Coordinated Activation of Notch, Wnt, and Transforming Growth Factor-β Signaling Pathways in Bone Morphogenic Protein 2-induced Osteogenesis

To examine early events in osteoblast differentiation, we analyzed the expression of about 9,400 genes in the murine MC3T3 cell line, whose robust differentiation was documented cytochemically and molecularly. The cells were stimulated for 1 and 3 days with the osteogenic stimulus containing bone morphogenic protein 2. Total RNA was extracted and analyzed by Affymetrix GeneChip oligonucleotide arrays. A regulated expression of 394 known genes and 295 expressed sequence tags was detected. The sensitivity and reliability of detection by microarrays was shown by confirming the expression pattern for 20 genes by radioactive quantitative reverse transcription-PCR. Functional classification of regulated genes was performed, defining the groups of regulated growth factors, receptors, and transcription factors. The most interesting finding was concomitant activation of transforming growth factor-β, Wnt, and Notch signaling pathways, confirmed by strong up-regulation of their target genes by PCR. The transforming growth factor-β pathway is activated by stimulated production of the growth factor itself, while the exact mechanism of Wnt and Notch activation remains elusive. We showed that bone morphogenic protein 2 stimulated expression of Hey1, a direct Notch target gene, in mouse MC3T3 and C2C12 cells, in human mesenchymal cells, and in mouse calvaria. Small interfering RNA-mediated inhibition of Hey1 induction led to an increase in osteoblast matrix mineralization, suggesting that Hey1 is a negative regulator of osteoblast maturation. This negative regulation is apparently achieved via interaction with Runx2: Hey1 completely abrogated Runx2 transcriptional activity. These findings identify the Notch-Hey1 pathway as a negative regulator of osteoblast differentiation/maturation, which is a completely novel aspect of osteogenesis and could point to possible new targets for bone anabolic agents.

by two main types of bone cells: bone-forming osteoblasts of mesenchymal origin and bone-resorbing osteoclasts of hematopoietic origin (1). A synchronized action of osteoblasts and osteoclasts enables balanced bone remodeling. If this balance is changed in a way that bone resorption exceeds bone formation, osteoporosis occurs, a disease prevalent in old age and characterized by bone loss and a high risk of fractures. Most treatments for osteoporosis available so far target osteoclasts and inhibit bone resorption. In osteoporosis, however, bone loss exceeds the degree of bone gain that can be restored by inhibitors of resorption. Therefore, there is a large need for anabolic agents that would accelerate osteoblast differentiation and promote bone formation. For that purpose, knowledge about molecular events involved in osteoblast differentiation is crucial.
Previously osteoblast differentiation was examined mostly at the cellular or single gene level. Many external regulating factors are known, but critical molecular steps in osteoblast differentiation and bone formation are largely unknown. One key player was identified recently as the Runx2 transcription factor (2). Another transcription factor, Osterix (Osx), which cooperates with and is genetically downstream of Runx2, has also been identified (3). However, more molecular players in osteoblast differentiation remain to be identified.
In vivo, bone forming osteoblasts develop from mesenchymal precursors, and this process can be mimicked in vitro. Different cellular phenotypes during osteogenesis were tentatively defined as osteoprogenitors, preosteoblasts, mature osteoblasts, and mineralizing osteoblasts, each of them characterized by an overlapping set of marker genes (4). Classical cytochemical markers of osteoblast differentiation are alkaline phosphatase and mineralized bone nodules. To examine osteoblast differentiation in vitro, a crucial step is the use of appropriate primary cells or cell lines and defined culture conditions, which allow an ordered stepwise differentiation process. To avoid many inherent problems observed with differentiation of osteosarcoma cell lines, which are transformed and may not be representative of the physiological situation, we used a non-transformed mouse calvarial cell line, MC3T3-E1. This cell line is also known for phenotypic variation in culture (5); therefore, we used a cell clone obtained at a low passage number from a laboratory that kept the cells at their original maintenance conditions (6). To ensure that the cells also show expected behavior at the molecular level, we selected from this MC3T3-E1 cell batch a further clone, which efficiently activated a Runx2-dependent reporter gene and activated Runx2 mRNA and protein. This clone of MC3T3-E1 cells exhibited very robust and fast differentiation properties. We then used this MC3T3-E1 cell clone to examine a genome-wide pattern of gene expression during early differentiation along osteoblastic lineage. The quality and * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
A total data set from microarray analyses has been submitted to the NCBI gene expression and hybridization array data repository (GEO) under GEO submission numbers GSE1131 and GSM18546 - GSM18560.
‡ To whom correspondence should be addressed: Novartis Institutes for BioMedical Research, WKL-125.9.12, CH-4002 Basel, Switzerland. Tel.: 41-61-696-44-49; Fax: 41-61-696-38-49; E-mail: mira.susa_spring@ pharma.novartis.com. refrozen in liquid nitrogen; 1 ml of TRIzol (Invitrogen) was added, and samples were rotated for 1 h. After 1 min of centrifugation at 13,000 rpm, the supernatant was collected, and phenol-chloroform extraction was performed. 0.2 ml of 1-bromo-3-chloropane (Sigma) was added to each tube, and tubes were strongly shaken and then incubated for 2-3 min at room temperature followed by a 15-min centrifugation at 12,000 ϫ g at 4°C. The upper colorless aqueous phase (RNA phase) was collected, RNA was precipitated with isopropanol, and the pellet was washed with 75% ethanol, air-dried, and dissolved in 40 l of water. RNA was cleaned and DNase I-treated using the RNeasy Clean-Up kit (Qiagen) according to the manufacturer's instructions.
