The cell adhesion molecule, GP116, is a new CD44 variant (ex14/v10) involved in hyaluronic acid binding and endothelial cell proliferation.

In this study we have found that endothelial cells from different origins all contain a CD44-related transmembrane glycoprotein, named GP116. Using a bovine aortic endothelial cell line and a standard pulse-chase protocol, we show that GP116 is synthesized as a 52-kDa nascent polypeptide precursor (p52) which is processed to GP116 as follows, p52 --> p63/65 --> p82 --> p100 --> GP116. GP116 contains approximately 8 N- and approximately 11 O-linked oligosaccharide chains (but lacks glycosaminoglycans) and interacts directly with the cytoskeletal protein, ankyrin, both in vitro (Kd approximately 1.2 nM) and in vivo. The results of GP116 amino acid composition, reverse transcriptase-polymerase chain reaction, Southern blot, Northern blot, cloning, and sequence analyses indicate that endothelial cells express this new CD44 variant that contains an exon having significant homology with human CD44 exon 14 (ex14/v10). GP116, designated as CD44 (ex14/v10), has been shown to be a major hyaluronic acid (HA) receptor (Kd approximately 0.5-0.8 nM) responsible for cell adhesion. Most importantly, we have found that the interaction between CD44(ex14/v10) and HA or a small fragment of HA (10-15 disaccharide units) induces a mitogenic response in endothelial cells. These findings suggest that this CD44 variant plays an important role in regulating endothelial cell proliferation.

CD44 denotes a family of glycoproteins that are expressed in a variety of cells and tissues derived from hemopoietic, epithelial, endothelial, and mesodermal origins (1)(2)(3)(4)(5). These CD44related glycoproteins are known to mediate both cell adhesion to extracellular matrix components (e.g. hyaluronic acid (HA), 1 fibronectin, and collagen) and homotypic cell aggregation (6 -13). Previously, we have demonstrated that CD44 interacts directly with ankyrin and contains an ankyrin binding region in its cytoplasmic domain (14 -16). Post-translational modification of CD44's cytoplasmic domain by protein kinase C (17), acylation (18), or GTP binding (19) has been shown to enhance the binding between CD44 and ankyrin. These observations suggest that CD44 not only functions as an adhesion protein but may also play an important role as a signal transducing molecule. The signaling properties of CD44 may be involved in a variety of cellular activities including lymphocyte activation, lymphocyte homing, hemopoiesis, cell migration, and tumor metastasis (1,20,21). Recent studies indicate that certain CD44 isoforms (or variants) are expressed on the surface of tumor cells during metastasis, particularly during the progression of various carcinomas (22)(23)(24)(25)(26).
One of the distinct features of CD44 isoforms is the enormous heterogeneity in the molecular masses of these proteins. It is now known that all CD44 isoforms are encoded by a single gene that contains 19 exons (27,28). Out of the 19 exons, 12 exons can be alternatively spliced (28). Most often, the alternative splicing occurs between exons 5 and 15 leading to an insertion in tandem of one or more variant exons (v1-v10, or exons 6 through exons 14 in human cells) within the membrane proximal region of the extracellular domain (28). The variable primary amino acid sequence of different CD44 isoforms is further modified by extensive N-and O-glycosylations and glycosaminoglycan (GAG) additions (10, 29 -31). In addition, a recent report indicates that one of the CD44 isoforms, containing exon 7 or v3 (ex7/v3), may bind heparin binding growth factor since the ex7/v3 exon-coded region contains heparan sulfate addition sites (29,30). Since the biological response(s) of these cells (expressing CD44(ex7/v3) isoforms) to heparin binding growth factor has not been determined, the functional role of CD44(ex7/v3) isoforms is not certain at the present time.
In order to gain a better understanding of CD44's biological functions, it is necessary to systematically isolate and characterize the various CD44 isoforms. In this paper we have used a bovine endothelial cell line (GM7372A) and a variety of techniques (including a standard pulse-chase protocol, chemical composition analysis, RT-PCR, one-step cloning, and sequencing) to isolate and characterize GP116, a new CD44 variant that contains a sequence encoded by exon 14 (variant exon v10). We have found that GP116, designated as CD44(ex14/ v10), is expressed in a variety of endothelial cells and is involved in HA-mediated cell adhesion events. Most importantly, CD44(ex14/v10) also appears to be required for the mitogenic responses stimulated by HA and specific fragments of HA in endothelial cells. These findings suggest that CD44(ex14/v10) may play an important role in regulating HA-mediated endothelial cell proliferation during normal vasculogenesis, tumor angiogenesis, and atherosclerosis.

