Regulation of Transforming Growth Factor-β-dependent Cyclooxygenase-2 Expression in Fibroblasts*

Abnormal transforming growth factor-β (TGF-β) signaling is a critical contributor to the pathogenesis of various human diseases ranging from tissue fibrosis to tumor formation. Excessive TGF-β signaling stimulates fibrotic responses. Recent research has focused in the main on the antiproliferative effects of TGF-β in fibroblasts, and it is presently understood that TGF-β-stimulated cyclooxygenase-2 (COX-2) induction in fibroblasts is essential for antifibroproliferative effects of TGF-β. Both TGF-β and COX-2 have been implicated in tumor growth, invasion, and metastasis, and therefore tumor-associated fibroblasts are a recent topic of interest. Here we report the identification of positive and negative regulatory factors of COX-2 expression induced by TGF-β as determined using proteomic approaches. We show that TGF-β coordinately up-regulates three factors, heterogeneous nuclear ribonucleoprotein A/B (HNRPAB), nucleotide diphosphate kinase A (NDPK A), and nucleotide diphosphate kinase A (NDPK B). Functional pathway analysis showed that HNRPAB augments mRNA and protein levels of COX-2 and subsequent prostaglandin E2 (PGE2) production by suppressing degradation of COX-2 mRNA. In contrast, NDPK A and NDPK B attenuated mRNA and protein levels of COX-2 by affecting TGF-β-Smad2/3/4 signaling at the receptor level. Collectively, we report on a new regulatory pathway of TGF-β in controlling expression of COX-2 in fibroblasts, which advances our understanding of pathophysiological mechanisms of TGF-β.

Transforming growth factor-␤ (TGF-␤) 3 is a multifunctional cytokine that regulates a multitude of physiological and pathological processes. TGF-␤ controls various cell activities ranging from cell differentiation, proliferation, and migration to extra-cellular matrix production (1)(2)(3)(4)(5). Therefore, abnormal regulation of TGF-␤-dependent signaling pathways often leads to diseases, such as fibrosis-related diseases (6,7), cancer (8 -11), cardiovascular disease (12,13), and autoimmune diseases (7). Among the wide range of biological functions regulated by TGF-␤, its effect on fibrosis remains one of the most extensively studied research areas because excessive TGF-␤ signaling is a critical contributor to the pathogenesis of various fibrotic diseases (2,7).
TGF-␤ directly promotes expression of type I collagen, a major component of the extracellular matrix, during fibrosis. TGF-␤ further induces connective tissue growth factor expression in fibroblasts, which synergizes with TGF-␤ to induce fibrosis. Epithelial to mesenchymal transition by TGF-␤ is another contributor to the fibrotic response. However, in addition to its well established profibrotic effects, several recent reports have demonstrated a key role of TGF-␤ as an antifibrotic regulator through regulation of cyclooxygenase-2 (COX-2).
Cyclooxygenase is a rate-limiting enzyme in the synthesis of prostaglandins, which are autocrine mediators of multiple cellular processes. Although cyclooxygenase-1 (COX-1) is constitutively expressed in most tissues and involved mainly in homeostasis, COX-2 is usually absent under basal conditions and induced by stimuli, such as growth factors and cytokines. Many lines of evidence have shown that the COX-2-prostanoid pathway is involved in many physiological and pathogenetic pathways, including those that regulate fibrosis, cancer, inflammation, angiogenesis, hemodynamics, and renal function (14 -17). Although COX-2 and its major product, prostaglandin E 2 (PGE 2 ), are generally considered potent proinflammatory mediators, they also possess antifibrotic effects (14). Increased secretion of PGE 2 in response to TGF-␤ has been documented in fibroblasts and shown to be responsible for its antiproliferative effects. Suppression of COX-2 up-regulation and subsequent PGE 2 production, which acts in an autocrine fashion to inhibit proliferation and production of collagen, results in loss of the antiproliferative effects of TGF-␤ (18). Further, COX-2deficient mice show enhanced susceptibility to pulmonary fibrosis (19) and cardiac fibrosis (20). Thus, TGF-␤-stimulated COX-2 induction in fibroblasts is thought to be essential for the antifibroproliferative effects of TGF-␤.
