Mesenchymal Stem Cell Laboratory, School of Medical Sciences, Faculty of Health Sciences, South Australia, Australia
Centre for Stem Cell Research, Robinson Institute, University of Adelaide, South Australia, Australia
Correspondence: Stan Gronthos, BSc, MSc, PhD, Medical School South Level 4, School of Medical Sciences, Faculty of Health Sciences, University of Adelaide, Adelaide 5005, South Australia, Australia. Telephone: 61-8-82223460; Fax: 61–882223139; e-mail: firstname.lastname@example.org
The methyltransferase, Enhancer of Zeste homology 2 (EZH2), trimethylates histone 3 lysine 27 (H3K27me3) on chromatin and this repressive mark is removed by lysine demethylase 6A (KDM6A). Loss of these epigenetic modifiers results in developmental defects. We demonstrate that Ezh2 and Kdm6a transcript levels change during differentiation of multipotential human bone marrow-derived mesenchymal stem cells (MSC). Enforced expression of Ezh2 in MSC promoted adipogenic in vitro and inhibited osteogenic differentiation potential in vitro and in vivo, whereas Kdm6a inhibited adipogenesis in vitro and promoted osteogenic differentiation in vitro and in vivo. Inhibition of EZH2 activity and knockdown of Ezh2 gene expression in human MSC resulted in decreased adipogenesis and increased osteogenesis. Conversely, knockdown of Kdm6a gene expression in MSC leads to increased adipogenesis and decreased osteogenesis. Both Ezh2 and Kdm6a were shown to affect expression of master regulatory genes involved in adipogenesis and osteogenesis and H3K27me3 on the promoters of master regulatory genes. These findings demonstrate an important epigenetic switch centered on H3K27me3 which dictates MSC lineage determination. Stem Cells2014;32:802–815
Multipotential bone marrow mesenchymal stem/stromal cells (MSC) are a heterogeneous population of stem cells and lineage committed progenitors, where a minor proportion of ex vivo expanded plastic adherent colony-forming unit-fibroblastic cells have the potential to differentiate into multiple mesodermal lineages, providing evidence of a stromal hierarchy of cellular differentiation [1-5].
Skeletal and adipose development during embryogenesis and postnatal homeostasis are dependent on key transcription factors runt-related transcription factor 2 (RUNX2) and peroxisome proliferator-activated receptor-γ (PPARγ2), respectively [6-8]. However, the differentiation of MSC is dictated by a complex network of signaling components of the bone morphogenetic protein (BMP)/TGFβ and Wnt pathways, which can act in a reciprocal manner to induce osteogenesis, while suppressing adipogenesis, and vice versa, depending on the presence and action of transcription factors such as MSX2, Twist-1, Dermo-1, TAZ, and C/EBP-β and δ, which are known to regulate Runx2 and PPARγ2 expression or activity [9-16].
It is now thought that stem cell populations undergo dynamic reprogramming of gene expression profiles during lineage commitment and maturation. This programming is partly mediated by Polycomb group (PcG) protein complexes through epigenetic modifications of chromatin . PcG proteins were initially identified in Drosophila melanogaster with mutations causing developmental defects due to misregulation of key transcription factors, such as the homeobox protein leading to typical homeotic phenotypes . PcG protein complexes act as transcriptional repressors manipulating chromatin and influencing the cells identity during development. Polycomb repressive complex 1 (PRC1) consisting of more than 10 subunits including Bmi-1, CBX7, and Ring1–2 interacts with chromatin remodelers, influencing a tightly repressive heterochromatin state . PRC2 contains three subunits, Enhancer of Zeste Homology 2 (EZH2), Suppressor of Zeste 12, and embryonic ectoderm development . The histone methyltransferase EZH2 trimethylates the histone 3 lysine 27 (H3K27me3) and this modification leads to the recruitment of PRC1 via the chromo domain protein CBX7 and other factors involved in condensing chromatin and repressing genes. It is thought that PRC1 blocks the transcriptional elongation of polymerase II (Pol II) as it is phosphorylated on its C terminal on serine 5. This results in RNA polymerase II being paused within the gene body, poised for activation upon differentiation stimuli .
Interrogation of microarray data from a recent study from our laboratory revealed an upregulation of the epigenetic modifying protein EZH2, in multipotential, long-lived MSC clones compared to uni-potential, short-lived MSC clones , suggesting a possible role for EZH2 in either the differentiation or life-span of human MSC. Previous studies have implicated EZH2 in maintaining the stem cell phenotype of hematopoietic stem cells, epidermal skin cells, myoblasts, preadipocytes, embryonic stem cells, and multipotent neural progenitor cells of the cerebral cortex in vitro [23-30]. Recently, an elegant study identified EZH2 role in repressing Wnt genes allowing the differentiation of mouse peripheral preadipocytes . With EZH2 involvement in regulating pathways that dictate peripheral preadipocytes differentiation, it is plausible that EZH2 may act as part of an epigenetic control switch to regulate cell fate determination of MSC.