MATERIALS AND METHODS
Cell Culture and Monoclonal Antibodies-The chemically transformed bovine aortic cell line (GM7372A) was obtained from the Institute for Medical Research (Camden, NJ). Bovine adrenal medulla endothelial cells (EJG) and bovine pulmonary endothelial cells (CPA47) were obtained from American Type Culture Collection. These endothe-lial cells (e.g. EJG and CPA47) were primary cultures with low passages and were derived from endothelium of different tissues. GM7372A, EJG, and CPA47 cells were grown in Eagle's minimum essential medium (EMEM) supplemented with Earle's salt solution, essential and nonessential amino acids, vitamins, and 10% fetal bovine serum. Monoclonal rat anti-CD44 antibody (isotype, IgG 2b ; obtained from CMB-TECH, Inc., Miami, FL) used in this study recognizes a common determinant of the CD44 class of glycoproteins including CD44s and other variant isoforms (32) and is capable of precipitating all CD44 variants. Monoclonal mouse anti-ankyrin antibody (ANK016) was purified from the hybridoma culture supernatants by ammonium sulfate fraction and DEAE-Sepharose chromatography.
Metabolic Labeling of Proteins-Confluent endothelial cells (Ϸ1 ϫ 10 6 /60-mm dish) were incubated in L-methionine-free DMEM (Life Technologies, Inc.) at 37°C for 4 h followed by pulse labeling with Tran 35 S-label (200 Ci/ml; ICN Pharmaceuticals, Costa Mesa, CA) at 37°C for various times as indicated. For pulse-chase experiments, the medium containing radioactive amino acid was replaced with DMEM containing 10 mM unlabeled methionine. The incubation was continued for various times as indicated. Alternatively, endothelial cells were incubated with sulfate-free Joklik medium for 6 h followed by ( 35 S0 4 ) 2labeling (300 Ci/ml). To study N-and O-glycosylation of CD44, endothelial cells were preincubated at 37°C for 4 h in L-methionine-free DMEM containing tunicamycin (2 g/ml) for 4 h or p-nitrophenyl-␣-Nacetyl-␣-D-galactosaminide (2 mM, Sigma) for 18 h, respectively, followed by Tran 35 S labeling.
Cell and Protein Iodination-Endothelial cells suspended in PBS were surface 125 I-labeled using IODO-GEN at 4°C for 1 h. Following labeling, cells were washed in PBS to remove free 125 I. Purified ankyrin and goat anti-mouse IgG were iodinated using IODO-GEN as described previously (33).
Immunoprecipitation-The radioactively labeled cells were washed in phosphate-buffered saline, pH 7.2 (PBS), and solubilized in RIPA buffer (10). The solubilized extracts were incubated with rat anti-CD44 antibody at 4°C for 15 h followed by incubation with goat anti-rat IgG agarose beads at 4°C for 90 min as described before (10). The immunoprecipitates were analyzed by 7.5% SDS-polyacrylamide gel electrophoresis followed by fluorography. In some cases, endothelial cells were solubilized in 50 mM Tris⅐HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100 buffer, and immunoprecipitated using rat anti-CD44 antibody followed by goat anti-rat IgG. The immunoprecipitated material was solubilized in SDS, electrophoresed, and blotted onto the nitrocellulose. After blocking the nonspecific sites with 3% bovine serum albumin, the nitrocellulose filter was incubated with mouse anti-ankyrin antibody (ANK016) (5 g/ml) at room temperature followed by incubation with horseradish peroxidase-conjugated rabbit anti-mouse IgG (1:10,000 dilution) at room temperature for 1 h. The blots were developed using Renaissance chemiluminescence reagent (DuPont NEN) according to the manufacturer's instructions.
Reverse Transcriptase PCR and One-step Cloning-Total RNA was extracted from various endothelial cells as described by Chomczynski and Sacchi (34). Approximately 3 g of total RNA was used to synthesize first strand oligo(dT)-primed cDNA, at 42°C for 1 h, using a reverse transcription system (Promega, Madison, WI) containing avian myeloblastosis virus reverse transcriptase. Following first strand synthesis, PCR amplification of cDNA was carried out with an initial melting of the RNA/cDNA hybrid at 94°C for 30 s, annealing at 60°C for 1 min, and polymerization at 72°C for 1 min. The PCR primers used in this study were designed to amplify the region of cDNA open reading frame involved in the alternative splicing of several exons (e.g. exons 6 -14). Specifically, the bovine CD44 cDNA primer pair was exon 5 (exon 5, 5ЈGCACTTCAGGAGGCTACGC3Ј) and exon 15 (5ЈAGCTGATTCAGAT-GCGTGAG3Ј) (35). The PCR products were one-step cloned using TAcloning kit (Invitrogen, San Diego, CA) and sequenced by dideoxy sequencing method.
Southern Blot Analysis-CD44 transcript(s) containing exon 14 sequence were amplified by RT-PCR using an exon 14 specific pair designed based on the human CD44 cDNA sequence. The left primer was designed to anneal at the junction of exons 5 and 14 and had the following sequence: 5ЈATCCATGAGTGGTATGGGAC3Ј. The right primer with sequence 5ЈTTAGAGTTGGAATCTCCAAC3Ј was designed such that it would anneal within exon 14. The RT-PCR reaction was carried out as described above with the exception that the annealing temperature was 50°C. The PCR products were separated on a 2% agarose gel, blotted on to the nitrocellulose filter, and hybridized to [ 32 P]dCTP-labeled PCR-amplified CD44(ex14/v10) cDNA. In a control experiment, CD44 transcripts were amplified using the exon 5 and 15 pair, and the products were analyzed by Southern hybridization using the mixture of PCR-amplified CD44s and CD44(ex14/v10) cDNAs as probes. The negative control included RT-PCR analysis carried out in the absence of the reverse transcriptase.
Northern Blot Analysis-Twenty g of total RNA isolated from bovine endothelial cells (GM7372A) was separated on a 1.2% agaroseformaldehyde gel, transferred to nylon membrane, and hybridized to either a mixture of PCR-amplified CD44s and CD44(ex14/v10) cDNA probes or CD44(ex14/v10) probe alone at 42°C for 16 h.
