containing medium CpG density are more likely to be regulated by methylation/demethylation (23,41,42). In this study, all of the six osteogenic genes examined (Dlx5, Hoxa10, Runx2, Bglap, Osterix, and Col1) were found to contain medium CpG density regions within their promoters. Among these six, the Dlx5 and Hoxa10 promoters also contain high density CpG islands. Therefore, the methylation status of medium CpG density regions in Runx2, Bglap, Osterix, and Col1 promotors and both medium CpG density regions and CpG islands in Dlx5 and Hoxa10 promoters were tested. As expected, five of the medium CpG density regions of Dlx5, Hoxa10, Runx2, Bglap, and Osterix promoters were highly methylated in undifferentiated ADSCs (the percent of methylated CpG is more than 80% in Dlx5, Hoxa10, Runx2, and Bglap promoters and 25% in Osterix promoter), with the exception of medium density region of Col1 (2%), whereas that in high density CpG islands of Dlx5 and Hoxa10 was unmethylated (Fig. 4). It was consistent with the results from previous high throughput experiments, suggesting CpG islands might not be involved in the methylation modification. After 6 days of osteogenic induction, the methylated regions within the Dlx5, Runx2, Bglap, and Osterix promoters were dramatically demethylated except Hoxa10. For example, the methylation level of the medium density regions of the Dlx5 promoter (Ϫ1133 to Ϫ668) was decreased from 86.1 to 56.5%, the Runx2 promoter (Ϫ868 to Ϫ509) from 80.6 to 66.7%, the Bglap promoter (Ϫ1060 to Ϫ781) from 85 to 77.5%, and the Osterix promoter (Ϫ912 to Ϫ549) from 25 to 1.4% in differentiated cells (Fig. 4A). Notably, the demethylation of CpGs arose mainly at specific sites, such as the 6th (Ϫ913) and 7th (Ϫ800) CpG in the Dlx5 promoter, the first three (Ϫ820 to Ϫ808) CpGs in the Runx2 promoter, the 2nd (Ϫ1003) CpG in the Bglap promoter, and the 4th to 12th (Ϫ727 to Ϫ632) CpGs in the Osterix promoter (Fig. 4B). Unexpectedly, the methylation level of the medium density regions of the Hoxa10 promoter was not significantly declined during the osteogenic differentiation (Fig. 4), despite the increased gene expression. Together with the observation that Col1 was less methylated, and neither Hoax10 nor Col1 expression could be enhanced by 5-AzadC-induced hypomethylation, it seems reasonable to suggest that the expression of Hoxa10 and Col1 gene might not be regulated by DNA methylation, and other regulatory mechanisms might be involved in this process. However, DNA methylation regulates the expression of the most osteogenic genes, including Dlx5, Runx2, Bglap, and Osterix, as evidenced by the decrease in promoter methylation concomitant with enhanced gene expression during terminal differentiation.
Knockdown of Gadd45a Depresses Osteogenic Differentiation-To determine whether Gadd45a was required for osteogenic differentiation of ADSCs, we initially examined whether ADSCs expressed Gadd45a and whether expression changed during differentiation. As expected, qRT-PCR analysis showed that Gadd45a expression gradually increased over the first 6 FIGURE 3. 5-AzadC promotes osteogenic gene expression. ADSCs pretreated with 10 M 5-AzadC for 48 h were induced for 6 or 9 days in osteogenic medium before analysis. Each gene expression pattern was normalized to Gapdh. 5-AzadC promoted Dlx5, Runx2, Osterix, and Bglap expression but had no effect on Hoax10 and Col1 expression.*, p Ͻ 0.05; **, p Ͻ 0.01. NOVEMBER 25, 2011 • VOLUME 286 • NUMBER 47 days of osteogenic induction (Fig. 5B), suggesting that Gadd45a may regulate the expression of osteogenic genes. To confirm this, Gadd45a levels in ADSCs were suppressed by RNAi. The oligonucleotides for generating Gadd45a shRNA were ligated into a lentiviral vector that also contain a CMV-EGFP expression cassette to produce EGFP-tagged infective viruses. After 3 days of infection with the constructed lentivirus vectors, the majority of ADSCs strongly expressed EGFP, demonstrating the high efficacy of the viral infection (Fig. 5A). In addition, Gadd45a expression was dramatically decreased (p Ͻ 0.01) in ADSCs infected with the Gadd45a shRNA lentivirus compared with cultures infected with the scrambled shRNA (Fig. 5C). Western blot analysis demonstrated the presence of GADD45A protein in scrambled shRNA-expressing ADSCs. However, this protein could not be detected in ADSCs infected with Gadd45a shRNA constructs (Fig. 5D).