Moreover, recent mounting evidence has confirmed that tumor-associated stromal fibroblasts play critical roles in tumor development and progression (9,10,21). Stromal fibro-blasts have been shown to be the predominant source of COX-2 in colon adenomas (22), and increased expression of COX-2 has been demonstrated in not only cancer cells per se but also the surrounding fibroblasts in invasive carcinomas (23). Considering that up-regulation of COX-2 and the subsequent prostaglandin cascade play crucial roles in tumorigenesis, tumor invasion, and metastasis (15), TGF-␤-induced COX-2 expression in fibroblasts is envisioned to be a critical contributory factor in the tumor microenvironment. However, the involved molecular mechanisms still remain a subject of intense investigation.
Thus, how TGF-␤-mediated COX-2 induction is controlled, especially in fibroblasts, is an important question to be addressed for elucidating the biology of these diseases. Here we show a new regulatory pathway of TGF-␤-dependent COX-2 expression as centered on coordinated regulation of three newly identified factors using a differential proteomic study. Our findings advance our understanding of the underlying mechanisms of TGF-␤ pathophysiology in various diseases, including fibrotic and cancer diseases.
Liquid Chromatography-Cell pellets before and 12 h after TGF-␤ stimulation were suspended in lysis buffer containing 6 M urea, 2 M thiourea, 10% glycerol, 50 mM Tris-HCl, 2% n-octyl glucoside, 5 mM Tris(2-carboxyethyl)phosphine hydrochloride, and 1 mM protease inhibitor (Sigma) and then centrifuged for 1 h at 18°C. The supernatants were recovered and stored at Ϫ80°C for further use. Cell extracts were analyzed using a two-dimensional liquid chromatography system (ProteomeLab PF2D, Beckman Coulter), according to the manufacturer's instructions and as previously described (24). Briefly, as the first dimension, chromatofocusing was performed on a high performance chromatofocusing column with a linear pH gradient generated with two buffers, the start buffer (pH 8.5 Ϯ 0.1) and the eluent buffer (pH 4.0 Ϯ 0.1). Fractions were collected, and then residual proteins were washed with 1 M NaCl. The collected fractions were automatically and sequentially applied to a high performance reversed phase column as the second dimension separation, which consists of a linear gradient of 0 -100% B against A for 30 min, where A is 0.1% trifluoroacetic acid in water and B is 0.08% trifluoroacetic acid in acetonitrile. Collected fractions were stored at Ϫ20°C for further analysis.
Matrix-assisted Laser Desorption Ionization Time-of-flight (MALDI-TOF) Mass Spectrometry-Fractions were reduced to a volume of 16 l using a vacuum concentrator, and 2 l of 1 M NH 4 HCO 3 , 1 l of 10 mM dithiothreitol, and 50 -150 ng of modified porcine trypsin (Promega) were added. After overnight incubation at 37°C, 1 l of 1% trifluoroacetic acid was added to stop digestion. Digests were desalted with C18 Zip-Tips (Millipore) and then eluted with 70% acetonitrile and 0.1% FIGURE 1. A differential display protein expression map for TGF-␤-treated and -untreated cells. After 24 h of growth arrest with serum-free medium, subconfluent cells were stimulated with recombinant human TGF-␤1 (1 ng/ml) for 12 h and then harvested. A, cellular protein separation of control cells (left) and TGF-␤stimulated cells (right) by two-dimensional liquid chromatographic fractionation. The x axis shows decreasing pH from 6.2 to 4.2, and the y axis displays increasing hydrophobicity (%B). The intensity of blue bands represents the relative protein amount as detected by UV absorption at 214 nm. The center panel is a two-dimensional differential map of the protein expression