Immunohistochemistry and RT in Situ PCR-Bovine aorta was purchased from Pel Freeze Co. (Rogers, AR). The tissue specimen was fixed in formalin, embedded in paraffin, and sectioned at 4.0-m thickness. The tissue section was subjected to immunoperoxidase staining using rabbit anti-factor VIII antibody (Accurate Chemicals, Westbury, NY) to visualize the endothelium. For RT in situ PCR analysis of CD44(ex14/ v10) transcript in bovine aortic endothelium, a protocol by Nuovo (36) was used with following modifications. In the PCR mixture we used the direct incorporation of fluorescein isothiocyanate-digoxigenin-11-dUTP and the CD44 exon 14-specific primers (described above), and 20 PCR cycles were carried out at 50°C for 2 min and 94°C for 1 min. The fluorescent signals were then analyzed by a laser scanning Confocal microscope (MultiProbe 2001 Invert CLSM System, Molecular Dynamics) using a 63 ϫ-oil immersion and an imaging processing device. Images were photographed with Kodak TMAX100 film.
Purification and Chemical Composition Analysis of GP116 -GP116 was purified from endothelial cell plasma membranes using nonionic detergent Triton X-100 extraction and sequential wheat germ agglutinin-Sepharose and anti-CD44 immunoaffinity chromatographies as described previously (10,19). Approximately 27 g of purified GP116 was hydrolyzed in 4 N methansulfonic acid at 100°C for 20 h. After neutralizing with sodium citrate, the sample was analyzed by a JEOL5AH amino acid analyzer using a single column and a ninhydrin detection method in the presence of amino acid standards.
Binding of 125 I-Labeled Ankyrin to GP116 (CD44(ex14/v10))-Ankyrin was purified from human erythrocyte ghosts as described by Bennett and Stenbuck (37). Ankyrin was 125 I-labeled using IODO-GEN (Ϸ5,000 cpm/ng of protein). Purified GP116 (CD44(ex14/v10)) protein (Ϸ10 ng) conjugated to anti-CD44 immunoaffinity beads was incubated with various concentrations of 125 I-labeled ankyrin in binding buffer (20 mM Tris⅐HCl, pH 7.4, 150 mM NaCl, 0.1% bovine serum albumin, and 0.05% Triton X-100) at 4°C for 5 h under equilibrium conditions. Following binding, the beads were washed and the bead-bound radioactivity was determined. The nonspecific binding was determined in the presence of 100-fold excess of unlabeled ankyrin. The nonspecific binding was 20 -30% of the total binding and was subtracted from the total binding.
Preparation of Hyaluronic Acid (HA) Fragments-Human umbilical cord HA (500 mg) was dissolved in 50 ml of 0.1 M acetate buffer, pH 5.4, containing 0.15 M NaCl and digested with 20,000 units of testicular hyaluronidase at 37°C. Ten milliliter aliquots were removed after 2-, 4-, 6-, 8-, and 24-h intervals, and the reaction was terminated by adding trichloroacetic acid at 10% final concentration (v/v). After standing at 4°C for at least 4 h, any precipitate was removed by centrifugation at 2500 ϫ g for 30 min. The supernatants were pooled, dialyzed extensively against distilled water, recentrifuged, and freeze-dried. The preparation was dissolved in 10 ml of 0.1 M acetic acid and applied to a column (2.0 ϫ 150 cm) of Sephadex G-50. The column was eluted in 0.1 M acetic acid at a flow rate of 10 ml/h, and 5-ml fractions were collected. Each fraction was assayed for hyaluronic acid content, and size ranges of the fragments were determined as described previously (38). Three fractions, F1, F2 and F3, were characterized and found to contain HA fragments of 10 -15, 2-3, and Ϸ2 disaccharide units, respectively.
[ 3 H]Hyaluronic Acid (HA) Binding to Endothelial Cells-[ 3 H]HA was prepared from rat fibrosarcoma cells by [ 3 H]D-glucosamine labeling as described previously (4). The final product had a specific activity of 1.2 ϫ 10 4 cpm/g and a molecular mass of Ϸ10 6 daltons. The molecular mass was determined by gel filtration column chromatography on Sephacryl S1000 column as described by West and Kumar (38). Endothelial cells (Ϸ4 ϫ 10 5 cells) grown in 6-well culture plates were incubated with various concentrations (0.3-4.8 g/ml) of [ 3 H]HA (1 ϫ 10 5 dpm/g) in binding buffer (PBS containing 0.2% bovine serum albumin) at 4°C for 4 h. Alternatively, cells were incubated with [ 3 H]HA (Ϸ1 g) in the presence of various concentrations of unlabeled HA or HA fragments in the binding buffer at 4°C for 4 h. In some cases, cells were first treated with various reagents such as monoclonal rat anti-CD44 antibody (50 g/ml), cytochalasin D (20 g/ml), colchicine (1 ϫ 10 Ϫ5 M), and W-7 (20 M) at room temperature for 30 min followed by [ 3 H]HA binding. Following binding, the cells were washed three times in the binding buffer and solubilized in 20 mM Tris⅐HCl, pH 7.4, solution containing 0.1% SDS. The radioactivity associated with solubilized cell extracts was counted in a liquid scintillation counter. The nonspecific binding was determined in the presence of 100-fold excess of unlabeled HA and was subtracted from the total binding.
Cell Adhesion Assay-Endothelial cells were metabolically labeled with Tran 35 S-label (20 Ci/ml) as described above. After labeling, the cells were washed in PBS and incubated in PBS containing 5 mM EDTA at 37°C to obtain a nonadherent single cell suspension. Labeled cells (Ϸ9.1 ϫ 10 5 cpm/10 5 cells) were incubated on HA-coated plates (prepared as described previously) (39) at 4°C for 30 min either alone or in the presence of soluble HA/HA fragments or rat anti-CD44 antibody (50 g/ml). Following incubation, the wells were washed three times in PBS, the adherent cells were solubilized in PBS containing 1% SDS, and the well-bound radioactivity was determined by liquid scintillation counting.
Proliferation Assay-Confluent endothelial cells were incubated in either serum-free or EMEM containing 5% fetal bovine serum and various concentrations of HA or HA fragments for 18 h. The concentrations of HA and HA fragments were expressed in terms of total uronate content but hyaluronate concentrations may be calculated by multiplying these values by 2.06 (38). Following incubation with HA or HA fragments, cells were pulsed with [ 3 H]thymidine (0.5 Ci/ml) for 2 h, and [ 3 H]thymidine incorporation was estimated as described before (40). Alternatively, cells were preincubated in serum-free EMEM in the presence or absence of rat anti-CD44 antibody (50 g/ml) for 8 h followed by incubation with HA and HA fragments in serum containing EMEM. The [ 3 H]thymidine incorporation assay was performed as described above.

