CREB-AP1 Protein Complexes Regulate Transcription of the Collagen XXIV Gene (Col24a1) in Osteoblasts*

Collagen XXIV is a newly discovered and poorly characterized member of the fibril-forming family of collagen molecules, which displays unique structural features of invertebrate fibrillar collagens and is expressed predominantly in bone tissue. Here we report the characterization of the proximal promoter of the mouse gene (Col24a1) and its regulation in osteoblastic cells. Using well characterized murine models of osteoblast differentiation, we found that the Col24a1 gene is activated sometime before onset of the late differentiation marker osteocalcin. Additional analyses revealed that Col24a1 produces equal amounts of two alternatively spliced products with different 5′-untranslated sequences that originate from distinct transcriptional start sites. Cell transfection experiments in combination with DNA binding assays demonstrated that Col24a1 promoter activity in ROS17/2.8 osteosarcoma cells is under the control of an upstream cis-acting element, which is shared by both transcripts and is recognized by specific combinations of c-Jun, CREB1, ATF1, and ATF2 dimers. Consistent with these results, overexpression of c-Jun, ATF1, ATF2, or CREB1 in transiently transfected osteoblastic cells stimulated transcription from reporter gene constructs driven by the Col24a1 promoter to different degrees. Moreover, chromatin immunoprecipitation experiments showed that these nuclear factors bind the same upstream sequence of the endogenous Col24a1 gene. Collectively these data provide new information about transcriptional control of collagen fibrillogenesis, in addition to implicating for the first time CREB-AP1 protein complexes in the regulation of collagen gene expression in osteoblasts.

Vertebrate collagens represent a very large superfamily of extracellular proteins that impart specific physical properties to the connective tissue of virtually every organ system (1)(2)(3). There are more than 42 collagen ␣-chains that form 27 distinct trimers or types, which in turn give rise to a large variety of specialized macroaggregates. The most abundant and ubiquitous collagen macroaggregates are the highly ordered banded fibrils made of the so-called fibrillar collagens (types I-III, V, and XI) (1)(2)(3). All members of the fibrillar collagen family share a common structure that consists of a long triple helical domain, which is made of uninterrupted Gly-X-Y triplets and flanked at both ends by noncollagenous propeptides (1)(2)(3). These structural features also characterize collagen molecules that form fibrils in the extracellular matrices of primitive invertebrates, such as sponges, annelids, echinoderms, and mollusks (4,5). Unlike the vertebrate counterparts, invertebrate fibrillar collagens display short interruptions in the triple helices and unique structural features in the amino-and carboxyl-terminal propeptides (4,5). Vertebrate fibrillar collagens are either widely distributed in soft and hard tissues (types I, III, and V) or are restricted predominantly to cartilage (types II and XI) (1). Genetic evidence from animal and human studies has indicated that the quantitatively minor types V and XI collagen regulate the diameter of the major types I and II fibrils, respectively, by participating in fibril assembly (2, 6 -9). These studies have also demonstrated the importance of heterotypic collagens I/V and II/XI fibrils in skeletal development and integrity.
As a result of the Human Genome effort, several new collagens have been recently identified which had escaped prior biochemical detection. Two among them (types XXIV and XXVII) bear the structural characteristics of fibrillar collagens and specifically, of invertebrate fibrillar collagens (10 -12). Gene expression analyses in the mouse have revealed that Col24a1 and Col27a1 display mutually exclusive patterns in the developing and adult skeleton. These studies have in fact shown that whereas Col24a1 transcripts accumulate at ossification centers of the craniofacial, axial, and appendicular skeleton, Col27a1 activity is instead confined to the cartilaginous anlagen of skeletal elements (10 -12). Additionally, structural considerations have suggested that collagens XXIV and XXVII are likely to form distinct homotrimers (11). Together these observations have been interpreted to indicate that these newly discovered fibrillar collagens may participate in the control of important physiological processes in bone and cartilage, such as collagen fibrillogenesis and/or matrix calcification and mineralization (10 -12).
