Fifteen putative transcriptional target genes regulated by the osteogenic transcription factor Runx2 were identified by cDNA microarray and differential hybridization techniques. Expression pattern and regulation of one gene, Pttg1ip, was analyzed in detail.
Introduction: The transcription factor Runx2 is a key regulator of osteoblast development and plays a role in chondrocyte maturation. The identification of transcriptional target genes of Runx2 may yield insight into how osteoblastic differentiation is achieved on a molecular level.
Materials and Methods: Using a differential hybridization technique (selective amplification through biotin and restriction-mediated enrichment [SABRE]) and cDNA microarray analysis, 15 differentially expressed genes were identified using mRNA from C3H 10T1/2 cells with constitutive and inducible overexpression of Runx2.
Results and Conclusions: Among the 15 genes identified, 4 encode the extracellular matrix proteins Ecm1, Mgp, Fbn5, and Osf-2, three represent the transcription factors Esx1, Osr1, and Sox9, whereas others were Ptn, Npdc-1, Hig1, and Tem1. The gene for Pttg1ip was upregulated in Runx2-expressing cells. Pttg1ip is widely expressed during development, but at highest levels in limbs and gonads. The Pttg1ip promoter binds Runx2 in a sequence specific manner, and Runx2 is able to transactivate the Pttg1ip promoter in MC3T3-E1 cells. Therefore, Pttg1ip is likely to be a novel direct transcriptional target gene of Runx2. In conclusion, the genes identified in this study are important candidates for mediating Runx2 induced cellular differentiation.
THE DEVELOPMENT OF the mammalian skeleton is a complex process involving coordinated patterning information as well as regulation of mesenchymal cell differentiation along osteoblast, osteoclast, and chondrocyte lineages. Whereas numerous transcription factors have been characterized that coordinate skeletal patterning (e.g., homeodomain transcription factors), only a limited number of transcription factors have been identified that regulate the development of the cellular components of the skeleton. Among these are SOX9, involved in chondrocyte maturation; fos in osteoclast development; and Runx2 and Osterix in osteoblast differentiation.(1–5) The role of Runx2 has especially been studied in detail after the finding that mutant mice deficient for the transcription factor Runx2 (formerly also called Cbfa1 or PEBP2αA) are devoid of bone. Analysis of Runx2 knockout mice revealed that they completely lack osteoblasts.(2,6) This finding and data indicating that Runx2 controls the expression of osteoblast-specific genes led to the conclusion that Runx2 is a key regulator of osteoblast development.(7) Furthermore, the transcription factor has also been implicated in the regulation of chondrocyte maturation in the growth plate.(8,9)
Runx2 is a member of the runt family of transcription factors. All three mammalian Runx proteins (Runx1, Runx2, and Runx3) share a highly conserved 128 amino acid DNA binding domain.(10) This domain was initially described in the gene product of the Drosophila pair rule gene runt.(11,12)
Mammalian runt-related proteins bind to a recognition motif in regulatory regions of several genes. The consensus sequence for this binding motif has been determined as 5′-ACCPuCPu-3′.(13–16) It was first identified in the polyoma virus enhancer.(13) Later the consensus motif could be identified in promoters of T-cell-specific genes such as TCRα, TCRβ, TCRγ, TCRδ, and CD3ε.(17–21) Furthermore, the cognate motif was found in promoters of genes for enzymes expressed in hematopoietic cells such as myeloperoxidase, neutrophil elastase,(22) and granzyme B serine protease,(23) as well as in promoters of genes for cytokines like granulocyte macrophage colony-stimulating factor (GM-CSF),(24,25) interleukin (IL)-3,(26) and colony-stimulating factor 1 (CSF-1).(27)
The identification of genes that are regulated in their transcriptional activity by Runx2 will provide insight into the way by which this transcription factor guides cellular differentiation toward a mature osteoblast phenotype. Some bone-related target genes of Runx2 have been described in recent years. These genes belong mainly to the group of extracellular matrix proteins such as osteocalcin, bone sialoprotein, osteopontin, α1(I) collagen, and ameloblastin.(7,28,29) Furthermore, the genes for galectin-3 and collagenase 3, as well as for growth factors and their receptors such as osteoprotegerin and transforming growth factor (TGF)-β receptor type I, have been reported to be regulated by Runx2.(30–33)
In this report, we describe the identification by a systematic approach of genes differentially expressed in cells overexpressing Runx2. We performed a differential screening technique, selective amplification through biotin and restriction-mediated enrichment (SABRE), comparing C3H10T1/2 mouse embryonic fibroblasts constitutively overexpressing Runx2 to wildtype cells, identifying a set of 10 differentially expressed genes. Furthermore, by using a cDNA microarray technique, we were able to identify and confirm another five genes differentially regulated in C3H10T1/2 cells with inducible Runx2 expression. Nine of these genes could be confirmed as being regulated by Runx2 in both cellular models. Finally we analyzed expression levels of these genes in several mesenchymal cell lines.
One of these novel Runx2 target genes, the yet poorly studied gene for pituitary tumor transforming 1 protein-interacting protein (Pttg1ip) was subject of further experiments. Expression analysis in mouse embryonic tissues revealed a wide tissue distribution pattern with high expression levels in limbs and gonads. Pttg1ip expression persists throughout organogenesis of murine embryos from embryonic day E9.5 to E16.5, with levels increasing slightly over time. Furthermore, we showed expression in osteoblast-like MC3T3 cells as well as in primary osteoblasts. In addition, Runx2 binds to sites in the 5′ flanking region of murine Pttg1ip gene and is able to transactivate a reporter gene driven by the Pttg1ip promoter.
