Characterization of Osf1, an Osteoblast-specific Transcription Factor Binding to a Critical cis-acting Element in the MouseOsteocalcin Promoters*

To elucidate the mechanisms of osteoblast-specific gene expression we are studying the regulation ofosteocalcin, the most osteoblast-specific gene. Previous studies of OG2, one of the two mouse osteocalcin genes, identified two osteoblast-specific cis-acting elements, OSE1 and OSE2, the latter being the binding site of Cbfa1, the only osteoblast-specific transcription factor known to date. Here we show that OSE1 is a cis-acting element as important as OSE2 for the osteoblast-specific expression of OG2 in cell culture and transgenic mice. We also show that OSE1 is present in the promoter of several osteoblast-specific genes including Cbfa1 itself. These biological features demonstrate the importance of OSE1 and led us to further characterize this site and the factor binding to it, provisionally termed Osf1. We first defined the core OSE1 sequence, 5′-TTACATCA-3′, which is necessary and sufficient for Osf1 binding to DNA. This sequence has no strong homology to any known transcription factor-binding sites. As a first step in identifying Osf1, we performed an analytical purification of this protein using nuclear extracts from two different osteoblastic cell lines. We purified Osf1 to homogeneity through a five-step procedure including a renaturation experiment and found that its apparent molecular mass is 40 kDa. In conclusion, this study indicates the existence of multiple osteoblast-specific cis-acting elements of equal importance in controlling OG2promoter activity and provides the first biochemical characterization of Osf1, a novel osteoblast-specific transcription factor.


From the Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, Texas 77030
To elucidate the mechanisms of osteoblast-specific gene expression we are studying the regulation of osteocalcin, the most osteoblast-specific gene. Previous studies of OG2, one of the two mouse osteocalcin genes, identified two osteoblast-specific cis-acting elements, OSE1 and OSE2, the latter being the binding site of Cbfa1, the only osteoblast-specific transcription factor known to date. Here we show that OSE1 is a cis-acting element as important as OSE2 for the osteoblast-specific expression of OG2 in cell culture and transgenic mice. We also show that OSE1 is present in the promoter of several osteoblast-specific genes including Cbfa1 itself. These biological features demonstrate the importance of OSE1 and led us to further characterize this site and the factor binding to it, provisionally termed Osf1. We first defined the core OSE1 sequence, 5-TTACATCA-3, which is necessary and sufficient for Osf1 binding to DNA. This sequence has no strong homology to any known transcription factor-binding sites. As a first step in identifying Osf1, we performed an analytical purification of this protein using nuclear extracts from two different osteoblastic cell lines. We purified Osf1 to homogeneity through a five-step procedure including a renaturation experiment and found that its apparent molecular mass is 40 kDa. In conclusion, this study indicates the existence of multiple osteoblast-specific cis-acting elements of equal importance in controlling OG2 promoter activity and provides the first biochemical characterization of Osf1, a novel osteoblast-specific transcription factor.
The osteoblast, a cell type of mesenchymal origin, is the only cell responsible for bone matrix deposition, also called bone formation (1). Bone formation is a physiological process of critical importance as it is involved in skeletal growth, bone remodeling, and fracture repair (2). Diseases affecting bone formation, whether they are genetically inherited like osteogenesis imperfecta, or acquired like osteoporosis, are common and very often debilitating (3)(4)(5)(6)(7). This underscores the biomedical importance of elucidating the genetic basis of osteoblastspecific gene expression and osteoblast differentiation. It is likely, as it is the case in many cell lineages (8 -11), that these two processes are related and that cell differentiation along the osteoblastic lineage is controlled, at least in part, by cell-specific transcription factors which need to be identified. In an effort to identify these factors we have embarked on a long-term study of the regulation of OG2, one of the two mouse osteocalcin genes, whose expression is strictly osteoblast-specific.
Previously, the analysis of the OG2 promoter defined a 147bp 1 fragment that confers osteoblast-specific expression to a reporter gene in vitro (12). Within this short OG2 promoter fragment we defined two osteoblast-specific cis-acting elements, termed OSE1 and OSE2 (originally designated as regions A and C, respectively), that are characterized by two important features. First, multimers of each of these elements confer osteoblast specific activity to a heterologous promoter. Second, OSE1 and OSE2 serve as binding sites for distinct factors that are only detected in nuclear extracts of osteoblasts but of no other cell types or tissues tested (12). These osteoblast specific binding activities were named Osf1 and Osf2, respectively. Further analysis of Osf2 revealed that it is the product of the Cbfa1 gene (13). We, and others have subsequently demonstrated that Cbfa1 acts as a transcriptional activator of osteoblast differentiation in mouse and human (13)(14)(15)(16)(17).