DNA Demethylation during ADSC Osteogenic Differentiation
To evaluate the role of GADD45A in osteogenic differentiation, osteogenic gene expression was examined in Gadd45adepleted ADSCs. As shown in Fig. 6D, the expression of Osterix, Dlx5, Runx2, and Bglap significantly decreased (p Ͻ 0.01) in Gadd45a-depleted cells compared with control cells infected with scrambled shRNA and uninfected osteogenic differenti- . Bisulfite sequencing analysis of DNA methylation. A, methylation levels of Dlx5, Hoxa10, Runx2, Bglap, and Osterix promoters were analyzed in undifferentiated ADSCs and after 6 days of osteogenic induction. In both undifferentiated and osteogenic differentiated ADSCs, the CpG islands in the Dlx5 promoter from ϩ190 to ϩ375 and the Hoxa10 promoter from Ϫ189 to ϩ31 were totally unmethylated, and the Col1 promoter from Ϫ1628 to Ϫ1334 was also unmethylated. In contrast, the percentage of methylated CpGs in medium CpG density regions of Dlx5, Hoxa10, Runx2, and Bglap promoters was more than 80% in undifferentiated ADSCs and that in the Osterix promoter was 25%. After differentiation, most methylated gene promoters were partly demethylated except Hoxa10. B, percentage of individual methylated CpGs (closed circles in A) in undifferentiated (solid line) or osteogenic differentiated ADSCs (dashed line). The demethylation of CpGs was largely restricted to particular sites, such as the 6th and 7th CpG in the Dlx5 promoter, the first three CpGs in the Runx2 promoter, the 2nd CpG in the Bglap promoter, and the 4th to 12th CpGs in Osterix promoter. CpG sites (tick marks) in the regions analyzed by bisulfite sequencing are underlined by black bars. Numbers are relative to the transcription start site (TSS). Each row represents one bacterial clone with one circle symbolizing one CpG. Closed circles, methylated CpG; open circles, unmethylated CpG. ated cells. In contrast, there was no notable change in Hoax10 or Col1 expression, indicating that GADD45A is involved in Osterix, Dlx5, Runx2, and Bglap gene regulation but not in Hoax10 and Col1 regulation, consistent with the 5-AzadC-induced hypomethylation and DNA demethylation assessment results. The osteogenic differentiation of Gadd45a-depleted ADSCs was also measured by ALP staining and calcium mineralization. As shown in Fig. 6, A and C, the amount of ALP was significantly decreased (p Ͻ 0.01) in Gadd45a-depleted ADSCs compared with control cells expressing the scrambled shRNA. In addition, calcium mineralization was dramatically inhibited (p Ͻ 0.01) by knockdown of Gadd45a (Fig. 6, B and C), confirming that Gadd45a expression is essential for osteogenic differentiation of ADSCs.
Knockdown of Gadd45a Depresses DNA Demethylation-To investigate the relationship between down-regulation of osteogenic genes and DNA methylation, we analyzed the methylation levels of Dlx5, Runx2, Bglap, and Osterix promoters using bisulfite sequencing after 6 days of osteogenic induction in Gadd45a-depleted ADSCs. Methylation levels of Dlx5, Runx2, Bglap, and Osterix promoters were markedly decreased (p Ͻ 0.01) in differentiating cells infected with the scrambled shRNA vector compared with undifferentiated ADSCs (Figs. 4 and 7). In contrast, the Dlx5, Runx2, and Osterix promoters remained highly methylated in Gadd45a-depleted cells, similar to undifferentiated cells (Figs. 4 and 7). These results clearly demonstrated that knockdown of Gadd45a inhibited demethylation of these gene promoters and that Gadd45a is an essential factor for the demethylation of Dlx5, Runx2, Bglap, and Osterix promoters during osteogenic differentiation of ADSCs.