of the treated versus untreated cells. The lanes in the map are obtained by point-to-point subtraction of corresponding data. The green bands indicate up-regulated proteins, and the red bands represent down-regulated proteins in TGF-␤ treated cells. Protein bands identified by subsequent MALDI-TOF mass spectrometry analysis are labeled in a corresponding manner with those in Table  1 and B. B, representative elution profiles of selected pI fractions (fractions 4 and 9). The blue line depicts cell extract of untreated cells, and the red line shows that of TGF-␤-treated cells. Three peaks higher in the TGF-␤stimulated cells (C, M, and LЈ) in these graphs were identified as NDPK B, HNRPAB, and vimentin, respectively. trifluoroacetic acid. The eluted peptide samples were mixed with a saturated matrix of ␣-cyanohydroxycinnamic acid, spotted onto a MALDI plate, and analyzed with a MALDI-TOF mass spectrometer (Voyager-DE STR; Applied Biosystems) in reflectron mode. All data sets were analyzed using the mass spectrometer software (Data Explorer, version 4.0, Applied Biosystems). Each spectrum was base line-corrected with noise reduction and smoothing before it was deisotoped, leaving only the monoisotopic masses. The spectra were internally calibrated using trypsin autolysis peaks (generated at 842.5099 Da and 2211.1046 Da). The peak detection threshold was manually adjusted over the background. Known contaminant peaks (e.g. trypsin and keratin) and matrix peaks were excluded manually from all mass lists.
Protein Identification-The mass list of each sample was entered into the Mascot public search engine (available at the Matrix Science web site) and used to query the Swiss-Prot data base 55.4 (385,721 sequences; 138,434,015 residues) using the following parameters: taxonomy, Mus musculus; enzyme, trypsin; one possible missed cleavage; monoisotopic masses; and peptide tolerance of 0.1 Da. Possible modifications included oxidation of Met and modification of Cys by carbamidomethylation. Results were scored using the probability-based Mowse score. Protein scores greater than 54 are considered significant (p Ͻ 0.05).
Preparation of Plasmid Constructs-The expression vector pCAG was previously described (25,26). Full-length Ndpk a (nucleotide diphosphate kinase A), Ndpk b (nucleotide diphosphate kinase B), and Hnrpab (heterogeneous nuclear ribonucleoprotein A/B) were PCR-amplified from aortic cDNA and subcloned into pCAG vector with an N-terminal FLAG tag (Sigma) to construct a mammalian expression vector for FLAG-tagged proteins.
Western Blot Analysis-Cell extracts 24 h after TGF-␤ stimulation were subjected to SDS-PAGE analysis and then immunoblotted with anti-COX-2 rabbit polyclonal antibody (Cayman Chemical) and anti-COX-1 mouse monoclonal antibody (Cayman Chemical). For analysis of Smad proteins, anti-phospho-Smad2 rabbit polyclonal antibody (Cell Signaling) and anti-Smad2/3 mouse monoclonal antibody (BD Biosciences) were used.
Immunocytochemistry-Cells plated on glass coverslips were analyzed 24 h after TGF-␤ stimulation. Cells were washed in phosphate-buffered saline, fixed with 4% paraformaldehyde for 20 min, permeabilized with 0.1% Triton X-100 in phosphatebuffered saline for 10 min, and then blocked with 1% bovine serum albumin for 30 min. Subsequently, cells were incubated with the same primary antibodies as those for Western blot analysis, exposed to Alexa Fluor 488-conjugated goat anti-rabbit or anti-mouse IgG antibody (Invitrogen), and counterstained with propidium iodide. Fluorescent images were collected with a confocal laser microscope (MRC 1024; Bio-Rad). COX-2, COX-1, or Smad2/3 was visualized using green fluorescence, and nuclei were visualized using red fluorescence.