Gene Expression Analysis by High Density Oligonucleotide Microarrays-The results shown in this study are derived from three independent experiments with MC3T3 cells. In each experiment, the cells were treated identically: no stimulation, day 0; GP treatment at days 1 and 3; and osteogenic stimulus treatment (GP/AA/BMP-2) at days 1 and 3. Total RNA from each sample was extracted and analyzed on oligonucleotide microarrays. Before microarray analysis, for each experiment we performed a marker gene analysis and cytochemical staining for alkaline phosphatase and mineralization to ensure that cells responded appropriately to the osteogenic stimulus. Microarray hybridizations were performed in the Pharmacogenomics Area, Novartis Pharma Development. Affymetrix GeneChip® Murine Genome U74Av2 arrays were used that consist of coated glass slides with series of oligonucleotide probes synthesized in situ. These arrays contain probes for ϳ9,400 genes (ϳ5,700 functionally characterized genes and ϳ3,700 expressed sequence tag (EST) clusters). Biotin-labeled cRNA probes were generated from each sample to be analyzed, starting from 5 g of DNase I-treated total cellular RNA prepared as described above. The cRNA probes were individually hybridized on the arrays, and the signals were detected according to the manufacturer's instructions (Affymetrix, Santa Clara, CA).
Hybridization data were analyzed using the MAS 5.0 (Affymetrix), NPGN (Novartis Pharmacogenetics Network), and Expressionist 3.0 (GeneData, Basel, Switzerland) software. Genes were considered as significantly expressed in a given experiment if they were classified as P (present), but not as M (marginal) or A (absent), at least at one time point. Twenty was chosen as the minimal significant hybridization signal value; all lower values were set to 20. All genes discussed and studied herein were detected with gene-specific probes but not with probe sets that recognize gene families.
Genes were selected as regulated by osteogenic stimulus if their expression deviated more than 2-fold from the corresponding timematched control, at any time point, in at least two of three experiments. Table I shows that mean relative expression levels (MRELs) for all three experiments were around 1, showing that most of the genes do not change their expression levels upon treatment. Standard deviation of the MREL represents a degree of variation in the expression level compared with the mean value for all the genes. Thresholds of 2-fold reflect approximately 1 standard deviation around the MREL for all experiments. In addition to this statistical criterion, biological data indicated that 2-fold is a meaningful difference since some of the known BMP-2-regulated genes (such as Dlx2 and Dlx5) were induced to a similar degree (see "Results" and Table II).
By analyzing osteoblast marker genes for each independent experiment, we observed that the kinetics of the differentiation process is not always the same, although the cells were treated exactly the same way. For example, PTHR was induced to a much higher degree at day 3 in experiment 3 as compared with experiments 1 and 2, suggesting that the differentiation process was faster and/or stronger in experiment 3 (data not shown). Higher standard deviations of MREL at day 1 in experiment 3 (Table I) also show that in this experiment more genes are regulated already at day 1, suggesting a faster differentiation process than in experiments 1 and 2. Because of this expected biological variability, we considered as significantly regulated only genes showing regulation in two of three experiments, a criterion that better tolerates the biological variability of the differentiation process. When calculating average expression from three experiments, the median value, which gives less significance to outlier values, was used.
Human mesenchymal stem cells (Poietics TM , Cambrex) obtained from human bone marrow withdrawn from the posterior iliac crest of the pelvic bone of a 19-year-old female donor were treated with 10 mM ␤GP, 50 M AA, and 1000 ng/ml BMP-2. Total RNA was extracted (RNeasy minikit, Qiagen), and the samples were analyzed on the Affymetrix GeneChip HG-U133A.
Real Time Quantitative RT-PCR Analysis-cDNA was synthesized using the High-Capacity cDNA Archive kit (product number 4322171 by Applied Biosystems), starting from 1 g of RNA, according to the manufacturer's protocol. For real time quantitative RT-PCR, an ABI Prism 7900HT sequence detection system (Applied Biosystems) was used. Reactions were performed in a 394-well format in a 10-l total volume using 5 l of 2ϫ Master Mix (TaqMan universal PCR Master Mix, Applied Biosystems), 0.5 l of 20ϫ primers, probes synthesized by Applied Biosystems, Assay-On-Demand (18 S rRNA, 4310893E; Hey1, Mm00468865_m1), and 2 l of cDNA (equivalent to 20 ng of RNA). Thermal conditions were as follows: 10 min at 50°C, 10 min at 94°C, followed by 40 cycles of 15 s at 94°C and 1 min at 60°C. Negative controls were included in each PCR experiment, with RTϪ instead of cDNA. -Fold inductions and expression ratios between two samples were calculated from differences in threshold cycles at which an increase in reporter fluorescence above a base-line signal could be first detected (Ct value). Results were averaged from triplicate determinations. 18 S rRNA was used as a normalization control.