To date Cbfa1 remains the only osteoblast-specific transcription factor identified (18). However, several lines of evidence suggest that other cell-specific transcription factors are involved in the control of gene expression in osteoblasts. For instance, multiple investigators have identified osteoblast-specific cis-acting elements in the promoter elements of the genes encoding ␣1(I) collagen, bone sialoprotein, insulin-like growth factor 1, and osteocalcin (19 -25). Likewise, as mentioned above, we have shown that OSE1 binds an osteoblast-specific factor that is not Cbfa1 (12).
Although OSE1 has been poorly characterized so far, Osf1, the factor binding to it, has one important property. Osf1 is present in nuclear extracts of primary osteoblasts at the earliest stage of cellular differentiation, but is absent in nuclear extracts of differentiated osteoblasts that are surrounded by a mineralized matrix (12). Thus, Osf1 is not only an osteoblastspecific but also a stage-specific transcription factor that could conceivably act upstream of Cbfa1, provided that OSE1 has a biologic function.
The present study was aimed at defining the functional importance of OSE1 and characterizing Osf1. We first demonstrate that OSE1 is as important as OSE2 for OG2 expression in cell culture and transgenic mice. We identified an 8-bp core OSE1 sequence that is also present in the promoter of other osteocalcin genes and of the mouse Cbfa1 gene. The absence of overt homologies between OSE1 and other binding sites of known transcription factors suggests that Osf1 may be a novel factor and led us to perform an analytical purification. This purification showed that Osf1 is a single polypeptide with an apparent molecular mass of 40 kDa. This is the first step toward identifying and eventually cloning a cDNA encoding Osf1, a second osteoblast-specific transcription factor.
Cell Culture and DNA Transfections-Cell culture media and supplements were purchased from Life Technologies, Inc. ROS17/2.8 osteoblasts were cultured in Dulbecco's modified Eagle's medium/F-12, F9 teratocarcinoma cells in Eagle's minimal essential medium, C1 mesenchymal cells, and C2 myoblasts in Dulbecco's modified Eagle's medium. All media were supplemented with 10% fetal bovine serum. Primary cultures of newborn mouse osteoblasts were prepared and maintained as described previously (12). DNA transfection experiments were carried out using the calcium phosphate coprecipitation method as described (12) with 10 g of reporter plasmid constructs and 2 g of pSV-␤Gal to quantitate the transfection efficiency. All transfections were repeated at least four times in triplicate with different DNA preparations.
Generation and Analysis of Transgenic Mice-The constructs p1314luc and p1314-mOSE1-luc were digested by BamHI. Inserts were purified by two rounds of agarose gel electrophoresis and injected into the pronuclei of fertilized B6D2F1 mouse oocytes (Charles River Laboratories). The injected oocytes were implanted in the oviducts of pseudopregnant CD1 foster mothers (Jackson ImmunoResearch Laboratories, Inc.) for development to term. Animals expressing the transgenes were identified by screening for luciferase activity in tails. Three founder animals carrying the wild-type construct and two founder animals carrying the mutant construct were obtained. The copy number of integrated DNA sequence was determined in F1 generation animals as described previously (26). Luciferase activities in tissues was assessed in F1 generation animals at 10 days of age. Organs were dissected and homogenized in buffer containing 100 mM Tris (pH 7.8) and 1 mM dithiothreitol. Protein homogenates were centrifuged, and supernatants were assayed for luciferase activity as described (26).
Nuclear Extracts and DNA Binding Assays-Nuclear extracts from cultured cells were prepared by the method of Dignam et al. (28). All buffers contained 0.5 mM dithiothreitol, 0.5 mM phenylmethylsulfonyl fluoride, and 10 g/ml leupeptin and pepstatin. Nuclear extracts from isolated tissues of newborn mice were prepared according to the method of Deryckere and Gannon (29). For electrophoretic mobility shift assays (EMSA) double-stranded oligonucleotides were labeled with [␥-32 P]ATP and T4 polynucleotide kinase, filled in with the Klenow fragment of DNA polymerase I, and purified on a polyacrylamide gel. Approximately 5 fmol of the probe was incubated with 1-5 g of nuclear extracts or 1 l of purified Osf1 in 10 l of buffer A (50 mM Tris, pH 7.5, 20% glycerol, 100 mM NaCl, 2 mM EDTA, 0.1% Nonidet P-40, 1 mM dithiothreitol) for 10 min at room temperature. 2 g of poly(dI-dC) and 250 fmol of single-stranded oligonucleotides were included as nonspecific competitors. Samples were separated by electrophoresis on a 5% polyacrylamide gel in 0.25 ϫ TBE at 160 V for 90 min. The gels were dried and autoradiographed.