Adenoviral Infection-Recombinant adenoviruses harboring FLAG-tagged Ndpk a, Ndpk b, Hnrpab p37, and Hnrpab p42 were prepared with the AdEasy system using homologous recombination in bacteria as previously described (27). Forty- a -Fold increase was estimated by calculating change in peak areas. ϩ, proteins whose peaks could not be detected in the control cell lysate.  2 Synthesis-Subconfluent cells maintained in Dulbecco's modified Eagle's medium (Sigma) with 10% fetal bovine serum were infected at 10 MOI with control empty adenovirus or adenovirus harboring HNRPAB. Culture medium was collected at 48 h after adenoviral infection. After the addition of indomethacin, a prostaglandin synthesis inhibitor, at a final concentration of 10 g/ml, the medium was centrifuged and stored at Ϫ80°C for further analysis. The concentration of PGE 2 in cell culture supernatant was estimated by using an enzyme-linked immunosorbent assay kit following the manufacturer's protocol (Cayman Chemical). The background concentration of PGE 2 in medium containing 10% fetal bovine serum was negligible (i.e. below the sensitivity of the assay). The enzyme-linked immunosorbent assay was performed in duplicate, and results are shown as means and S.E. Shown are representative data from one of three independent experiments.

Regulation of TGF-␤-dependent COX-2 Expression
RNAi Transfection-Silencer predesigned small interfering RNA targeted to NDPK A, NDPK B, and HNRPAB and Silencer Negative Control small interfering RNA was purchased from Ambion. Cells were harvested for RT-PCR or Western blot analysis at 24 or 48 h after transfection of double-stranded small interfering RNA using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions.

Differential Protein Expression Profile for TGF-␤-treated and -untreated Cells Generated Using the Two-dimensional Liquid Chromatographic Fractionation System with Subsequent Protein Identification Using MALDI-TOF Mass
Spectrometry-Extracts of control and TGF-␤-treated cells were first applied to liquid chromatography, which separated proteins with chromatofocusing based on pI in the first dimension. Subsequently, selected fractions separated with chromatofocusing were sequentially and automatically applied to reversed phase chromatography as the second dimension, thus making a twodimensional protein map based on pI and hydrophobicity (Fig. 1A). The center panel in Fig. 1A is a differential display between the treated and untreated cells, calculated by point-to-point subtraction of corre-  1 versus lane 4). The treatment condition was the same as that of E, but cells were harvested 24 h after TGF-␤ stimulation. Note that neither TGF-␤ nor HNRPABs affected the expression level of COX-1 as demonstrated by the lower panel. DECEMBER 18, 2009 • VOLUME 284 • NUMBER 51 JOURNAL OF BIOLOGICAL CHEMISTRY 35865 sponding data. As shown in this panel, comparison before and after TGF-␤ stimulation revealed dozens of protein peaks, whose intensity increased or decreased after stimulation. These peaks were identified with MALDI-TOF mass spectrometry, based on the second dimensional elution profiles, whose representative data are presented in Fig. 1B. The identified proteins are listed in Table 1 along with the corresponding band label in Fig. 1, protein name, accession number in the Swiss-Prot data base, Mascot score, and percentage of coverage.

Regulation of TGF-␤-dependent COX-2 Expression
Up-regulation of NDPK B, HNRPAB, and Their Related Factors as Confirmed by RT-PCR Analysis and Increased COX-2 by Western Blot and Immunocytochemistry-Heterogeneous nuclear ribonucleoprotein A/B (HNRPAB) and nucleotide diphosphate kinase B (NDPK B) were among the proteins identified as being up-regulated by TGF-␤ treatment. HNRPAB has been previously reported to bind to the COX-2 3Ј-untranslated region and stabilize an exogenous COX-2 3Ј-untranslated region mRNA reporter (28). On the other hand, NDPK B belongs to the NDPK/NM23 gene family, whose member protein NM23-H1, a human orthologue of NDPK A, up-regulates the COX-2 promoter activity in reporter assays (29). Because the effect of these factors on endogenous COX-2 expression level is hitherto unknown, we focused our subsequent studies on understanding the mechanisms underlying the actions of these factors on COX-2 regulation.
