Glycosylation of Human Protein C Affects Its Secretion, Processing, Functional Activities, and Activation by Thrombin*

Human protein C (HPC) is an antithrombotic serine protease that circulates in the plasma as several glycoforms. To examine the role of glycosylation in the function of this protein, we singly eliminated each of the four potential N-linked glycosylation sites by site- directed mutagenesis of Asn to Gln at amino acid po-sitions97,248, and 313 (HPC derivatives Q097,Q248, and Q313) or at the unusual consensus sequence Asn-X-Cys at 329 (HPC derivative Q329). The cDNAs for wild type and each derivative were inserted into expression vectors and expressed both transiently and stably in human 293 and hamster AV12-664 cells. We demonstrate that N-linked glycosylation at position 97 in the light chain of HPC is critical for efficient secretion and affects the degree of core glycosylation at Asn-329. Glycosylation at position 248 affects the in- tracellular processing of the internal Lys-Arg (KR) KR cleavage site, and partial glycosylation at the sequence Asn-329-X-Cys is responsible for the natural a-glyco-form. Altering the glycosylation pattern of the protein had no significant effect on the level of fully y-carbox-ylated HPC secreted from the 293 cell line. However, elimination of glycosylation sites in the heavy chain resulted in a 2- to 3-fold increase in anticoagulant activity. Utilizing synthetic substrate, both the K,,, and kcat were Rates rates were determined using bovine thrombin (0.20 nM) alone or in complex with a 10-fold molar excess of rabbit thrombomodulin. The reaction mix contained 20 mM Tris, 0.15 M NaC1, 0.1 mg/ml BSA, 0.02% NaN3, 3 mM CaC12, pH 7.4. Protein C, both wild type and glycosylation derivatives, was at a concentration of 1.61 pM in the activation reaction with thrombin alone and at 0.32 pM in the activation with TMT complex. Fully activated PC was obtained either by incubating the reaction mix overnight with additional TMT or by activating material with 0.05 p~ thrombin in complex with soluble recombinant human thrombomodulin TMD1-75 (29). Activation rates were determined by removing aliquots from the activation reaction mix at various time points, to a 96-well plate containing hirudin at a concentration in 50-fold molar excess of thrombin. The chromogenic sub- strate, S-2366, was added to each well at a final concentration of 0.75 mM, and amidolytic activity was measured as the change in absorb- ance uaits/min at 405 nM in a ThermoMax kinetic microtiter plate reader (Molecular Devices). Rates were determined by converting change in OD/min to pg of aPC generated by using the specific activities determined for each of the glycosylation mutants and plotting uersus activation time. kcat activity using S-2366 at a final concentration of 1 mM. The tlh in plasma was determined from decay curves generated using the EnzFitter Software.

thelia1 cell surface, it is converted to its active form by limited proteolysis with a-thrombin in complex with a cell surface membrane protein, thrombomodulin (1)(2)(3). The activated form of protein C (aPC)' has potent anticoagulant activity due to its ability to inactivate factors Va and VIIIa (4), and it has been reported to have profibrinolytic activity (5, 6). Protein C plays a critical role in the regulation of thrombin generation (reviewed in Refs. [7][8][9] and may be effective in the treatment of a number of thrombotic diseases (7,10).
The native HPC molecule is shown schematically in Fig. 1. HPC circulates predominantly as a disulfide-linked heterodimer (7,11) composed of a light chain (Mr -25,000) and a heavy chain (Mr -41,000) containing the serine protease domain. The remaining material circulates in a single chain form, lacking the removal of an internal Lys-Arg (KR) dipeptide (12). The light chain contains the region of y-carboxyglutamic acid (Gla) residues that is highly conserved among vitamin K-dependent proteins. This specialized post-translational modification is required for calcium-dependent membrane binding and functional activity (reviewed in Ref. 8). In addition to this complex modification, HPC contains -1 residue of P-hydroxyaspartic acid (Asp-71) (13)(14)(15) that is believed to be involved in a Gla region-independent calciumbinding site (16, 17).