Purification of Osf1-The analytical purification of Osf1 was performed from 30 ml of nuclear extract of ROS17/2.8 or C1 cells. The purification was monitored at every stage by EMSA using the ⌬OSE1 oligonucleotide as a probe. One Osf1 binding unit was defined as the amount of protein required to retard 0.25 fmol of the probe. Total protein concentrations were measured by the Bio-Rad protein assay. The nuclear extract was brought to 25% ammonium sulfate saturation at 4°C over a 30-min period. Precipitated proteins were spun down at 15,000 ϫ g for 15 min, dissolved in 5 ml of buffer A, and dialyzed against the same buffer for 3 h with two buffer changes. Dialyzed proteins were applied onto a 5-ml phosphocellulose column (Whatman P11). After washing the column with 5 volumes of buffer A containing 0.2 M NaCl, bound proteins were eluted in 1-ml fractions with 0.5 M NaCl in buffer A. Fractions containing Osf1 binding activity were pooled, dialyzed as described above, and applied onto a MonoQ column that was run using a fast pressure liquid chromatography system (Amersham Pharmacia Biotech). After a washing step with 0.2 M NaCl the proteins were eluted by a 0.2-1 M NaCl gradient in buffer A over 10 min. Osf1 binding activity was eluted at a NaCl concentration between 0.35 and 0.4 M. The corresponding fractions were pooled, dialyzed against buffer A, and incubated for 10 min with 14 g/ml poly(dI-dC), 200 fmol/l singlestranded OSE1 oligonucleotide, and 100 fmol/l double-stranded mutant ⌬OSE1 oligonucleotide (⌬OSE1-mut4). This mixture was applied to a 1-ml DNA affinity column containing multimerized ⌬OSE1 oligonucleotide that was prepared as described by Kadonaga and Tjian (30). After a washing step at 0.2 M NaCl, the Osf1 activity was eluted with buffer A containing 0.5 M NaCl. To verify the purity of the eluted Osf1 the final EMSA was performed with both wild-type and mutant ⌬OSE1 oligonucleotide in the absence of nonspecific competitor DNA.
SDS-PAGE and Renaturation-SDS-polyacrylamide gel electrophoresis was carried out as described by Laemmli (31). Proteins were visualized by silver staining according the method of Heukeshoven and Dernick (32). 5-mm gel slices were cut out from an adjacent lane that was run on the same gel, but not stained. These gel slices were crushed with a pestle and soaked in 500 l of elution buffer (50 mM Tris, pH 7.5, 0.1 mM EDTA, 0.1% SDS, 5 mM dithiothreitol, 150 mM NaCl, 0.5 mM phenylmethylsulfonyl fluoride, 100 g/ml bovine serum albumin) for 3 h at room temperature. Eluted proteins were precipitated with 80% acetone, dissolved in 50 l of buffer A containing 6 M guanidine-HCl, and denatured for 30 min. After removal of the guanidine by a 10desalting gel spin column (Bio-Rad) the proteins were allowed to renature for 2 h at room temperature. 3 l of renaturated proteins were assayed for Osf1 binding activity by EMSA using the wild-type and mutant ⌬OSE1 oligonucleotides as probes.

OSE1 and OSE2 Are Equally Important for OG2 Promoter
Activity in Cell Culture-To determine the relative importance of OSE1 compared with OSE2 for OG2 expression we introduced small substitution mutations into either one of these two elements in the context of a p147-luc reporter gene construct that contains the first 147 bp of the mouse OG2 promoter fused to the luciferase gene (luc) (12). These mutations abolish the DNA binding activity of Osf1 and Cbfa1 to their respective binding sites ( Fig. 8B and Ref. 12). Using the wild-type and mutated p147-luc constructs we performed DNA transfection experiments in ROS17/2.8 osteoblasts (Fig. 1A). We chose ROS17/2.8 cells to perform these experiments because these cells were used in the initial characterization of OSE1 and OSE2 (12).