Genome-wide chromatin immunoprecipitation (ChIP) sequencing in embryonic stem cells  and human embryonic lung fibroblasts  has shown that EZH2 occupies promoters of various genes associated with development, lifespan, and differentiation. It is now thought that the process of H3K27me3 removal may play a pivotal role in lineage determination of many cells types. The discovery of the histone H3K27me3 demethylases, lysine demethylase 6A (KDM6A) (UTX), and 6B (JMJD3) has increased our understanding of how methylation dynamics are regulated. Initial studies found that KDM6A was localized at the transcriptional start sites of Hox genes, where knockdown of KDM6A in zebrafish resulted in misregulation of Hox genes and a posterior developmental defect . Functional studies have shown that KDM6A homozygous mutations in mouse cause severe midgestational defects, developmental delay, neural tube closure, yolk sack, and heart defects . In addition, KDM6A-deficient embryonic stem cells display a failure of the cardiac differentiation program . KDM6A has also been found in association with the mixed lineage leukemia histone methyltransferase and trithorax group proteins involved in gene activation . Drosophila dKDM6A mutants have similar phenotypes to trithorax mutants and show that loss of dKDM6A results in increased proliferation due to activation of the Notch pathway and inactivation of the retinoblastoma pathway . To date, one study has examined the genome-wide occupancy of KDM6A in fibroblasts, which reported that the most represented genes based on Gene Ontology analysis was a network of 49 genes centered around the retinoblastoma pathway . Apart from skeletal muscle differentiation, very little is known about the role of KDM6A in adult stem cell differentiation. However, another H3K27me3 demethylase, KDM6B (JMJD3), was recently shown to work with H3K9 demethylase KDM4B in regulating osteogenic differentiation of human bone marrow cells (hBMCs) . This study reported that KDM6B and KDM4B may have a role in regulating the BMP pathway during osteogenesis; however, KDM6B was not recruited to RUNX2, a major driver of osteogenic differentiation, suggesting that KDM6B regulates RUNX2 by an indirect mechanism or another demethylase might be involved. Up to now the direct involvement of EZH2 and its relationship with its counterpart KDM6A have not been examined in human MSC cell fate determination.
In this study, we assessed a potential epigenetic switch involving histone methylation and demethylation by EZH2 and KDM6A in human MSC differentiation using over-expression, siRNA knockdown, and enzymatic inhibition. We also examined the ability of EZH2 and KDM6A to regulate promoter methylation and expression of mesodermal tissue master genes involved in lineage commitment and maturation.
Isolation of MSCs
Bone marrow aspirates were isolated from the posterior iliac crest of healthy human adult donors (17–35 years of age), with informed consent (IMVS/SA Pathology normal bone-marrow donor program RAH#940911a). STRO-1+ selected MSCs were cultured in Alpha Modification of Eagle's (αMEM) (Sigma Aldrich Inc., Sydney, NSW, Australia, http://www.sigmaaldrich.com/australia.html) supplemented with 10% fetal calf serum (FCS; SAFC Biosciences, Sydney, NSW, Australia, http://www.safcglobal.com/safc-global/en-us/home.html), 2 mM l-Glutamine, 1 mM sodium pyruvate (Sigma Aldrich Inc.), 10 mM HEPES buffer (Sigma Aldrich Inc.), and 50 U/ml penicillin-streptomycin (Sigma Aldrich Inc.) as previously described . Equal numbers of male and female donors were used where possible.
In Vitro Differentiation Assay
Culture expanded human MSCs were cultured in either normal growth conditions or osteogenic media (αMEM supplemented with 5% FCS, 2 mM l-Glutamine (Sigma Aldrich Inc.), 1 mM sodium pyruvate, 10 mM HEPES buffer, 50 U/ml penicillin-streptomycin, 0.1 µM dexamathasone, and 2.6 mM KH2PO4; Asia Pacific Speciality Chemicals Limited, Seven Hills, NSW, Australia, http://www.apschem.com.au/) or adipogenic media (αMEM supplemented with 10% FCS, 2 mM l-Glutamine, 1 mM sodium pyruvate, 10 mM HEPES buffer, and 50 U/ml penicillin-streptomycin, 0.1 µM Dexamathasone, and 120 mM indomethacin, Sigma Aldrich Inc.) for up to 28 days as previously described . Mineralized bone matrix (mineral) formation was identified with Alizarin red (Sigma Aldrich Inc.) staining of hydroxyapatite mineral deposits . Extracellular calcium was measured in triplicate samples and normalized to DNA content per well as previously described . Lipid formation was assessed by oil red O (MP Biomedicals, Solon, OH, http://www.mpbio.com/) staining as previously described . Quantitation of lipid was performed by oil red O extraction in triplicate wells or by Nile red (Sigma Aldrich Inc.) fluorescence staining (for limited cell numbers) normalized to DAPI (Invitrogen/Life Technologies Australia, Mulgrave, VIC, Australia) stained nuclei per field of view in triplicate wells as previously described [15, 40].