Both plasma-derived and recombinant HPC have several forms of the heavy chain, designated a, p, and y, with apparent M, of -41,000, 37,000, and 32,000, respectively. Previously, we demonstrated that recombinant HPC secreted from tunicamycin-treated cells (18) contained one heavy chain band, which also was observed by Yan et al. (15) following treatment of purified HPC with N-glycanase. These data suggested that the various subforms represented differences in glycosylation pattern. Of the four potential sites for N-linked glycosylation in HPC, one is in the EGF domain of the light chain at amino acid position 97, and the remaining three are in the heavy chain at positions 248, 313, and 329. Interestingly, the site at 329 contains the unusual sequence Asn-X-Cys that has been shown to be glycosylated in bovine protein C (19) and in von Willebrand Factor (20). In this paper, we examine the role of glycosylation at each of the four potential sites in HPC. We demonstrate that glycosylation of HPC at specific sites influences the secretion, proteolytic processing, and rate of activation by thrombin.

Materials
Restriction endonucleases, T4 DNA ligase, and T4 polynucleotide kinase were obtained from New England Biolabs, mutagenesis was The abbreviations used are: aPC, activated human protein C; HPC, human protein C zymogen; rHPC, recombinant human protein C; SDS, sodium dodecyl sulfate; PAGE, polyacrylamide gel electrophoresis; Gla, y-carboxyglutamate; wt, wild type; al-AT, el-antitrypsin; EGF, epidermal growth factor; TM, thrombomodulin; T, thrombin; BSA, bovine serum albumin. performed using Bio-Rad Muta-Gene in vitro mutagenesis kit (catalog #170-35711, agarose was from Bethesda Research Laboratories, and chemicals were purchased from Boehringer Mannheim and/or Sigma. Ham's F-12 and Dulbecco's modified Eagle's medium were purchased from GIBCO, and fetal bovine serum was from HyCIone. Eli Lilly supplied bovine insulin and hygromycin B. Hirudin (H-7016) and human a-1-antitrypsin (A9024) were purchased from Sigma. Vitamin K1 (Aquamephyton) was purchased from Merck Sharp and Dohme. Bovine thrombin ( 1 0~ NIH units/mg), bovine serum albumin, fatty chased from Miles Scientific. Rabbit thrombomodulin was obtained acid free bovine serum albumin, and human transferrin were purfrom N. Esmon (Ok~ahoma Medical Research Foundation) or was purchased from American Diagnostica. The chromogenic substrate S-2366 was obtained from KabiVitrum. Sheep polyclonal antibody to HPC was from American Diagnostica. Peroxidase conjugated goat anti-mouse IgG was purchased from Tago, and 2,2'-azino-di-{3-ethylbenzthiazoline-sulfonate] was purchased from Bionetics. Biotinylated antibodies and the avidin-biotin complex detection kit were obtained from Vector.