A 2-bp mutation introduced into OSE2 decreased the activity of p147-luc 85% confirming the critical role of OSE2 for OG2 expression (Fig. 1A). Likewise, a 2-bp mutation in OSE1 led to a 76% decrease of p147-luc activity (Fig. 1A). These results indicate that OSE1 is virtually as important as OSE2 for OG2 expression. We also introduced the same 2-bp mutation into OSE1 in the context of a longer 1316-bp OG2 promoter fragment and performed DNA transfection experiments in ROS17/ 2.8 cells with the wild-type and mutated p1316-luc constructs (Fig. 1B). The OSE1 mutation led to a 51% decrease of p1316luc activity indicating that even in this larger OG2 promoter fragment the presence of OSE1 is required for maximal activity in cell culture.
OSE1 Is Critical for OG2 Promoter Activity in Vivo-As a further demonstration of the functional importance of OSE1 for osteocalcin expression, we generated transgenic mice carrying either the wild-type p1316-luc or the mutated p1316-luc chimeric genes. Three founder animals carrying the wild-type and two founder animals carrying the mutant chimeric gene were obtained. Transgenic mice of the F1 generation from these founder animals were analyzed for luciferase expression in various organs at 10 days of age (Fig. 2). In all mice containing the wild-type p1316-luc transgene the relative luciferase activities in bone were between 3000 and 5000 per 100 g of total protein ( Fig. 2A). In these mice luciferase activities in other tissues were less than 1% compared with the values we measured in bone (2-30/100 g of protein). This bone-specific luciferase expression was also observed in all mice carrying the mutated p1316-luc transgene (Fig. 2B). However, the luciferase activities in long bones of these latter mice were in average 75% decreased (700 -1300/100 g of protein) compared with what was observed in long bones obtained from mice carrying the wild-type transgene (Fig. 2B). Taken together, these data demonstrate that OSE1 is functionally as important as OSE2 for OG2 expression, both in cell culture and in vivo. The functional importance of OSE1 was an incentive to better characterize this element and the factor binding to it, thereafter termed Osf1.
Osf1 Binds to an 8-bp Core OSE1 Sequence-The OSE1 oligonucleotide that was used for EMSA in our previous study covered a relatively large region of the OG2 promoter consisting of 35 bp (12). Originally, we did not find any motifs within this region sharing similarities to binding sites of known transcription factors.
To abolish Osf1 binding to DNA we originally used a large mutation affecting 6 bp (Table I). By definition this mutation of a large number of base pairs did not allow us to determine precisely the nucleotides critical for protein binding. We reasoned that by defining a shorter Osf1-binding site we may be in a better position to identify key nucleotides within OSE1 and to recognize binding sites of known transcription factors.
To achieve this goal we first assayed shorter OSE1 oligonucleotides for Osf1 binding activity by EMSA (data not shown). This led us to the definition of a shorter oligonucleotide spanning only 22 bp of the OG2 promoter that we termed ⌬OSE1 (Table I). To demonstrate that the same nuclear activity is binding to the OSE1 and the ⌬OSE1 oligonucleotide, direct binding and DNA competition experiments were performed by EMSA using nuclear extracts of ROS17/2.8 osteoblasts as a source of protein (Fig. 3). We previously described the presence of three protein-DNA complexes formed upon incubation of ROS17/2.8 nuclear extracts with the OSE1 oligonucleotide in EMSA (12). Osf1, defined by its osteoblast-specific presence and its heat lability, is part of the complex of intermediate mobility. Using the ⌬OSE1 oligonucleotide as a probe in EMSA the same three complexes were detected (Fig. 3A, lane 1). The complex of intermediate mobility was specifically competed by the wild-type, but not by the mutant ⌬OSE1 and OSE1 oligonucleotides (Fig. 3A, lanes 2-5). Likewise, when using the OSE1 oligonucleotide as a probe in EMSA the Osf1-DNA complex was specifically competed by the wild-type, but not by the mutant ⌬OSE1 and OSE1 oligonucleotides (Fig. 3B, lanes 2-5). These results suggest that the same nuclear protein binds to both the OSE1 and ⌬OSE1 oligonucleotide. This was further demonstrated by the osteoblast-specific nature of the factor binding to ⌬OSE1 (Fig. 5A). Thus, we used the ⌬OSE1 oligonucleotide as a probe for the DNA binding experiments described below.
To identify the critical base pairs within the OSE1 sequence that are required for Osf1 binding to DNA we mutagenized individually 8 bp within the ⌬OSE1 oligonucleotide (Fig. 4A). The use of these various ⌬OSE1 oligonucleotides in EMSA allowed us to define an 8-bp core OSE1 sequence: 5Ј-TTA-CATCA-3Ј (Fig. 4B).
