Structural Characterization of the Fibroblast Growth Factor-binding Protein Purified from Bovine Prepartum Mammary Gland Secretion*

A novel heparin-binding protein was purified to ho-mogeneity from bovine prepartum mammary gland secretion using heparin-Sepharose chromatography and reverse-phase high performance liquid chromatography successively. Structural information obtained by N-ter-minal amino acid sequencing of a series of proteolytically generated peptides permitted the cloning of the corresponding cDNA. The isolated cDNA was 1170 base pairs long and consisted of an 83-base pair 5 * -untrans-lated region followed by a 702-base pair coding region and a 385-base pair 3 * -untranslated region. The open reading frame resulted in a protein comprising 234-amino acid residues, including a signal sequence. Instead of Lys 24 as the predicted N terminus, Edman degradation of the native protein revealed N-terminal processing at two sites as follows: a primary site between Arg 31 –Gly 32 and a secondary site between Arg 51 –Ser 52 . The amino acid sequence showed a significant similarity with that of human (60%) and mouse (53%) fibroblast growth factor-binding protein (FGF-BP). Accordingly, ligand blotting experiments revealed that bovine FGF-BP bound FGF-2. The theoretical mass of the The eluate was collected and aliquots from each fraction were taken before freeze-drying. The aliquots were subjected to SDS-PAGE and electroblotting following standard procedures. A selection of consecu- tive fractions from the heparin affinity chromatography was dissolved in 0.1% trifluoroacetic acid and passed over a 1-ml Resource RPC column (Amersham Pharmacia Biotech) equilibrated with 0.1% trifluoroacetic acid (flow 1 ml/min). Bound material was eluted using a linear gradient of 0–55% 2-propanol in 0.1% trifluoroacetic acid for the initial 65 min and then with a 55–80% gradient for the following 7 min. Collected fractions were evaporated under vacuum (Hetrovac, Heto Denmark). The fractions were submitted to SDS-PAGE and electroblotting. N-terminal amino acid sequencing of excised bands of Coomassie-stained components was carried out using an ABI 477A/120A automated protein sequencing system (Applied Bio- systems, Inc., Santa Clara, CA). Spectrophotometric scanning of SDS-PAGE was done with a Shimadzu CS-930 dual wavelength scanner. Peptide Mapping and Analysis— Purified FGF-BP, either unmodified or carboxymethylated by standard procedures, was subjected to proteo- lytic digestion with trypsin (3 h) or S. aureus V8 protease (18 h) at 37 °C in 50 m M NH 4 HCO 3 , pH 8.1, with a protein-to-protease ratio of 50:1 (w/w). The resulting peptides were purified by reverse-phase chromatography on a SMART-system equipped with a 2.1 3 100-mm C2/C18 m RPC column (Amersham Pharmacia Biotech) using a linear gradient from 0 to 60% of acetonitrile in 0.1% trifluoroacetic acid at 25 °C (flow 0.15 ml/min).

The fibroblast growth factor (FGF) 1 family, with its prototype members FGF-1 and FGF-2, comprises structurally related heparin-binding proteins involved in a variety of biological processes including morphogenesis, angiogenesis, and tissue remodeling (1)(2)(3)(4)(5)(6). Signaling by members of the FGF family is dependent upon a dual-receptor system, consisting of four high affinity tyrosine kinase receptors, termed fibroblast growth factor receptors (FGFRs), and of low affinity heparan sulfate proteoglycan (HSPG) that enhances ligand presentation to the FGFRs (7)(8)(9). Despite the ubiquitous presence of FGFs throughout the body, their interaction with the signaltransducing receptors is tightly controlled. Sequestration of FGF by the heparan sulfate-rich extracellular matrix (ECM) seems to fulfill this role by inflicting a local differential distribution of FGFs and FGFRs. Furthermore, the binding and release from ECM seem important as means of protection and optimizing the biological effect of FGF-2 (10,11). Several mechanisms have been proposed for the release of active FGFs from the ECM reservoir. The classical route is release of the growth factors upon digestion of HSPG by glycosaminoglycan-degrading enzymes or protease activity (9,(13)(14)(15). However, recent transfection studies have revealed the existence of an alternate mode that does not require matrix degradation but instead involves an FGF-binding protein (FGF-BP). The heparin-binding protein FGF-BP is distinct from cellular receptor molecules and binds to FGF-1 and FGF-2 in a non-covalent, reversible manner. Besides decreasing the affinity or accessibility of FGF for ECM-HSPGs, FGF-BP has been shown to protect FGF-2 from degradation and preserves its mitogenic activity (16 -20).
