Isolation and Characterization of a Glycoprotein Secreted Endothelial Cells in Culture APPARENT IDENTITY WITH PLATELET THROMBOSPONDIN * by Aortic

A high molecular weight glycoprotein, reported to be secreted by endothelial cells (Sage, H., Crouch, E., and Bornstein, P. (1979) Biochemistry 18, 5433-5442), has been purified to apparent homogeneity from culture medium of adult bovine aortic endothelial cells. Purification was achieved by ammonium sulfate fractionation and successive chromatography on gelatin-Sepharose, Sephacryl 5-300, and hydroxylapatite. The glycoprotein was found to be a disulfide-linked oligomer with a subunit molecular weight of 190,000, as judged by its mobility on sodium dodecyl sulfate (NaDodSO& polyacrylamide gels. The endothelial cell-derived protein is distinct from high molecular weight serum glycoproteins such as fibronectin and at-macroglobulin. However, immunological and structural studies indicate that the M, = 190,000 glycoprotein is either identical with or closely related to thrombospondin, a glycoprotein contained in platelet granules and released in response to thrombin-induced aggregation.

by Schwartz (12). The proteins synthesized by these cells were metabolically labeled with either ~-[2,3-~H]proline or ~-[2-~H]mannose according to the methods described by Sage et al. (6). Briefly, confluent cultures were exposed to radiolabeled proline or mannose (50 pCi/ml of medium or 80 pCi/lO" cells) for 24 h in serum-free DMEM' supplemented with sodium ascorbate and P-APN. Radiolabeled culture medium was added to a protease inhibitor mixture to produce a find concentration of 0.2 mM PhCHzSOzF, 10 mM MalNEt, and 2.5 mM EDTA.
Purification of the Glycoprotein-The glycoprotein was purified from metabolically labeled endothelial cell culture medium utilizing a combination of ammonium sulfate fractionation and chromatography on gelatin-Sepharose, Sephacryl S-300, and hydroxylapatite. Initially, the radiolabeled culture medium, pooled from 25 tissue culture dishes (150 X 25 mm) of confluent BAE cells, was made 20% (w/v) in ammonium sulfate, and the solution was stirred overnight at 4 "C. The precipitated protein was removed by centrifugation at 40,000 X g for 30 min. Ammonium sulfate was subsequently added to the supernatant to a final concentration of 50% (w/v), the solution was stirred overnight at 4 "C, and the precipitated protein was removed as described above. The supernatant was discarded, and the pelleted material was dissolved in 10-15 ml of PBS containing 2.5 mM EDTA and 0.2 mM PhCh2S02F (Buffer A). The sample was then dialyzed against 1 liter of this buffer with a total of three changes over an 18-h period. The retentate was loaded onto a gelatin-Sepharose column (1.5 X 25 cm) (prepared according to the method of Engvall and Ruoslahti (13)), equilibrated in Buffer A, and incubated at least 4 h prior to elution with Buffer A. The flow-through material from the gelatin-Sepharose column was dialyzed against 1 liter of 50 mM ammonium bicarbonate containing 2.5 mM EDTA and 0.2 mM PhCHZS02F and then was lyophilized. The lyophilized powder was dissolved in 5-10 ml of PBS containing 3 M urea, 2.5 mM EDTA, and 0.2 mM PhCHzSOzF (Buffer B). Subsequently, the sample was applied to a Sephacryl S-300 column (2 X 100 cm) which had been equilibrated in Buffer B. The sample was eluted from the S-300 column and fractions containing the glycoprotein (identified by slab gel electrophoresis) were pooled, dialyzed against 1 liter of 10 mM Nap04 (pH 6.5) containing 3.2 M urea, 2.5 mM EDTA, and 0.2 mM PhCHzSOzF (Buffer C) for 24 h with three buffer changes. This retentate was applied to a hydroxylapatite (Bio-Rad) column (0.9 X 12 cm) which had been equilibrated in Buffer C. The column was washed with 50 ml of equilibration buffer and developed with a linear gradient (between 10 mM and 0.3 M) of sodium phosphate (pH 6.5) which contained 3.2 M urea, 2.5 mM EDTA, and 0.2 mM PhCH2SOzF. Fractions containing the glycoprotein were pooled and concentrated on an Amicon pressure cell using an XM-100A fdter.