Methods
Si~e-d~rec~ed ~u~~e~s~ of Protein C-Changes in each of the glycosylation sites were made by site-directed mutagenesis using the cDNA coding sequence previously described (21). A 730-base-pair SaEI/SacI fragment corresponding to a region from just N-terminal of the first EGF domain through approximately half of the serine protease domain was isolated from the plasmid pLPC (13) and cloned into both M13mp18 and M13mp19. These clones, designated MP18-nHPC.730 and MPl9-NHPC.730, were dtimately used as the source of single-stranded DNA template for generation of glycosy~at~on mutants MP18-QO97 and MPf9-Q248, respectively. In addition, a 2016-base-pair SacI/EcoRI fragment containing the remainder of the serine protease domain was isolated from pLPC and cloned into M13mp18. This clone, designated MP18-NHPC.2016, was ultimately used as the single-stranded DNA template source for generat.ion of glycosylation mutants MPWQ313 and MP18-Q329. Oligonucleotide primers were made with an Applied Biosystems model 380 DNA synthesizer and had the following sequences: Pm-Q097, 5'-GGACAACGGC-3'; Pm-Q248,5'-CGGTGGTGCTCTTGCTGTAC-TGAGGGTGCACGAAGACCTCCTTGATGTCC-3'; Pm-Q313, 5'-

GCCGAGAGAAGGAGGCCAAGCGCCAACGCAC~TTCGTCCT"
CAAC-3'; PM-Q329, 5"GATTCCCGTGGTCCCGCACCAAGAAT-GCAGCGAGGTCATGAGC-3'. These primers were designed not only to change the codon of their corresponding asparagine residue to one coding for glutamine, but also to introduce silent mutations facilitating restriction endonuclease screening prior to actual DNA ApaLI and DdeI (Q248), HaeII (6313) and BsmI (Q329). Mutagenesis sequencing. Introduced restriction sites were HindIII and PstI (Q097), was performed by the method originally described by Kunkel (22) with modifications to the protocol in the Bio-Rad in vitro mutagenesis kit. The clones MP18-~HPC.730~ MP19-NHPC.73U, and MP18-NHPC.2016 were transformed into competent CJ236 (dut-ung-) cells, and uracil-containing single-stranded phage DNA was isolated for use as template. Approximately 250 ng of template was annealed to 4 pmol phosphorylated primer in 10 p1 of buffer containing 20 mM Tris-HC1 (pH 7.41, 2 mM MgC12, and 50 mM NaCl at 85 "C, allowed to slow-cool to room temperature over a period of approximately 2 h, and then placed on ice. The primer extension reaction was modified as follows: initial incubation was for 15 min at 4 "C, 15 min at 25 "C, and 90 min at 37 "C, followed by the addition of 5 gI of deionized/ distilled water, 1 pl of ligation buffer (10 X as described by New England Biolabs), and 1 pl of T4 DNA ligase (400 units/pl) for a final volume of 20 pl. Incubation was continued overnight at 16 "C, and then 80 pl of deionized/distilled water was added as diluent, and the reaction was frozen briefly at -20 "C. Competent MV1190 (dut+ ung+) cells were transformed with 7.5 gl of the above reaction, and individual resultant plaques were picked for analysis.
Preliminary screening of phage was by DNA miniprep procedure and restriction endonuclease digestion. Candidate positives were then verified by dideoxy nucleotide sequencing using an AB1 370A Automated Fluorescent Sequencer. Following sequence confirmation of mutant phage, double-stranded phage DNA was isolated from bacterial cells and the protein C insert was isolated. Reconstitution of the complete protein C coding sequence was accomplished via a threepart ligation involving the 3788-base-pair EcoRIISalI fragment from pLPC, a 730-base-pair SalIISacI fragment (from pLPC, MP18-QO97, or MP19-Q248) and a 2016-base-pair SacI/EcoRI fragment (from pLPC, MP18-Q313, or MP18-Q329). DNA from the resultant clones (designated pLPC-Q097, pLPC-Q248, pLPC-Q313, and pLPC-Q329) was used to transform Escherichia coli strain GM48 (dam-) to facilitate removal of the entire protein C coding sequence on a BclK fragment.