In rodents in situ hybridization and Northern analysis have shown that FGF-BP expression is prominent in skin and intestine during the perinatal growth (19). The down-regulation of FGF-BP in the adults is, however, reversed in tumor samples, cell lines derived from squamous cell carcinomas (SCC), and some colon cancers (16,20,21). The involvement of FGF-BP in skin cancer is supported by the findings that the chemotherapeutic agent all-trans-retinoic acid reduces FGF-BP expression in SCC xenografts, inhibits their angiogenesis, and leads to a decrease of the tumor growth rate (22,23 mRNA expression is up-regulated by direct transcriptional mechanisms in phorbol ester-promoted skin cancer (19,24). The significance of FGF-BP expression in tumors has also been assessed in vitro and in situ by transfection studies. Overexpression or conversely reduced expression by ribozyme targeting suggests that FGF-BP mobilizes and activates extracellular stored FGF-2 and promotes angiogenesis and growth of xenografted tumors in mice (16,17,19).
Previously, FGF-BP cDNAs from human and mouse have been cloned, and their predicted open reading frames encode proteins of 234-and 251-amino acid residues, respectively. A comparison of their deduced amino acid sequences shows 63% similarity, including a conserved location of all cysteine residues (19,20). The purified form of FGF-BP is a 17-kDa heparinbinding protein (HBp17), derived from culture medium conditioned by the human epidermal carcinoma cell line A431. HBp17 binding of FGF-1 and FGF-2 can be reversed by heparin, which in turn binds a cluster of basic amino acids in the C-terminal half of the molecule (20,25). Because of its scarcity, very little information is available regarding structural features of FGF-BP so far.
In bovine mammary gland the growth-promoting activity peaks in the early stage of the last trimester of gestation. Accordingly, mammary gland secretion drawn from this period is a rich source of different growth-promoting substances (26,27). In the present study, we report the purification and structural characterization of bovine FGF-BP recovered from bovine prepartum mammary gland secretion (BPMS). Peptide mapping, cDNA cloning, and N-terminal amino acid sequencing allowed disclosure of the primary structure of a member of the FGF-BP family for the first time.

EXPERIMENTAL PROCEDURES
Miscellaneous-FGF-1 and FGF-2 (purified from bovine brain) were from R & D Systems (Minneapolis, MN). Trypsin (L-1-tosylamido-2phenylethyl chloromethyl ketone-treated), Staphylococcus aureus V8 protease, and chymotrypsin were all from Worthington. Reagents used for protein sequencing were purchased from Applied Biosystems (Foster City, CA). Polyvinylidene difluoride (PVDF) filters were from Millipore Corp. (Bedford, MA). Proteins were iodinated with Na 125 I using chloramine T as the oxidizing agent. The reaction was stopped with metabisulfite, and the ligands were desalted on a PD10 column (Amersham Pharmacia Biotech) equilibrated with 50 mM sodium phosphate buffer, pH 7.4. M r markers used were SeeBlue Pre-Stained standards (NOVEX, San Diego, CA) or SDS molecular weight standard from Bio-Rad. The amount of purified protein was assayed with a kit provided by Bio-Rad.
Preparation of Bovine Prepartum Mammary Gland Secretion-Bovine prepartum mammary gland secretion (BPMS) was collected 3-5 weeks before parturition by hand milking. BPMS was acidified by lowering the pH to 4.6 and centrifuged 30 min at 40,000 ϫ g. The resulting fatty overlayer and the precipitate material were discharged, and the clear interface was readjusted to pH 7.4. The processed secretion was transferred to dialysis tubing (Spectrum, Laguna Hills, CA) and dialyzed overnight at 4°C against 20 liters of 50 mM NH 4 HCO 3 , pH 8.1.