A n Endothelial Cell
Glycoprotein Related to Thrombospondin observed on the film were quantitated using a scanning densitometer (Beckman DU-8).
Amino Acid Analysis-Glycoprotein samples (60-100 pg), to be used for amino acid analysis, were desalted on a Bio-Rad P-2 column (1 X 25 cm) in the presence of 50 mM NHIHCO:~ (pH 7.8) and then were lyophilized. Amino acid analyses were performed on a Durrum (model D-500) amino acid analyzer following 24-h hydrolysis in U~C U O in 5.7 N HCI a t 110 "C. Cysteine was determined as cysteic acid following performic acid oxidation (18).
Purification of Thrombospondin-Human thrombospondin was purified from 2 units of platelet concentrate (obtained from the Puget Sound Blood Center) essentially according to the method of Lawler et al. (IO). Minor modifications of the method included the addition of 2.5 mM EDTA to the final resuspension buffer prior to thrombin activation and the termination of the thrombin reaction with 5 mM PhCH,SO,F rather than hirudin. Phillips et al. (11) have recently shown that addition of EDTA to the final resuspension buffer significantly enhances the yield of thrombospondin.
Bovine thrombospondin was purified in a similar manner although with more extensive modifications. Bovine blood was collected at the slaughterhouse and rapidly mixed with 0.1 volume of anticoagulant citrate dextrose (3.8 g of trisodium citrate and 2 g of glucose/100 ml of H,O). The sample was then centrifuged a t 900 X g for 6 min a t room temperature (as were all subsequent centrifugations). The platelet-rich plasma was withdrawn and recentrifuged a t 300 X g for 6 min in order to remove contaminating red cells. Platelets were isolated by centrifugation a t 2,200 X g for 9 min. Platelets were resuspended in 10% anticoagulant dextrose, 20 mM Tris (pH 7.5). 145 mM NaCI. 5 mM KCI, and 5 mM glucose (Buffer D) and recentrifuged a t 300 X g for 6 min to remove residual red cells. Platelets were again pelleted by Centrifugation a t 2,200 X g for 9 min and subsequently washed two times in Buffer D. Following the final centrifugation, the platelets were resuspended in 20 mM Tris-HCI, pH 7.5, 145 mM NaCI, 5 mM KCI, 5 mM glucose, and 2.5 mM EDTA (5 ml/unit of platelets). Highly purified human a-thrombin (a gift from William Canfield, University of Washington, Seattle, WA) was added to a final concentration of 7 units/ml and incubated at room temperature for 15 min in order to initiate platelet aggregation. The reaction was terminated by the addition of PhCH,SO,F to a final concentration of 5 mM. Platelet aggregates were removed by centrifugation a t 22,000 X g for 20 rnin at 4 "C. Eight M urea was added to the supernatant to a final concentration of 3 M and the sample was applied to a Sephacryl S-300 column as previously described for the purification of the glycoprotein. Protein fractions eluting in the void volume of this column were pooled and chromatographed on hydroxylapatite as previously described. Protein fractions eluting at a conductivity between 5.5 and 6.5 mmho were pooled and concentrated with an Amicon pressure cell using an XM-l00A filter.
Protein Transfers-Antibodies to the glycoprotein, purified from bovine aortic endothelial cell media, were raised in rabbits and the serum IgG fractions were enriched by ammonium sulfate fractionation. The antiserum did not react with fibronectin or types I11 or IV (pro)collagen as judged by an enzyme-linked immunosorbent assay.
Protein transfer from NaDodS04-polyacrylamide gels to nitrocellulose paper for immunochemical cross-reactivity experiments was performed according to the methods of Towbin et al.