First, to validate the proteomic screening analysis, mRNA levels of HNRPAB and NDPK B were analyzed by conventional RT-PCR and real-time PCR assays. As shown in Fig. 2A, the transcription levels of both factors were augmented after TGF-␤ stimulation in a time-dependent manner. mRNA levels of their related factors, NDPK A and COX-2, were also increased in a similar time-dependent manner. Next, because HNRPAB exists as two isoforms, heterogeneous nuclear ribonucleoprotein A/B (Q99020) and S1 protein C2 (Q20BD0) that differ by a 47-amino acid insertion close to the C terminus (28), we examined which isoform of HNRPAB is induced by TGF-␤, using a PCR primer pair that discriminates the two isoforms. As shown in the lowest left panel in Fig.  2A, TGF-␤ led to the expression of both isoforms of HNRPAB. We refer to the smaller isoform as HNR-PAB p37 and to the larger isoform as HNRPAB p42 hereafter. Further, effects on COX-2 expression were examined by Western blot, which showed a resultant increase in COX-2 protein in contrast to the constitutive expression of COX-1 (Fig.  2B) and immunocytochemistry (Fig. 2C) with further documentation of characteristic prominent staining of the nuclear envelope (30).
HNRPAB Increases the mRNA and Protein Levels of COX-2 and Leads to the Production of PGE 2 -To understand the effects of HNRPAB on COX-2, we overexpressed HNRPAB by adenoviral infection. As indicated in Fig. 3A, both isoforms of HNRPAB augmented the mRNA level of COX-2. HNRPAB p42 was more potent in inducing transcription levels of COX-2 than HNRPAB p37. These effects were accompanied by similar increases in the expression level of COX-2 protein as shown by Western blot analysis (Fig. 3B) and immunofluorescence (Fig. 3C). Also, the secretion of PGE 2 , a down-  , lanes 1-4). Note that, in contrast, adenoviral transfer of HNRPAB contributed to prolonged stabilization of COX-2 mRNA over 120 min after DRB treatment (upper and middle panels, lanes [5][6][7][8][9][10][11][12]. The three panels show the results of PCR with identical samples and increasing cycle numbers (24, 26, and 29 cycles). B, degradation curves of COX-2 mRNA as shown by real-time PCR assay. Although cells with empty adenovirus showed rapid decrease in COX-2 mRNA (solid line and squares), overexpression of HNRPAB resulted in no significant change in mRNA levels of COX-2 for 120 min (dotted lines and circles for HNRPAB p37 and triangles for HNRPAB p42). The expression level of COX-2 mRNA was normalized to that of 18 S rRNA and subsequently that at 0 min. C, augmented degradation of COX-2 mRNA by RNAi silencing of HNRPAB as revealed by real-time PCR assay. Twenty-four hours after RNAi transfection, transcription was stopped by the addition of 100 mM DRB. RNA samples were isolated at 0 and 60 min following DRB treatment. Cells transfected with RNAi targeting HNRPAB (dotted line and circles) showed enhanced degradation of COX-2 mRNA, as compared with cells transfected with negative control RNAi (solid line and squares). The expression level of COX-2 mRNA was normalized to that of 18 S rRNA and subsequently that at 0 min. stream product of COX-2 metabolism, in cell culture supernatant was measured by enzyme immunoassay. Although the prominent induction of COX-2 protein by HNRPAB p42 did not lead to proportional accumulation of PGE 2 , the significant induction of PGE 2 in culture medium of cells overexpressing HNRPAB was seen (Fig.  3D). Thus, overexpression of HNR-PAB augmented COX-2 transcript, protein, and enzymatic activity.
In addition, we examined the effect of HNRPAB on TGF-␤-induced COX-2 mRNA and protein.
As shown in Fig. 3, E and F, TGF-␤ and HNRPAB showed a synergistic effect on COX-2 expression.