Purification of the FGF-BP-300 ml of BPMS dialysate was batchincubated for 2 h with 400 ml of heparin-agarose (Amersham Pharmacia Biotech) equilibrated in 50 mM NH 4 HCO 3 , pH 8.1. The heparinagarose was extensively washed with 50 mM NH 4 HCO 3 and packed on a column (diameter 50 mm). The column was developed with a linear gradient of 50 mM to 1.5 M NH 4 HCO 3 at 4°C with a flow rate of 50 ml/h. The eluate was collected and aliquots from each fraction were taken before freeze-drying. The aliquots were subjected to SDS-PAGE and electroblotting following standard procedures. A selection of consecutive fractions from the heparin affinity chromatography was dissolved in 0.1% trifluoroacetic acid and passed over a 1-ml Resource RPC column (Amersham Pharmacia Biotech) equilibrated with 0.1% trifluoroacetic acid (flow 1 ml/min). Bound material was eluted using a linear gradient of 0 -55% 2-propanol in 0.1% trifluoroacetic acid for the initial 65 min and then with a 55-80% gradient for the following 7 min. Collected fractions were evaporated under vacuum (Hetrovac, Heto Laboratory Equipment, Denmark). The fractions were submitted to SDS-PAGE and electroblotting. N-terminal amino acid sequencing of excised bands of Coomassie-stained components was carried out using an ABI 477A/120A automated protein sequencing system (Applied Biosystems, Inc., Santa Clara, CA). Spectrophotometric scanning of SDS-PAGE was done with a Shimadzu CS-930 dual wavelength scanner.
Peptide Mapping and Analysis-Purified FGF-BP, either unmodified or carboxymethylated by standard procedures, was subjected to proteolytic digestion with trypsin (3 h) or S. aureus V8 protease (18 h) at 37°C in 50 mM NH 4 HCO 3 , pH 8.1, with a protein-to-protease ratio of 50:1 (w/w). The resulting peptides were purified by reverse-phase chromatography on a SMART-system equipped with a 2.1 ϫ 100-mm C2/C18 RPC column (Amersham Pharmacia Biotech) using a linear gradient from 0 to 60% of acetonitrile in 0.1% trifluoroacetic acid at 25°C (flow 0.15 ml/min). Selected tryptic peptides were subdigested with chymotrypsin (3 h) at 37°C, and the generated peptides were purified using the described conditions. Cystine-containing fractions were selected by amino acid analysis of aliquots treated with performic acid prior to hydrolysis. N-terminal amino acid sequences were obtained by sequential Edman degradation as described above. Carbohydrate composition analysis was carried out using a Dionex pulsed-electrochemical detector and a polymeric anion-exchange CarboPac PA1 column (Dionex Corp., Sunnyvale, CA) as described (28). The oligonucleotides were purchased from DNA Technology, Aarhus, Denmark. All oligonucleotides were [␣-32 P]ATP-labeled at the 5Ј using T4 polynucleotide kinase (New England Biolabs) and incubated with a bovine mammary gland cDNA library derived from a Holstein cow, lactating for 12 days, and 1-day post-estrus (Stratagene, La Jolla, CA). Approximately 400,000 plaques were screened by in situ hybridization. Prehybridization was performed in 6ϫ SSPE (1ϫ ϭ 10 mM sodium phosphate, pH 7.4, 0.15 M NaCl, 1 mM EDTA), 5ϫ Denhardt's solution (1ϫ ϭ 0.02% bovine serum albumin, 0.02% Ficoll-400, 0.02% polyvinylpyrrolidone), 0.5% SDS, 0.1 mg/ml salmon sperm DNA for 2 h at 42°C. Hybridization was done in 6ϫ SSPE, 1ϫ Denhardt's solution, 0.25% SDS, 0.2 mg/ml salmon sperm DNA for 36 h at 42°C. The filters were washed twice at room temperature with 2ϫ SSC (1ϫ ϭ 0.15 M NaCl, 15 mM sodium citrate, pH 7.0) containing 0.1% SDS and once at 42°C with 6ϫ SSC containing 0.1% SDS. The air-dried filters were exposed to Fuji RX film. cDNA inserts from positive clones were subcloned and sequenced on both strands using Big Dye sequencing kit and an ABI prism 310 Genetic analyzer (Perkin-Elmer). Both the cDNA and the amino acid sequences were used in a FASTA search looking for structural homologs in the EMBL data base.