OSE1 Appears to be a Unique Sequence-Analysis of the core OSE1 sequence failed to show a strong homology to any known   Fig. 4A.
transcription factor-binding sites. However, a closer inspection of this 8-bp sequence showed a weak similarity with the cAMPresponsive element (CRE) whose sequence is 5Ј-TGACGTCA-3Ј (33). This led us to explore the possibility that Osf1 could be a member of the CREB (CRE-binding proteins) family of tran-scription factors.
The guanosine present in position 5 of the CRE is known to be critical for the binding of CREB proteins to DNA (34). Importantly, when we used a mutant ⌬OSE1 oligonucleotide carrying an A to G mutation in position 5 of the core OSE1 sequence (⌬OSE1-mut5) as a probe in EMSA with ROS17/2.8 nuclear extracts, the Osf1-DNA complex was absent and replaced by a complex of slower mobility (Fig. 4, lane 6). To determine wether the factor binding to the ⌬OSE1-mut5 oligonucleotide was osteoblast-specific, EMSA with nuclear extracts of various tissues and of primary osteoblasts was performed (Fig. 5). While Osf1, the factor binding to the ⌬OSE1 oligonucleotide, was present only in nuclear extracts of non-mineralized primary osteoblasts (Fig. 5A, lane 8), the factor binding to the ⌬OSE1-mut5 oligonucleotide was present in nuclear extracts of every tissue tested (Fig. 5B). Moreover, this latter factor was heat-stable (Fig. 5B, lanes 9 -13) as described for CREB proteins (35), unlike Osf1, whose binding to DNA is abolished upon heating ( Fig. 7 and Ref. 12). In summary, the ubiquitous nature and the heat stability of the factor binding to the ⌬OSE1-mut5 oligonucleotide clearly distinguishes it from Osf1. Thus, Osf1 is most likely not a member of the CREB family of transcription factors, and the core OSE1 sequence appears to be unique.
The Core OSE1 Sequence Is Present in the Mouse Cbfa1 Promoter-The definition of the core OSE1 sequence allowed us  to search for OSE1 sites in other genes that are expressed in osteoblasts. OSE1 is present at the same location in the promoter of OG1, the other mouse osteocalcin gene (36). We were not able to locate additional copies of OSE1 in more distal promoter regions of the mouse osteocalcin genes. When we searched the 5Ј-region of the mouse gene encoding the osteoblast-specific transcription factor Cbfa1 (kindly provided to us by Dr. P. Ducy), we found one OSE1 site located 3700 bp upstream of the transcriptional start site. This site is homologous, but not identical to the OSE1 site present in the OG2 promoter. In particular the surrounding sequences are totally different. Nevertheless, given the restricted expression of Cbfa1, we analyzed whether this sequence, termed OSE1 Cbfa1 , could act as an osteoblast-specific cis-acting element.
Six copies of the wild-type or mutant OSE1 Cbfa1 oligonucleotide were cloned upstream of a minimal OG2 promoter fragment in a luciferase expression vector (p34-luc). This minimal promoter fragment lacks OSE1 and OSE2 and has virtually no transcriptional activity in DNA transfection experiments (12). The multimerized wild-type OSE1 Cbfa1 oligonucleotide conferred osteoblast-specific activity to this promoter fragment in DNA transfection experiments as the luciferase expression was at least 30-fold higher in ROS17/2.8 cells compared with F9 teratocarcinoma cells and C2C12 myoblasts, two cell lines that do not express Cbfa1 (Fig. 6A). In contrast, multimerized mutant OSE1 Cbfa1 oligonucleotide failed to increase the activity of the minimum OG2 promoter.
Additionally, DNA binding and competition experiments by EMSA were performed to determine if OSE1 Cbfa1 serves as a binding site for Osf1. When using nuclear extracts of different tissues and non-mineralized primary osteoblasts as a source of protein, and the OSE1 Cbfa1 oligonucleotide as a probe, an osteoblast-specific complex migrating at the same location as the Osf1-⌬OSE1 complex was observed (Fig. 6B). This protein-DNA complex was competed by the wild-type ⌬OSE1 and OSE1 Cbfa1 oligonucleotides (Fig. 6B, lanes 2 and 3). When using the ⌬OSE1 oligonucleotide as a probe in EMSA the Osf1-DNA complex was also competed by both the ⌬OSE1 and the OSE1 Cbfa1 oligonucleotide (Fig. 6C, lanes 2 and 3). Taken together, these functional and biochemical lines of evidence indicate that the Cbfa1 promoter contains at least one OSE1-like site whose sequence is partially identical to the one present in the OG2 promoter and that can act as an osteoblast-specific element.