Interaction of GADD45A with Promoters of Osteogenic Genes-A subcellular localization analysis and DNA binding assay of GADD45A was performed to provide further evidence that GADD45A participates in the demethylation of osteogenic gene promoters. Immunofluorescence staining showed that GADD45A is distributed in the nucleus of ADSCs (Fig. 8A). This result suggests that GADD45A plays a role in the nucleus and has the potential to bind to genomic DNA. ChIP-qPCR analysis demonstrated that GADD45A binds to the promoters of the Dlx5, Runx2, Osterix, and Bglap genes during the osteogenic differentiation of ADSCs but not to the Hoxa10 promoter (Fig. 8B). These observations are consistent with those from the Gadd45a RNAi experiments because Gadd45a knockdown decreased Dlx5, Runx2, Osterix, and Bglap expression and inhibited the demethylation of these gene promoters but had little effect on Hoxa10 expression. GADD45A binds to Dlx5, Osterix, and Runx2 promoters in undifferentiated ADSCs. However, the binding effect increased more than 10-fold after osteogenic induction. In contrast, GADD45A binds to the Bglap promoter effectively in both the undifferentiated ADSCs and osteogenic differentiated cells. The demethylation process might need the participation of other proteins aside from the involvement of GADD45A because the Bglap promoter was methylated in the undifferentiated ADSCs.
Effect of Gadd45b and Gadd45g on Osteogenic Differentiation-Examination of whether ADSCs express Gadd45b and Gadd45g and whether their expression change during differentiation was conducted to determine whether these two homolog genes of Gadd45a are also required for the osteogenic expression of ADSCs. Both Gadd45b and Gadd45g were expressed in ADSCs, but neither of them was up-regulated after osteogenic induction (supplemental Fig. 1). This result suggests that Gadd45b and Gadd45g might not largely contribute to the regulation of osteogenic differentiation. To confirm this suggestion, the changes in the osteogenic differentiation of the Gadd45b-and Gadd45g-depleted ADSCs were further examined. For this purpose, ADSCs were infected with the Gadd45b and Gadd45g shRNA-constructed lentiviruses. In addition, osteogenic differentiation was determined by the expression of the osteogenic markers (Dlx5, Runx2, Bglap, Osterix, Hoxa10, and Col1). The results show that after infection with the constructed lentiviruses, the expression of Gadd45b and Gadd45g in the ADSCs dramatically decreased compared with that infected with the scrambled shRNA (supplemental Fig. 2A). However, the expressions of Dlx5, Runx2, Bglap, Osterix, Hoxa10, and Col1 were not decreased in the Gadd45b-and Gadd45g-depleted cells (supplemental Fig. 2B). This result suggests that the knockdown of Gadd45b and Gadd45g does not interrupt the osteogenic differentiation of ADSCs. Therefore, Gadd45b and Gadd45g, unlike Gadd45a, are not essential in the osteogenic differentiation of ADSCs.

DISCUSSION
Methylation of genomic DNA has long been viewed as a stable epigenetic mark. However, studies in the past decade have revealed that methylation is not as static as once thought. It can be regulated by passive or active demethylation. The passive process takes place in the absence of methylation of newly synthesized DNA strands by DNMT1 during several replication rounds (26). However, the active process occurs via active removal of the methyl group (7). Methylation plays an important role in maintaining the stem cell properties of self-renewal and multipotency. Analyses of global DNA methylation show that mouse ESCs genomes are less methylated than those of differentiated somatic cells (43). Gene markers of pluripotency, such as Oct4 and Nanog, are unmethylated in ESCs but methylated in somatic cells. As ESCs differentiate, however, Oct4 and Nanog become methylated and silenced (44 -46). In addition, repression of DNA methylation by DNMT1 knock-out causes ESCs to lose pluripotency and can no longer differentiate (47). Unlike DNA methylation, DNA demethylation occurs rarely during stem cell differentiation (24,48). It was reported that several hundred promoters, including pluripotency and germline-specific genes, became methylated when ESCs differentiated to neuronal progenitors. Only 22 genes, which were all brain-specific, became demethylated, and few DNA methylation changes occurred as neuronal progenitors differentiate into pyramidal glutamatergic neurons (24). In that report (24), the majority of differentiation-coupled methylation events occurred in the multipotent progenitor state, with few changes occurring during subsequent terminal differentiation. These results suggest that active DNA demethylation mainly occurs in