(20) and Burnette (21) with some modifications. Proteins were transferred From 5% NaDodS04-polyacrylamide gels to nitrocellulose paper (Schleicher and Schuell Co., BA85) by electrophoresis for 18 h a t 200 mA (-15 volts) in a buffer composed of 25 mM Tris, 192 mM glycine, and 20% methanol (pH 8.3). Following the electrophoretic transfer, the nitrocellulose paper was cut into strips, which corresponded to the lanes of electrophoresis on the polyacrylamide gel. The position of the various lanes on the nitrocellulose strips was determined by the addition of pyronine red to the original polyacrylamide gel electrophoresis sample buffer. This dye was transferred to the nitrocellulose paper during electrophoresis from the polyacrylamide gel and, thus, could be used to mark the dye front of the various lanes on the paper. The strips of nitrocellulose paper, containing the proteins of interest, were washed for 60-90 min in 10 ml of 0.15 M NaCI, 0.05 M Tris (pH 7.4), and 5% BSA + 2.5% gelatin (Buffer E). The nitrocellulose strips were then incubated for 2 h a t room temperature, with gentle agitation, in Buffer E containing a 1/10 final dilution of the antiglycoprotein antibody. At the end of this incubation period, the nitrocellulose strips were washed in 50 ml of Buffer E for 10 min, followed by four IO-min washes in Buffer E containing 0.05% Nonidet P-40, and finally for 10 min in Buffer E. Subsequently, the nitrocellulose strips were incubated in 10 ml of Buffer E containing ""I-protein A (10' cpm/ml) for 20 min a t room temperature. The strips were then washed as described above following exposure to the antibody. The nitrocellulose strips were dried and exposed to Kodak BB-5 fdm using Chronex Hi-Plus enhancing screens. Strips were routinely exposed for 18 h prior to developing the film.
Comparative Peptide Mapping-Proteins of interest were resolved by NaDodSOr-polyacrylamide gel electrophoresis and compared by a two-dimensional mapping technique described by Elder et al. (22) and modified by Sage et al. (23).

RESULTS
NaDodS0,-polyacrylamide gel electrophoresis of the proteins secreted by bovine aortic endothelial cells in culture demonstrated the presence of several metabolically labeled proteins (Fig. 1A). These secreted products included fibronectin, type I11 procollagen (6), a unique pepsin-sensitive collagen (EC) (7), and a high molecular weight protein that was pepsinsensitive and collagenase-resistant (6). This latter protein was clearly labeled when cells were incubated with ["Hlproline as well as with ["Hlmannose, thus identifying it as a glycoprotein ( Fig. 1, A and B ) . The glycoprotein migrated with an apparent molecular weight of 190,000 on NaDodS04-polyacrylamide gel electrophoresis under reducing conditions ( Fig. 2), while under nonreducing ccnditions, it migrated as a disulfide-bonded oligomer. Estimation of the molecular weight of the oligomeric glycoprotein was difficult; its mobility on a NaDodS04-polyacrylamide gel was considerably slower than that of unreduced fibronectin ( M , = 440,000; see Fig. 1A). Purification of this M , = 190,000 glycoprotein from serum-free culture media necessitated its separation not only from other proteins synthesized and secreted by the endothelial cells, but also from relatively large amounts of serum proteins that were bound to the endothelial cell layer during subculture and slowly released into the culture medium during the labeling period (Fig. 1 0 . The two principal contaminating serum proteins were a2macroglobulin and serum albumin (Fig. 1C).