Osf1 Can Be Purified to Homogeneity-The lack of overt homology of the core OSE1 sequence to any known transcription factor-binding sites excluded a polymerase chain reaction based or low-stringency hybridization approach to cloning a cDNA encoding Osf1. Therefore we decided to purify Osf1 from osteoblast nuclear extracts in order to obtain amino acid sequence information. In a survey of various osteoblastic and mesenchymal cell lines we found one mesenchymal, pluripotent cell line, C1 (37), expressing high levels of Osf1. When performing EMSA with nuclear extracts of C1 cells, we observed only one complex, unlike what was seen with ROS17/2.8 nuclear extracts (Fig. 7, lanes 1 and 2). As it was the case for the Osf1 activity present in ROS17/2.8 nuclear extracts, this complex was heat-sensitive (Fig. 7, lane 3) and was not observed when using as probes any of the eight ⌬OSE1 oligonucleotides harboring single base pair mutations that were used to define the core OSE1 sequence (Fig. 7, lanes 4 -11) indicating that this protein-DNA complex contains Osf1.
Given the higher level of Osf1 expression and the existence of a single protein-DNA complex in EMSA, C1 nuclear extracts were chosen as a source of protein for our purification approach. However, we also confirmed the results obtained with C1 cells by purifying Osf1 with the same strategy from ROS17/ 2.8 nuclear extracts. Fig. 8A summarizes the strategy used to purify Osf1 on an analytical scale. The presence and integrity of Osf1 was monitored at each step by EMSA using the ⌬OSE1 oligonucleotide as a probe. Enrichment of Osf1 in the purified fractions was estimated by comparing the Osf1 binding activity to the total amount of protein.
Nuclear extracts were first subjected to a 25% ammonium sulfate precipitation. Osf1 activity was only detectable in the precipitated fraction. The precipitate was dissolved in buffer A containing 0.1 M NaCl, dialyzed against the same buffer, and applied to a phosphocellulose cation exchange resin where Osf1 activity was eluted at 0.5 M NaCl. These two steps led to a 20-fold enrichment of Osf1 binding activity (Fig. 8A). Following dialysis against buffer A containing 0.1 M NaCl, the Osf1 containing fractions were applied onto a Mono Q anion exchange column. Osf1 was eluted from this column at a salt concentration between 0.35 and 0.4 M NaCl. After these three initial FIG. 6. Presence of a functional OSE1 element in the mouse Cbfa1 promoter. A, DNA transfection experiments in ROS17/2.8 osteoblasts, F9 teratocarcinoma cells, and C2C12 myoblasts were performed with reporter gene constructs containing six copies of either the wild-type or the mutant OSE1 Cbfa1 oligonucleotide cloned upstream of a minimal OG2 promoter-luciferase chimeric gene (p34-luc). Data represent the ratios of luciferase to ␤-galactosidase activities and are the average of five independent experiments. Error bars represent the standard deviation. B, EMSA was performed with nuclear extracts of non-mineralized primary osteoblasts (lanes 1-3) and of various tissues (lanes 4 -9). Competition experiments were performed using a 50-fold molar excess of either the OSE1 Cbfa1 (lane 2) or the ⌬OSE1 oligonucleotide (lane 3). Nuclear extracts used were from non-mineralized primary osteoblasts (nmi ob), mineralized primary osteoblasts (mi ob), skin (sk), kidney (ki), brain (br), muscle (mu), and liver (li). The Osf1-DNA complex is indicated by the arrow. C, EMSA was performed with nuclear extracts of non-mineralized primary osteoblasts using the ⌬OSE1 oligonucleotide as a probe. Competition experiments were performed using a 50-fold molar excess of either the OSE1 Cbfa1 (lane 2) or the ⌬OSE1 oligonucleotide (lane 3). The Osf1-DNA complex is indicated by the arrow.
purification steps we estimated that Osf1 was enriched more than 75-fold compared with nuclear extracts (Fig. 8A). The last step of the purification scheme was a DNA affinity column using the multimerized ⌬OSE1 oligonucleotide coupled to a Sepharose matrix (30). The fractions containing Osf1 were dialyzed against buffer A containing 0.1 M NaCl and applied onto the DNA affinity resin. Nonspecific interactions of other proteins with this resin were significantly reduced by preincubating the protein sample with poly(dI-dC), single-stranded OSE1 oligonucleotide, and a double-stranded mutant oligonucleotide (⌬OSE1-mut4). Specific Osf1 binding activity was eluted from this column at a concentration of 0.5 M NaCl.