the differentiation of progenitor cells from ESCs but rarely during the terminal differentiation of adult stem cells. In this study, however, we investigated whether active DNA demethylation occurs in the terminal differentiation of adult stem cells by using the ADSC-derived osteogenic specification model. The ADSCs were cultured to over 90% confluence before osteogenic induction to ensure that contact inhibition would prevent DNA replication. Under these conditions, DNA demethylation represents an active rather than a passive (replication-dependent) process. From six up-regulated bone-specific genes (Dlx5, Runx2, Osterix, Hoxa10, Bglap, and Col1), we found five genes (Dlx5, Hoxa10, Runx2, Bglap, and Osterix) in undifferentiated ADSCs that were highly methylated, and four (Dlx5, Runx2, Bglap, and Osterix) that underwent demethylation during differentiation accompanied by an increase in gene expression. Thus, only specific osteogenic genes appear to be regulated by methylation/demethylation during terminal differentiation. These methylation-modified osteogenic genes may be essential for maintaining the pluripotency of ADSCs. Active DNA demethylation, however, may play a crucial role in terminal osteogenic differentiation from ADSCs. The precise mechanisms underlying these gene-specific DNA methylation or demethylation events remain to be elucidated.
An accumulating body of work has demonstrated that downregulation of gene expression by DNA methylation depends on CpG density (23). Methylation occurs either at CpG islands with high CpG density or at upstream CpG islands in regions called CpG "island shores" with medium CpG density (41,42). In most cases, the methylation patterns of CpG in medium density regions show a negative correlation with gene expression (42). For example, promoters with medium CpG density genes, such as Oct4, Nanog, Tcl1, Tdgf1, Gdf3, and Dppa4, the six best characterized pluripotency-maintaining genes, became methylated and silenced after ESCs differentiate to neuronal progenitors (24). In our present osteogenic differentiation model, we also found that there was a close relation between demethylation of most osteogenic genes and terminal differentiation, and the medium density areas rather than the high density islands of these osteogenic gene promoters were preferentially regulated by DNA methylation/demethylation. Furthermore, demethylation in osteogenic differentiation occurs mainly at some specific CpGs, suggesting site-specific demethylation mechanisms might exist in this process. Further investigations are needed to reveal the exact role of these site-specific modifications and the relationship between site-specific demethylation and other epigenetic modifications. Besides, further clarification on the remaining questions, such as why Hoxa10 remains methylated despite increasing expression and how different forms of regulation interact to determine the final level of gene expression, is required for a more in-depth understanding of epigenetic regulation in adult stem cell specification.
Active DNA demethylation might be a complex reaction involving many enzymes and factors; however, details of this process are still poorly understood (26,49). The role of GADD45A in this process also remains controversial. For example, several recent studies reported that GADD45A triggers genome-wide active DNA demethylation in Xenopus oocytes and zebrafish models (15,16). However, Gadd45a knock-out did not alter demethylation in the mouse genome (33). Here, we provide evidence that GADD45A plays a crucial role in active DNA demethylation during osteogenic differentiation of mouse ADSCs. Knockdown of Gadd45a by shRNA dramatically suppressed the expression of Dlx5, Runx2, Bglap, and Osterix genes. The promoter of each gene was methylated in undifferentiated stem cells. By bisulfite sequencing analysis, we demonstrated that knockdown of Gadd45a resulted in significant inhibition of DNA demethylation of Dlx5, Runx2, Bglap, and Osterix promoters under differentiating culture conditions. This concomitant suppression of Dlx5, Runx2, Bglap, and Osterix promoter demethylation and gene expression resulting from Gadd45a knockdown was accompanied by the inhibition of osteogenic differentiation, indicating that demethylation-mediated up-regulation of these genes is a prerequisite for terminal differentiation and that GADD45A activity is essential for DNA demethylation during osteogenic differentiation. In animals, four possible mechanisms for active DNA demethylation have been proposed (Fig. 9) (26). They include nucleotide excision repair (NER) (12,50), deamination of 5-methylcytosine (5meC) to T followed by base excision repair (BER) of the