RESULTS
Purification of Bovine FGF-BP-Bovine prepartum mammary gland secretion collected 3-5 weeks before parturition was centrifuged to remove debris and subjected to heparin affinity chromatography (Fig. 1A). The bound proteins were eluted with a linear gradient of ammonium bicarbonate, and aliquots from each fraction were collected, pooled, and freezedried. Proteins present in each pool were resolved by SDS-PAGE and electrotransferred to PVDF membrane. Coomassiestained protein bands were then dissected and subjected to N-terminal amino acid sequence analysis. Homology searches based on the retrieved amino acid sequences revealed that the predominant 83-kDa protein recovered was lactoferrin. Based on spectrophotometric scanning, we estimated that lactoferrin and lactoferrin-derived fragments comprised more than 90% of all the heparin-binding protein isolated from the secretion (Fig. 1C). Among the minor protein components, eluting in the 1.0 -1.2 M ammonium bicarbonate interval, was a previously unidentified protein migrating as a 36-kDa band in SDS-PAGE. The 36-kDa band resembled the amino acid sequence of FGF-BP as previously deduced from human and mouse cDNA (19,20). Thus, to purify the putative bovine FGF-BP homolog further, the 36-kDa enriched fractions were pooled, freezedried, and processed by reverse-phase HPLC (Fig. 1B). As revealed by SDS-PAGE and confirmed by N-terminal amino acid sequencing, the 36-kDa protein and its 34-kDa N-terminal truncated derivative coeluted in the peak fraction at approximately 38% solvent B (Fig. 1B). By using a colorimetric assay, we estimated that about 30 g of purified FGF-BP were extracted from 300 ml of BPMS.
cDNA Cloning and Sequencing-To obtain the complete protein sequence, bovine FGF-BP was cleaved using trypsin, and/or chymotrypsin, and the amino acid sequences of reversephase HPLC-purified peptides were determined. A bovine mammary gland cDNA library was screened using degenerated oligonucleotides based on the amino acid sequence derived from the bovine FGF-BP digests. The resulting 1170-bp cDNA and its derived amino acid sequence are shown in Fig. 2. The deduced amino acid sequence was verified by the finding that all proteolytically generated peptides analyzed by N-terminal amino acid sequencing were encompassed in the open reading frame. A putative initiation codon was found at position 85, representing the first in-frame ATG codon in the 5-prime region. Two consecutive stop codons commenced from position 787. Assuming the methionine codon to be the initiator, the open reading frame encodes 234 amino acids with a calculated molecular mass of 22,648 Da. Only one potential N-glycosylation site with the canonical (Asn-X-(Ser/Thr), X Pro) sequence was found (residues 155-157). A prediction of the secretory signal sequence using the neural network-based SignalP program (30) suggests that the bovine FGF-BP has a cleavage site between Ala 23 and Lys 24 . However, instead of the predicted Lys 24 as the leading amino acid residue, Edman degradation of the purified FGF-BP protein revealed N-terminal processing at two sites as follows: a primary site between Arg 31 and Gly 32 and a secondary site between Arg 51 and Ser 52 . In addition, a more thorough analysis of the amino acid sequencing data revealed a background sequence accompanying the nonreduced FGF-BP. As deduced from the cDNA sequence, the background sequence, FLSMVQGSSC, represents the C-terminal part (residues 225-234) of the purified FGF-BP (Fig. 2).