MATERIALS AND METHODS
C2C12 pre-myoblasts were obtained from the American Type Culture Collection (Manassas, VA, USA). Cells were grown in DMEM, supplemented with 10% fetal calf serum (FCS) and 4 mM L-glutamine (all from Life Technologies, Karlsruhe, Germany). As outlined in the supplier's manual, cells were subcultured before reaching confluence.
Murine embryonic calvaria MC3T3-E1 cells were cultured in α-MEM, supplemented with 10% FCS and 2 mM L-glutamine (Life Technologies) as previously described.(33) Murine ATDC5 pre-chondrocytes were a gift from Chisa Shukunami (Department of Molecular Interaction and Tissue Engineering, Kyoto University), kindly provided by Gerd Scherer (Institute of Human Genetics, University of Freiburg). These cells, NIH3T3 fibroblasts (gift from Mateusz Kolanczyk, University of Freiburg Medical Center, Department of Hematology/Oncology) and murine embryonic fibroblastic C3H10T1/2 cells obtained from American Type Culture Collection, were maintained in DMEM supplemented with 10% FCS and 2 mM L-glutamine. Subculturing of C3H10T1/2 cells was performed before reaching confluency.
Stably transfected cell lines C3H10T1/2-Runx2 and C3H10T1/2 with inducible Runx2 expression were established and cultured as described before.(33) Osteogenic differentiation of MC3T3-E1 was performed and monitored as reported previously.(33)
Primary mouse calvarial cells were isolated from 6-day-old mice according to a protocol from Bächner et al.(34) Calvariae were prepared, cleaned from residual noncalvarial tissue, and digested for a total of five rounds of shaking at 37°C for 20 minutes in a collagenase solution (1 mg/ml collagenase, 18.4 mg/ml sorbitol, 1 mg/ml chondroitin sulfate in DMEM). Cells from the first two rounds of digestion, containing mainly fibroblasts, were discarded. Cells from rounds 3-5 were collected, washed in PBS, and plated in DMEM with 10% FCS and 2 mM L-glutamine at a density of 6000 cells/cm2. Medium was changed the next day, and cells were cultured under standard conditions until they reached confluency.
RT-PCR and generation of cDNA libraries
Total RNA was isolated by acid guanidinium thiocyanate-phenol-chloroform extraction method.(35) For RT-PCR, 4 μg of total RNA were used as template for standard cDNA synthesis using Superscript II reverse transcriptase (Life Technologies). In parallel approaches, reactions with and without reverse transcriptase were performed to exclude genomic DNA contamination. PCR with gene-specific primers was carried out in a standard approach using TaqPCR Core Kit (Qiagen, Hilden, Germany). cDNA libraries were generated from poly A+ RNA extracted from wildtype C3H10T1/2 and C3H10T1/2-Runx2 cells using Oligotex Direct mRNA Minikit (Qiagen). First-strand cDNA synthesis was performed by standard methods using Superscript II reverse transcriptase and 2 μg of poly A+ RNA as template. After second-strand synthesis and purification by phenol extraction, 2 μg ds cDNA was cut with SauIII AI restriction enzyme (New England Biolabs, Frankfurt, Germany).
Primers and probes
Hybridization probes for GAPDH, Runx2, and osteocalcin were generated as reported previously.(33) T3 and T7 primers were used to amplify inserts from candidate clones from the SABRE experiments. These PCR products served as hybridization probes for the respective clone. Some of the cDNAs identified by microarray technique to be differentially expressed in inducible clone 18/17 were ordered as IMAGE clones from Incyte (San Diego, CA, USA). Inserts were amplified by PCR using T7 (5′-TAATACGACTCACTATAGGG-3′) and T3 (5′-ATTAACCCTCACTAAAGGGA-3′) primers to serve as hybridization probes. IMAGE cDNA clones used in this study include those for pleiotrophin (Clone 478168) and osf-2 (Clone 403071), All other cDNAs were amplified by RT-PCR and cloned into pCR II-TOPO. Gene specific primers used for this study were as follows; osr1-f, 5′-TTTCCGGAGGCAAGACCAC-3′; osr1-r, 5′-CCGCCCAGGCTGTAGGAGT-3′; Slpi-f, 5′-AGTCCTGCGGCCTTTTACCT-3′; Slpi-r, 5′-ATGCGTTTATTTATTTGCTCTCCA-3′; Hig1-f, 5′-CAAGAAATCACAATGTCAACCAAC-3′; Hig1-r, 5′-AGACGTCTAACTAAAGCAAGCACT-3′; Pttg1ip-f, 5′-CCCCGCCGACGACGCACTCACC-3′; Pttg1ip-r, 5′-TGGCCCGCTCATCGCTCTTGTCTGG-3′.
Probes were generated by PCR, using the respective clone as template and vector specific primers T7 and M13rev (5′-CAGGAAACAGCTATGAC-3′). Sequences of all hybridization probes were confirmed by automated sequencing (Taq DyeDeoxy Sequencing Sytem; ABI, Weiterstadt, Germany). Pttg1ip primers were used for analytical RT-PCR but not for generation of a hybridization probe. A Pttg1ip-specific hybridization probe was directly obtained from the respective SABRE clone.