E x~r e s s w~ of ~e c o r n b i~n~ Protein C-The 1425-base-pair BclI fragment of each of the pLPC clones above was excised and inserted into the eukaryotic expression vector pGT-hyg at its unique BclI site. Plasmid pGT-h derivatives contain the following elements beginning at the EcoRI site and proceeding counterclockwise. The EcoRI to blunt-ended NdeI fragment of pBR322 containing the ampicillin resistance gene and origin of replication, the PvuII to blunt-ended BanHI fragment ofpSV2-hyg' (derivative of pSV2-h (13) constructed by A. Smith and P. Berg) containing a hygromycin phosphotransferase expression cistron, the blunt-ended NdeI (nt 2297) to AccI (nt 2246) restriction fragment of pBR322, the AccI (nt 4339) to StuI (5122) restriction fragment of BKVP2 (23), the synthetic GT element previously described (24), the StuI (5122) to AurII restriction fragment of BKVP2 containing the enhancer sequence (23), the Nsp74251 (nt 5931) to PvuII (nt 6071) of human adenovirus type 2 containing the major late promoter, a synthetic sequence comprising the spliced tripartite leader sequence of human adenovirus type 2, HindIII and Bcll linker, the Ban1 to PstI fragment of the cDNA coding sequence for human protein C (21) flanked by BclI linkers, the 610-bp MboI fragment of SV40 containing the small t splice junction, and the 988base-pair BclI to EcoRI fragment of SV40 containing the polyadenyl-pGTH-QO97, pGTH-Q248, pGTH-Q313 and pGTH-Q329, where Qn ation signal. The completed expression plasmids were designated designates the Asn to Gln substitution at the indicated amino acid position (n). Each expression plasmid was purified on a CsCi density gradient, linearized with FspI, and used to transfect 293 cells.
Cell ~u~t u r e , DNA Transfection, and Drug Selection-The adenovirus-transformed human kidney 293 (ATCC CRL 1573) was grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 50 pg/ml gentamicin, and 10 pg/ml vitamin K1. One day prior to transfection, cells were plated at a density of lo6 cells/ 55 cm' . Calcium phosphate"DNA precipitates were prepared (25) with 10-50 pg of expression plasmid DNA and no carrier DNA. Four hours after transfection, the culture medium was replaced, and 2-3 days after transfection the culture medium was replaced with selection medium containing 200 pg of hygromycin B/ml. Clonal cell lines resistant to hygromycin B were isolated 4-5 weeks after applying drug selection. These recombinant cell lines were expanded, and the presence of HPC in the conditioned culture medium was determined by the enzyme-linked immunosorbent assay described below.
Detection of HPC Antigen-A solid phase sandwich enzyme-linked immunosorbent assay, with a sensitivity of 5 ng/ml, was used to measure the level of recombinant HPC (rHPC) antigen in culture media. 96-well plates were coated either with sheep anti-HPC polyclonal antibody or a murine monoclonal antibody. HPC antigen sandwiched between the polyclonal antibodies was followed by a biotiny~ated rabbit anti-sheep IgG and developed with avidin-conjugated horseradish peroxidase and 0-phenylenediamine color indicator. Alternatively, the antigen was sandw~ched between m o n~l o n a l antibodies designated HPCl and HPC3 and developed using horseradish peroxidase-conjugated goat anti-murine IgG. For quantitation of HPC E n the culture medium, the polyclonal assay was used because of differences in immunoreactivity of the variants with the two monoclonal antibodies.