Bone formation is a complex and tightly regulated genetic program that involves two distinct pathways at different anatomical locations (13)(14)(15). In intramembranous ossification, mesenchymal cells condense and differentiate directly into collagen I-producing osteoblasts. In endochondral bone formation, cells at condensation sites differentiate into chondrocytes that form a cartilage (collagen II-rich) anlagen, which is replaced by a bony (collagen I-rich) matrix and bone marrow following chondrocyte hypertrophy, matrix calcification, and vascular invasion. At the same time, cells around the condensations form the perichondrial layer that gives rise to the osteoblast-forming periosteum and ulti-mately to cortical bone. Distinct transcriptional codes control osteoblastogenesis and chondrogenesis and thus, assembly of the collagen I-rich bone matrix and the collagen II-rich cartilage matrix (14,15). The canonical Wnt signaling pathway has been shown recently to direct differentiation of mesenchymal cells toward either the osteoblast or chondrocyte lineages (16,17). Previous investigations, on the other hand, have implicated the transcription factors Runx2 and Osterix in the progression of osteoblastogenesis during intramembranous and endochondral ossification, as well as the Sox5, 6, and 9 nuclear proteins in the regulation of chondrogenesis (18 -22). A number of ubiquitous transcription factors have been also involved in osteoblast differentiation and function, including Msx proteins, Dlx5, Twist, and members of the AP1 complexes (23)(24)(25)(26)(27)(28). Similarly, several studies have identified DNA cis-acting elements and nuclear trans-acting factors that regulate cartilage-specific expression of the collagen II and XI genes (29 -32). By contrast, significantly less is known about the regulation of fibrillar collagen genes in osteoblasts.
One of our research interests is the study of the regulation and function of fibrillar collagen genes in normal and diseased conditions. As part of this ongoing effort, the present study was designed to characterize the proximal promoter of the mouse Col24a1 gene using a combination of cell transfection and DNA binding assays. The results of these experiments suggest that Col24a1 is activated during the mid to late phase of osteoblast differentiation mostly through the binding of CREB 2 -AP1 complexes to an upstream sequence, which is shared by two alternative transcription start sites. This study therefore extends our knowledge of the transcriptional regulation of collagen fibrillogenesis, in addition to implicating for the first time CREB-AP1 protein complexes in the expression of a fibrillar collagen gene in osteoblasts.
Cell Transfection Assays-Col24a1 promoter-luciferase (LUC) reporter gene constructs were derived from clone pBeroBAC RP23-205C6 using PCR amplification. Amplified products were cloned into the pGEM-T Easy vector (Promega, Madison, WI) and sequenced. Internal deletions and nucleotide substitutions were generated by sitedirected mutagenesis as described previously (34). Transient transfections were performed using the Lipofectamine Plus reagent system (Invitrogen), and luciferase activity was assayed 48 h later using the Dual-luciferase TM reporter assay system (Promega). The pRL-TK Renilla reniformis luciferase expression vector was used as an internal control for transfection efficiency. Results were expressed as the mean Ϯ S.E. of five to seven independent experiments and evaluated by Student's t test. Expression vectors for ATF1, ATF2, CREB1, and c-Jun expression vectors were kindly provided by Drs. Gerard Karsenty (Baylor College of Medicine, Houston, TX) and Lionel Ivashkiv (Hospital for Special Surgery, New York, NY).
DNA Binding Assays-Preparation of nuclear extracts and DNA binding assays were carried out according to the published protocols (34,35). Wild-type and mutant oligonucleotide probes were generated by PCR amplification using HindIII site-linked primers. PCR products were subcloned into pGEM-T Easy vector, cleaved with HindIII, and radiolabeled with [␣-32 P]dCTP using the Klenow enzyme (34,35). DNA-nuclear protein binding was carried out at 25°C for 30 min in 25 l of reaction buffer containing 3 g of poly(dI-dC). DNA-bound protein complexes were separated in a 4.5% nondenaturing polyacrylamide gel in 0.25% TBE buffer. For competition and antibody interference assays, unlabeled probes or antibodies were added to the reaction mixture for 1 h at 4°C before the addition of labeled probe. The anti-CREB1 antibody was purchased from Upstate Biotechnology (Lake Placid, NY), and the other antibodies were from Santa Cruz Biotechnology (Santa Cruz, CA). Chromatin immunoprecipitation (ChIP) assays were performed using a commercial kit (Upstate Biotechnology) (34). Quantitative PCR was carried out for 35 cycles using 5 l of sample DNA solution/50-l reaction, and amplification products were separated in 2.5% agarose gel in 1 ϫ TAE buffer.