Adenoviral Transfer of HNRPAB Stabilizes COX-2 mRNA Levels-HNRPAB has been previously reported to stabilize an exogenous COX-2 3Ј-untranslated region mRNA reporter in HeLa cells (28). However, it is unknown whether overexpression of HNRPAB can stabilize endogenous mRNA of COX-2. To evaluate this potential post-transcription regulation, HNRPAB was overexpressed in cells using adenoviral transfer with further inhibition of transcription by the addition of DRB at 48 h. Quantitative RT-PCR assay (Fig. 4A) and real-time PCR (Fig. 4B) showed rapid degradation of COX-2 mRNA after DRB treatment in cells infected with control empty adenovirus (lanes 1-4 in Fig. 4A and solid line in Fig. 4B) and prolonged stabilization of COX-2mRNAincellsoverexpressing HNRPAB, especially p42 (lanes 5-12 in Fig. 4A and dotted lines in Fig. 4B). On the other hand, knockdown of HNRPAB using RNAi led to enhanced degradation of COX-2 mRNA (Fig. 4C). In addition, HNR-PAB did not show any transactivation effects on a murine Cox-2 promoter in cotransfection reporter assays (data not shown). Collectively, the mechanism of HNRPABdependent up-regulation of COX-2 is mainly due to post-transcriptional stabilization of COX-2 mRNA.

JOURNAL OF BIOLOGICAL CHEMISTRY 35867
COX-2-Next, we assessed whether NDPK A and NDPK B increase COX-2 expression as is consistent with a previous report using reporter assays (29). Unexpectedly, however, ectopic overexpression of these factors diminished TGF-␤-dependent induction of COX-2 mRNA (Fig. 5A). Suppressed COX-2 protein levels were also confirmed by Western blot analysis (Fig. 5B) and immunofluorescence (Fig. 5C). RNAi analysis was further used to confirm the effect of NDPK A and NDPK B on COX-2 by depleting these factors. As shown in Fig. 5, D and E, knockdown of either NDPK A or NDPK B led to increased COX-2 expression at both the mRNA and protein levels. Thus, NDPK A and NDPK B repress COX-2.

NDPK A and NDPK B Impair TGF-␤-dependent Induction of COX-2 Not at the Transcriptional Level but at the TGF-␤ Receptor
Level-With these apparently conflicting results, we conducted reporter assays similar to that used in a previous report to rule out cell typedependent effects, which showed that NDPK A and NDPK B can transactivate the Cox-2 promoter (supplemental Fig. S1). It was therefore thought unlikely that NDPK A and NDPK B act on the Cox-2 promoter to impair transcription. We thus speculated that an alternative mechanism might be that NDPK A and NDPK B attenuate COX-2 expression not at the transcription level but at the receptor level, because a recent report has shown that NM23-H1, a human orthologue of NDPK A, interacts with serine-threonine kinase receptorassociated protein, a TGF-␤ receptor-interacting protein, and negatively regulates a number of TGF-␤ target genes in other cell lines (31).
TGF-␤ receptors can activate both Smad-dependent pathways and Smad-independent pathways (32,33). First, to address whether TGF-␤-induced COX-2 expression is mediated by Smad proteins, a chromatin immunoprecipitation assay using anti-Smad2/3 antibody was done. PCR analysis was performed to amplify the region of the Cox-2 promoter that contains possible Smad binding sites (34), including multiple Smad binding elements (AGAC), one AP-1-like site (TGCGTGG), and one Sp1 site (GGGCGG) (35). In vivo binding of Smad2/3 to the Cox-2 promoter was clearly induced in response to TGF-␤ stimulation (Fig. 6A). In addition, real-time PCR analysis showed that recruitment of Smad2/3 on the promoter is increased by the addition of TGF-␤ in a time-dependent manner (Fig. 6B). These findings suggest that TGF-␤-dependent COX-2 expression is at least in part regulated by Smad2/3/4 pathways.
In the Smad-dependent pathways, the TGF-␤ receptor phosphorylates Smad2 and Smad3. Phosphorylated Smad2 and Smad3 then form oligomers with or without Smad4 that translocate to the nucleus, where they regulate the transcription of downstream genes (32)(33)(34). Thus, we assessed whether NDPK A and NDPK B can suppress each step of these pathways. First, adenoviral overexpression of these factors decreased TGF-␤mediated phosphorylation of Smad2 as shown by Western blot (Fig. 6C). Next, subsequent nuclear translocation of Smad2/3 was impaired as demonstrated by immunofluorescence (Fig.  6D). Moreover, chromatin immunoprecipitation showed that resultant TGF-␤-induced recruitment of Smad2/3 onto the Cox-2 promoter was clearly attenuated by forced expression of NDPK A or NDPK B (Fig. 6E). Collectively, these findings confirmed that NDPK A and NDPK B can suppress TGF-␤-stimulated induction of COX-2 by down-regulating the Smad pathways at the receptor level.