SDS-PACE and Western Blot Analysis-The recombinant glycosytation derivatives were subjected to ~~S-polyacry~amide gel electrophoresis (26) on 10% SDS-polyacrylamide gels (acrylamide/bis 300.8) under either reducing or nonreducing conditions. Protein was detected by Coomassie staining or by electrophoretic transfer of protein from the gel to a nitrocellulose membrane for Western blot analysis. After transfer to nitrocellulose, the membrane was blocked for 1 h in 5% nonfat dry milk in phosphate-buffered saline, pH 7.4, and incubated with a sheep anti-HPC polyclonal antibody followed by biotinylated rabbit anti-sheep IgG. The blot was developed using the Vector avidin-biotin complex kit and 4-chtoro-~-naphtho~, Functional Activity of rHPC-The zymogen HPC was converted to the activated protein C (aPC) by treatment with rabbit thrombomodulin-bovine thrombin essentially as described previously (27). The anticoagulant activity of the recombinant aPC was measured with an activated partial thromboplastin time clotting assay. The amidolytic activity of the recombinant aPC was determined by the hydrolysis of a tripeptide substrate Glu-Pro-Arg-p-nitroanilide (S-2366) at pH 7.4, Glycosylation of Human Protein C 25 "C. One unit of aPC was defined as the amount of aPC required to release 1 pmol of p-nitroanilide in 1 min at 25 "C, pH 7.4, using an extinction coefficient for p-nitroanilide at 405 nm of 9620 M" cm" (28). Plasma-derived aPC was purified, as previously described (13), and used as a standard for both functional assays. Purijication and Calcium Dependency of rHPC-To obtain material for purification and the determination of calcium-dependent elution from anion exchange (15), the recombinant cell lines were grown in a modified mixture of Dulbecco's modified Eagle's and Ham's F-12 media (1:3) containing 1 pg/ml human insulin, 1 pg/ml human transferrin, and 10 pg/ml vitamin K1. Conditioned media were collected, adjusted to a final concentration of 5 mM benzamidine and 4 mM EDTA, pH 7.4, and then adsorbed to an anion-exchange column (Pharmacia Fast Flow Q resin) as described by Yan et al. (15). After washing with 3 column volumes of 20 mM Tris, 0.15 M NaCI, 5 mM benzamidine, 2 mM EDTA, pH 7.4, and 3 column volumes of 20 mM Tris, 0.15 M NaCl, 5 mM benzamidine, the bound HPC was eluted with 20 mM Tris, 0.15 M NaCI, 10 mM Cacl,, 5 mM benzamidine, pH 7.4. Residual material was eluted in a high salt buffer of 20 mM Tris, 0.5 M NaC1, and 5 mM benzamidine. The partially purified protein was desalted in a Centricell unit with a 30,000 M, retention filter (Polysciences, Inc. #Y18674-8) following dilution with a buffer containing 20 mM Tris, 0.15 M NaCl, and 0.02% NaN3, according to the recommended procedure of the manufacturer. The concentration of protein was determined by the BCA assay method (Pierce Chemical Co.) using FFQ-purified wt HPC as a standard. Quantitation of the wt HPC reference using BSA as a standard corresponded to values obtained measuring UV absorbance at 280 nm with less than 5% error. Protein purity was determined by densitometer scanning of Coomassie-stained protein separated by SDS-PAGE using a Shimadzu dual wavelength TLC scanner. Concentration of each glycosylation mutant was determined as the percentage of total protein that was HPC, relative to the reference standard wt HPC.
Activation Rates and Kinetics-Activation rates were determined using bovine thrombin (0.20 nM) alone or in complex with a 10-fold molar excess of rabbit thrombomodulin. The activation reaction mix contained 20 mM Tris, 0.15 M NaC1, 0.1 mg/ml BSA, 0.02% NaN3, 3 mM CaC12, pH 7.4. Protein C, both wild type and glycosylation derivatives, was at a concentration of 1.61 p M in the activation reaction with thrombin alone and at 0.32 p M in the activation with TMT complex. Fully activated PC was obtained either by incubating the reaction mix overnight with additional TMT or by activating material with 0.05 p~ thrombin in complex with soluble recombinant human thrombomodulin TMD1-75 (29). Activation rates were determined by removing aliquots from the activation reaction mix at various time points, to a 96-well plate containing hirudin at a concentration in 50-fold molar excess of thrombin. The chromogenic substrate, S-2366, was added to each well at a final concentration of 0.75 mM, and amidolytic activity was measured as the change in absorbance uaits/min at 405 nM in a ThermoMax kinetic microtiter plate reader (Molecular Devices). Rates were determined by converting change in OD/min to pg of aPC generated by using the specific activities determined for each of the glycosylation mutants and plotting uersus activation time.