The M , = 190,000 glycoprotein was purified to apparent homogeneity using ammonium sulfate fractionation and chro- NaDodS04-polyacrylamide gel electrophoresis of proteins secreted b y bovine aortic endothelial cells in culture. Confluent cultures were labeled with L-12.3-'Hlproline or D-[%'H] mannose for 24 h in serum-free DME medium supplemented with sodium ascorbate and 8-APN. Secreted proteins were concentrated by precipitation with ammonium sulfate and, after resuspension and dialysis, were subjected to electrophoresis on 5% NaIlodSO,-polyacrylamide gels either in the presence (+) or absence (-)   glycoprotein as estimated by electrophoresis in a 6% Na-DodS0.-polyacrylamide gel. Molecular weight standards included unreduced fibronectin (440,000), reduced thyroglobulin (330,000). reduced fibronectin (220,000). reduced a2-macroglobulin (185.000). unreduced IgG heavy chains (150,000), reduced phosphorylase (97,000), reduced bovine serum albumin (67,000), and reduced catalase (60,000 from the majority of the type I11 procollagen. Following the ammonium sulfate fractionation, the sample was applied to a gelatin-Sepharose affinity column (Fig. 3A). The M , = 190,OOO glycoprotein did not bind to this column, and greater than 70% of the fibronectin which contaminated the preparation was removed at this point. Subsequent elution of the affinity column with 6 M urea indicated that only fibronectin had bound to the column. The unbound fraction of the gelatin-Sepharose column was pooled and concentrated by lyophilization. The lyophilized protein was redissolved in Buffer B and applied to a Sephacryl S-300 molecular sieve column (Fig.   3B). The M, = 190,OOO glycoprotein eluted near the void volume of the column and was clearly separated from serum albumin and from other lower molecular weight contaminants. Chromatography of the pooled Sephacryl S-300 fractions on hydroxylapatite (Fig. 3 0    ' Endothelial cells (approximately 1.5 X 10' ) were radiolabeled with [3H]proline ( 5 0 pCi/ml), and the medium was collected after a 24-h incubation as previously described (6).
Coomassie blue staining, although, on occasion, very low levels of contaminating protein were detectable by fluorescence autoradiography.
Endothelial cell medium pooled from 25 culture dishes (150 X 25 mm) provided approximately 0.3 mg of purified M , = 190,000 glycoprotein with a final estimated recovery of 8% based on radioactivity (Table I). It became increasingly difficult to maintain high yields at the later purification steps. This problem was partially overcome by the addition of urea to the buffers used in the Sephacryl S-300 and hydroxylapatite columns and the use of siliconized glass tubes for the collection and storage of the protein. In spite of these precautions, handling of the purified M , = 190,000 glycoprotein usually resulted in poor yields. This low recovery was due either to the nonspecific adhesion of the glycoprotein to surfaces or to the aggregation and precipitation of the glycoprotein from solution.
The amino acid composition of the purified M , = 190,000 glycoprotein, shown in Table 11, indicates that it contains relatively high levels of cysteine, aspartic acid (or asparagine), and glutamic acid (or glutamine). The amino acid composition ~ ~~~~ clearly distinguished the M , = 190,000 glycoprotein from a2macroglobulin. A recent report by Doyle et al. (24) indicated that human umbilical vein endothelial cells produced a M , = 145,000 glycoprotein that corresponded to thrombospondin, a glycoprotein released by platelets in response to thrombin. A comparison of the amino acid composition of the M , = 190,000 glycoprotein with that of human thrombospondin indicated a high degree of similarity (Table  11). This observation prompted us to purify thrombospondin from bovine platelets in order to compare the two glycoproteins more carefully. It was apparent that the major thrombin-released protein from bovine platelets (thrombospondin) co-migrated with the M , = 190,OOO glycoprotein, purified from BAE cell culture media, on NaDodS0,-polyacrylamide gel electrophoresis (Fig. 6). In addition, a M, = 160,000 glycoprotein (carbohydrate content detected by periodic acid-Schiff staining) which co-purified with thrombospondin was noted. Similar results were obtained when the migration of human platelet thrombospondin was compared with that of purified M , = 190,000 glycoprotein (data not shown); however, the M , = 160,000 glycoprotein was not detected in the thrombin-released material from human platelets. Purification of the thrombin-released material from bovine platelets indicated that both thrombospondin and the M, = 160,OOO glycoprotein eluted from the Sephacryl S-300 and hydroxylapatite columns a t positions identical with that of the M, = 190,000 glycoprotein secreted by bovine aortic endothelial cells. Lanes 4 and 8 of Fig. 6 illustrate that the combination of Sephacryl S-300 chromatography and hydroxylapatite chromatography provided a high degree of purification of both thrombospondin and the M , = 160,000 glycoprotein.