Col24a1
Contains Two Alternative Promoters-Previous work established the entire coding sequence of the human ␣1(XXIV) collagen (COL24A1) gene and of only part of the mouse Col24a1 gene (11). We used this information to complete the primary structure of the mouse ␣1(XXIV) collagen chain by identifying mouse expressed sequence tags in the GenBank TM and by generating PCR amplification products covering sequence gaps. As a result, we found a 19-amino acid insertion in the noncollagenous amino-terminal sequence of the mouse compared with the human chain (see GenBank TM accession numbers AY244357 and DQ157748). These experiments were also instrumental in identifying the foremost exon of Col24a1 as consisting of a 5Ј-untranslated region (UTR) of undetermined length and a coding segment corresponding to the first 94 amino acid residues of the ␣-chain. The oligonucleotide-capping RACE approach was therefore employed to determine the Col24a1 start site of transcription and implicitly, the 5Ј-boundary of exon 1. As the source of template for the reaction, we utilized RNA purified from the eye and bone in which accumulation of Col24a1 transcripts has been found to be the highest (11). Sequencing of nearly 40 independent cDNA clones from each set of RNA samples revealed the presence of two different 5Ј-UTRs that upon subsequent analysis of genomic clones, were accounted for by a combination of alternative splicing and transcriptional start sites. To be precise, half of the cDNA clones contained a 353-nucleotide long 5Ј-UTR that is continuous with the genomic sequence of the exon originally identified as the first of exon of the human gene and which was now renamed exon 1a (Fig. 1A) (11). The other half of the cDNA clones instead contained the 87 nucleotides immediately upstream of the ATG codon, in addition to an 80-nucleotide long 5Ј-UTR corresponding to an upstream exon (named exon 1b) that is separated from exon 1a by a 152-bp intervening sequence (Fig. 1A). Both transcripts 1a and 1b are spliced correctly into exon 2, leaving the open reading frame unaffected, and consequently, they are predicted to translate into identical ␣1(XXIV) chains (Fig. 1A). In summary, Col24a1 contains two alternative start sites of transcription, thereby identified as Ϫ1 (exon 1a) and ϩ232 (exon 1b), two alternatively spliced transcripts with different 5Ј-UTRs, the shortest of which (transcript 1b) splices into nucleotide ϩ509 of exon 1a, and the same start site of translation, located at nucleotide ϩ586 (Fig. 1A). The functional significance of Col24a1 alternative promoters and 5Ј-UTR heterogeneity was not addressed in the present study. Comparison of the 5Ј-end sequences of the COL24A1 and Col24a1 genes revealed three segments of high homology which span from nucleotides Ϫ100 to ϩ133, from ϩ198 to ϩ256, and from ϩ470 to ϩ632 (Fig. 1A).
Collagen XXIV has been estimated to represent about 4% of the amount of collagen I in bone, thus slightly less than collagen V, the regulator of collagen I fibrillogenesis (11). The estimate was based on gene expression analyses that in addition documented coexpression of collagens I and XXIV genes at ossification centers in the mouse embryo (11). Osteoblasts were therefore chosen as the experimental system in which to study the transcriptional regulation of Col24a1. Reverse transcription-PCR amplifications were used to assess the levels of Col24a1 expression in ROS17/2.8 and ROS25 osteosarcoma cell lines, which represent late and early stages of osteoblast differentiation, respectively (33). As positive and negative controls, PCR amplifications were also performed with RNAs purified from MCC and NIH-3T3 fibroblasts. Osteoblast-specific genes included Col1a2 (early osteoblast differentiation marker) and osteocalcin (late osteoblast differentiation marker), whereas the ubiquitous GAPDH gene served as the normalizing control. The results of these experiments suggested that the onset of Col24a1 expression occurs sometime after Col2a1 and prior to osteocalcin gene activation (Fig. 1B). Implicitly, they also identified ROS17/2.8 cells as the most suitable model in which to study the anatomy of the minimal Col24a1 promoter.