Drug Inhibition of EZH2 by 3-Deazaneplanocin A
3-Deazaneplanocin A (DZNep) was obtained from (Sequoia Research, Pangbourne, U.K., http://www.seqchem.com/) and resuspended in dimethyl sulfoxide (DMSO) to stock concentration of 25 mM (Chem Supply, Adelaide, SA, AUS, http://www.chemsupply.com.au/). DZNep was tested over the concentration range of 0.1–20 µM as previously described .
Western Blot Analysis
Culture expanded MSCs were seeded at 3 × 105 per dishes (55 cm2), and cultured in either regular growth medium, adipogenic, or osteogenic inductive media as described above. Whole cell lysates (50 µg) were separated on SDS gel as previously described [15, 42]. Membranes were probed with anti-mouse (Merk Millipore, Kilsyth, VIC, Australia, http://merck.com.au/en/index.html) EZH2 (Acc; Cell Signaling Tecnology Inc./Genesearch Pty. Ltd., Arundel, QLD, Australia, http://www.cellsignal.com/, 1/1,000 dilution), anti-rabbit KDM6A (Abcam/Sapphire Bioscience Pty. Ltd., Waterloo, NSW, Australia, http://www.abcam.com/, ab36938, 1/1,000 dilution), anti-H3K27me3 rabbit IgG (Merk Millipore, Kilsyth, VIC, Australia, 1/1,000 dilution), anti-β-actin, and anti-mouse IgG (Sigma Aldrich Inc., 1/2,500) antibodies. Secondary detection was performed using anti-Rabbit-Alk Phos (Millipore, 1/10,000) and anti-Mouse-Alk-phos (Millipore, 1/10,000) antibodies.
Measurement of Apoptosis
MSCs were detached by trypsin/EDTA (Sigma Aldrich Inc.) treatment, washed, and then assessed for Annexin V (Invitrogen/Life Technologies Australia, Mulgrave, VIC, Australia, http://www.lifetechnologies.com/) staining as previously described . Flow cytometric analysis was used to detect early apoptotic Annexin-positive/7AAD-negative cells and late apoptopic Annexin positive/7AAD-positive cells (Ambion/Life Technologies, Mulgrave, VIC, Australia).
Real-Time Polymerase Chain Reaction Analysis
Culture expanded MSCs were seeded at 5 × 104 per well (9.4 cm2) in the presence of either noninductive growth media, osteogenic, or adipogenic inductive. Total RNA was extracted using TRIzol reagent (Sigma Aldrich Inc.) and converted to cDNA by reverse transcription as previously described . Transcription was assessed by real-time polymerase chain reaction (RT-PCR) amplification using RT2 Real-Time SYBR Green/Rox PCR master mix (Qiagen, Doncaster, Chadstone Centre, VIC, Australia, http://www.qiagen.com/), using the Rotor-Gene 6000 Real-Time Thermal Cycler (Corbett Research, Mortlake, NSW, Australia, http://www.corbettlifescience.com/). Primer sets used in this study: β-actin: Fwd 5′gatcattgctcctcctgagc3′; Rev 5′gtcatagtccgcctagaagcat3′ 157bp, Runx2: Fwd 5′-gtggacgaggcaagagtttca-3′; Rev 5′catcaagcttctgtctgtgcc3′, Osteopontin (OPN): Fwd 5′ acatccagtaccctgatgctacag 3′, Rev 5′ gtgggtttcagcactctggt 3′, Osteocalcin: Fwd 5′ atgagagccctcacactcctcg 3′, Rev 5′ gtcagccaactcgtcacagtcc 3′, PPARγ2: Fwd 5′ ctcctattgacccagaaagc 3′, Rev 5′tcaaaggagtgggagtggtc 3′, Adipsin: Fwd 5′ gacaccatcgaccacgac-3′, Rev 5′ ccacgtcgcagagagttc 3′, C/EBP-α: Fwd5′ gggcaaggccaagaagtc 3′, Rev5′ ttgtcactggtcagctccag 3′, Ezh2, Fwd 5′ ccggagacctagatgtcattg 3′, Rev 5′ ggcctgtcttctcgctttctctt 3′, Kdm6a, Fwd 5′ gagggaagctctcattgctg 3′, Rev 5′ agatgaggcggatggtaatg 3′. ChIP primers; Runx2, Fwd 5′ aggccttaccacaagccttt 3′, Rev 5′ agaaagtttgcaccgcactt 3′, Osteocalcin, Fwd 5′ caaatagccctggcagattc 3′, Rev 5′ gagggctctcatggtgtctc 3′.