Detection of Surface CD44 Molecules in Endothelial Cells
Previously, we have found that bovine aortic endothelial cells (GM7372A) express a CD44-type surface glycoprotein with a molecular mass of 116 kDa (designated as GP116) (Fig. 1, lane B) (4). In this study we report that this molecule also occurs on the surface of a variety of endothelial cells derived from different tissues including bovine adrenal medulla capillaries (EJG) (Fig. 1, lane C) and bovine lung arteries (CPA47) (Fig. 1, lane D). The detection of GP116 in these endothelial cells by anti-CD44-mediated immunoprecipitation is specific since no protein is immunoprecipitated in the presence of normal IgG (Fig.  1, lane A). Therefore, GP116 appears to be a common CD44 isoform present on many bovine endothelial cells.

Biosynthesis and Processing of Endothelial Cell GP116
In order to investigate the biosynthetic processing of GP116, we have utilized a bovine aortic endothelial cell line (GM7372A) and a standard pulse-chase approach. Initially, cells were pulse-labeled with Tran 35 S-label for 10 min followed by various chase times (t ϭ 0, 5, 10, 15, 20, and 30 min) in growth medium containing unlabeled methionine. The 35 Slabeled cell extracts were then processed for anti-CD44-mediated immunoprecipitation. As shown in Fig. 2A, after a 10-min pulse followed by 0 or 5-min chase, a doublet (p63/p65 (mass of 63 and 65 kDa, respectively)) and a single polypeptide of Ϸ82 kDa (p82) are the major CD44 species present in the anti-CD44 immunoprecipitated material (lanes 1 and 2). At 10-, 15-, and 20-min chase times, all of the p63/p65 doublet appears to be converted to p82 suggesting that p63/p65 is a precursor of p82 ( Fig. 2A, lanes, 3, 4 and 5). Densitometric analysis indicates that the t1 ⁄2 of p63/p65 conversion to p82 is Ϸ7.5 min (data not shown). An additional 30-min chase in unlabeled methionine results in the appearance of GP116 suggesting that GP116 is generated from p82 by further post-translational processing ( Fig. 2A, lane 6). These findings are corroborated by a longer pulse-chase analysis. For example, when endothelial cells are pulse-labeled for 20 min followed by a 0-min chase, p82 is the major CD44 species immunoprecipitated by anti-CD44 antibody (Fig. 2B, lane 1). A minor transient precursor, p100 (Ϸ100 kDa), is also immunoprecipitated during the same time interval (Fig. 2B, lane 1). Subsequently, both p82 and p100 appear to be completely converted to GP116 by a 30-min chase (Fig.  2B, lane 2). At this time, newly synthesized GP116 also appears on the cell surface since it is sensitive to trypsin treatment of whole cells (data not shown). In addition, we have found that the turnover rate of GP116 is relatively slow. As shown in Fig.  2B (lanes 3-5), Ͼ90% of the newly synthesized GP116 is still detected on the cell surface after a 6-h chase in the unlabeled medium. Preliminary data indicate that the t1 ⁄2 of GP116 turnover is Ϸ10 h and that GP116 is degraded via the lysosomal pathway (data not shown).

Post-translational Modifications of Endothelial Cell GP116
CD44 isoforms are known to be extensively modified posttranslationally. Some of the post-translational modifications involve N-and O-glycosylations and the additions of GAG (e.g. heparan sulfate and chondroitin sulfate) (10, 29 -31). To characterize the N-glycosylation of GP116, endothelial cells were labeled with Tran 35 S-label in the presence of tunicamycin, an N-glycosylation inhibitor (38). As shown in Fig. 3A (lanes 1 and  2), tunicamycin effectively blocks the formation of both p82 and GP116 and induces the formation of two new polypeptides, p84 and p52 (Fig. 3A, lane 2). The reason for the apparent double band close to 52 kDa (Fig. 3A, lane 2) is not clear at the present time. It is possible that the smaller protein is a degradative product of the larger band. The molecular mass difference between these two bands appears to be only Ϸ2 kDa. It is unlikely that the oligosaccharide differences of these two bands (the mass of each oligosaccharide is Ϸ3-4 kDa) are responsible for the size differences in these two proteins. The concentration of tunicamycin used in this experiment is within the standard range known to inhibit N-glycosylation (41). The fact that the molecular mass of GP116 is decreased to 84 kDa, which contains O-linked oligosaccharides (Fig. 3C) in the presence of tunicamycin (2 g/ml), suggests that this inhibitor does not block O-glycosylation.
In an earlier study, GP116 was found to be converted to a 70-kDa product following treatment with N-glycanase (an enzyme that removes the N-linked oligosaccharide from the gly- coproteins) (4). In this study we have found that the apparent molecular mass of the GP116 species following tunicamycin treatment is 84 kDa. Since the commercial preparations of N-glycanase do not have consistent quality control and often contain other unknown degradative enzymatic activities, we feel that the 84-kDa molecular mass resulting from tunicamycin treatment is more accurate than that obtained by N-glycanase treatment. This is further supported by the chemical composition data of GP116 (Table I, discussed below) that indicates that this protein contains Ϸ8 -9 N-linked oligosaccharide chains. The molecular mass of each N-linked oligosaccharide chain is assumed to be Ϸ3-4 kDa. Therefore, the Nglycosylation accounts for Ϸ32 kDa molecular mass of GP116.
To determine the O-glycosylation of GP116, endothelial cells were treated with p-nitrophenyl-␣-N-acetylgalactosaminide, an O-glycosylation inhibitor (42), followed by metabolic labeling for 20 min and a chase in unlabeled medium for 60 min. As shown in Fig. 3B (lanes 1 and 2), the O-glycosylation inhibitor blocks the formation of GP116 and induces formation of a new species with molecular mass Ϸ86 kDa (p86). Thus, inhibition of O-glycosylation results in a reduction of the molecular mass of GP116 by Ϸ30 kDa (e.g. p86).
A number of CD44 isoforms have been shown to contain sulfated oligosaccharides (5). In this study we labeled endothelial cells with ( 35 S0 4 ) 2Ϫ in the presence or absence of either tunicamycin or p-nitrophenyl-␣-N-acetylgalactosaminide. As shown in Fig. 3C, GP116 is labeled with ( 35 S0 4 ) 2Ϫ (Fig. 3C, lane  1). In the presence of tunicamycin, a sulfate-labeled 84-kDa protein (Fig. 3C, lane 2) is detected indicating that GP116 does not contain sulfated N-linked oligosaccharides. However, no sulfate labeling (Fig. 3C, lane 3) is observed in the presence of the O-glycosylation inhibitor suggesting that GP116 contains sulfated O-linked oligosaccharides. In addition, our results show that GP116 does not contain any sulfated GAGs since the sulfate labeling of GP116 or its precursors is completely blocked in presence of the O-glycosylation inhibitor. This finding is further confirmed by treating sulfate-labeled GP116 with heparitinase and chondroitin ABC lyase. No loss of label was observed following these enzyme treatments (data not shown).

Molecular Biological Analysis of Bovine Endothelial Cell GP116
As described above, GP116 is synthesized from a 52-kDa polypeptide precursor, whose molecular mass is clearly larger than that of the lymphocyte nascent polypeptide precursor of CD44(GP85) (Ϸ42 kDa) (10). This suggests that GP116 is an isoform of CD44 that differs from CD44(GP85) in amino acid sequence. To test this possibility, we have used both molecular biological and cytochemical approaches as described below.