When the whole protein sequence of bovine FGF-BP was submitted to a data base search, only human and mouse FGF-BP showed close resemblance (Fig. 3). Similar to the bovine sequence the open reading frame of human FGF-BP translates to 234 residues (20), and the mouse protein was extended with an additional 17 residues (19). The amino acid sequence of the bovine protein was 60 and 53% identical to that of FGF-BP from human and mouse, respectively. Notably, the positions of all the 10 cysteines present in each of the three proteins are conserved.
Analysis of Disulfide Bridges-To localize disulfide bridges in bovine FGF-BP, the purified protein was cleaved with trypsin, and the resulting fragments were separated by reversephase HPLC. Aliquots of the eluted peptide fractions were subjected to amino acid analyses, and retrieved fractions containing cysteine were analyzed by N-terminal sequencing and MALDI-TOF-MS (Table I). Sequence analysis revealed the tryptic peptides T6 and T9 to coelute. Taking into account that both peptides contain only one cysteine, the finding suggests that T6 and T9 are covalently connected, via a disulfide bond between Cys 71 and Cys 88 . Mass spectrometry confirmed this conclusion, as the obtained mass peaks at m/z 1160.2, 734.8, and 1893.6 agree with the calculated weight of the T6 and T9, and a disulfide-linked complex between the two peptides. Trypsin digestion also produced a cysteine-containing fraction that comprised T10, T15, and T17ϩT18. Subdigestion with chymotrypsin followed by reverse-phase HPLC resulted in a chromatogram where T15 (containing Cys 130 ) coeluted with and a new peptide T10.I containing Cys 97 . Mass spectroscopy of the fraction revealed a mass corresponding to the size of T15 linked to T10.I, demonstrating that Cys 97 is bound to Cys 130 . The chymotrypsin cleavage also resulted in the formation of a T10.II peptide, which contained Cys 106 . N-terminal sequencing revealed that T10.II coeluted with the tryptic peptide T17ϩT18, which includes Cys 142 . In support of a disulfide bond between Cys 106 and Cys 142 , the detection of the molecular mass of 1571.7 Da corresponded well to the size of a disulfide-linked complex between T10.II and T17ϩT18. Both amino acid sequencing and mass spectrometry showed the presence of T29.I,

TABLE I Peptide mapping analysis of bovine FGF-BP using N-terminal sequencing and MALDI-TOF-MS
Tryptic and chymotryptic fragments of bovine FGF-BP were isolated by reverse-phase HPLC. Obs., observed mass estimated by MALDI-TOF-MS as described under "Experimental Procedures." Cal., calculated mass based on the assigned amino acid sequence. which originates from a tryptic cleavage following Lys 208 and an intrinsic C-terminal processing between Phe 224 and Phe 225 . The existence of an internal disulfide bond between the included Cys 214 and Cys 222 was demonstrated by identification of a di-phenylthiohydantoin-cystine in the 14th cycle of the Edman degradation of T29.I. This also agrees with the finding that reduction of T29.I resulted in a mass increase from m/z 1890.7 to 1893.5, corresponding to the reduction of one cystine for two cysteines. The disulfide bond between Cys 198 and Cys 234 was not directly identified but deduced by elimination. Moreover, the existence of a disulfide bond between Cys 198 and Cys 234 explains why the observed background sequence of T29.II accompanies the N-terminal amino acid sequence of unreduced bovine FGF-BP (see Fig. 2).
Glycosylation-Tryptic peptide fragments purified by reverse-phase HPLC were analyzed for the presence of the amino sugars GalNAc and GlcNAc. GlcNAc was detected in the tryptic fraction T22, which contained the only consensus sequence for N-glycosylation in the binding protein. A carbohydrate composition analysis of T22 performed by high pH anion-exchange chromatography showed the presence of GlcNAc, GalNAc, galactose, mannose, and sialic acid in a molar ratio of 2:1:1:5:1 (results not shown). In favor of N-glycosylation, amino acid sequencing of T22 did not detect an asparagine in the third step, and finally, instead of the calculated 915.1-Da molecular mass of T22, mass spectrometry revealed a peak at m/z 2790.2 (Table I). Collectively our results warrant the attachment of a carbohydrate to Asn 155 .