Small Interfering RNA (siRNA) Transfection-MC3T3 cells were plated on 6-well plates (0.5 ϫ 10 5 cells/well in 2 ml of medium). After 24 h, siRNA transfection was performed in a total volume of 1 ml using Oligofectamine (Invitrogen) according to the manufacturer's instructions. siRNA concentration was 0.1 M, and the amount of Oligofectamine used was 4 l/well. Transfection was stopped after 4 h by adding 0.5 ml of medium containing 30% serum and the osteogenic stimulus. RNA was isolated after 1, 2, 3, or 4 days. For Alizarin Red S staining, medium containing 10% fetal calf serum and the osteogenic stimulus was changed twice weekly. Staining was performed after 17 days. The siRNA sequence of sense strand Hey1 siRNA was GCTA-GAAAAAGCTGAGATC with dTdT overhangs purified by ion exchangehigh pressure liquid chromatography (Xeragon Inc.). The sense strand sequence of control siRNA was AGAAGGAGCGGAATCCTCG with dTdT overhang (provided by François Natt, Novartis).
Transient Co-transfections and the Luciferase Assay-Runx2 and Hey1 cDNA were cloned into the pcDNA3.1 (ϩ) expression vector and used in a luciferase reporter assay. Primers used for Hey1 cDNA cloning were: sense, GACCCTCCTCGGAGCCCAC; antisense, TTAGAAAGCT-CCGATCTCTGTCC. The OG2luc plasmid containing the minimal osteocalcin promoter in front of the luciferase gene was used as a reporter construct. This luciferase expression vector (without OG2 promoter) was a kind gift from Roland Schule, Freiburg, Germany. Eight times repeated wild-type (8XOSE2 wt) or mutated (8XOSE2 mut) Runx2binding sites have been cloned upstream of the OG2 minimal promoter (Johann Wirsching, Novartis). Cells were seeded in 96-well plates (4 ϫ 10 3 cells/well in 200 l of medium) and grown for 24 h. Transfection was performed using LipofectAMINE Plus reagent (Invitrogen) according to the manufacturer's protocol. An MTS assay for cell number normalization and luciferase assays were performed 24 h after transfection. For the MTS assay, 20 l/well of MTS solution (Cell Titer 96 Aqueous One Solution Reagent, Promega) was added to the medium (100 l/well). Cells were incubated for 20 min at 37°C. Absorbance was measured at 490 nm in a microplate reader. For the luciferase assay, cells were washed twice with phosphate-buffered saline, and 50 l/well 1ϫ lysis buffer (Promega) was added and incubated for 15 min at room temperature. Subsequently the cultures were shaken for 5 min at room temperature and frozen for 1 h at Ϫ70°C. Upon thawing, 20 l of lysate was transferred into a white 96-well plate (Costar £3912, opaque plate), 2 ϫ 50 l of Luciferase Assay Reagent (Promega) was added, and luciferase luminescence was measured (MicroLumat LB 96P, Berthold).

Characterization of Osteoblastic Differentiation in MC3T3-1b Cell Line
In response to osteogenic stimulus (BMP-2, AA, and GP), the MC3T3-1b clone of MC3T3-E1 cells (referred to hereafter as MC3T3) showed a strong increase in alkaline phosphatase activity at day 3 (Fig. 1A). Runx2 protein levels were also increased by day 3 (Fig. 1B). Furthermore the cells produced bone nodules, which mineralized by days 11-14 ( Fig. 1C). After a longer exposure to osteogenic stimulus, the number of nodules increased dramatically, covering more than 50% of the dish area. Thus, the whole osteoblast differentiation/maturation process is reproduced sequentially within 2 weeks, allowing the exact study of osteoblast biology.
To study the differentiation process at the transcriptional level, we analyzed mRNA expression of a number of molecular markers of osteoblast differentiation. MC3T3 cells were stimulated with osteogenic stimulus for 1 and 3 days and compared with non-stimulated time-matched controls. Eight markers of osteoblast differentiation were analyzed: ALP, transcription factors Msx2 and Runx2, PTHR, and extracellular matrix proteins OCN, OPN, SPARC, and ColI␣1 (Fig. 1D). mRNA levels of five osteoblast markers were up-regulated in response to the osteogenic stimulus. ALP was strongly induced already at day 1 and further increased at day 3. Msx2 was transiently upregulated at day 1, and Runx2 was induced at day 3. PTHR and OCN, late markers of osteoblast differentiation, were strongly induced at day 3. For OPN, OCN, and ColI␣1, high basal levels of expression were already detected in non-stimulated cells and did not change upon osteogenic treatment. As MC3T3 cells are already committed to the osteoblast lineage, high expression levels of these extracellular matrix proteins is not surprising. Similar levels of ribosomal RNA (18 S rRNA) and GAPDH control mRNAs were found in all treatment conditions. C2C12 premyoblastic and primary mouse calvarial cells were also considered as alternative cellular systems to the MC3T3 cell line. However, they expressed a smaller number of regulated markers, and the variability was greater than in MC3T3 cells (data not shown). We concluded that MC3T3 cells are an appropriate system for studying changes in gene expression during osteoblastic differentiation.