Identification of CD44 Variant Transcripts Expressed in Bovine Endothelial Cells by RT-PCR, Southern Blot Analyses, cDNA Cloning, and Nucleotide Sequencing
The RT-PCR analysis of total bovine endothelial RNA revealed two specific amplification products, 156 and 351 bp using the exon 5 and 15 primer pair (that amplifies all CD44 splice variants) (Fig. 5A, lane 2). These two PCR products were subjected to one-step cloning into pCRII vector and sequenced. Nucleotide sequence data indicate that 156 bp represents the hemopoietic or CD44 standard (CD44s) isoform (Fig. 4A, a). The 351-bp amplimer has a sequence similar to human CD44 exon 14 (or v10) (with 52% sequence homology) (Fig. 4, A and  B). This particular exon has not been identified previously in bovine CD44 cDNA.
Further analysis using the exon 14-specific primers reveals a 186-bp PCR product in endothelial cells (GM7372A) that hybridizes to the CD44(ex14/v10) cDNA (Fig. 5A, lane 1). The use of the exon 5 and 15 primer pair also confirms the presence of two PCR products, 156 and 351 bp, in endothelial cells (GM7372A) (Fig. 5A, lane 2). As a control, we have carried out the RT-PCR reaction in the absence of the reverse tran- scriptase. No PCR products are detected in these samples (Fig. 5A, lane 3).

Detection of CD44 Variant Transcripts Expressed in Bovine Endothelial Cells by Northern Blot and RT in Situ PCR
Northern Blot Analysis-In the Northern blot analysis, a mixture of both CD44s and CD44(ex14/v10) cDNA probes was used to detect all of the CD44-related transcripts expressed in endothelial cells (GM7372A). Our data indicate that these probes detect two transcripts of 2.2 kb (indicated by b in Fig.  5B(i), lane 2) and 2.8 kb (indicated by a in Fig. 5B(i), lane 2).
Additional Northern blot analysis using the exon 14 (v10)specific probe also reveals two similar transcripts (e.g. 2.2-and 2.8-kb messages) as shown in Fig. 5B(i), lane 1. Among these two transcripts, the 2.8-kb message appears to be the major one as it hybridizes more strongly than the 2.2-kb transcript. In these experiments, equal loading of RNA samples (revealed by the presence of 28 S and 18 S rRNA, respectively) in each gel lane (Fig. 5B(ii), lanes 1 and 2) was routinely monitored by ethidium bromide staining. These results suggest that a CD44 exon 14 (v10)-containing transcript is expressed in bovine endothelial cells and is probably the major CD44-related mRNA in these cells.  14 protein GP116 (Ϸ27 g) was hydrolyzed and analyzed in an amino acid analyzer as described under "Materials and Methods." The amounts of cysteine, leucine, phenylalanine, valine, and glycine could not be estimated accurately in this analysis. The chemical compositions of putative CD44 exon 14 protein were deduced from the cDNA sequence data. The * indicates amino acids and sugar residues in GP116 whose composition is significantly different between CD44s and CD44(ex14/v10) proteins. The compositions of these amino acids and sugar residues in GP116 match more closely to those present in CD44(ex14/v10). In order to calculate the number of N-linked oligosaccharide chains, we have first assumed a tri-and tetra-antenary structure for each oligosaccharide chain. The tri-to tetra-antenary structures of oligosaccharides contain approximately 5 and 6 N-acetyl-D-glucosamine residues, respectively. Based on this assumption, the GP116 molecule is proposed to contain 8 -9 N-linked oligosaccharides. The structure of the O-linked oligosaccharide chains contains a single ␤-D-galactosamine residue initially added onto a Ser/Thr residue followed by the addition of other variable sugar residues. Thus, there is only one ␤-D-galactosamine residue per O-linked oligosaccharide chain.
Since the chemical composition analysis shows the presence of 10.6 D-galactosamine residues in GP116 molecule, it indicates that this protein contains 10 -11 O-linked residues.

Immunohistochemistry and RT in Situ PCR-The in vivo
expression of CD44(ex14/v10) transcript was also analyzed in bovine aortic endothelium by RT in situ PCR analysis. The location of endothelial lining in bovine aorta is established by immunoperoxidase staining using anti-factor VIII antibody (Fig. 6A). The RT in situ PCR using the exon 14-specific primer pair shows the presence of strong fluorescent signal in the endothelium lining (Fig. 6B). This signal is absent in the negative control samples where the reverse transcriptase was eliminated during the RT reaction (Fig. 6C).

Chemical Composition Analysis of GP116
Further chemical composition analysis indicates that purified GP116 is different from CD44s and shares a great deal of similarity with CD44 exon 14(v10) ( Table I). For example, similar to CD44 exon 14 (or v10) protein, GP116 is rich in Ser and Thr and contains 8 N-linked and 11 O-linked oligosaccharide chains (assuming a tri-tetra-antenary structure for Nglycosylation and 1 galactosamine residue per O-linked oligosaccharide chain).
Taken together, the results of GP116 amino acid assays, RT-PCR, Southern blot, Northern blot, and RT in situ PCR analyses strongly suggest that endothelial cells express a new CD44 variant containing an exon that shares significant homology with human CD44 exon 14(ex14/v10). GP116 is therefore designated as CD44(ex14/v10).