Retroviral Transduction Over-Expression Studies
The human Ezh2 and Kdm6a genes were ligated into pRUF-IRES-GFP using PCR primers to amplify the coding region. The pRUF-IRES-GFP, pRUF-IRES-GFP-Ezh2, and/or pRUF-IRES-GFP-Kdm6a constructs were transfected into the HEK293T viral packaging cell line together with Pol and GAG protein (PGP) and vesicular stomatitis virus G-protein (VSVG) (viral envelope proteins, SBI System Biosciences, Mountain View, CA, http://www.systembio.com/). Viral supernatant was used for infection of MSC as previously described . Stable lines were generated by sorting for GFP-positive cells using fluorescence-activated cell sorting.
MSCs were seeded at 3 × 103 cells per well (24-well plate) the day before siRNA knockdown. siRNA (12 pmol; Ambion/Life Technologies, Mulgrave, VIC, Australia, http://www.lifetechnologies.com/) was used per siRNA to achieve around 80% knockdown of Ezh2 and Kdm6a. siRNA used in this study: Ezh2 s4916 and s4918, Kdm6a s14737 and s14735, Negative siRNA #1, and positive siRNA GAPDH (Ambion Life Technologies). Briefly 12 pmol of siRNA was added with 2 µl of RNAi MAX lipofectamine (Ambion/Life Technologies, Mulgrave, VIC, Australia) per well and incubated with the cells for 72 hours then removed before adding the differentiation media.
MSCs were seeded at 8 × 103 cells per flask (75 cm2) and induced under normal growth medium, adipogenic, and osteogenic differentiation media for 14 days. ChIP analysis was performed using the Magna ChIP kit (Millipore Corporation, Billerica, MA, http://www.millipore.com/index.do) and carried out according to manufacturer's instructions. Anti-mouse H3K27me3 and anti-rabbit IgG control antibodies (5 µg) were used for the immunoprecipitation. Immunoselected genomic DNA was used for RT-PCR using primers targeting the promoters of genes of interest as previously described .
Ectopic Bone Implants
Approximately 5 × 106 ex vivo expanded MSCs were mixed with 40 mg of hydroxyapatite/tricalcium phosphate (HA/TCP) ceramic powder (Zimmer, Inc., Warsaw, IN, http://www.zimmer.com/) and then transplanted subcutaneously into the dorsal surface of 8-week-old immunocompromised NOD/SCID mice for 8 weeks as previously described . Harvested implants were fixed with 4% paraformaldehyde (Sigma Aldrich Inc.) and then processed for paraffin embedding. Representative 5 µm sections were stained with hematoxylin and eosin (Sigma Aldrich Inc.). These procedures were performed in accordance to specifications of an approved animal protocol (University of Adelaide Animal Ethics Committee Number M-2012-207/141c.12). New bone formation area was calculated using OsteoMeasurexp V3.3.02 (Osteometric, Decatur, GA, http://www.osteometrics.com/software_update.htm) software on an Olympus BX53 Microscope (Olympus, Melbourne, VIC, Australia, http://www.olympusaustralia.com.au/).
Data analysis was carried out using Microsoft GraphPad Prism 5 (GraphPad Software, LA Jolla, CA, http://www.graphpad.com/), which was used for the generation of graphs and statistical analysis. One-way ANOVA with a Dunnett's post-test and statistical significance at p value <.05 was used to analyze the apoptosis assays. For differentiation and RT-PCR studies, a paired Student's t test was used to compare MSC treated with control media and differentiation inductive media. Unpaired Student's t test was used to compare over-expressing, siRNA knockdown MSC lines with corresponding controls for RT-PCR, differentiation, and ChIP studies.
Ezh2 and Kdm6a Are Transcriptionally Regulated During MSC Differentiation into Adipocytes and Osteoblasts
The potential of MSC to differentiate into functional adipocytes in vitro was evident by oil red O staining of lipid droplets (Fig. 1A). Quantitation of oil red O stained lipid confirmed the steady increase in lipid formation between 14 and 28 days of adipogenic differentiation compared to control media treated cells (Fig. 1A). In parallel experiments, the potential of MSC to differentiate into functional osteoblasts in vitro was assessed by Alizarin red staining of mineralized deposits in the extracellular matrix (Fig. 1B). Quantitation of extracellular calcium levels confirmed the formation of mineral deposits at 28 days induction compared to control media for different MSC lines (Fig. 1B).
We next analyzed expression of master regulators of MSC differentiation. The adipogenic transcription factor, PPARγ2, was upregulated under adipogenic conditions from 14 to 28 days compared with nondifferentiation control media (Fig. 1C). Consistent with adipogenic differentiation, the expression of the late adipogenic marker, Adipsin, was upregulated from 14 to 28 days under adipogenic inductive conditions (Fig. 1D). Expression of the histone H3 lysine 27 methyltransferase, Ezh2, was upregulated early within 14 days following adipogenic induction compared to control culture conditions; however, by 28 days, the levels had rapidly decreased (Fig. 1E). The early upregulation of Ezh2 transcript coincided with the initial upregulation of PPARγ2 and the onset of lipid formation, suggesting Ezh2 might be a positive regulator of adipogenesis. Conversely, transcript levels of the histone H3 lysine 27 demethylase, Kdm6a, decrease dramatically at both 14 and 28 days under adipogenic conditions (Fig. 1F).