LPS Does Not Up-regulate HNRPAB and NDPKs, and Their Effects on LPS-induced COX-2 Are Limited-To examine the generality and specificity of the involved mechanisms, we assessed whether LPS can up-regulate HNRPAB and NDPKs (Fig. 7A). Although LPS induced transient increase of COX-2, LPS did not show significant effects on the expression levels of HNRPAB, NDPK A, or NDPK B.
Further, we overexpressed HNRPAB and NDPKs in the presence of LPS. As demonstrated in Fig. 7B, although HNRPAB showed no significant effect on the peak induction level of COX-2 (lanes 4 -6), HNRPAB slightly increased COX-2 mRNA at later phases (lanes 7-9), suggesting its mRNAstabilizing effect. However, as a whole, its regulatory ability on LPS-induced COX-2 seems to be limited in this setting. Moreover, overexpression of NDPKs resulted in no suppression or only mild suppression of COX-2 induction as shown in Fig. 7C. This observation that NDPKs did not efficiently attenuate the expression level of LPS-induced COX-2 is consistent with the proposed mechanism that they impair COX-2 expression mainly at the TGF-␤ receptor level.

DISCUSSION
A carefully controlled balance of positive and negative signaling pathways regulates normal physiological cellular activities. Their inappropriate alteration often leads to human diseases. Thus, understanding regulatory signaling networks, especially those of growth factors, is a critical step in elucidating mechanisms that regulate development of pathophysiological conditions. Although the complexity of TGF-␤ signaling is very challenging to analyze, a differential proteomic approach allowed us to discover both positive and negative components of TGF-␤stimulated regulation of COX-2 expression. Although few previous reports have described that TGF-␤ induces COX-2 in several cell lines (18,36,37), surprisingly, almost nothing has been described about underlying molecular mechanisms. This is the first report that shows that HNRPAB is a TGF-␤-induced factor that directly increases the expression level of COX-2 protein. Further, our findings on NDPKs are intriguing, considering their possible roles as negative regulatory components of TGF-␤-dependent pathways. Moreover, this is the first to report on a direct link between TGF-␤-Smad2/3/4 signaling and COX-2 expression (supplemental Fig. S2).
Implications in Fibroproliferative Diseases-Fibroproliferative diseases, including cardiac fibrosis, pulmonary fibrosis, systemic sclerosis, liver cirrhosis, progressive kidney disease, and macular degeneration, are one of the leading causes of morbidity and mortality and can affect all tissues and organ systems (6). TGF-␤ plays a pivotal role in regulating fibrosis and can exert both pro-and antiproliferative effects on fibroblasts. TGF-␤ induces a biphasic response in lung fibroblasts with stimulation of proliferation at low concentrations and inhibition at high concentrations, the latter of which is accounted for by autocrine synthesis of PGE 2 (18). Interestingly, although inappropriate activation of TGF-␤ signaling leads to excess fibrosis in the pathogenesis of various fibrotic diseases, genetic ablation of Cox-2, one of the downstream targets of TGF-␤, also results in excessive fibrosis, as shown in Cox-2 null mice (19,20). Although antifibrotic therapies are considered to be promising, an incomplete understanding of the molecular machinery of TGF-␤-dependent pathways has hindered the development of such therapies.