Expression of Osteoblast Marker Genes on Microarrays
Having selected an appropriate in vitro osteoblast differentiation system, we wanted to identify genes potentially involved in this process by performing a genome-wide analysis of gene expression. MC3T3 cells were stimulated with osteogenic factors, and total RNA was extracted at days 0, 1, and 3 from three independent experiments. Gene expression was analyzed on Affymetrix GeneChip microarrays representing about 10,000 ESTs and genes with known function.
To validate the microarray data for each independent experiment, expression of osteoblast marker genes on microarrays was compared with the qrRT-PCR data. The GeneChip contained oligonucleotide probes for seven of eight chosen osteoblast markers and for GAPDH control gene. For these genes, both methods produced a very similar pattern of regulation. The results from a representative experiment are shown in Fig.  2. Reliable hybridization microarray data were obtained even for weakly expressed genes, such as those encoding transcription factors (i.e. Msx2 and Runx2, Fig. 2), indicating a good sensitivity of detection. We concluded that microarray detection was sensitive enough to detect expression and regulation of osteoblast markers and thus should be a good method to detect regulation of novel genes as well.

Non-hierarchical Clustering of Genes Regulated during Osteoblast Differentiation
To get an insight into the transcriptional events involved in the osteoblast differentiation, we investigated genes whose expression levels changed upon treatment with the osteogenic stimulus. Genome-wide gene expression levels were compared between treated samples and time-matched non-stimulated controls. We detected a significant regulation (2-fold in at least two of the three experiments) of 394 genes with known function and of 295 ESTs. In further analyses described here, we focused on the genes with known function. This subset of genes was further analyzed by a non-hierarchical clustering method, which groups genes according to their temporal regulation patterns (Fig. 3). One hundred seventy-two genes were upregulated, and 222 genes were down-regulated. There was a very good concordance between the general gene expression patterns observed in three independent experiments and in their median (Fig. 3).

Regulated Genes
We defined several groups according to the cellular function of the up-and down-regulated genes. A large number of regulated genes encode extracellular matrix proteins and adhesion molecules (data not shown). This is consistent with the fact that osteoblasts are adherent cells responsible for production of the bone matrix. The regulation of many genes involved in the cell cycle and DNA replication is consistent with our observation that the cells, although almost confluent prior to stimulation, still continue to proliferate to a small degree up to 3 days after stimulation with the osteogenic stimulus (data not shown). This regulation is also consistent with the process of exiting the cell cycle and switching to differentiation program. We focused, however, on three groups of genes, whose products could have major contributions to the differentiation process: growth factors, receptors, and transcription factors.
Growth Factors-The growth factor group contained genes expected to be regulated in osteoblasts but also novel genes (Table II). Among the expected genes, we found known modulators of osteoblast proliferation: platelet-derived growth factor ␣, vascular endothelial growth factor, fibroblast growth factor, IGFbinding proteins (IGFBPs), and PTH-rP (8 -11). The most prominent event was regulation of four members of the TGF-␤ superfamily: TGF-␤1, TGF-␤3, and Activin were up-regulated, while TGF-␤2 was down-regulated (Table II). TGF-␤ family members are well known and potent modulators of the osteoblasts and bone (12), but their regulation by osteogenic stimulus containing BMP-2 has not been reported yet. We have shown that BMP-2 within the osteogenic mixture is responsible for activation of TGF-␤1 (data not shown). The activation of the TGF-␤ signaling pathway by the osteogenic stimulus was further indicated by the regulation of several TGF-␤ target genes: the transcriptional repressor TIEG, the extracellular matrix protein tenascin C, the adhesion molecule keratoepithelin, and the type 2 somatostatin receptor (SSTR2) ( Table V) (13)(14)(15)(16). Apart from the TGF-␤ pathway, worth noting is a 2-3-fold up-regulation of the BMP antagonist Gremlin2, which points to a negative feedback mechanism. These results shed a new light on the interplay between different growth factors in osteoblast differentiation.
Among down-regulated genes, we will mention IGFBP2 and -6, known modulators of IGF-induced osteoblast proliferation. Furthermore angiotensinogen, a precursor of angiotensin, was progressively down-regulated at days 1 and 3. Since angiotensin II was reported to have a role in osteoblast differentiation, this could be of relevance (17). Two additional down-regulated genes were small inducible cytokine A7 (SCYA7) and PTH-rP, both of which can stimulate differentiation or function of boneresorbing osteoclasts (18,19), suggesting that differentiating osteoblasts have a reduced ability to stimulate osteoclasts.
Receptors-Up-regulated genes in the receptor group included those encoding for the receptors of many factors implicated in osteoblast differentiation or function such as PTHR, leukemia-inhibitory factor receptor, leptin receptor, prostaglandin F receptor, fibroblast growth factor receptor 2, urokinase receptor, and thrombomodulin (Table III) (11, 20 -25). Some of the up-regulated receptor-encoding genes, such as the ephrin receptor EPHA2 or the somatostatin receptors SSTR4 and SSTR2, have not been reported previously to play a role in osteoblasts. The highest up-regulated receptor genes were those encoding PTHR, an established osteoblast marker, and surprisingly a co-receptor for calcitonin receptor-like receptor (RAMP1). Interestingly osteoprotegerin (OPG), a decoy receptor that inhibits the signaling of receptor activator of NFB ligand (RANKL), the main cytokine in osteoclastogenesis, was up-regulated almost 4-fold. This result, together with the down-regulation of osteoclast-stimulating SCYA7 and PTH-rP described in the previous section, strongly suggests that differentiating osteoblasts have a reduced ability to stimulate boneresorbing osteoclasts function and differentiation (Table V).