SABRE was performed as previously reported by Lavery et al.(36) In summary, SauIII AI cleaved double-stranded cDNA from wildtype C3H10T1/2 (sample 1) or C3H10T1/2-Runx2 (sample 2) cells was ligated to adaptor A (hybridization product of Oligo A1: 5′-GGTCCATCCAACC-3′ and Oligo A2: 5′-phosphate-GATCGGTTGGATGGACCGT-3′). This DNA was used as template for tester and driver PCR reactions. To generate tester DNA, each sample was amplified by PCR using adaptor-ligated cDNA as template and Oligo T (5′-biotin-CCAGGATCCAACCGATC-3′) harboring a BamHI restriction site as upstream and downstream primer. For driver DNA, Oligo D (5′-GGTCCATCCAACCGATC-3′) was used as primer. PCR programs were optimized to achieve similar length distributions in tester and driver DNAs by adjusting the number of cycles (usually to between 24 and 36 cycles) and Mg2+ concentration to 2 mM. Each cycle consisted of three segments: 1 minute 94°C; 2 minutes 52°C; 2 minutes 72°C. After purification, 333 ng tester DNA derived from sample 1 and 10 μg driver DNA from sample 2 were hybridized by phenol emulsion DNA reassociation technique (PERT).(37) After this step, only tester-tester and tester-driver hybrids—in contrast to driver-driver hybrids—should contain biotin residues. Biotinylated hybrids were bound to paramagnetic streptavidin beads (Dynal, Oslo, Norway). Because of a point mutation in Oligo D1, only tester-tester hybrids should contain a functional BamHI restriction site derived from Oligo T1. Therefore, only tester-tester hybrids are released from the beads by cleavage with restriction enzyme BamHI. Retrieved DNA was amplified with Oligo T under the same conditions as outlined above to obtain tester DNA for the next round of enrichment. Second round driver DNA for sample 1 tester DNA was Oligo D amplified DNA derived from PERT hybridization of sample 2 tester and driver DNA. The second to fourth rounds were performed the same way as the first. After four rounds, the resulting DNA was amplified in a final PCR reaction using Oligo T. Amplified DNA fragments were digested with Sau3AI purified by gel elution and cloned into pBlueskript SK+ (Stratagene). For prescreening, inserts of 480 clones were amplified using vector specific T7 and T3 primers, and PCR products were resolved in an agarose gel and subjected to Southern blotting. Blots were sequentially hybridized with enriched total cDNA from samples 1 and 2 (“reverse Northern blot”), and clones showing different intensity after the different hybridizations were analyzed further.
Expression profiling by cDNA microarray
Total RNA from C3H10T1/2 clone 17/18 cells was isolated in induced and noninduced states. Induction was performed by supplementing media with 30 nM mifepristone (dissolved in ethanol) and incubation of cells for 38 h. Noninduced cells were treated in parallel with the same volume of carrier ethanol. Hybridization and reading of a Mouse UniGEM 1 cDNA microarray chip (Incyte Genomics, Palo Alto, CA, USA) with 9596 cDNA clones was conducted by Incyte.
Expression analysis: Northern and Western Blot
Northern blotting was performed as described previously.(33) Protein gel electrophoresis and immunoblotting were performed as outlined earlier.(33) Rabbit polyclonal anti-Runx2 antibody (anti-AML3 Ab-1; Calbiochem, San Diego, CA, USA) as primary antibody and goat anti-rabbit IgG-horseradish peroxidase conjugate (Santa Cruz Biotechnology, Santa Cruz, CA, USA) were used for immunodetection of Runx2 protein. Visualization was performed by chemoluminescence using ECL system (Amersham).
Electromobility shift assays
Electromobility shift assays (EMSA) studies were performed as reported previously using in vitro translated Runx2 protein or nuclear extract from in vitro differentiated MC3T3-E1 cells.(33) Nuclear extracts were prepared as outlined in a protocol from Andrews and Faller.(38) Briefly, cells were washed in PBS and incubated for 10 minutes on ice in hypotonic buffer A (10 mM HEPES-KOH, 1.5 mM MgCl2, 10 mM KCl, 0.5 mM dithiothreitol, and 0.2 mM PMSF, pH 7.9). After brief vortexing, reactions were spun down, and pellets were resuspended in high salt buffer C (20 mM HEPES-KOH, 25% glycerol, 420 mM NaCl, 1.5 mM MgCl2, 0.2 mM EDTA, 0.5 mM dithiothreitol, and 0.2 mM PMSF, pH 7.9) and incubated for an additional 20 minutes on ice. Lysates were centrifuged for 2 minutes, and supernatants containing nuclear proteins were used for EMSA. Runx2 binding double-stranded oligonucleotide Oligo A (5′-GAACTCTGTGGTTGCG-3′), adopted from Tahirov et al.,(39) was radiolabeled and incubated with Runx2 protein or nuclear extract from in vitro differentiated MC3T3-E1 cells with or without nonlabeled double-stranded competitor oligonucleotides containing putative Runx binding sites derived from Pttg1ip 3′ flanking genomic region in binding buffer (1 μl pepstatin/20 μl reaction, and 1 μl poly [dI-dC; 1 OD260/μl] in 2% glycerol, 5 mM Tris-HCl, 0.2 mM EDTA 0.01% NP-40, 0.1 mM DTT, 17.5 mM NaCl, and 10 μg/ml BSA, pH 7.5). For oligonucleotide sequences, refer to Fig. 5A. Reactions were resolved on a 5% polyacrylamide gel in 0.5× TBE and submitted to autoradiography.