Surface GP116, CD44(ex14/v10) Interaction with Hyaluronic Acid (HA) and HA Fragments Binding and Adhesion between GP116, CD44(ex14/v10), and HA/HA Fragments
It is now well established that HA is one of the ligands recognized by surface CD44 molecules. Furthermore, surface CD44 mediates adhesion of cells to HA-coated surfaces (16,24,43). To characterize the binding and adhesion interactions between surface GP116, CD44(ex14/v10), and HA, we initially performed [ 3 H]HA equilibrium binding studies with bovine aortic endothelial cells (GM7372A). As shown in Fig. 7A, [ 3 H]HA binds to endothelial cells specifically and in a dose-dependent manner that is saturable. Scatchard plot analysis (Fig.  7B) indicates that there is single class of high affinity HA binding sites on endothelial cells. Further data analysis indicates that there are Ϸ10,000 HA binding sites each on bovine aortic endothelial cell (GM7372A) with a dissociation constant (K d ) of Ϸ0.8 nM. To determine whether GP116, CD44(ex14/ v10), is the major HA receptor involved in HA binding and the adhesion of endothelial cells to HA-coated surfaces, cells were preincubated with monoclonal rat anti-CD44 antibody followed by the addition of [ 3 H]HA or incubation of cells on the HAcoated dishes. As shown in Table II, monoclonal rat anti-CD44 antibody (but not normal rat IgG control) significantly inhibits [ 3 H]HA binding (Ϸ70% inhibition) and adhesion to HA-coated surfaces (Ϸ75% inhibition). These results indicate that GP116, CD44(ex14/v10), must be the major HA receptor on the endothelial cell surface.
It has been shown that HA fragments, between 3 and 25 disaccharide units in length, promote angiogenesis by inducing endothelial cell proliferation and migration (38,44). In this part of the study we decided to test both the binding and adhesion interaction between surface GP116, CD44(ex14/v10), and various HA fragments (e.g. F1, F2, and F3 that consist of 10 -15, 2-3, and Ϸ2 disaccharide unit fragments, respectively). First, we examined the binding of [ 3 H]HA to endothelial cells in the presence of various concentrations of unlabeled HA fragments F1, F2, and F3. As a positive control, unlabeled undigested HA (HA polymer) was also included in the competition studies. As shown in Fig. 8 (Fig. 8). The dissociation constants (K d ) for the binding of HA polymer or F1, F2, and F3 HA fragment to endothelial cells, as determined from the binding competition curves, are 0.5, 0.8, 35, and 90 nM, respectively. It has been shown that at least a hexasaccharide (i.e. three disaccharide units) is required to efficiently compete with native HA to bind cell surface HA receptor (45). The dissociation constants of F2 fragment (2-3 disaccharides units tetra/hexasaccharides, 35 nM) and the F3 fragment (Ϸ2 disaccharide units tetrasaccharides, 90 nM) are 70 and 180 times higher than native HA (K d Ϸ0.5 nM) for HA binding. However, the dissociation constant (K d 0.8 nM) of F1 fragment binding to endothelial cells is very similar to that of native HA. These results suggest that F1 (but not F2 and F3) fragment is an efficient competitor of native HA for the binding to endothelial cells.
To determine whether HA fragments are involved in HAmediated cell adhesion events, bovine endothelial cells (GM7372A) were incubated on HA-coated plates in the presence or absence of the three HA fragments (e.g. F1, F2, and F3).
Our results indicate that the F1 fragment (but not F2 and F3) blocks HA-mediated cell adhesion (Ϸ 90% inhibition) as effectively as the intact soluble HA polymer (data not shown).

FIG. 5. RT-PCR, Southern, and Northern blot analyses of CD44 transcripts in bovine endothelial cells.
A, RT-PCR and Southern blot analysis. Total RNA isolated from bovine endothelial cells (GM7372A) was reverse-transcribed and subjected to PCR using PCR primer pairs designed to amplify either CD44 exon 14/v10 or CD44 variants. Subsequently, RT-PCR products were analyzed by Southern blot hybridization as described under "Materials and Methods." Lane 1, RT-PCR product generated by CD44 exon 14/v10-specific primer pairs. Lane 2, RT-PCR product generated by CD44 exon 5 and 15 primer pairs. Lane 3, as a control, RT-PCR was carried out in the absence of reverse transcriptase. B, Northern blot analysis. Twenty g of total RNA isolated from bovine endothelial cells (GM7372A) was separated on a 1.2% agarose-formaldehyde gel, transferred to nylon membrane, and hybridized to PCR-amplified CD44 cDNA probes as described under "Materials and Methods." (i) Lane 1, hybridization with CD44 ex14/v10-specific cDNA probe as described in Fig. 4A. Lane 2, hybridization with CD44s and CD44 ex14/v10 cDNA amplified by exon 5 and 15 primer pair as described in

Effect of HA and HA Fragments on GP116, CD44(ex14/ v10)-associated Endothelial Cell Mitogenic Response
To investigate the functional significance of HA polymer or HA fragment binding to endothelial cells, we determined whether the intact HA polymer or F1, F2, and F3 fragments can stimulate a mitogenic response in endothelial cells. The mitogenic response was examined using [ 3 H]thymidine incorporation as a measure of DNA synthesis in either serum-containing or serum-free medium. As shown in Fig. 9A, in the serum-containing medium both the intact HA polymer and F1 fragment induce DNA synthesis in endothelial cells in a dosedependent manner. The intact HA polymer and F1 fragment induce a maximum increase in DNA synthesis of approximately 1.8-and 3.5-fold, respectively (Fig. 9A). It is interesting to note that the concentrations of the HA polymer (Ϸ0.5 nM) and F1 fragment (Ϸ0.2 nM) required for a maximum mitogenic response are within the range of values obtained for the dissociation constant for their binding with bovine aortic endothelial cells (GM7372A). Conversely, F2 and F3 HA fragments that bind endothelial cells with low affinity (K d Ϸ35 and 90 nM, respectively) display little mitogenic response in these cells. It also may be noted that the HA polymer inhibits endothelial cell growth at concentrations Ն1 g/ml that has been reported previously (44). The F1 fragment also induces a mitogenic response in serum-free medium causing Ϸ2-fold increase in the [ 3 H]thymidine incorporation (Fig. 9B). To test whether CD44(ex14/v10) is involved in mediating HA-induced mitogenic response in serum-free medium, we examined whether anti-CD44 antibody is able to inhibit F1 fragment-induced mitogenic response. As shown in Table III, anti-CD44 antibody   FIG. 6. RT in situ PCR analysis of CD44(ex14/v10) in bovine aorta. Bovine aorta tissue specimen was fixed with formalin, paraffinembedded, sectioned (4 m), and processed for RT in situ PCR and immunoperoxidase staining as described under "Materials and Methods." A, immunoperoxidase staining of factor VIII in bovine aorta. B, detection of CD44 (ex14/v10)-specific transcripts by RT in situ PCR using an ex14/v10-specific primer pair and fluorescein isothiocyanate-digoxigenin-11-dUTP as described under "Materials and Methods." (Arrowheads indicate endothelium or endothelial cells in tunica intima of aorta). C, as a control, RT in situ PCR was carried out in the absence of reverse transcriptase. (E, endothelium or endothelial cells; M, tunica media; ϫ 600). blocks F1 fragment-induced mitogenic response suggesting that GP116, CD44(ex14/v10), is involved in mediating HAinduced endothelial cell proliferation.
Transmembrane Interaction between CD44(ex14/v10) and Ankyrin-We have previously shown that both HA binding and adhesion of lymphoma cells to HA require the presence of an intact cytoskeletal network (39). To investigate whether cytoskeletal proteins are needed in HA binding to endothelial cells, we pre-treated bovine aortic endothelial cells (GM7372A) with various drugs, such as cytochalasin D (a microfilamentdisrupting agent known to prevent actin polymerization), W-7 (a calmodulin inhibitor known to block myosin light chain kinase that is required for actomyosin contraction), and colchicine (a microtubule-disrupting agent). Following drug treatment, the endothelial cells were examined for HA binding or adhesion to HA-coated surface. As shown in Table IV, both cytochalasin D and W-7 (but not colchicine) significantly inhibit both HA binding and cell adhesion to HA. These results suggest that the microfilament network (but not microtubules) is required to promote HA binding and HA-mediated cell adhesion.
To study whether the cytoskeletal protein, ankyrin, is associated with GP116, CD44(ex14/v10), in vivo, we have used the nonionic detergent Triton X-100 to extract 35 S-labeled cells followed by anti-CD44 antibody-mediated immunoprecipitation. As shown in Fig. 10A, anti-CD44 antibody immunoprecipitates two proteins from the 35 S-labeled endothelial cell extracts, one is CD44(ex14/v10) of 116 kDa (GP116) and the other is a 216-kDa protein. The 216-kDa protein is recognized by monoclonal anti-ankyrin antibody (ANK016) indicating that this protein (Fig. 10B) is similar to erythrocyte ankyrin (Fig.  10C). These results suggest that GP116, CD44(ex14/v10), and ankyrin form a complex in vivo.
In order to determine whether a direct interaction occurs between GP116, CD44(ex14/v10), and ankyrin, an in vitro binding assay using highly purified GP116, CD44(ex14/v10) and ankyrin was employed. Our data indicate that 125 I-labeled ankyrin binds CD44(ex14/v10) specifically, saturably, and in a dose-dependent manner (Fig. 11A). Scatchard plot analysis indicates that GP116, CD44(ex14/v10), contains a single class of binding sites for ankyrin with an apparent dissociation constant (K d ) of 1.2 nM (Fig. 11B). We believe that the cytoplasmic domain of GP116, CD44(ex14/v10), contains a region that strongly binds to ankyrin similar to that found in lymphocyte CD44(GP85).