Notch receptors 1 and 3 were prominently down-regulated. The role of Notch signaling in osteoblast differentiation has not been much investigated; only two conflicting reports indicate that Notch signaling may influence osteoblast differentiation (26,27). Two low density lipoprotein-related proteins LRP1 and LRP5 were also down-regulated. This result is intriguing because LRP5 is an important player in bone metabolism and has been identified as a high bone mass gene (28 -30). Transcription Factors-The transcription factor group of regulated genes is the largest (Table IV). Despite the fact that transcription factors are generally weakly expressed, we detected 33 regulated transcription factor genes, all with reliable hybridization signals. Furthermore we confirmed regulation patterns for 10 of these genes by qrRT-PCR (Fig. 4). The regulation of a large number of transcription factors suggested that their orchestrated regulation is crucial for the osteoblast differentiation process.
Three genes encoding Smads, major components of the BMP-2 signaling pathway, were shown to be up-regulated 2-3fold. The inhibitory Smads, Smad6 and -7, were transiently up-regulated at day 1 (Table IV), exposing another component of a negative feedback loop for regulation of BMP signaling. This result is in agreement with another study (31). A BMP receptor-specific stimulatory Smad1 was up-regulated at day 3, indicating that BMP induces components of its signaling pathway as a means of signal enhancement.
Similarly to Smads, many other regulated transcription factor genes have an already established function during osteoblast differentiation. Some genes that were previously identified as BMP-2 target genes in osteoblasts by Locklin and colleagues (31) were also found to be regulated in our study (Table IV). JunB, a component of AP-1 transcription factor complex that is involved in osteoblast differentiation (32), was strongly up-regulated at both time points studied. Id1, -2, and -3, encoding helix-loop-helix transcriptional inhibitors shown to be direct BMP-2 target genes (33), were strongly up-regulated, while Id4, another member of this gene family, was down-regulated. The homeobox transcription factors of the Msx and Dlx families are important in skeletal development (34 -36). Msx2, a known BMP-2 target gene and osteoblast differentiation marker, and Dlx2 and Dlx5, other known BMP-2 targets (37), were all up-regulated in our osteoblast differentiation system. Another member of the Dlx family, Dlx1, was also up-regulated, and this is, to our knowledge, the first report of BMP-2 regulation of this gene. These confirmatory findings further strengthen the relevance of our novel findings. Interestingly a Wnt pathway target gene Tcf7 was very strongly up-regulated (about 10-fold at day 3). LEF1 is another transcription factor mediating Wnt signaling and was also up-regulated (2-fold at day 3). This is a novel and intriguing finding taking into account a central role of the Wnt pathway in osteoblast differentiation unraveled through LRP5, a Wnt coreceptor and a high bone mass gene (28 -30). However, we did not detect a consistent regulation of Wnt genes in our system in contrast to a recent report (38). In one of three experiments, we detected up-regulation of Wnt6 and Wnt10a at day 3 (data not shown). Weak up-regulation of Wnt6 was confirmed by qrRT-PCR (data not shown). However, in the other two experiments, Wnts were not significantly regulated. We conclude that the Wnt pathway is activated, but the exact mechanism of activation remains to be determined. Regulated components of the Wnt pathway are shown in Table V.
As one of the most striking novel findings, we point to a strong up-regulation of the Hey1 transcription factor (about 3-fold at day 1 and about 5-fold at day 3, Table IV). Hey1 belongs to a family of transcription factors named HES, which belong to a superfamily of basic helix-loop-helix transcription factors. The Hey1 subfamily has only recently been described, and it comprises three members: Hey1, Hey2, and HeyL (39). Together with other HES family members they are thought to be direct targets of the Notch signaling pathway (40,41). Strong up-regulation of the Hey1 transcription factor suggested that the Notch signaling pathway is activated in MC3T3 cells during osteogenic differentiation. Regulated components of the Notch signaling pathway are shown in Table V.

Confirmation of Selected Gene Profiles by qrRT-PCR
We selected a few representative genes from the three gene groups described above and tested their expression by an independent method: qrRT-PCR (Fig. 4). Selected genes were growth factors TGF-␤1 and TGF-␤3, (co)receptors LRP5 and RAMP1, and transcription factors Smad1, Smad6, JunB, Id2, Dlx2, TIEG, Tcf7, and Hey1. For all the genes analyzed, the expression profiles obtained by qrRT-PCR closely corresponded to the profiles obtained by microarray analysis (Tables II-IV and Fig. 4). Together with osteoblast markers and GAPDH, we analyzed and confirmed the expression levels of 20 genes, again indicating the relevance of gene expression analysis in these experiments.