Runx2 expression vector and Pttg1ip promoter cloning
Runx2 cDNA was cloned into pCR2.1 as described earlier.(33) This cDNA was subcloned into pCMVβ (BD Clontech), replacing the lacZ gene. The resulting Runx2 expression vector was designated as pCMV-Runx2.
The murine Pttg1ip promoter from −1648 to +109 nt relative to the predicted transcription start site was cloned into pBlue-TOPO (Invitrogen) 5′ of the lacZ gene. Taq polymerase and PCR reagents were from Qiagen. Primers were designed based on the genomic sequence of the murine Pttg1ip gene obtained from Genbank (Genbank accession no. gi 20862904). The following oligonucleotides were used as primers: 5′-TTTCCCTCACCCCCACAACCAT-3′ (sense) and 5′-GGGGACCTCAGGCGACTCAAC-3′ (antisense). The 1757-bp PCR product was introduced into pBlue-TOPO (Invitrogen) by TOPO-TA cloning according to the manufacturer's instructions. The construct containing the Pttg1ip promoter in the same orientation as the lacZ gene of the vector is referred to as pPttg1ipProm-f. The vector with an inverted Pttg1ip promoter fragment was designated as pPttg1ipProm-r.
Promoter reporter assays
For reporter assays, MC3T3-E1 or ROS17/2.8 cells were transfected using Fugene6 transfection reagent (Roche). Cells were plated in 6-well format at a density of 1.5 × 105 cells per well. The next day, cells were transfected with 18 μl Fugene6 and 4 μg DNA per well according to the manufacturer's protocol. Cells were co-transfected with the indicated promoter-reporter construct, Renilla luciferase control vector phRL-SV40 (Promega), and pCMV-Runx2 or pBluescript II SK+ (Stratagene Europe, Amsterdam, The Netherlands) in equal amounts. Luciferase activity driven from phRL-SV40 provided a means of normalizing β-galactosidase activity from pPttgipProm with respect to transfection efficiency and total protein content of each sample analyzed. Each transfection was performed in triplicate. Twenty hours after transfection, cells were washed three times in PBS and lysed in 250 μl lysis buffer (Roche) per well at room temperature for 30 minutes. Luciferase activity was determined using the dual luciferase assay system from Promega according to the manufacturer's protocols. Roche β-galactosidase chemiluminescence assay system was used for detection of β-galactosidase activity. Nonspecific luminescence was detected by performing β-galactosidase and luciferase assays, respectively, with lysis buffer instead of sample. Values from nonspecific control reactions were subtracted from those of the samples. Relative β-galactosidase activity was calculated by dividing the luminescence value of the β-galactosidase assay by that of the luciferase assay. Arithmetic mean and SD of the triplicates were calculated.
Table Table 1.. Putative Runx2 Target Genes
Identification of potential Runx2 target genes by SABRE
We reasoned that to isolate Runx2-dependent genes, we had to chose a cell line that represents the lineage and differentiation stage of the cell in which Runx2 expression is first detected in embryonic development. Forced overexpression of Runx2 in such a cell line should allow to recapitulate—at least partially—its effects in embryogenesis. C3H10T1/2 is a mesenchymal precursor cell line of murine embryonic origin. Its pluripotency, that is, its ability to differentiate toward different mesenchymal lineages like adipocytes, myoblasts, chondrocytes, and osteoblasts, encouraged us to use this cell line to study Runx2-mediated target gene activation.(40–42) The generation of a C3H10T1/2 clone constitutively overexpressing Runx2 (C3H10T1/2-Runx2) was described previously.(33) Figure 1 shows that only transfected C3H10T1/2 cells express Runx2 to a degree detectable by Western blotting. Although these cells constitutively overexpress Runx2, they do not express alkaline phosphatase at a level detectable by histochemical staining, indicating that these cells are not in a more osteoblast-like differentiation state than wildtype cells (Fig. 1). To identify transcriptional targets of Runx2, we compared C3H10T1/2-Runx2 cells and wildtype C3H10T1/2 cells applying a cDNA subtraction method (SABRE). The enrichment of differentially expressed cDNAs was performed in both directions, that is, cDNA species more abundant in Runx2 overexpressing cells (forward reaction) were identified in a parallel approach to those occurring more frequently in wildtype cells (reverse reaction). After four rounds of enrichment for differentially expressed species, cDNAs were cloned into pBlueskript. A total of 480 clones for each reaction were analyzed in a primary screen employing “reverse Northern blotting” as described in the Materials and Methods section. This screen revealed a total number of 167 candidate clones for the forward reaction and 105 candidates for the reverse reaction (data not shown). The inserts of candidate clones were sequenced, and 61 different potential candidates from forward and 31 from reverse reactions were used as hybridization probes for Northern blot analysis of RNA from wildtype and Runx2 overexpressing cells. Among those, 31 did not yield a detectable signal in Northern analysis, and 51 clones showed no difference in expression levels between wildtype and Runx2 overexpressing cells. However, 10 clones showed prominent differences in expression levels depending on Runx2 overexpression. Four of these were upregulated and six were downregulated in Runx2-overexpressing cells compared with wildtype cells (Table 1; Fig. 2).