To determine the K,,, and kcat values for wt and 6313 HPC at 3 mM CaCI2, concentrations from 1 to 20 p M of each were activated with 0.2 nM thrombin/2.0 nM thrombomodulin, and aliquots were taken over time to determine rates of activation by measuring amidolytic activity as described above. K, and Vmex were determined by plotting the slopes of activated PC over time versus the concentration of substrate. Nonlinear regression analysis was performed with EnzFitter software (Elsevier Biosoft, United Kingdom).
For a-1-antitrypsin (u,-AT) inhibition studies, 100 pg of each purified rHPC glycosylation derivative was activated in a 1-ml reaction with 1 NIH unit of bovine thrombin and 1.5 pg of 75-kDa soluble recombinant thrombomodulin (29) in 0.5 mM CaCI2, 150 mM NaC1, 20 mM Tris, pH 7.4, for 45 min. Fully activated rHPC at a concentration of 20 nM was incubated with from 0 to 71 p M ul-AT in 3 mM CaC12, 150 mM NaCl, 20 mM Tris, pH 7.4, and 1 mg/ml BSA. At selected times, aliquots were removed and the activated protein C activity was determined by amidolytic activity using 5-2366 at a final concentration of 1 mM. As previously described (30), apparent firstorder rate constants, kl, were calculated from the slopes of plots of activated protein C over time in the presence of molar excess of inhibitor. Association rate constants, k~, were calculated as k2 = k,/ (I] (inhibitor concentration), and half-life of protein C was calculated at tH = 0.69/k2(I]. The inhibition of each HPC derivative in plasma was determined by incubating normal human plasma (citrated) with 10 nM activated wt rHPC or each rHPC derivative. The plasma concentration was 90% (v/v) in the final reaction, with the remaining volume consisting of buffer containing 3 mM CaCl,, 150 mM NaC1, 20 mM Tris, pH 7.4, and 1 mg/ml BSA. At selected times, aliquots were removed and activated protein C activity was determined by amidolytic activity using S-2366 at a final concentration of 1 mM. The tlh in plasma was determined from decay curves generated using the EnzFitter Software.

RESULTS
Expression and Secretion of Glycosylation Derivatives-Shown schematically in Fig. 1 are the four potential N-linked glycosylation sites in HPC at amino acid positions 97, 248, and 313 and at the unusual consensus site Asn-X-Cys at 329. Each of these sites was singly eliminated by substitution of Gln (Q) for Asn (N) using site-directed mutagenesis (Fig. 1B).
Vectors for the expression of wt HPC and each derivative were constructed and introduced into the adenovirus-transformed cell lines 293 (human kidney) and AV12-664 (Syrian hamster tumor). Previously, we have shown that at low levels of expression, both of these cell lines will secrete fully processed rHPC (13,31) and that stable recombinant 293 cell clones will perform each of the complex post-translational modifications of the protein even at high expression levels (15,32). To determine if the alteration in glycosylation pattern affected the secretion of HPC, we initially performed transient expression experiments in the two cell lines as described under "Methods." As shown in Fig. 2, the levels of rHPCs Q248, Q313, and Q329 secreted from both cell lines were essentially the same as those observed with the wt HPC. However, the level of rHPC Q097, containing the deletion of glycosylation site 97 in the light chain, was significantly reduced in both cell lines. These data indicate that glycosylation of HPC in the EGF domain is important for efficient secretion of the protein.