Hey1 Expression in Mouse and Human Osteoblastic Cells and Mouse Calvaria
In further work we focused on expression and function of Notch target gene transcription factor Hey1. To determine which component of the osteogenic stimulus is responsible for Hey1 up-regulation, we analyzed Hey1 mRNA levels after treating MC3T3 cells with separate components of the osteogenic stimulus (Fig. 5A). A time course was performed during 4 days after addition of the stimulus. The result revealed that Hey1 is a BMP-2-induced gene, although ascorbic acid had a small additive effect especially noticeable at day 4 (Fig. 5A). Next we have analyzed Hey1 expression in several osteoblast differentiation systems. Hey1 is strongly induced by BMP-2 in murine C2C12 cells, which have a potential to differentiation into either myoblasts or osteoblasts (Fig. 5B). The induction was measured by qrRT-PCR at days 1 and 3 and reached about 10-fold. Human Hey1 expression was determined by microarrays in human mesenchymal stem cells stimulated by osteogenic stimulus containing BMP-2 (Fig. 5C). The induction persisted up to day 15.
We analyzed Hey1 expression in vivo in mouse calvarial bone by real time PCR. BMP-2 was injected subcutaneously over calvariae, and RNA was isolated after 7 and 14 days. Hey1 mRNA levels were already high in non-treated bone tissue comparable to the BMP-2-induced levels in MC3T3 cells (Fig. 5,  D and E). This is evident from similar ⌬Ct values for Hey1 detection and is likely due to endogenous calvarial BMP-2.

FIG. 4. qrRT-PCR confirmation of selected up-regulated genes identified by GeneChip microarrays.
Microarray-derived expression profiles for selected genes were confirmed by qrRT-PCR. Confluent MC3T3 cells were cultured in medium containing GP without (Ϫ) or with (ϩ) AA/BMP-2 for 0, 1, and 3 days. Total RNA was extracted and used either for microarrays analysis or for qrRT-PCR. The radioactive PCR products were analyzed by polyacrylamide gel electrophoresis and visualized by PhosphorImager. The genes were grouped based on their function.

TABLE V Expression profiles of regulated genes of the TGF-␤, Wnt, and Notch signaling pathways and genes involved in osteoclastogenesis control
Genes from TGF-␤, Wnt, and Notch signaling pathways as well as genes involved in regulation of osteoclastogenesis whose expression changed Ն2-fold upon stimulation with osteogenic stimulus are shown. PSN, Affymetrix probe set number; Acc no., GenBank TM sequence accession number; REL, relative expression level (median value) compared with the non-stimulated, time-matched control; GPCR, G-protein-coupled receptor; bHLH, basic helix-loop-helix. Treatment of calvariae with exogenous BMP-2 induced a trend of Hey1 induction visible through a reduction of Ct values at day 14. Induction of Hey1 was strongest in MC3T3 cells treated by BMP-containing osteogenic stimulus, reaching over 1,000fold difference in expression levels at day 4 as measured by real time PCR (Fig. 5E). We conclude that Hey1 is induced by BMP-2 in both mouse and human cells and is expressed in mouse bone.

Down-regulation of Hey1 mRNA by siRNA Stimulates Mineralization
Since members of the HES-Hey family work predominantly as transcriptional repressors (42) and since Hey1 inhibits myogenic differentiation (43), we tested whether the induction of Hey1 could inhibit osteogenic differentiation. For this reason we blocked BMP-2-induced Hey1 up-regulation with a siRNA specific for Hey1 (Fig. 6A). Hey1 was efficiently down-regulated at day 1, and this down-regulation persisted until day 3 (Ͼ70% inhibition). At day 4, Hey1 mRNA levels in siRNA-treated samples returned back to the control levels (Fig. 6A). Then we tested the effect of this transient Hey1 down-regulation on several osteoblast early marker genes (Fig. 6B). However, all these genes (ALP, Osx, Id2, Dlx2, and Fra-1) were unaffected by the treatment by siRNA for Hey1. This result indicated that Hey1 was not involved in regulation of genes expressed early in osteoblastic lineage. Therefore, we next tested the effect of Hey1 down-regulation on late events in osteoblastic differentiation, such as bone nodule formation and mineralization (Fig. 6C). Treatment with osteogenic stimulus induced intense mineralization in MC3T3 cultures (Fig. 6C).
The mineralization capacity of MC3T3 cells treated with siRNA for Hey1 further increased this mineralization, while control siRNA had a somewhat inhibiting effect compared with mock-treated cultures (Fig. 6C). These results showed that Hey1 has an inhibitory role during matrix mineralization by osteoblasts and that even transient removal of Hey1 is sufficient to influence this process.

Hey1 Inhibits Runx2 Transcriptional Activity
The above results prompted us to further analyze the molecular basis of enhanced osteoblast activity due to the inhibition of Hey1 induction. Runx2 is a transcription factor that plays an essential role in osteoblast differentiation, and mutations interfering with its function correlate with defects in ossification in humans and mice (44). Many signal transduction pathways that affect osteoblast differentiation modulate Runx2 activity (45). Therefore, we examined the effect of Hey1 on Runx2 transcriptional activity using a luciferase reporter gene driven by an artificial promoter containing eight Runx2-binding sites (OSE2) in front of a minimal osteocalcin promoter, mOG2. As shown in Fig. 7A, ectopic expression of Runx2 strongly stimulated transcription of the reporter gene controlled by the intact, wild type (Fig. 7A, 8XOSE2 wt, Runx2). As expected, Runx2 did not stimulate reporter gene activity that was controlled by the promoter without OSE2 sites, which contained only minimal osteocalcin promoter (OG2luc, Runx2). Similarly no activation by Runx2 was detected with the promoter containing mutated OSE2 sites (8XOSE2 mut, Runx2). Co-expression of Hey1 with Runx2, however, almost completely abrogated Runx2-driven transcription (Fig. 7A, Runx2 ϩ Hey1). Control co-transfections showed that empty vector pcDNA3.1 had a small effect on reporter genes and that pcDNA3.1 did not affect Runx2-induced reporter gene activity. Fig. 7B shows increased expression levels of Runx2 and Hey1 in co-transfection experiments. These results suggested that Hey1 inhibits bone matrix mineralization by osteoblasts by controlling Runx2 activity.