Identification of potential Runx2 target genes by cDNA microarray
C3H10T1/2 cells have the potential to differentiate during culture. Random differentiation events might therefore contribute to differential gene expression when subclones of the cell line are cultured separately as the wildtype and the Runx2 overexpressing clones used in our SABRE experiments. To overcome this problem, we established an inducible Runx2 expression system in C3H10T1/2 cells as reported previously.(33) These cells were used in a second set of experiments employing cDNA microarrays. We used one clone (clone 18/17) to obtain mRNA from noninduced and mifepristone-induced cells overexpressing Runx2 in a microarray experiment. The microarray analysis was performed by Incyte on a mouse UniGEM1 microarray. The data we obtained included a large number of genes upregulated after induction of Runx2 expression. The respective clones were obtained from Incyte and sequenced. Inserts amplified by PCR were used as probes in Northern blot. However, a large number of these genes turned out to contain repetitive sequences that had produced apparent expression differences on the microarray. Accordingly, differential expression of these cDNA species could not be confirmed by Northern blotting. For five genes, however, differential regulation was confirmed by Northern blot. These are listed in Table 1. To exclude clonal artifacts, a C3H10T1/2 clone (11/2) different from the one (18/17) used for the microarray experiment was used for Northern blot confirmation (Fig. 3). Like cells constitutively overexpressing Runx2, induced cells expressed Runx2 as shown by Northern and Western blot, but no alkaline phosphatase activity was detectable by histochemical staining (Figs. 1 and 3).
Novel genes under transcriptional control of Runx2
One of the cDNA species upregulated by Runx2 (Rig 2) that was identified by the SABRE approach has not been functionally characterized so far and exhibits nucleotide sequence similarity to expressed sequence tag clones or predicted genes only. The other upregulated genes characterized by SABRE were those for extracellular matrix protein 1 (Ecm1), matrix Gla protein, and Pttg1ip. Genes downregulated by Runx2 were fibulin 5 (Fbn5); extraembryonic, spermatogenesis, homeobox 1 (Esx1); the transcription factor Sox9, a key regulator of chondrogenesis; neural proliferation, differentiation, and control gene 1 (Npdc1); Tem1, the murine counterpart for human tumor endothelial marker 1 (TEM1); and a cDNA similar to angiopoietin-like 4 (4/44; Table 1; Figs. 2 and 3).
Genes identified by cDNA microarray expression profiling and confirmed by Northern blot analysis are comprised of osteoblast specific factor 2 (Osf-2), pleiotrophin (Ptn), odd skipped related 1 (Osr1), secretory leukocyte protease inhibitor (Slpi), and hypoxia induced gene 1 (Hig1; Table 1; Figs. 2 and 3).
To investigate whether these genes identified as being putative transcriptional Runx2 targets play a role in osteogenesis, their expression was analyzed in several mesenchymal cell lines including undifferentiated and differentiated MC3T3-E1 (pre-) osteoblasts. Osteogenic differentiation of MC3T3-E1 cells was performed as described earlier and confirmed by staining cells for alkaline phosphatase expression and Northern blot for osteocalcin expression (Fig. 1 and data not shown).(33) While the C3H10T1/2 cell line was used to represent undifferentiated mesenchymal precursor cells, NIH3T3 cells represented fibroblast, C2C12 myoblast, and ATDC5 chondrocyte lineages, respectively.
Thus far, most Runx2 target genes identified are under positive transcriptional control of Runx2, and the large majority of these genes are coding for components of the extracellular bone matrix.(16) Thus, it is not surprising that this study revealed three more genes for extracellular matrix proteins to be transcriptionally activated by Runx2: Ecm1, matrix Gla protein, and osf-2. Ecm1 is expressed in pre-osteoblasts, and mutations in the (human) ECM1 gene are associated with the human autosomal recessive skin and mucosa disorder lipoid proteinosis.(43) Matrix γ-carboxyglutamate (Gla) protein (Mgp), initially isolated from bovine bone matrix, is associated with Keutel Syndrome, a human autosomal recessive disorder comprising short phalanges, cartilage calcification, and other features.(44–46) The genes for these extracellular matrix proteins were induced by Runx2 expression (Figs. 2 and 3). The expression patterns of Ecm1 and Mgp in mesenchymal cell lines were similar, with the exception of C3H10T1/2 cells, where only Ecm1 was markedly expressed. Osteoblast-like MC3T3-E1 cells expressed both Ecm1 and Mgp (Fig. 4). Osf-2 expression was predominantly present in differentiated MC3T3-E1 cells, supporting earlier findings that this adhesion molecule is associated with bone formation.(47)
Expression of another gene for an extracellular matrix protein, fibulin 5, was downregulated by Runx2 in both cellular systems (Figs. 2 and 3). Fibulin 5 (Fbn5) is a secreted factor that is associated with the development of elastic fibers. Mutations in the Fbn5 gene cause the human autosomal recessive connective tissue disorder cutis laxa type I.(48,49) The expression pattern in mesenchymal cell lines shows strong Fbn5 expression in primitive C3H10T1/2 cells only, pointing to a transcriptional control that is coupled to a differentiation state in mesenchymal cells (Fig. 4).