A Short Upstream Sequence Promotes Col24a1 Transcription in Osteoblasts-Cell transfection experiments were initially employed to delineate the shortest promoter sequence of Col24a1 capable of direct- ing transcription in ROS17/2.8 cells. To this end, we engineered two distinct sets of LUC reporter gene constructs representative of the alternative promoters of Col24a1. The first set of Col24a1 promoter-LUC constructs shared the same 3Ј-end at position ϩ 509 and included both start sites of transcription, whereas the 3Ј-ends of the second set of Col24a1 promoter-LUC constructs was located at ϩ80 and excluded the start site of transcript 1b ( Fig. 2A). Both sets of promoter-LUC constructs included progressive 5Ј-deletions of the upstream Col24a1 sequence ( Fig. 2A). Irrespective of the 3Ј-end of the promoter-LUC construct, cell transfection assays assigned maximal transcriptional activity to the region between nucleotides Ϫ144 and ϩ80 (Fig. 2B). This promoter segment contains one of the three homology sequences of the COL24A1 and Col24a1 genes (Fig. 1A).
Next, an electrophoretic mobility shift assay (EMSA) was employed to identify potential DNA-nuclear protein interactions within the Ϫ144 to ϩ81 segment of the Col24a1 promoter. To this end, four overlapping probes (p1-p4) spanning from nucleotide Ϫ163 to nucleotide ϩ116 were each incubated with ROS17/2.8 nuclear extracts (Fig. 3A). Specific band shifts were only obtained with overlapping probes p2 and p3, which cover together the sequence between nucleotides Ϫ98 and ϩ51 (Fig. 3A). To be precise, p2 yielded four retarded bands (b1-b4) of which p3 appeared to migrate as band b2 of probe p3 (Fig. 3A). In support of this postulate, band b2 disappeared from the p2 EMSA pattern when competed with a molar excess of unlabeled probe p3; conversely, formation of the p3 retarded band was eliminated by competition with a molar excess of the p2 sequence (Fig. 3B). Taken together, these results mapped the b1, b3, and b4 binding sites between nucleotides Ϫ98 and Ϫ33 and the binding site of b2 between nucleotides Ϫ33 and Ϫ15 (Fig. 4).
The functional contribution of the segment encompassing the b1-b4 binding sites was evaluated by cell transfection experiments using the Ϫ144 ϩ 509/LUC plasmid bearing internal deletions of the Ϫ98 to Ϫ33 sequence (p2D; b1,3,4 binding sites), the overlapping Ϫ33 to Ϫ15 sequence (p2/3D; b2 binding site), or both of them (p2,3D; b1-b4 binding sites) (Fig. 4A). Unlike elimination of the b2 binding site in the p2/3D construct, deletion of the upstream sequence that gives rise to retarded bands b1, b3, and b4 led to a drop in luciferase activity of construct p2D nearly equal to the Ϫ144 ϩ 509/LUC construct with the internal dele- tion of all binding sites (p2,3D) or the shorter Ϫ52 ϩ 509/LUC plasmid (Fig. 4B). We therefore focused on the characterization of the factors binding to the sequence between Ϫ98 and Ϫ33 because this cis-acting element appears to drive most of Col24a1 promoter expression in ROS17/2.8 cells.
Binding of CREB-ATF and AP1 Complexes Is Required for Col24a1 Transcription-A series of shorter oligonucleotides encompassing the Ϫ98 to Ϫ33 sequence was used in the EMSA to compete in vitro binding to the p2 probe to narrow down the site(s) of nuclear protein interaction within this cis-acting element. The results clearly demonstrated that all p2 retarded complexes (b1, b3, and b4) were competed by the p2b oli-gonucleotide that extends from nucleotides Ϫ55 to Ϫ38 (Fig. 5A). Inspection of the Ϫ55 to Ϫ38 sequence identified within it the TGACGTCA sequence, which represents a perfect match of the cyclic AMP-responsive element (CRE) protein/ATF binding site (Fig. 5B) (36). Indeed competition experiments using a molar excess of a mutated CRE oligonucleotide failed to interfere with formation of the b1, b3, and b4 complexes (Fig. 5B). The significance of the binding assays was corroborated further by cell transfection experiments that showed a loss of promoter activity in the Ϫ144 ϩ 509/LUC plasmids bearing the same p2b mutation (Fig. 4).