Microarray Analyses of Genome-wide Gene Expression in
Osteoblasts-There is a strong interest in discovering new players in the osteoblast differentiation process, which is still far from being completely understood. One approach to unraveling the molecular players in the osteoblast differentiation process is to analyze genome-wide gene expression profiles during differentiation, and the results of several such studies have recently been reported (31, 46 -49). These studies, using different model systems of osteoblast differentiation and different differentiation agents, all highlighted some new genes involved in the differentiation process, proving the usefulness of microarray experiments.
In this study, we identified 394 genes with known function and 295 ESTs to be regulated during osteoblast differentiation. The features of this study are (a) extensive cellular and molecular characterization of the osteoblast differentiation process for each individual experiment analyzed by microarray, (b) normalization to time-matched controls to eliminate gene expression changes spontaneously occurring during cell culture, (c) stringent selection criteria, (d) extensive qrRT-PCR validation of regulated genes, and (e) annotation and classification of regulated genes according to their cellular functions. As the most interesting and novel finding we report the concomitant regulation of genes encoding components of the TGF-␤, Wnt, and Notch signaling pathways. Thus, although we used technology similar to that in studies described above, the precision of our cellular system and extensive control of experimental and analysis conditions enabled us to unravel several new sets of genes that are regulated during osteogenesis and to come to a more compete picture of osteogenesis.
TGF-␤ Pathway-We showed that activation of the TGF-␤ pathway by osteogenic stimulus and more precisely by BMP-2 is achieved by increasing the autocrine production of ligands TGF-␤1 and -3 and of related ligand Activin followed by the up-regulation of corresponding target genes (TIEG, tenascin, keratoepithelin, and SSTR2). Biological evidence for a role for TGF-␤ in osteoblast regulation is ample (12), but its mechanisms of action are complex and poorly understood. Differences in the ultimate effect appear to depend on TGF-␤ concentration, duration of exposure, and cell differentiation status (50). TGF-␤ inhibits apoptosis in osteoblasts (51), and it enhances osteoclast differentiation (52). It has been shown that pretreatment of osteoblastic cells with BMP-2 changes the binding of TGF-␤ to its receptors by increasing binding to the type I receptor (53). Depending on the status of the cells being tested, this binding shift FIG. 6. The effects of Hey1-specific siRNA on Hey1, early marker gene expression, and matrix mineralization by MC3T3 cells. A, inhibition of osteogenic stimulus-induced Hey1 expression by Hey1-specific siRNA. MC3T3 cells were not transfected (untreat and treat), mock-transfected (untreat ϩ mock and treat ϩ mock), transfected with Hey1-specific siRNA (Hey1siRNA), and transfected with non-silencing siRNA (control (Ctrl) siRNA). After 4 h of siRNA transfection cells were either not treated (untreat and untreat ϩ mock) or treated with GP/AA/BMP-2 (treat, treat ϩ mock, Hey1siRNA, and control (Ctrl) siRNA). Total RNA was isolated after 1, 2, 3, or 4 days, and Hey1 expression was analyzed by qrRT-PCR. B, the effects of Hey1 siRNA on expression of early osteoblast marker genes. Total RNA isolated as described in A (day 2) was used for the analysis of expression of the indicated genes by qrRT-PCR. C, the effect of Hey1 siRNA on matrix mineralization by MC3T3 cell cultures. MC3T3 cells were either mocktransfected (untreat ϩ mock and treat ϩ mock) or transfected with Hey1 siRNA or with non-silencing siRNA (Control siRNA). After 4 h of transfection cells were either not treated (untreat) or treated with GP/AA/ BMP-2 (treat ϩ mock, Hey1siRNA, and Control siRNA). Alizarin Red S staining was performed 17 days after transfection and is visible as dark areas on the photos shown.

FIG. 7. Hey1 inhibits Runx2 transcriptional activity.
A, activity of OSE2-luciferase in MC3T3 cells after co-transfection of Runx2, Hey1, and OSE2-luciferase vectors. Runx2 and Hey1 cDNAs in pcDNA3.1 (ϩ) expression vector were transiently co-transfected with OSE2-luciferase reporter gene containing a minimal osteocalcin gene promoter (OG2) and eight wild-type (8XOSE2 wt) or mutated (8XOSE2 mut) Runx2binding sites. Transfection was done for 4 h, and luciferase activity was measured after 24 h. B, overexpression of transfected Runx2 and Hey1 shown by qrRT-PCR. Transfection was done as in A, and after 24 h total RNA was extracted, and qrRT-PCR was performed. RLU, relative luciferase units. enhances TGF-␤-induced collagen synthesis or alkaline phosphatase activity (53). Therefore, our result that BMP-2 induced the TGF-␤ pathway suggests a way to broaden and diversify the direct action of BMP-2 signaling on osteoblast.