Pleiotrophin, a secreted growth factor, was identified as a transcriptional target of Runx2 (Figs. 2 and 3). Pleiotrophin, also termed HB-GAM, Osf-1, HBGF-8, or HBNF, is a secreted heparin-binding growth factor associated with neurite outgrowth.(50) As discussed below, this gene may be an in vivo target of Runx1 or Runx3, as well as of Runx2.
Three transcription factors were identified to be regulated by Runx2 in this study. These comprise odd skipped related 1 (Osr1), extraembryonic, spermatogenesis, homeobox 1 (Esx1), and Sox9. So far, only one other transcription factor, C/EBPδ, has been described as regulated by Runx2.(51) Osr1 is a zinc finger transcription factor related to Drosophila melanogaster pair rule gene odd; extraembryonic, spermatogenesis, homeobox 1 (Esx1) is related to Drosophila pair rule gene paired and implicated to play a role in placental development; and the transcription factor Sox9 is associated with chondrocyte differentiation.(1,52,53) All these transcription factors were downregulated by Runx2 in our study (Figs. 2 and 3). Esx1 and Sox9 showed only low levels of expression in MC3T3-E1 cells. Consistent with Sox9 being a key regulator of chondrogenesis, Sox9 was expressed in ATDC5 pre-chondroblasts at the highest level of all mesenchymal cell lines tested (Fig. 4).
Neural proliferation, differentiation, and control gene 1 (Npdc-1) is supposed to be involved in neural differentiation.(54) Runx2 repressed Npdc-1 expression in vitro (Fig. 2). However, its in vivo expression pattern comprises specifically neural tissues, thus suggesting that in vivo Npdc-1 gene expression is controlled by the other two Runx factors, as discussed below.(55)
Secretory leukocyte protease inhibitor (Slpi) is a secreted protease inhibitor with specificity to trypsin, chymotrypsin, cathepsin G, and neutrophil elastase found in several mucous fluids.(56,57) Therefore, again, the expression pattern of Slpi points to a regulation in vivo by Runx1 or Runx3 rather than by Runx2.
Expression of Tem1, the murine homolog of human tumor endothelial marker 1 (TEM 1), which has been described to be expressed in human tumor epithelium and has sequence similarity to thrombomodulin, was decreased by constitutive Runx2 expression in C3H10T1/2(58) (Fig. 2). Runx2-dependent repression of Tem1 gene expression could not be confirmed by inducible expression of Runx2 (data not shown). Expression pattern analysis in mesenchymal cell lines revealed strong Tem1 expression in primitive C3H10T1/2 cells (Fig. 4). These findings point to a differentiation stage dependent transcriptional control of the Tem1 gene.
Pttg1ip is a direct transcriptional target of Runx2
The nucleotide sequence of the cDNA fragment we termed Pttg1ip exactly matched the sequence of an mRNA previously described as being similar to pituitary tumor transforming 1 protein interacting protein (Genbank accession no. gi 19343823 gb BC025533.1). In Northern blot experiments, this cDNA fragment revealed the strongest difference in expression between C3H10T1/2-Runx2 and wildtype cells of all genes identified in this study (Fig. 2). This and the fact that Pttg1ip function has only preliminarily been studied encouraged us to perform further investigations on this gene.
Pttg1ip mRNA sequence was screened against the mouse genome using the BLAST software revealing a single homologous sequence on chromosome 10 (ref NW_000028.1 Mm10_WIFeb01_212). The gene consists of six exons with an open reading frame spanning all exons and has an overall size of ∼17.5 kb (Fig. 5A). Pttg1ip mRNA contains an open reading frame coding for a polypeptide with 75% homology to human PTTG1IP protein (GenBank accession no. gi 18088698 gb BC020983.1 BC020983), suggesting that Pttg1ip is the murine ortholog of human PTTG1IP (Fig. 5B).
To find out whether Pttg1ip expression is under direct control of Runx2, we screened the genomic sequence flanking the predicted transcription start of Pttg1ip for Runx consensus binding sites. Within 3000 bp 5′ to the predicted transcription start, we identified seven RUNX consensus sites, with the most proximal at −1172 nt (Fig. 5A). To determine the potential of Runx2 protein to physically interact with these potential binding sites, EMSA studies were performed. As shown in Fig. 6A, in vitro translated Runx2 protein bound to a labeled standard oligonucleotide OligoA adopted from Tahirov et al.(39) To investigate the affinity of putative binding sites identified in the Pttg1ip 5′ flanking genomic region, unlabeled oligonucleotides O1-O7 containing putative binding sites 1-7, respectively, were used as competitors. Binding to OligoA by Runx2 protein was strongly competed for by Oligos O2, O6, and O7, implying physical interaction of Runx2 protein with binding sites 2, 6, and 7 in the Pttg1ip promoter region (Fig. 6A). Similar results were obtained in EMSA using O1-O7 as labeled oligonucleotides (data not shown). Oligonucleotides O2, O6, and O7 were also tested for binding to protein from nuclear extracts of differentiated MC3T3-E1-cells. Figure 6B shows an EMSA with binding of nuclear extract protein to Oligo A (lane 3), O6 (lane 7), and O7 (lane 9). Interaction of in vitro translated Runx2 with Oligo A indicated the correct size of Runx2/oligonucleotide dimers in this EMSA (lane 1). Oligonucleotide O2, however, yields only a weak EMSA band at the size expected for a Runx2/oligonucleotide dimer. Instead, O2 seems to bind more efficiently to a different, smaller protein from the nuclear extract, as indicated by a strong band of lower size. Abolishing the appearance of EMSA bands by competition with an excess of unlabeled oligonucleotide (lanes 2, 4, 6, 8, and 10) indicated that these bands were specific for protein/oligonucleotide interactions.