To demonstrate that the purified protein using this procedure was Osf1 we performed EMSA using the wild-type and mutant ⌬OSE1 and OSE1 Cbfa1 oligonucleotides as probes in the absence of nonspecific competitor DNA (Fig. 8B). The purified protein bound specifically to the wild-type ⌬OSE1 and OSE1 Cbfa1 oligonucleotides, but not to their mutant counterparts (Fig. 8B). Knowing the amount of radiolabeled oligonucleotide added to the binding reaction, we estimated the final concentration of the purified Osf1. The yield of Osf1 starting from 30 ml of nuclear extract was approximately 150 fmol. Thus, the four-step purification procedure led at least to a 5000-fold purification of Osf1 (Fig. 8A). Using the same procedure we were also able to purify the specific Osf1 binding activity from ROS 17/2.8 cells. Osf1 purified from ROS 17/2.8 nuclear extracts behaves like Osf1 purified from C1 cells in EMSA (Fig. 8B, lanes 5 and 6).
The Apparent Molecular Mass of Osf1 is 40 kDa-When we performed an SDS-PAGE and silver staining at this stage of the purification, we observed that the eluate of the DNA affinity resin contained two distinct polypeptides migrating at 40 and 45 kDa, respectively (Fig. 9A). To determine which of these polypeptides contains the Osf1 binding activity, a renaturation experiment was performed. Aliquots of the DNA affinity eluate fraction were subjected to SDS-PAGE, the proteins were eluted from 5-mm gel slices, precipitated by acetone, and subjected to denaturation with 6 M guanidine. After removal of the guanidine the proteins were allowed to renature and finally assayed for Osf1 binding activity by EMSA using the wild-type or mutant ⌬OSE1 oligonucleotide as a probe (Fig. 9B). Specific Osf1 binding activity was recovered from the gel slice containing the polypeptide migrating at 40 kDa (Fig. 9B, gel slice B). This binding activity was only observed when using the wild-type ⌬OSE1 oligonucleotide as a probe, but was absent when using the mutant ⌬OSE1 oligonucleotide as a probe (Fig. 9B, lanes 5  and 6). Additionally, the protein-DNA complex migrated at the same location as the complex observed with purified Osf1 that was not subjected to an SDS-PAGE (Fig. 9B, lane 1). In contrast, a DNA binding activity that was recovered from the gel slice containing the polypeptide migrating at 45 kDa (gel slice A) differed from the Osf1 binding activity as the observed protein-DNA complex had a slower mobility in EMSA (Fig. 9B,  lane 3). Other fractions of the SDS gel did not contain detectable binding activities.

DISCUSSION
Taken together, the results of this study indicate that OSE1 is a cis-acting element as important as OSE2 in controlling osteoblast-specific expression of osteocalcin in vivo and that it may be involved in regulating Cbfa1 expression in cells of the osteoblast lineage. We also show that the core OSE1 sequence has no overt homologies to binding sites of known transcription factors. Last, this study provides the first biochemical characterization of Osf1 and thereby expands the spectrum of regulatory elements and factors involved in osteoblast-specific gene expression.
To date only one osteoblast-specific transcription factor, Cbfa1, has been cloned and functionally analyzed using molecular and genetic means (13)(14)(15)(16)(17)(18). All the evidence accumulated recently demonstrate that Cbfa1 acts as a differentiation factor in the osteoblast lineage by binding to the OSE2 sites that are present in the promoters of most genes expressed in osteoblasts (13). Yet, other osteoblast-specific cis-acting elements have been identified in the genes encoding ␣1(I) collagen, BSP, insulin-like growth factor 1, and osteocalcin (12, 19 -25) suggesting the existence of additional osteoblast-specific transcription factors. OSE1, the element that we analyzed in this study, is of particular interest as it has been shown to bind a factor termed Osf1 present only in nuclear extracts of non-mineralized osteoblasts (12).
We have shown previously that OSE1 could confer osteoblast-specific activity to a heterologous promoter. However, these studies did not address the respective importance of OSE1 and OSE2 in the control of OG2 expression. This is an issue of critical importance that needed to be addressed before embarking in any further characterization of Osf1. The mutagenesis of OSE1 and OSE2 presented here demonstrates that these cis-acting elements are equally important in controlling OG2 expression. Surprisingly OSE1 appears to be more important for OG2 expression in transgenic mice than in cell culture experiments. This may reflect the genetic differences between osteoblasts in vivo and ROS17/2.8 cells.