The early osteogenic transcription factor, Runx2, was upregulated from 14-day until 28 days under osteogenic differentiation conditions compared with control media (Fig. 1G). Expression of the late osteoblastic marker, OPN, was slightly elevated by 14 days and more evident by 28 days (Fig. 1H). Under osteogenic inductive conditions, Ezh2 transcript levels decreased dramatically at 14 days and more so by 28 days in differentiation conditions relative to control conditions (Fig. 1I). In contrast, the transcript levels of Kdm6a increased steadily relative to control conditions between 14 and 28 days (Fig. 1J) in osteogenic differentiation conditions suggesting that Kdm6a might be a promoter of osteogenesis. Overall, these results demonstrate that Ezh2 and Kdm6a gene expression levels are inversely regulated during MSC differentiation and the trend suggested that Ezh2 is an early promoter of adipogenesis and a repressor of osteogenesis whereas the converse applies for Kdm6a.
Ezh2 Promotes Adipogenesis and Inhibits Osteogenesis
Retroviral-mediated enforced expression of Ezh2 in MSC demonstrated increased levels of both Ezh2 transcripts and EZH2 protein levels, which was associated with increased H3K27me3 compared to vector controls (Fig. 2A). When cultured under adipogenic inductive conditions, Ezh2 over-expressing cells exhibited a significantly higher potential to differentiate into adipocytes and form lipid compared to vector control cells (Fig. 2B). Over-expression of Ezh2 resulted in a significant decreased potential to form an Alizarin red positive mineral matrix, quantified by measuring extracellular calcium levels, compared with the corresponding vector control MSC (Fig. 2C).
During adipogenic differentiation, the transcript levels of the two major regulators of adipogenesis, PPARγ2 and C/EBPα, were significantly higher in Ezh2 over-expressing cells compared to empty vector infected cells (Fig. 2D, 2E). In agreement with Ezh2 promoting adipogenesis, transcript levels of the adipocyte marker, Adipsin, were also significantly elevated when Ezh2 was over-expressed compared to empty vector control MSC under adipogenic differentiation conditions (Fig. 2F).
Over-expression of Ezh2 significantly reduced Runx2 transcription and the late osteoblast markers, OPN and Osteocalcin, under osteogenic differentiation conditions compared to empty vector infected cells (Fig. 2G-2I). To verify our findings based on Ezh2 over-expression, we used the methyltransferase inhibitor DZNep to reduce the activity of EZH2. The titration range of 0.3125–20 µM was used to determine the optimal concentration in which EZH2 activity is reduced in MSC based on previous studies .
Initial studies showed that DZNep did not induce significant levels of apoptosis in MSC over a wide concentration range, based on dual Annexin V and 7AAD staining by flow cytometric analysis (Fig. 3A). Importantly, Western blot analysis using MSC lysates treated with 20 µM DZNep showed a dramatic reduction in H3K27 trimethylation compared to the DMSO vehicle control (Fig. 3B). Functional studies showed that DZNep treatment reduced the numbers of oil red O-positive cells under adipogenic differentiation conditions compared to the vehicle control (Fig. 3C). Quantitative analyses demonstrated a significant decrease in oil red O-positive lipid formation when MSCs were treated with 20 µM DZNep (Fig. 3C). In parallel studies, the effect of DZNep-mediated inhibition of Ezh2 on osteoblast differentiation was also examined. Alizarin Red staining of hydroxyapatite deposits showed a higher increase in extracellular mineral deposit formation in the presence of DZNep compared with DMSO vehicle control under osteogenic induction conditions (Fig. 3D). Quantitative analyses measuring extracellular calcium levels indicated a significant increase in mineral deposits when MSCs were treated with 20 µM DZNep (Fig. 3D).
To further confirm that EZH2 is a positive regulator of adipogenesis and a negative regulator of osteogenesis, we performed Ezh2 knockdown studies using siRNA (Fig. 4A). Knockdown of Ezh2 using two different siRNA oligonucleotides resulted in a significant reduction in oil red O-positive lipid producing cells compared to the scrambled control siRNA-treated cells under adipogenic inductive conditions. This was verified by quantifying the number of lipid-positive cells in a total population of cells verifying that knockdown of Ezh2 results in a decrease in lipid positive cells (Fig. 4B). Conversely, knockdown of Ezh2 using two siRNA oligonucleotides showed a significant increase in Alizarin Red mineral-positive cells compared to the scrambled negative control under osteogenic inductive conditions (Fig. 4C). In order to determine the mechanism of EZH2 regulation of MSC differentiation, we examined expression of master regulatory genes. Under adipogenic differentiation conditions, PPARγ2 expression was greatly reduced when Ezh2 was knocked down (Fig. 4D). The same was evident for C/EBPα (Fig. 4E). In agreement with effects on adipogenic differentiation, expression levels of adipogenic marker, Adipsin, were greatly reduced when Ezh2 was knocked down under adipogenic differentiation conditions (Fig. 4F). Under osteogenic differentiation conditions, Runx2 expression levels were increased when Ezh2 was knocked down (Fig. 4G) as was the mature osteoblast markers, OPN and Osteocalcin (Fig. 4H, 4I). Taken together, Ezh2 over-expression, enzymatic inhibition, and siRNA knockdown studies demonstrated that Ezh2 is a positive regulator of adipogenesis and negative regulator of osteogenesis.