Wnt Pathway-The next interesting finding was the activation of the Wnt pathway by osteogenic stimulus documented by up-regulation of two target genes, transcription factors Tcf7 and LEF1. During the course of our work, the Wnt pathway came into focus in bone biology through the finding that LRP5, a co-receptor for Wnt, is a high bone mass gene and is mutated in osteoporosis-pseudoglioma syndrome (28 -30). Interestingly we showed that LRP5 was down-regulated by the osteogenic stimulus. Recently it was shown that BMP-2-induced activation of alkaline phosphatase, an early marker of osteoblast differentiation, depends on Wnt/LRP5 signaling and that BMP-2 induces Wnt1 and Wnt3a expression, resulting in autocrine Wnt pathway activation (38). We have also observed transcriptional up-regulation of some Wnt family members; however, they were not identical to those reported (Wnt6 and Wnt10a), and this activation was not reproduced in two of three experiments. Therefore, we concluded that this up-regulation may not be significant and that mechanisms of Wnt pathway activation by BMP-2 need further evaluation.
Notch Pathway and Hey1-Another novel finding was a strong up-regulation of Hey1, a direct Notch target gene. Notch signaling is an evolutionarily conserved mechanism used by metazoans to control cell fates through local cell interactions (54). Initially the Notch pathway was linked to bone biology by observations that mutations in the genes encoding a Notch ligand Delta homologue (Dll-3) and a Notch signaling molecule presenilin-1 both cause axial skeletal phenotypes (55,56). Recently it was shown that generation of hematopoietic stem cells in bone marrow is supported by activation of Notch pathway by a ligand, Jagged1, produced by osteoblasts, pointing to a Notchmediated functional interaction between bone and bone marrow (57). So far there have been two studies investigating Notch signaling in osteoblasts; both used exogenous overexpression of the constitutively active Notch1 intracellular domain, and both produced conflicting results (26,27). We argue that the Notch pathway is activated by osteogenic stimulus and BMP-2 based on a strong up-regulation of Hey1 transcription factor. However, we did not detect mRNA for the Notch ligands Delta and Jagged in MC3T3 cells (data not shown), indicating that they are weakly expressed or absent. As Notch activation is achieved via proteolytic cleavage (58), a high protease expression/activity could be responsible for its activation. Another possibility for enhancing of Notch activation is association with gene NOV (nephroblastoma overexpressed gene), a growth factor from the CCN (constituted of connective tissue growth factor CTGF, cysteine-rich 61 Cyr61, NOV, and other related genes) gene family (59). Recently it was shown that NOV associates with Notch1 in C2C12 cells and thereby activates Notch, leading to induction of promoters of immediate target Notch genes HES1 and -5 (60). This activation led to an inhibition of myoblastic differentiation, an effect similar to the effect of BMP-2. Interestingly we found that NOV expression is high in MC3T3 cells, and it is strongly down-regulated in cells undergoing differentiation in parallel with Notch1 and -3. A graphical summary of gene activation on TGF-␤, Wnt, and Notch pathways is shown in Fig. 8.
What is the consequence of Notch pathway activation in various cell types? Expression of constitutively active Notch inhibits differentiation of neural and myogenic cells and promotes generation of hematopoietic stem cells (60 -62). Very recent studies have shown that BMP and Notch signaling do collaborate during differentiation of myogenic and endothelial cells and that Hey1 is a crucial player in this collaboration (63,64). Our data in osteoblast cells show that even transient inhibition of Hey1 expression leads to enhanced matrix mineralization, a sign of their maturation, by osteoblast. Therefore, in the osteoblastic lineage as well, the Notch pathway seems to have a role of preserving pluripotent cell phenotype. Down-regulation of Notch1, Notch3, and NOV later during osteogenesis suggests that cells are down-regulating this pathway to advance their differentiation. Thus, the Notch pathway seems to play a transient role only at the beginning of the differentiation.
Finally what is the mechanism of action of Hey1? One of the central players in the osteoblast differentiation process is the Runx2 transcription factor, which coordinates multiple signals involved in osteoblast differentiation (45). So far, two transcription factors were shown to form inhibitory complexes with Runx2 and to inhibit osteoblast differentiation: Stat1 (65) and Twist (66). Our result showing that Hey1 can almost completely abrogate Runx2 transcriptional activity indicates that Hey1 is a novel inhibitory partner of Runx2.
In summary, in this study we identified a number of genes that were regulated in osteoblastic differentiation and that were either known for their involvement in this process or were novel. The analysis of activated genes showed that BMP-2containing osteogenic stimulus, in addition to the expected activation BMP-2 pathway, concomitantly induces three other signaling pathways: those of TGF-␤, Wnt, and Notch (Fig. 8). Furthermore we highlighted a negative role of a Notch target gene, Hey1, in the osteogenic differentiation. Thus, these data enabled novel insights in osteoblast biology. porter gene and Cbfa1 expression plasmid and for the advice in cloning experiments; and Reinhard Moschitz, a trainee in our laboratory, for help with real time PCR experiments.