DISCUSSION
CD44 is a cell surface adhesion molecule that occurs as several isoforms that are widely distributed in different cells and tissues (1)(2)(3)(4)(5). The isoforms of CD44 (e.g. CD44s, CD44E, and CD44 variants) arise from differential splicing of 1-10 variable exons (v1-v10) that encode portions of the membrane proximal extracellular domain (27,28). The molecular diversity of CD44 isoforms is further compounded by differential posttranslational modifications within either the invariant region of the extracellular domain encoded by the constitutively spliced exons or the variant region encoded by the variable exons (9, 10, 29 -31). Although some correlation between CD44 variant isoforms and tumor metastasis has been established (22-26, 32, 46), very limited information with regard to the functional properties of these isoforms is currently available.
Previously, it has been shown that the human endothelium contains very little CD44 expression (23,47). Recently, Bennett and co-workers (29) reported that only a single CD44 isoform (CD44s) was found at the transcript and protein (Ϸ90 kDa) levels in both resting and activated human umbilical vein endothelial cells and pulmonary artery endothelial cells. No CD44 variants were detected in these cells. In this paper we have used biochemical, cell biological, and molecular biological techniques to identify a new CD44 isoform, GP116 (also designated as CD44(ex14/v10)), in bovine endothelial cells (4). Chemical composition analysis of GP116 indicates that this protein is different from CD44s and shares a great deal of structural similarity with human CD44(ex14/v10) protein (Table I). Furthermore, the results of RT-PCR and Southern blot analyses indicate that, in fact, the CD44(ex14/v10) transcript (but not CD44s) is the major CD44-related transcript expressed in bovine endothelial cells (Fig. 4A, lane 2). The CD44(ex14/ v10) message is translated into a 52-kDa precursor that is then converted to GP116 by post-translational modifications.
The RT in situ PCR data reveal the presence of CD44(ex14/ v10) transcript in bovine aorta endothelial cells in vivo (Fig. 6). Furthermore, the results of anti-CD44-mediated immunoprecipitation analysis show that other primary endothelial cell cultures from different origins (e.g. EJG and CPA47) express CD44(ex14/v10) (Fig. 1). At the present time, it is not clear why bovine endothelial cells but not human endothelial cells ex- H]HA binding, endothelial cells were either untreated (control) or treated with rat anti-CD44 IgG (50 g/ml) or rat IgG (50 g/ml) followed by an incubation with [ 3 H]HA (1 g/ml) at 4°C for 2 h in PBS. The nonspecific binding was determined in the presence of unlabeled HA (100 g/ml) and was subtracted. The specific binding (394 Ϯ 25 dpm) observed in no treatment (control) is designated as 100%.
b Cell adhesion, endothelial cells labeled with Tran 35 S-label (9.1 ϫ 10 5 cpm/10 5 cells) were pretreated with either rat anti-CD44 IgG (50 g/ml) or rat IgG (50 g/ml) as described above. Following treatment, the cells were incubated with HA-coated plates at 4°C for 30 min. The nonspecific binding of cells to HA-coated wells was determined in the presence of soluble HA and was subtracted. The specific adhesion (2.7 ϫ 10 5 cpm or 30,000 cells) observed in no treatment control sample is designated as 100%. Each experiment was performed in duplicate, and the results represent an average of three separate experiments. press CD44 ex14 (or v10).
The 52-kDa precursor of GP116 in endothelial cells (GM7372A) is the nascent polypeptide chain of CD44(ex14/v10) protein that contains a total of 435 amino acids (65 additional amino acids encoded by the new bovine exon, exon 14 (v10) (Fig. 4A, b)). The estimated 52-kDa molecular mass of a protein of 435 amino acid residues is based on the assumption that the molecular mass of each amino acid is approximately 120 Da. However, the molecular mass can also be estimated to be 48 kDa if the molecular mass of each amino acid is assumed to be 110 Da. Since the standard SDS-polyacrylamide gel electrophoresis analysis does not provide enough resolution to distinguish 48 from 52 kDa, it is possible that the actual molecular mass of the GP116 precursor is between 48 and 52 kDa. This 48-/52-kDa precursor of GP116 is different from CD44s' precursor. CD44s has been shown to contain a polypeptide chain of 42 kDa (370 amino acids) that is N-glycosylated and does not contain any exon insertion. The deduced amino acid sequence reveals that the exon 14/v10 of GP116 also encodes a sequence rich in serine (Ser) and threonine (Thr) amino acids. In fact, 28% of the total Ser/Thr found in the complete amino acid    (Table I). O-Glycosylation accounts for Ϸ28 kDa of mass for this molecule (Fig. 3). It is worth noting that the structure of the 11 O-linked oligo-saccharides in GP116, CD44(ex14/v10), appears to differ from those in CD44s (with 4 -5 O-linked oligosaccharide chains) (10). In addition, CD44(ex14/v10) (but not CD44s/GP85) is resistant to neuraminidase treatment indicating the lack of a terminal sialic acid residue on CD44(ex14/v10). 2 However, CD44(ex14/ v10) (GP116) appears to contain sulfated O-linked oligosaccharides (Fig. 3C) similar to mucin-like molecules (48), but it does not contain any sulfated GAGs as described for the CD44v3 isoform (29,30). These selective post-translational modifications of GP116's structure may be required for the function of this molecule in cell adhesion, membrane-cytoskeleton interaction, and/or cell proliferation.