To explore the functionality of these Runx binding sites, we performed reporter assays with a 1.8-kb promoter fragment cloned upstream of a lacZ reporter gene (pPttg1ipProm-f). This fragment included the two most proximal Runx binding sites (sites 6 and 7). Promoter activity was assayed by measuring β-galactosidase activities of lysates from transfected cells. All transfections were carried out in triplicate, and normalization was performed by co-introducing phRL-SV40 for constitutive expression of Renilla luciferase in transfected cells. Pttg1ip promoter activity was determined in MC3T3-E1 cells and rat osteosarcoma ROS17/2.8 cells in the presence or absence of Runx2 (Fig. 6C). Interestingly, promoter activity was significantly higher in developmentally further advanced ROS17/2.8 (rat) cells compared with MC3T3-E1 (mouse) cells, although the promoter sequence introduced was of murine origin (Fig. 6C). In MC3T3-E1 cells, promoter activity could be increased by co-transfection with the Runx2 expression vector pCMV-Runx2 (Fig. 6C, left). However, co-transfection of pCMV-Runx2 did not significantly affect promoter activity in pPttg1ipProm-f in ROS17/2.8 cells (Fig. 6C, right). This implies that, in rat osteosarcoma cells, further advanced in lineage development the Pttg1ip promoter is already maximally activated by intrinsic Runx2 protein. Specificity of Runx2-dependent promoter activation was assayed by co-transfection of pPttg1ipProm-r with or without Runx2. pPttg1ipProm-r has the same design as pPttg1ipProm-f; however, it harbors a Pttg1ip promoter fragment in inverted orientation. This construct yielded significantly lower β-galactosidase activities compared with pPttg1ipProm1-f, and β-galactosidase activity could not be augmented by Runx2 under conditions tested. These results suggest a direct transcriptional control of Pttg1ip gene expression by Runx2.
Runx2-dependent regulation of Pttg1ip expression was substantial in the constitutive Runx2 expression system (Fig. 2). However, C3H10T1/2 cells with mifepristone-inducible expression of Runx2 showed only a moderate increase of Pttg1ip expression after induction of Runx2 expression (Fig. 3). Previously we have shown that the histone deacetylase inhibitor trichostatin A (TSA) can further increase Runx2-dependent induction of the direct Runx2 target gene LGALS3.(33) We investigated whether TSA could augment Runx2-dependent Pttg1ip mRNA expression in mifepristone inducible Runx2-expressing cells. C3H10T1/2 cells were pretreated with TSA before Runx2 induction by mifepristone. Similar to the LGALS3 gene expression, Runx2-dependent induction of Pttg1ip expression could be increased by TSA (Fig. 7A).
Human PTTG1IP has been reported to be ubiquitously expressed in human adult tissues.(59) Assuming that Pttg1 is under transcriptional control of Runx2, we analyzed Pttg1ip gene expression in several cell lines, including MC3T3-E1 cells in which Runx2 is expressed at high levels. Pttg1ip expression was present in NIH3T3 fibroblasts, C3H10T1/2 embryonic fibroblasts, ATDC5 pre-chondrocytes, and undifferentiated MC3T3-E1 osteoblastic cells. After osteogenic differentiation of MC3T3-E1 osteoblasts, Pttg1ip expression was slightly enhanced, an effect potentially mediated by Runx2. In contrast, Pttg1ip expression was barely detectable in pre-myoblastic C2C12 cells (Fig. 7B).
Because in vivo Runx2 is expressed predominantly in developing bone, we further analyzed time- and tissue-specific expression patterns of Pttg1ip in mouse embryos. Expression of Pttg1ip in primary calvarial cells from 6-day-old mice was assayed by RT-PCR and Northern blot (Figs. 7C and 7D). These experiments revealed a slightly lower expression level of Pttg1ip in calvarial cells compared with measured in vitro differentiated MC3T3-E1 cells (Fig. 7D). Levels of expression observed in whole embryo during the organogenesis stage of mouse embryogenesis were slowly increasing from embryonic day 8.5 to 16.5 (Fig. 7E). During organogenesis, Pttg1ip expression is distributed in all organs as revealed by Northern blot analysis. Whereas expression was weakest in developing brain, the strongest signal was detected in limbs and gonads (Fig. 7F).
These findings are consistent with a regulation of Pttg1ip by Runx2, but they also indicate that other transcription factors must be involved in the transcriptional control of this gene.