Proteins solubilized from whole bovine platelets by Na-DodSO, buffer as well as proteins released by thrombin activation were subjected to electrophoresis on NaDodS04-polyacrylamide slab gels, together with purified M , = 190,000 endothelial cell glycoprotein. These proteins were transferred to nitrocellulose paper and reacted with an antiserum to the M , = 190,OOO glycoprotein. As shown in Fig. 7, only throm-

An Endothelial Cell
Glycoprotein Related to Thrombospondin bospondin and the M , = 160,000 glycoprotein, among the complex mixture of proteins (see Fig. 6 , lunes Gand 7), reacted with the antibody. A cross-reaction with human thromhospondin was also observed (data not shown). washed bovine platelets (lane 2). and thrombin-released protein from bovine platelets (lane 3 ) were resolved on 5$ NaDodSOI-polyacrylamide gels in the presence of 50 mM dithiothreitol, as illustrated in Fig. 6. Following electrophoresis, protein was transferred from the polvacrylamide gel to nitrocellulose paper and subsequently exposed to anti", = 190,OOO glycoprotein antibodies and '"'I-labeled protein A as described under "Materials and Methods." Following this treatment, the nitrocellulose paper was dried and exposed to x-ray film. TLC. thin laver chromatography the two-dimensional peptide maps of the two glycoproteins ( Fig. 8, A and C ) . Fig. 8B illustrates the high degree of similarity between bovine and human thromhospondin. The two-dimensional peptide map of the M , = 160,000 glycoprotein ( Fig. 80) indicates that it too is very similar to thromhospondin and is perhaps derived from it.

DISCUSSION
We have previously reported that bovine aortic endothelial cells in vitro synthesize and secrete a large noncollagenous glycoprotein which is distinct from fibronectin (6). The apparent molecular weight of the glycoprotein monomer, based on its mobility on NaDodS04-polyacrylamide gel electrophoresis, is 190,000. Under nonreducing conditions, the glycoprotein migrates as a disulfide-linked oligomer. Accurate estimates of the molecular weight of the disulfide-bonded glycoprotein are extremely difficult due to the lack of suitable standards in its molecular weight range. However, since its migration is significantly slower than that of the fihronectin dimer (440,000), we suggest that the glycoprotein may exist as a trimer or tetramer under nonreducing conditions. The glycoprotein has been purified to homogeneity from the culture medium of bovine aortic endothelial cells by a combination of ammonium sulfate fractionation, gelatin-Sepharose affinity chromatography, Sephacryl S-300 molecular sieve chromatography, and hydroxylapatite chromatography. This protocol led to a 10-to 20-fold purification of the glycoprotein with an 8-10% recovery. The variability in the extent of purification required to attain homogeneity was a function of the concentrations of contaminating proteins in the culture media. Several problems were encountered during attempts to purify the M , = 190,000 glycoprotein. Although the glycoprotein represents between 15 and 25% of the total protein synthesized by bovine endothelial cells in culture, significant quantities can he obtained only by working with large numhers of cells. In addition, as the glycoprotein was purifi?d, the yields decreased considerably. This loss appeared to be due to the strong tendency of the glycoprotein to adsorb to glass and plastic. These problems were partially circumvented by using siliconized glass and by adding urea to buffers. After lyophilization of pure M , = 190,000 glycoprotein from volatile buffers such as NH4HC03, the glycoprotein was largely insoluble in aqueous buffers. Lyophilization of crude preparations of M , = 190,000 glycoprotein did not seem to affect its solubility appreciably.
We noted early in our studies that the M, = 190,000 glycoprotein migrated in close proximity to reduced a2-macroglobulin on NaDodS04-polyacrylamide gel electrophoresis. In fact, under certain conditions, the two proteins appeared to comigrate (Fig. 1). However, M , = 190,000 glycoprotein and a2macroglobulin could be distinguished from each other on the basis of their mobility under nonreducing conditions (Fig. 1); while a2-macroglobulin migrated as a dimer under these conditions, the mobility of the M , = 190,000 glycoprotein indicated that it was a larger complex. Additional evidence that the M , = 190,000 glycoprotein and a2-macroglobulin are different proteins was obtained from the lack of immunological cross-reactivity, from comparison of two-dimensional peptide maps (data not shown), and from the significant differences in their amino acid compositions (Table 11). Similar considerations distinguished the M, = 190,000 glycoprotein from fibronectin, another glycoprotein synthesized and secreted in large amounts by bovine aortic endothelial cells in culture. Lack of immunological cross-reactivity as well as significant differences in one-dimensional peptide maps of the two glycoproteins argued against the possibility that one was derived from the other. The amino acid composition of the M , = 190,000 glycoprotein was also clearly distinct from that of angiotensinconverting enzyme (25), another high molecular weight glycoprotein reported to be synthesized by endothelial cells in culture (26,27). Doyle et al. (24) have reported recently that human umbilical vein endothelial cells synthesize a M , = 145,000 glycoprotein which closely resembles platelet thrombospondin.