Members of the CREB-ATF family of nuclear factors bind with high affinity to CRE, but AP1 dimers can also interact with CREB-ATF sites, depending on the composition of their flanking sequences (36). Antibodies specific for AP1 and CREB family members were therefore used in a screen to identify which of the possible protein combinations bind the p2b element of Col24a1. The results showed supershifts or binding interferences only with c-Jun, CREB1, ATF1, and ATF2 antisera (Fig. 6). Importantly, each of the antibodies was noted to affect formation of different retarded bands, implying that a distinct combination of AP1 and CREB-ATF proteins binds to the p2b element that directs Col24a1 transcription in ROS17 cells (Fig. 6, A and B). This last conclusion was corroborated by EMSAs in which different combinations of antibodies suggested that band b1 corresponds to c-Jun/ATF2 heterodimers, band b3 to CREB1 homodimers, and band b4 to CREB1/ATF1 heterodimers (Fig. 6, C and D).
Expression vectors for c-Jun, ATF1, ATF2, or CREB1 were each cotransfected into ROS17/2.8 cells together with the wild-type or p2a mutant Ϫ144 ϩ 509/LUC plasmid to correlate DNA-protein binding with promoter function. These functional assays demonstrated that each of the four recombinantly expressed nuclear factors was capable of stimulating Col24a1 promoter activity in a dose-dependent manner and only when the integrity of the p2b sequence was preserved (Fig. 7A). Independent confirmation of the functional assays was obtained by a ChIP assay that documented in vivo occupancy by CREB1, ATF1, c-Jun, and ATF2 of the p2b element in the endogenous Col24a1 promoter (Fig.  7B). Consistent with the EMSA data, this in vivo assay also provided evidence for CREB-AP1 specificity by showing lack of JunD and ATF4 binding to the collagen but not the TIMP1 or osteocalcin promoters (Fig. 7B) (37,38).

DISCUSSION
The present study demonstrated that Col24a1 is a marker of late osteoblast differentiation which is positively regulated by the binding of specific combinations of CREB-AP1 proteins to an upstream cis-acting element, which is shared by the two alternative promoters of the gene. These findings advance knowledge of the transcriptional pathways that regulate formation of fibrillar collagen assemblies in the skeleton. They will also inform the characterization of genetic programs that may be negatively affected by the loss of collagen XXIV in the developing and adult mouse bone.
The role of collagen I fibrils in bone physiology is well established as is the contribution of the minor collagen V to guiding nucleation of collagen I fibrils in the skin and the eyes and by extrapolation, in bone tissue (1-3, 6, 8, 9). The expression pattern of Col24a1 in the developing bone and the eye is at least consistent with a similar role of this macromolecule in collagen I fibrillogenesis, even though the lack of suitable antibodies has hampered experimental confirmation of this hypothesis. Our preliminary findings indicate that Col24a1 is inactive in ROS25 cells, which correspond to early differentiating osteoblasts, but not in ROS17/ 2.8 cells, which represent late differentiating osteoblasts, or in calvarial osteoblasts, albeit at seemingly lower levels than in ROS17/2.8 cells.
Within the limitations of this experimental system, these results nonetheless suggest that collagen XXIV is an integral part of the genetic program of osteoblast terminal differentiation (39). That collagen XXIV is expressed at a lower level in nonskeletal tissues, such as the brain and the eye, also suggests a potentially broader role in organogenesis (11).