Kdm6a Promotes Osteogenesis and Inhibits Adipogenesis
Retroviral-mediated enforced expression of Kdm6a in MSC demonstrated increased levels of both kdm6a transcripts and KDM6A protein levels, which was associated with suppression of H3K27me3 compared to vector controls (Fig. 5A). Functional studies showed that Kdm6a over-expressing in MSC exhibited a decreased potential to form lipid compared with the corresponding empty vector control MSC cells (Fig. 5B). Quantitation studies revealed significantly lower levels of lipid formation in Kdm6a over-expressing MSC compared with vector control MSC under adipogenic conditions (Fig. 5B). Kdm6a over-expressing MSC demonstrated a significant increased potential to differentiate into mineral forming osteoblasts compared to empty vector infected cells (Fig. 5C).
Under adipogenic differentiation conditions, the levels of PPARγ2 and C/EBPα were found to be significantly decreased in Kdm6a over-expressing MSC compared to empty vector infected cells (Fig. 5D, 5E). Similarly, transcript levels of Adipsin were significantly decreased under adipogenic inductive conditions when Kdm6a was over-expressed (Fig. 5F). Upon osteogenic differentiation, the levels of Runx2 were dramatically higher in Kdm6a over-expressing MSC compared to empty vector infected cells (Fig. 5G). In agreement with Kdm6a promoting osteogenesis, transcript levels of the osteoblast markers, OPN and Osteocalcin, were also significantly elevated when Kdm6a was over-expressed compared to empty vector infected cells under osteogenic inductive conditions (Fig. 5H, 5I). To verify the role of KDM6A in adipogenesis, Kdm6a was knocked down in MSC using siRNA (Fig. 6A). Compared to scrambled negative control siRNA, two different siRNA oligonucleotides targeting Kdm6a showed a significant increase in lipid formation when MSCs were cultured under adipogenic differentiation conditions (Fig. 6B).
The function of KDM6A in osteogenesis was also confirmed by knocking down Kdm6a in MSC using siRNA. Compared to scrambled negative control siRNA, two different siRNA oligonucleotides targeting Kdm6a showed a significant decrease in mineral formation compared to the scrambled negative control under osteogenic differentiation conditions (Fig. 6C). To further understand how KDM6A is regulating MSC differentiation, we examined expression of key master regulatory genes. Under adipogenic differentiation conditions, PPARγ2 transcript levels were increased when Kdm6a was knocked down (Fig. 6D). The same was evident for master regulator C/EBPα (Fig. 6E). In agreement with its effect on differentiation, transcript levels of Adipsin were also increased when KDM6A was knocked down (Fig. 6F). Under osteogenic differentiation conditions, transcript levels of Runx2 were greatly decreased when Kdm6a was knocked down (Fig. 6G). The same was evident for OPN and Osteocalcin (Fig. 6H, 6I). Overall, Kdm6a enforced expression or knockdown in human MSC demonstrated that Kdm6a is a negative regulator of adipogenesis via the suppression of adipogenic genes, whereas it acts as a positive regulator of osteogenesis by activating expression of osteogenic genes.
Differential Capacity of Ezh2 and Kdm6a to Regulate Bone Formation In Vivo
We investigated the potential of over-expressing Ezh2 or Kdm6a MSC or vector controls to form ectopic bone when seeded onto osteoconductive HA/TCP carrier particles and then implanted subcutaneously into NOD/SCID mice. Histological assessment at 8 weeks post-transplantation demonstrated the development of new bone in transplants containing Kdm6a over-expressing MSC and vector control MSC only (Fig. 7A). Quantitative analyses found that Kdm6a over-expressing MSC produced significantly higher levels of bone formation compared to vector control MSC, in contrast to transplants containing Ezh2 over-expressing MSC which were devoid of any new bone formation (Fig. 7D).