T-G mismatch (15), oxidative demethylation (51), and radical S-adenosylmethionine (AdoMet)-based demethylation (52). Here, we propose that demethylation of osteogenic gene promoters might be through NER because several previous experiments excluded the others. For example, DNA demethylation can be achieved by deamination of 5meC to produce T, followed by BER to replace the mismatched T with unmethylated C. The deamination of 5meC to T is catalyzed by activation-induced deaminase (AID) (15). Therefore, AID is a key factor for DNA demethylation in this mechanism (15,53). However, AID was found to be highly expressed in embryonic cells but expressed at low levels in most differentiated cells except B lymphocytes (54 -57). In our study, we found no detectable AID expression in either ADSCs or differentiated osteogenic cells (data not shown), suggesting that that AID is expressed mainly by embryonic cells, and functions exclusively in embryonic genome demethylation (53,58). In oxidative demethylation, the ten-eleven translocation (TET) proteins were recently found to play a crucial role in DNA demethylation by hydrolyzing 5meC to 5-hydroxymethylcytosine (5hmC) (51,59). However, the TET-produced 5hmC is abundant in neurons and ESCs, but it is not found in other cell types and tissues, including bone, indicating that TET does not participate in osteogenic differentiation (59,60). In radical AdoMetbased demethylation, the mammalian elongator complex pro-tein 3 (ELP3) is important for active DNA demethylation of the zygotic paternal genome rather than gene-specific demethylation (52). Because gene-specific demethylation occurs in our osteogenic differentiation model, it seems reasonable to suggest radical AdoMet-based demethylation was not involved. More recently, it was reported that GADD45A can be recruited to the rDNA promoter by TAF12 and then trigger demethylation of promoter DNA by recruiting the NER complex, including XPA, XPG, and XPF. The hypomethylated state of the rDNA promoter allows for the binding of RNA polymerase I and the transcription of rRNA (12). This result suggests that GADD45A triggered DNA demethylation through an NER-dependent mechanism and points to this mechanism as the most probable for demethylation of osteogenic gene promoters. Demethylation might enhance the binding of RNA polymerase II and the transcription of mRNA. Thus, GADD45A may play a role in the regulation of both rRNA and mRNA transcription by changing the methylation patterns of different types (pol I and II) of promoters.
ADSCs possess multilineage differentiation potential and thus provide a powerful system for studying commitment and differentiation toward various lineages. They may prove useful for cell therapy (61) and to identify molecular pathways involved in stem cell self-renewal, plasticity, and differentiation (62). We established this model for the study of DNA demethylation events during osteogenic differentiation. Our data show that gene-specific active demethylation occurs during osteogenic differentiation of ADSCs and that GADD45A contributes to this modification. In summary, active DNA demethylation FIGURE 9. Proposed models for active DNA demethylation. A, GADD45A-mediated NER. GADD45A interacts with the NER complex, including XPG, XPA, and XPF, to excise methylated nucleotides and replace them with normal nucleotides by repair synthesis. B, deamination and BER. Deamination of 5meC to T can be catalyzed by AID, and the T-G mismatch can be repaired by methyl-CpG-binding domain protein 4 (MBD4) or T DNA glycosylase (TDG), which allows BER to regenerate an unmethylated cytosine. C, in oxidative demethylation, TET proteins catalyze the conversion of 5meC to 5hmC, which may be further processed by several mechanisms as follows. 1) NER or BER, 5hmC may be recognized by NER or BER and then replaced with cytosine by repair synthesis. 2) Deamination and BER, 5hmC may undergo deamination to produce 5hmU, which is repaired by BER. 3) 5hmC may directly release formaldehyde and be converted to cytosine by DNMT. 4) 5hmC may be processed by consecutive hydroxylation reactions followed by a decarboxylation reaction that ultimately generates C. Alternatively, 5hmC itself may be a functional modification. D, radical AdoMet-based demethylation, elongator complex protein 3 (ELP3) contains an Fe-S radical AdoMet domain that uses AdoMet to generate a 5meC radical and then forms 5hmC in water; finally, formaldehyde is released under nucleophilic attack and results in production of unmethylated cytosine. occurs in terminal differentiation of stem cells to promote lineage-specific gene expression, and GADD45A plays an essential role in this demethylation process.