Thrombospondin has also been referred to as glycoprotein G (28) and thrombin-sensitive protein (8). Several values for the apparent molecular weight of this protein, based on its mobility on NaDodSOl gels under reducing conditions, have been reported in the literature. Lawler et al. (10) reported that the protein is a disulfide-linked trimer with a subunit molecular weight of 142,000. Hagen et al. (9) determined a subunit molecular weight of 147,000 for the glycoprotein. In contrast, Phillips and Agin (29) and Baenziger et al. (8) reported molecular weights of 185,000 and 190,000, respectively, for the subunit of the disulfide-linked oligomeric protein. The similarity of the latter molecular weight estimates to that determined for the M , = 190,000 endothelial cell glycoprotein, together with the report by Doyle et al. (24), led us to examine the relationship between the M , = 190,000 glycoprotein and platelet thrombospondin.
We noted that the amino acid composition of the bovine M, = 190,000 glycoprotein is very similar to that reported for thrombospondin by Lawler et al. (lo), albeit that the latter was obtained from human platelets. Further evidence that the M, = 190,000 glycoprotein and thrombospondin are closely related came from immunological and structural studies. Antibodies directed against the M , = 190,000 glycoprotein crossreacted with bovine thrombospondin and a M, = 160,000 glycoprotein, also released by bovine platelets, as well as with human thrombospondin. The fact that the antibody reacted against both the M, = 190,000 and M , = 160,000 bands of thrombin-released material from bovine platelets suggests that the M, = 160,000 band may be a derivative of thrombo-spondin. However, if this is the case, processing or partial degradation of thrombospondin must have occurred prior to thrombin activation since both bands were present in whole platelets (Fig. 6) and seemed to disappear from the platelet pellet following thrombin-induced aggregation.
Two-dimensional peptide maps of the M , = 190,000 glycoprotein, bovine thrombospondin, and human thrombospondin were essentially indistinguishable, providing further evidence for their close similarity (Fig. 7). The map of the M , = 160,000 glycoprotein from bovine platelets shares many features with that of thrombospondin, indicating that it is either a derivative of thrombospondin or a closely related protein. The occurrence of a similar relationship for the human platelet protein could account for reports indicating lower molecular weights for thrombospondin (9, 10, 24). Although we have not observed the M , = 160,000 glycoprotein in human platelets, it is possible that differences in the methods of blood collection and platelet purification were responsible for the variation between the human and bovine preparations. We have noted that under certain conditions the glycoprotein purified from endothelial cell culture medium is partially converted to a form that migrates near the position of the M, = 160,000 protein observed in platelets.
Several important questions remain regarding thrombospondin. One relates to whether endothelial cells are uniquely responsible for the synthesis of this glycoprotein. In that case, platelets might acquire the protein from plasma in a manner similar to the uptake of fibrinogen. Alternatively, megakaryocytes may also synthesize the glycoprotein. A second question deals with the function of thrombospondin. Lactoperoxidase-mediated iodination of bovine aortic endothelial cells in culture indicated that the M , = 190,000 glycoprotein was associated with the surface of these cells2 Since thrombospondin is also released from platelets upon thrombin-induced aggregation, this glycoprotein could be an integral component in the control of some aspect of the clotting process that involves platelet aggregation and thrombus formation. Thrombospondin could also function as a specific protease inhibitor of one of the enzymes of the coagulation cascade to delay clotting or as an activator to promote thrombus formation.