To date several HA binding proteins have been identified on endothelial cells of different tissue origins. For example, a specific HA receptor is found on liver sinusoidal endothelial cells (49), and a HA binding chondroitin sulfate proteoglycan has been detected on aortic endothelial cells (50). In addition, certain HA binding proteins appear to be involved in endothelial cell migration and tubule formation (51). However, the detailed structural and functional properties of these HA receptors are largely unknown. In this study, we have determined that GP116, CD44(ex14/v10), is the major HA receptor present on bovine aortic endothelial cells.  35 S-labeled endothelial cells (GM7372A) were immunoprecipitated using rat anti-CD44 antibody under nondenaturing conditions as described under "Materials and Methods." The immunoprecipitated material was analyzed by gel electrophoresis and fluorography. B and C, immunoblot analysis of CD44(ex14/v10) immunoprecipitates using mouse anti-ankyrin antibody (ANK016). Endothelial cell extracts were immunoprecipitated using rat anti-CD44 antibody. The anti-CD44-immunoprecipitates (B) and purified erythrocyte ankyrin (C) were further analyzed by immunoblotting using mouse anti-ankyrin antibody (ANK016) as described under "Materials and Methods. "   FIG. 11. Binding of 125 I-labeled ankyrin to CD44(ex14/v10). Various concentrations (10 -400 ng/ml) of 125 I-labeled ankyrin were incubated with purified CD44(ex14/v10) protein (Ϸ5 ng) at 4°C for 5 h under equilibrium conditions. Nonspecific binding was determined in presence of a 100-fold excess of unlabeled ankyrin. A, equilibrium binding of 125 I-labeled ankyrin to CD44(ex14/v10). B, Scatchard plot analysis of the equilibrium binding data presented in A.
possible explanation for the variation in K d measurement may be the different [ 3 H]HA preparations used in these experiments. Since the K d measurements for HA binding to endothelial cells in these studies are all within the 0.1-1 nM range, there clearly exists a high affinity interaction between HA and endothelial cells. In this regard, CD44(ex14/v10) appears to be similar to CD44s that is the major HA receptor in a number of other cell types (7,13,24).
HA has been shown to play an important role in several important physiological functions such as maintaining cartilage integrity, balancing homeostasis of water and plasma proteins in the intercellular matrix, and promoting/inhibiting mitosis and cell migration (52). In addition, degradation products of HA containing 3-25 disaccharide units have been found to promote angiogenesis that involves endothelial cell proliferation, migration, and tubule formation (38,44,51). The results presented in this study show a correlation between high affinity binding of HA polymer or F1 HA fragment (Ϸ10 -15 disaccharide units) to endothelial cells and the occurrence of a mitogenic response. These findings suggest that a specific cell surface HA receptor may be responsible for the induction of ligand-induced mitogenic signals (Figs. 7 and 8). Since anti-CD44 antibody inhibits F1 HA fragment-mediated mitogenic response (Table III), it is possible that CD44(ex14/v10) is a CD44-related HA receptor involved in HA-mediated function. Furthermore, the mitogenic response of endothelial cells to F1 fragment is observed both in serum-containing and serum-free media, although the response is somewhat lower under the latter condition (Fig. 9). It is also possible that HA interacts with growth factors present in serum, and this interaction synergistically regulates endothelial mitogenic response. However, we do believe that HA fragments are authentic modulators of endothelial cell growth since the fragments generated have been extensively purified to remove any impurities that might be present in the original HA preparation. In addition, similar effects of HA and HA fragments on endothelial cell growth have been observed previously (38).
Both HA binding and HA-mediated cell adhesion functions of endothelial cells are inhibited by two cytoskeletal drugs, cytochalasin D and W-7, suggesting that microfilament-associated components regulate these events (Table IV). This observation is in agreement with our earlier findings that CD44(GP85)-mediated HA binding and cellular adhesion by lymphocytes are dependent on an intact microfilament network (39). Further analyses indicate that GP116, CD44(ex14/v10), binds ankyrin directly and with high affinity (Fig. 11) in vitro. Interestingly, CD44(ex14/v10) and ankyrin also appear to form a stable complex in vivo since ankyrin can be co-immunoprecipitated with GP116 using an anti-CD44 antibody (Fig. 10A). In addition, some of the biosynthetic precursors of GP116 also associate with ankyrin in vivo. 2 Recently, it has been reported that certain CD44 isoforms, such as the CD44E protein, interact directly with the ERM family of cytoskeletal proteins. However, it should be noted that these observations were solely based on co-immunoprecipitation experiments (53). A direct binding interaction between purified CD44E and these ERM cytoskeletal proteins in vitro has not yet been demonstrated (53). In this study, using a similar co-immunoprecipitation approach designed to preserve the membrane-cytoskeleton complexes, we have failed to observe any other co-immunoprecipitated materials such as the ERM cytoskeletal proteins (Fig.  10A). Thus, at least in endothelial cells, GP116 (CD44(ex14/ v10)) appears to directly associate only with ankyrin and not other cytoskeletal proteins. Furthermore, we have shown that the ankyrin binding domain in the CD44s cytoplasmic tail region is required for cell surface HA binding and HA-mediated cell adhesion (16). The fact that the ankyrin binding domain is Ն90% conserved in all of the CD44 isoforms studied so far implies that GP116, CD44(ex14/v10), also contains an ankyrinbinding site. Most importantly, we believe that the binding of ankyrin to GP116 (CD44(ex14/v10)) may not only be important for HA-mediated cell adhesion but also is needed for the onset of HA/HA fragment-mediated endothelial cell proliferation.