In this study, we report a systematic approach to isolate genes regulated in their expression levels by the osteoblast-specific transcription factor Runx2. We used two different, but complementary, techniques to compare mRNA from Runx2-overexpressing versus -nonexpressing cell lines. While comparative hybridization to a cDNA microarray allowed us to screen a large number of genes in a short period of time, no such microarray was available that preferentially represented cDNA species expressed in osseous tissue. Therefore, we turned to a second method: the PCR-based differential hybridization technique SABRE. This second approach allowed us to potentially identify Runx2 target genes that exhibit a tissue-specific pattern of expression and may not be represented on the microarray used. Indeed, there was no overlap between the sets of Runx2 target genes identified by the two different approaches.
We used two genetically modified sublines of the mesenchymal precursor cell line C3H10T1/2 for our experiments. One of them stably overexpressed Runx2 at high levels, and the other carried a vector system allowing Runx2 expression to be turned on by adding mifepristone to the cell culture medium. We hypothesized that, by overexpressing Runx2 in a cell line that has most features of the type of cell showing the first onset of Runx2 expression during embryonic development, we might be able to induce the expression of Runx 2 transcriptional target genes. Runx2 binds to the consensus sequence 5′-(A/G/T)ACCPuCPu-3′, a property it shares with the other two members of the family of core binding factors, Runx1 and Runx3.(16) Therefore, in the somewhat artificial cell line model, Runx2 might be able to bind to the regulatory elements and change the expression levels of genes that are transcriptional targets of Runx1 or Runx3 in vivo. For that reason, we studied the genes identified in this study with respect to their expression pattern in different cell lines of mesenchymal origin and screened the literature for data on expression levels in nonosseous tissues. Judging from these data, it seemed likely that some of the genes (i.e., Ecm1, Fbn5, Npdc-1, pleiotrophin, and Slpi) might also or exclusively be regulated by Runx1 or Runx3. While both these Runx proteins are expressed in cells of the hematopoietic lineage and specific neurons in dorsal root ganglia, Runx1 is expressed in the epidermis, especially epidermal appendages, whereas Runx3 can be detected in the mesenchyme underlying these structures.(60,61) Furthermore, Runx1 is expressed in glandular epithelium (e.g., of the salivary glands).(60) Mutations in the ECM gene cause a heritable disease (lipoid proteinosis) that affects skin and mucosa, pointing to a potential regulatory influence of Runx1, while expression in osseous cells is most likely driven by Runx2.(43) However, Fbn5, the gene mutated in cutis laxa, is potentially a target gene of Runx3 in connective tissue.(48,49) The expression patterns of Npdc-1 and pleiotrophin comprise predominantly neural tissues and might therefore be under transcriptional control of Runx1 and/or Runx3.(50,54)Slpi was reported to be detected in mucous fluids, thus pointing to a transcriptional regulation by Runx1.(56)
Different transcription factors have been characterized as key control elements of different mesenchymal lineages: whereas MyoD, myogenin, and others direct the development of mesenchymal precursor toward the myoblast lineage, Sox9 is a factor important for chondrocyte, peroxysome proliferator-activated receptor (PPAR) γ for adipocyte, and Runx2 and Osterix for osteoblast development.(1–3,62,63) Thus far, no transcription factor has been identified that governs the decision for fibroblast fate and possibly the fibroblastic cell lineage represents the default fate in absence of activity of the transcription factors mentioned above. It has been shown in the past that one mechanism to stabilize the commitment to a certain mesenchymal lineage is autoregulation of the transcription factors involved by binding to their respective regulatory elements.(64) A second important mechanism might involve transcriptional repression of the factors representing alternative lineages be a lineage-specific transcription factor once activated. The negative regulatory effect of Runx2 on the chondrocyte lineage-specific transcription factor Sox9, as shown in this study, might be one example to show this mechanism. The reciprocal situation, regulation of Runx2 gene activity by Sox9, might also exist, given a number of evolutionarily conserved Sox-binding sites in the Runx2 promoters.(65) Downregulation of Runx2 by PPARγ has been described in the literature.(63) Thus, a network of regulatory interdependence might exist between key transcription factors that guide development along and are responsible for commitment to the different mesenchymal lineages.
Runx2 is a homolog of the Drosophila transcription factor runt.(12) Runt (run) together with hairy (h) and even-skipped (eve) constitute the group of primary pair-rule genes expressed in a periodic pattern in the Drosophila embryo that specifies body segmentation. The primary pair-rule genes are thought to modulate the expression of another set of transcription factors named secondary pair-rule genes. Among the latter are odd-skipped (odd) and paired (prd), two genes that are also known to be involved in the regulation of runt.(66,67) A striking observation made in our study is the regulatory of Runx2 of the odd-homolog osr1 and the prd-homolog esx1. It is tempting to speculate that a regulatory network that acts in early Drosophila embryogenesis has been conserved during evolution but is used for a different purpose in vertebrates, namely cell lineage specification.
Taken together, our study has revealed several new aspects on how Runx2 mediates its effect on cell differentiation and maturation. The novel target genes identified here will stimulate further work into regulatory networks and pathways.
We thank K. Warnatz for help with the SABRE technique. We are grateful to G. Gross for providing C3H10T1/2-Runx2 cells and for help with mouse calvarial cell extraction. M. Kolanczyk, G. Scherer, and C. Shukunami are acknowledged for providing cell lines. We thank I. Huber and S. Enger for excellent technical help. This work was supported by Deutsche Forschungsgemeinschaft Grant DFG 134/2-2 to FO.