Here we have demonstrated that TGF-␤ up-regulates HNR-PAB. Of note, HNRPAB is a single protein with multiple functions. Although HNRPAB acts as an RNA-binding protein, it was also identified independently as CArG box-binding factor-A on the basis that it binds to single-stranded and double-  10 versus lanes 11 and 12). Lanes 1-4 are input (1%), confirming that the applied chromatin amounts were the same. The graph on the right shows results of real-time PCR assay. Data were normalized by input DNA.
stranded DNA in a sequence-specific manner as a transcriptional regulator (38 -40). A recent report has demonstrated the key role of this factor in epithelial-mesenchymal transition, a main contributor to tissue fibrosis (40). Thus, HNRPAB is a unique protein that shows bimodal actions; as a profibrotic factor, it is a DNA-binding protein that promotes epithelial-mesenchymal transition, and as an antifibrotic factor, it is an RNA-binding protein that stabilizes mRNA of COX-2. Therapeutic modification of its RNA-binding ability that does not affect its DNA-binding capacity or vice versa may be potentially exploitable for targeted therapeutic intervention against fibroproliferative diseases.
Possible Roles in Carcinogenesis and Metastasis-The significance of TGF-␤ signaling in carcinogenesis and metastasis has been widelyinvestigated.Ingeneral,TGF-␤ is a tumor suppressor early in carcinogenesis, inhibiting growth and promoting differentiation, and then acts as a tumor promoter later in cancer development to promote growth, survival, invasion, and metastasis (2,8,9,11). COX-2 is also involved in cancer progression. Clinical and experimental data have demonstrated that increased COX-2 expression correlates with poor survival in various cancers, and selective inhibitors of COX-2 inhibit tumor growth (15). Stromal fibroblasts as well as cancer cells per se have been shown to be the source of COX-2 (15,22,23). Given that tumor-associated fibroblasts are important contributors to the tumor microenvironment (9, 10, 21), TGF-␤-mediated COX-2 expression in fibroblasts is a topic that deserves further attention.
An earlier proteomic report identified HNRPAB as one of eight up-regulated genes in the gene expression signature associated with various tumor metastases (41). Notably, the authors suggested that a considerable proportion of the gene expression signature seems to be derived from stromal components of tumors. Thus, not only TGF-␤ and COX-2 but also HNR-PAB is probably up-regulated in the tumor microenvironment. Although nothing has been shown concerning their relationship in tumor-associated fibroblasts, our findings lead us to propose a straightforward pathway by which TGF-␤ stimulates HNRPAB, which in turn increases COX-2 in tumor-associated fibroblasts, thus resulting in tumor development and progression.
Bifunctional Effects of NDPKs-NDPK A and NDPK B are members of the NDPK/NM23 family, which have multiple activities ranging from DNA-binding in the nuclei to proteinprotein interaction in the cytosol. These family proteins have been shown to be involved in many cellular processes, including cell growth, differentiation, metastasis, and development (42)(43)(44)(45). Although our findings in the present study are based on experiments using fibroblasts, this novel pathway potentially impacts on the biology of tumor cells. NDPK was first characterized as a metastasis suppressor in melanoma (46). Its high expression was associated with less metastatic potential in various tumors, including melanoma, breast cancer, colon cancer, and liver cancer. In contrast, increased expression in prostate cancer, neuroblastoma, and lymphoma is related to tumor aggressiveness (42,47). Despite the widely accepted notion of close involvement of NDPKs in tumor progression, no clear consensus has been made regarding its precise roles in tumor biology, or rather, NDPKs are considered to have multifaceted roles in a variety of tumors in a highly context-dependent manner. We note that an earlier study showed that NM23-H1, a human orthologue of NDPK A, up-regulates COX-2 promoter activity in reporter assays (29). This is in agreement with the tumor-progressive property of NDPKs, considering the effect of COX-2 on tumor progression. On the other hand, our finding that NDPKs attenuate COX-2 expression is consistent with classically proposed functions of NDPK as a metastasis suppressor and possible tumor suppressor. The molecular basis of multiple NPDK-dependent signaling pathways is, however, more complex, and further investigation is warranted to better understand the precise underlying mechanisms.
In conclusion, we have shown a new regulatory pathway of TGF-␤-dependent COX-2 expression by HNRPAB and NDPKs, which provides a new perspective in understanding the biology of various diseases, including fibrotic and cancer diseases.