Our study adds AP1 and CREB-ATF proteins to the list of transcription factors that are involved in the regulation of fibrillar collagen genes, particularly in osteoblastic cells. A large body of work has demonstrated the critical contributions of these two families of basic leucine zipper proteins to bone formation and remodeling (40 -42). For example, genetic alterations in functions of AP1 and related proteins have been shown to affect negatively osteoblast differentiation and function as well as bone development (40). Similarly, transgenic interference of CREB protein activity greatly impairs the normal process of endochondral bone formation (43). Members of the AP1 or CREB-ATF family of nuclear factors can form homodimeric or heterodimeric protein complexes, which transduce distinct signals and exert discrete transcriptional responses on various promoter targets (36). The heterogeneity in dimer composition is the main determinant of the functional diversification of AP1 and CREB-ATF complexes, which include dimers within and between selected members of each family of transcription factors (36,40,41). Obligatory combinations of CREB-ATF proteins recognize the octameric TGACGTCA element, as do heterodimers between ATFs and specific Jun and/or Fos proteins (36). In line with this last consideration, our antibody interference experiments indicate that CREB1, CREB1/ATF1 and c-Jun/ATF2 dimers can specifically bind in vitro to the evolutionarily retained TGACGTCA sequence (p2b element) of the Col24a1 and COL24A1 promoters. Further confirmation of this in vitro finding was provided by the ChIP assay, which showed that the endogenous p2b site of the rat gene is occupied in ROS17/2.8 cells by the CREB1, ATF1, ATF2, and c-Jun proteins. Consistent with this result, we showed that overexpression of each of these four nuclear proteins in osteoblasts stimulates transcription from the wild-type Col24a1 promoter, but not from the same promoter harboring four nucleotide substitutions in the CREB-ATF binding site.
Although our experiments left unresolved whether and how the various CREB-AP1 complexes may compete for 2b binding, they clearly FIGURE 4. Functional contribution of the b1,3,4 and b2 binding sites to minimal promoter activity. Luciferase activity of various mutant promoter sequences was evaluated in transiently transfected ROS17/2.8 cells in relationship to that of the wild-type Ϫ144 to ϩ509 promoter construct in which the relative positions of the b1,3,4 and b2 binding sites are shown. Mutations include deletions of the sequence encompassing probes p2 and p3 (p2,3D plasmid), only probe p2 (p2D plasmid), only probe p3 (p3D plasmid), or the overlap between probes p2 and p3 (p2/3D plasmid), as well as single nucleotide substitutions in the nuclear protein binding sites of probe p2 (plasmid p2bmt, see also top of Fig. 5B). Bars indicate the S.E. of the means. FIGURE 5. Mapping of the nuclear protein binding sites within the probe p2 sequence. A, EMSA using the p2 probe in the absence (Ϫ) or in the presence (ϩ) of a 100-fold molar excess of unlabeled oligonucleotides p2a-c that cover the p2 sequence (nucleotide positions shown above the autoradiographs). B, top, the CREB-ATF binding site within the p2b sequence is highlighted, and the nucleotide substitutions (mt) are indicated (p2bmt, see also relevant functional assay of Fig.  4); bottom, EMSAs showing formation of retarded complexes between the p2b probe and ROS17/2.8 nuclear extracts in the absence (Ϫ) or in the presence (ϩ) of increasing (10 -100-fold) molar excess of unlabeled wild-type (wt) or mutant (mt) p2b oligonucleotides.
indicate that binding of these nuclear proteins stimulates transcription from the minimal Col24a1 promoter in osteoblasts. This conclusion is based on the absolute requirement of element 2b integrity for promoter activity and on the positive effect on promoter activity of each of the four nuclear proteins overexpressed in ROS17/2.8 cells. The expression profiles of the proteins that bind element 2b during osteoblast differentiation in vitro are also consistent with the time of Col24a1 onset estimated by the present study. Jun proteins are in fact highly expressed in differentiating osteoblasts prior to matrix production and mineralization, and phosphorylated CREB reaches its highest level in the early mineralization stage (44,45). Irrespective of whether or not these findings are functionally correlated, the characterization of the Col24a1 promoter further support the emerging notion that distinct regulatory pathways coordinate expression of different fibrillar collagen genes in bone tissue. Transgenic studies have in fact indicated that cooperation between an Sp1 binding site in the proximal promoter and uncharacterized complex(es) interacting with a far upstream enhancer directs bone-specific expression of the human COL1A2 gene (46). The same kind of approach has identified positive osteoblast-specific elements in the upstream promoter of the mouse and rat Col1a1 genes, as well as a ␦EF1/ZEB1 binding site further upstream which represses transcription of the mouse promoter in osteoblasts (47)(48)(49). Work in progress is extending the present work to the characterization of possible interactions of the CREB-AP1 complexes with other nuclear proteins, in addition to evaluating these in vitro findings within the physiological context of the transgenic mouse model.