OSE1 is also present in the promoter of other osteocalcin genes and of another gene specific for the osteoblast lineage, Cbfa1. This suggests that Osf1 may be one regulator of Cbfa1 expression in vivo and is consistent with the presence of Osf1 in nuclear extracts of non-mineralized primary osteoblasts, but not of fully differentiated osteoblasts. In contrast, OSE1 is not present in the promoter of the human osteocalcin gene (38). Several possibilities could explain this finding. We know from previous studies that the regulation of osteocalcin expression by steroid hormones such as 1,25-dihydroxyvitamin D 3 differs in mouse and human, indicating a lack of conservation of regulatory mechanisms between these two species (39,40). We also know that the serum level of osteocalcin in humans is significantly lower than in mice (41), suggesting that the human gene is expressed at a lower level. This could well be explained by the absence of an important regulatory element such as OSE1 in the promoter of human osteocalcin. Regardless of the reasons explaining this discrepancy the functional importance of OSE1 for osteocalcin expression in mouse and the exquisite cell distribution of Osf1 are incentives to study this factor further.
To determine wether Osf1 belongs to a known family of transcription factors we defined the 8-bp core OSE1 sequence through an extensive mutagenesis approach. This sequence, 5Ј-TTACATCA-3Ј, shows no strong homology to any binding sites of known transcription factor families, but a weak similarity to the cAMP-responsive element, the binding site of CREB proteins (42,43). However, several important biochemical characteristics distinguish Osf1 from CREB proteins. First, the CREB consensus-binding site is palindromic whereas the core OSE1 sequence is not. Second, Osf1 is heat-sensitive whereas CREB proteins are heat-stable (34). Third, and more importantly, an A to G mutation in the core OSE1 sequence that generates a better CRE-like sequence abolishes binding of Osf1 and results in the binding of an ubiquitously expressed heat-stable protein, most likely a CREB family member. The exact molecular nature of Osf1 will be defined when a cDNA becomes available.
The fact that the core OSE1 sequence shows no overt homology to known transcription factor-binding sites led us to use a classical protein purification approach to obtain amino acid sequence information. This approach has been successfully used as an initial step toward the cloning of several transcription factors (44,45). Moreover, the significant progress in protein sequencing by nanospray mass spectrometry enables researchers to sequence unknown proteins at femtomole quantities (46). This is a critical issue as osteoblasts are adherent cells, thus precluding us from obtaining large amounts of pure protein.
Initially, Osf1 was identified in ROS17/2.8 osteoblasts (12). However, nuclear extracts of these cells also contain other factors binding to the ⌬OSE1 oligonucleotide in EMSA. Although we were able to purify Osf1 from ROS17/2.8 nuclear extracts, we used C1, another osteocalcin-expressing cell line (37), as a source of protein for this purification effort because Osf1 is the only activity present in C1 nuclear extracts binding to the ⌬OSE1 oligonucleotide. Moreover, Osf1 is more abundant in C1 than in ROS17/2.8 nuclear extracts. After a fourstep purification, Osf1 migrated as a single band at 40 kDa in an SDS-PAGE and formed a single complex with the ⌬OSE1 oligonucleotide in EMSA, even in the absence of nonspecific competitor DNA.
One important question that will need to be addressed in the future is where Osf1 resides in the genetic cascade controlling osteoblast-specific gene expression and, possibly, osteoblast differentiation. The evidence we have gathered to date can support two opposite views equally well. The presence of OSE1 in the osteocalcin promoters and our failure to detect them in the promoter of other extracellular matrix encoding genes (data not shown) may indicate that Osf1 acts downstream of Cbfa1 and is required solely for osteocalcin expression. Alternatively, the fact that an OSE1 site is present in the Cbfa1 promoter, along with the presence of Osf1 only in nuclear extracts of nonmineralized osteoblasts, would rather suggest that Osf1 acts upstream of Cbfa1. Cbfa1 and Osf1 could also regulate the expression of each other as it has been shown to be the case for transcription factors involved in adipogenesis and myogenesis (8,47). The distinction between these possibilities will have to  4) or B (lanes 5 and 6) as indicated, were renatured and analyzed for Osf1 binding activity by EMSA using the wild-type (lanes 1, 3, and 5) or the mutant ⌬OSE1 (lanes 2, 4, and 6) oligonucleotide as a probe. The binding activity renatured from the indicated portion of the SDS gel is compared with purified Osf1 that was not subjected to an SDS-PAGE (lanes 1 and 2). Nonspecific competitor DNA, poly(dI-dC), was excluded in this experiment.
await the molecular cloning of Osf1 and the analysis of its spatial and temporal expression pattern.