In contrast to N-glycosylation, no consensus amino acid sequence for O-glycosylation exists. However, evidence for an O-linked glycosylation on Ser 172 was found. Following digestion with S. aureus V8 protease, GalNAc was observed in a fraction containing the octapeptide 168 LMEPSPMD 175 . Thus, no Ser 172 was detected upon Edman degradation, and the mass spectrum of the octapeptide showed a peak at m/z 2121.9. Since the calculated mass of the unmodified peptide was 920.1 Da, this suggests that the carbohydrate moieties amount to 1201.8 Da.
Binding of Bovine FGF-BP to FGF-The previously reported presence of an FGF binding domain within human FGF-BP prompted us to investigate the binding specificity of bovine FGF-BP. We tested the ability of FGF-BP to bind FGF-1 and FGF-2, using ligand blotting analysis. Following electrotrans-fer of the two growth factors, bovine serum albumin, and a crude preparation of BPMS to PVDF, the membrane was probed with radiolabeled bovine FGF-BP. The analysis revealed 125 I-FGF-BP binding affinity towards FGF-2, whereas we observed no detectable binding to FGF-1 (Fig. 4). As controls, the bovine serum albumin and crude BPMS did not bind 125 I-FGF-BP, and the binding of radiolabeled FGF-BP could be blocked with an excess of the non-radiolabeled ligand. The binding activity of FGF-BP toward FGF-2 was not observed in the presence of 10 g/ml heparin (data not shown). DISCUSSION This paper describes the purification and characterization of a novel heparin-binding protein secreted from the bovine mammary gland. Partial amino acid sequences obtained from purified material enabled isolation of a full-length cDNA from a mammary gland library and deduction of the amino acid sequence of bovine FGF-BP. The derived amino acid sequence was confirmed by peptide mapping, covering 74% of the purified protein. Glycosylation sites and disulfide bridges were also assigned (Fig. 5).
Inspection of the cDNA sequence predicts that the unprocessed bovine FGF-BP is 234 residues long, which is equivalent to the human counterpart. Based on the known structural requirements it is expected that signal peptidase would cleave between residues 23 and 24 during the secretion process, resulting in Lys 24 at the N terminus. Nevertheless, the predominant forms of the purified protein observed in the present study started at Gly 32 and Ser 52 , respectively. The enzyme responsible for this processing remains to be elucidated. However, it is noteworthy that a dibasic sequence is found in the C terminus of the removed peptides (Arg 30 -Arg 31 and Lys 48 -X-X-Arg 51 ), suggesting that one or more subtilisin-like endopeptidases may be responsible for the processing (31). A similar phenomenon is seen in human FGF-BP, where it has been reported that the purified protein starts with Lys 34 (20). However, no dibasic sequence prior to an eventual processing site is found in the human precursor. Whether secondary modification of the N-terminal is a common phenomenon linked to FGF-BPs must await further investigations as well as elucidation of the proteases involved.
Applied methods for estimating the molecular weight of the isolated FGF-BP gave different results. Based on the amino acid composition, the molecular mass of the predominant 203residue form was calculated to 22,648 Da. Measurement by mass spectrometry estimated the molecular mass to 28,549 Da, whereas the protein migrates as a 36-kDa protein in SDSpolyacrylamide gels. The discrepancy in mass between the isolated protein and the translated cDNA sequence was at least partly due to post-translational modifications. A search for glycosylation sites revealed the presence of glycans on Asn 155 and Ser 172 , resulting in an additional mass of about 3074 Da. The collected data do not explain the remaining difference between the calculated and observed molecular weights. However, the apparently anomalous behavior of the bovine FGF-BP in SDS-polyacrylamide gels is most likely a result of the basic nature of the protein (estimated pI is 9.3) and the attached carbohydrate. The literature provides no data on glycosylation of the other FGF-BPs; however, the bovine N-linked glycosylation site has no counterpart in the predicted amino acid sequences of human and mouse FGF-BP. Although the O-glycosylated residue in the bovine protein is conserved in human