Histone H3K27 Trimethylation on Adipogenic and Osteogenic Gene Promoters Decreases Upon Differentiation and Levels Are Influenced by Ezh2 and Kdm6a
Studies were performed to determine whether enforced expression of Ezh2 or Kdm6a influences H3K27 methylation along the master regulator genes that drive MSC lineage determination. H3K27me3 marks present on the Runx2 and Osteocalcin transcription start sites (TSS) were found to decrease dramatically in control MSC under mineral inductive conditions, correlating with increased transcription normally seen during osteogenesis (Fig. 7C, 7D). However, enforced expression of Ezh2 in MSC resulted in increased H3K27me3 on the Runx2 and Osteocalcin TSS under osteogenic inductive conditions compared to empty vector infected MSC (Fig. 7C, 7D). While, H3K27me3 marks present on the PPARγ2 or C/EBPα TSS were found to decrease during adipogenic differentiation, retroviral-mediated over-expression of Ezh2 in MSC failed to significantly alter the levels of H3K27me3 along the adipogenic TSS under control or adipogenic conditions (Fig. 7E, 7F).
We then examined H3K27me3 in response to enforced expression of Kdm6a in MSC. H3K27me3 marks present on the Runx2 and Osteocalcin TSS decreased during osteogenesis in vector control MSC (Fig. 7G, 7H). However, enforced expression of Kdm6a resulted in a decrease in H3K27me3 for both Runx2 and Osteocalcin TSS compared to empty vector controls (Fig. 7G, 7H). While, H3K27me3 marks present on either the PPARγ2 or C/EBPα TSS decreased during adipogenic differentiation (Fig. 7I, 7J), over-expression of Kdm6a in MSC failed to significantly alter the levels of H3K27me3 along the adipogenic TSS under control or adipogenic conditions (Fig. 7I, 7J). Collectively, these studies revealed that the H3K27 methylation levels of osteogenic gene promoters are directly dependent on the expression of Ezh2 and Kdm6a, whereas H3K27me3 methylation patterns for adipogenic gene promoters were largely unaffected by Ezh2 and Kdm6a.
Studies on embryonic stem cell commitment have proposed that the epigenetic state of pluripotent cells influences fate determination, where lineage-specific promoters that are associated with terminal differentiation often contain methylated DNA . This process is thought to impede improper or premature differentiation toward a specific lineage, in order to maintain pluripotency. However, identification of an epigenetic switch that mediates stem cell fate decisions remains to be determined. Previous studies [6, 45] have shown that the differentiation of human MSC is associated with the upregulation of Runx2, Osteocalcin, and OPN during osteoblast differentiation and PPARγ2, C/EBPα, and Adipsin during adipocyte differentiation, respectively  where activation of PPARγ2 is known to inhibit osteogenesis. This study revealed that the histone H3K27 methyltransferase, Ezh2, and the histone H3K27 demethylase, Kdm6a, exhibited an inverse expression pattern during human MSC osteogenic or adipogenic differentiation. Functional over-expression studies determined that EZH2 was a negative regulator of osteogenesis, whereas KDM6A was found to promote the osteogenic commitment of human MSC. This in vitro data were supported by ectopic bone implants which confirmed that Ezh2 over-expressing MSCs were deficient in their ability to from bone, while Kdm6a over-expressing MSC formed greater amounts of bone compared with the vector controls. Conversely, Ezh2 was shown to be a positive regulator of adipogenesis, as previously described for murine peripheral fat preadipocytes , while this study found that KDM6A acted as a negative regulator of adipocyte differentiation by human MSC. These observations suggest that EZH2/KDM6A may function as an epigenetic switch for human MSC fate determination. Confirmatory studies using siRNA knockdown and chemical inhibition of methyltransferase activity demonstrated that EZH2 and KDM6A exhibited opposing functions during human MSC differentiation toward adipocytes or osteoblasts. We found that the mechanism of action is partly due to transcriptional regulation of lineage determination and marker genes. These findings support a recent study that demonstrated that the histone demethylases KDM4B and KDM6B promote osteogenic commitment of MSC by removing H3K9me3 and H3K27me3. KDM4B or KDM6B significantly increased osteogenic differentiation and decreased adipogenic differentiation through the removal of repressive methylation marks on early transcription factor such as HOX genes and DLX, respectively . However, we have found that Kdm6b is expressed at much lower levels than Kdm6a in MSC undergoing differentiation and its expression is not as dynamic when MSC differentiate toward the osteoblast or adipocyte lineage. In our study, we have analyzed expression from day 1 to day 28 following osteogenic and adipogenic differentiation, respectively, since mineralization and lipid production generally occur after 7 days of induction and expression of master regulatory genes occurs from 7 days onward. In contrast, a previous study analyzed expression of all demethylases for a few hours following BMP-induced differentiation . It is therefore possible that KDM6B is playing a role early during differentiation and would be interesting to determine if both KDM6A/B work together to cooperatively remove H3K27me3.
During osteogenic differentiation, we found that the master regulator, Runx2, was repressed in the presence of EZH2 due to a dramatic increase of H3K27me3 on the Runx2 TSS. Similarly, expression of Ezh2 was found to cause repression of transcription and increased H3K27me3 for the osteoblast-associated marker Osteocalcin. Conversely, Runx2 and Osteocalcin transcript levels were upregulated by Kdm6a coinciding with downregulation of H3K27me3, revealing that KDM6A opposes the actions of EZH2 at the molecular and functional levels. Moreover, given the known inverse relationship between adipose and bone formation by MSC, a recent report found that cyclin-dependent kinase 1 negatively regulates Ezh2 in MSC to increase osteogenic differentiation by adding an inhibitory phosphorylation on residue threonine (Thr) 487 onto EZH2, causing the disassociation of the PRC2 and loss of EZH2 methyltransferase activity . However, these findings were largely correlative and failed to demonstrate a direct effect of EZH2 to inhibit osteogenesis using gene knockdown or over-expression studies. Interestingly, EZH2 was found to directly target β-catenin signaling pathway components such as Wnt1, −6, −10a, −2B, −3a, −8a, −2, −11, and −10b in peripheral preadipocytes and MSC [30, 46]. Given that the canonical Wnt pathway is known to upregulate Runx2 expression and activate Runx2  we predict that EZH2 action represses osteogenesis at multiple levels by directly affecting Wnt genes, Runx2, and its downstream targets such as OPN and Osteocalcin. Since Ezh2 is more highly expressed in long-lived, multipotent MSC and is associated with an increased life-span , it is plausible that uncommitted MSC may have a default pathway toward adipogenesis in the absence of appropriate osteogenic inductive stimuli. We further predict that Kdm6a will also act on the Wnt genes, Runx2, and downstream targets, OPN and Osteocalcin, to remove the H3K27 methylation mark and activate transcription. Collectively, these studies together with our findings suggest that EZH2 directly methylates H3K27 on the promoters of Wnt genes, Runx2, and downstream Runx2 target promoters such as OPN and osteocalcin to induce H3K27me3, leading to inhibition of gene expression, whereas KDM6A opposes this action by removing H3K27me3, promoting osteogenic differentiation.
A previous report showed that Ezh2 knockout of primary murine preadipocytes extracted from Ezh2Flox/Flox mice resulted in reduced lipid formation . Another study examining the DNA methylation profile of both adipogenic and nonadipogenic genes in MSC-like cells derived from human adipose tissue found that the promoters for four adipogenic genes (PPARγ2, leptin, fatty acid-binding protein 4, lipoprotein lipase) were hypomethylated in freshly harvested human MSC . This study also found that promoter regions for housekeeping genes such as GAPDH were hypomethylated, whereas nonadipogenic lineage-specification gene promoters were hypermethylated. Furthermore, in vitro analyses have correlated the demethylation of various adipogenic promoters, including that of PPARγ2, with adipogenic differentiation in murine cell lines . Collectively, these findings suggest that the commitment of MSC to the adipogenic lineage may be reflected by a particular epigenetic signature in which adipogenic gene promoters are hypomethylated while nonadipogenic promoters are methylated. This study showed that over-expression of Ezh2 resulted in an increase in transcription of the master adipogenic regulators PPARγ2, C/EBPα, and the adipocyte-associated marker Adipsin, where over-expression of Kdm6a resulted in a decrease in transcription of these adipogenic factors. The lack of significant change in H3K27me3 levels on the adipogenic promoters we tested when Ezh2 or Kdm6a is over-expressed suggests that EZH2 and KDM6A directly affect other genes such as the Wnt pathway which then leads to changes in other histone modifications on adipogenic promoters to alter gene expression. This however needs to be assessed in more detail including other histone modifications that occur on adipogenic promoters in response to over-expression of Ezh2 and Kdm6a to gain a better understanding.
Overall, this study has identified, a potential epigenetic switch involving histone H3K27 methylation and demethylation orchestrated by EZH2 and KDM6A which determines the differentiation of MSC into adipocytes and osteoblasts. We demonstrated that EZH2 and KDM6A regulated osteogenic differentiation by directly methylating or demethylating the promoter regions osteogenic genes, however the effects of EZH2 and KDM6A were found to be indirect on adipocyte promoters suggesting the potential role of other histone modifiers in regulating adipogenic differentiation of MSC. Understanding the molecular mechanisms of epigenetic modifiers such as EZH2 and KDM6A in regulating MSC lineage determination is pivotal for understanding bone cell differentiation and diseases, in circumstances where a decrease in bone mass result from a qualitative or quantitative alteration in MSC pool.
Funding in part from NHMRC project Grants #626910 and #1046053.
S.H. and D.C.: conception and design, collection and/or assembly of data, data analysis and interpretation, and manuscript writing; S.I., L.C., and D.M.: collection and/or assembly of data and data analysis and interpretation; A.Z.: data analysis and interpretation, provision of study material, and manuscript writing; S.G.: conception and design, financial support, administrative support, provision of study material, data analysis and interpretation, manuscript writing, and final approval of manuscript. D.C. and S.G. are Co-senior authors.
Disclosureof Potential Conflictsof Interest
The authors indicate no potential conflicts of interest.