For further characterization, stable recombinant 293 cell lines secreting w t HPC and each of the glycosylation derivatives were created by the isolation of hygromycin-resistant clones as previously described (13). For further study, recombinant clones secreting from 5 to 10 pg/m1/24 h in serum-free medium were chosen. To analyze the secreted product, samples of serum-free conditioned medium from representative clones were subjected to SDS-PAGE, and the rHPCs were detected by Western blotting. As shown in Fig. 3, elimination of the glycosylation site in the light chain (Q097) resulted in a reduction in the apparent M , of the light chain, and elimination of the glycosylation sites in the heavy chain (Q248, Q313, Q329) resulted in alterations in the distribution of the a , p, and y subforms. From an examination of the mobility of the heavy chain subforms from each derivative, it was apparent that the M , of the carbohydrate side chain at each site must be different (Asn-329 -4000, Asn-248 -3000, Asn-313 -2000). Elimination of glycosylation at Q248 resulted in a 3-4-fold increase in the amount of single chain rHPC secreted, indicating that glycosylation at this site is important for the intracellular removal of the KR dipeptide. Removal of the glycosylation site at 329, with the unusual consensus sequence Asn-X-Cys, resulted in the loss of a-form heavy chain. These data suggest that partial glycosylation at 329 is responsible for a-form HPC. Interestingly, the elimination of glycosylation at 313 resulted in the secretion of a mixture of modified 0and y-form rHPC in a ratio similar to that of wt HPC.
The minor band migrating faster than the major HPC doublet in Q313 comigrated with nonglycosylated heavy chain obtained by treatment of HPC-secreting cells with tunicamycin (data not shown).
To obtain a more accurate determination of the relative proportion of the heavy chain glycoforms in the derivative HPCs, purified material (described below) was separated by SDS-PAGE and stained with Coomassie, and the resulting bands were quantitated by densitometric scanning as described under "Experimental Procedures." The data, summarized in Table I, show that the percentages of wt HPC CY, 8, and y subforms were identical to Q313 modified @, y, and nonglycosylated subforms. These data suggest that the site at 313 is always glycosylated and that the CY, /3, and y subforms represent tri-, di-, and monoglycosylated heavy chain, respectively. Upon elimination of either Q329 or Q248, we did not

TABLE I Distribution of hrauy chain R1ycoforrn.q in tot and drriuatiup H I T S
The amount of each glycoform was determined bv scanning densitometry of Coomassie-stained SDS-PAGE as described under "Methods."The results are the average of scans from five independent gels using from 2 to 10 pg of each HPC isolated from the calcium fraction (see Fig. 4 Calcium Dependency and Functional Activities of H I T GI?cosylation Deriuatiues-Using the procedure of Yan rt al. (15). the calcium dependency profiles of wt HPC and each of the glycosylation derivatives were compared. In this procedure, fully y-carboxylated HPC (containing 9 residues of Gla) elutes from an anion-exchange resin in 10 mM ca(II, (CaCI, frac-tion), and incompletely carboxylated HPC (6-7 Gla) remains bound, but can he subsequently eluted with 0.5 M NaCl (NaCI fraction). Very poorly carboxylated molecules do not hind t.o the column (15, 18). Serum-free conditioned culture medium from each of the recombinant 293 lines was adsorbed to the anion exchange resin, washed and sequentially eluted as described under "Methods." As shown in Fig.  4A, the elution profile for each of the derivatives was essentially the same as obtained with wt HPC; from 85 to 95% of the total material eluted in the 9-Gla calcium fraction. Thus, alternation in glycosylation pattern of HPC had no significant effect on the y-carboxylation of the light chain. Following this one-step procedure, the purity of each preparation ranged from 75 to 95%. Shown in Fig. 4 R is a Coomassie-stained gel of the calcium-eluted material from the ion exchange resin before (-) and after (+) incubation with thromhin in complex with thrombomodulin (TM/T). The activation peptide of each derivative could be removed as indicated by the increase in migration of each of the heavy chain glycoforms.
Following activation to a P C by thrombin/thrombomodulin, the functional activities of the wt and derivative rHPCs were determined using the fully y-carboxylated material (calcium fraction). The amidolytic activity was assessed using the synthetic tripeptide substrate Glu-Pro-Arg-p-nitroanilide (S-2366), and the anticoagulant activity was measured by the activated partial thromboplastin time. As shown in Table 11, the amidolytic and anticoagulant activities of the heavy chain glycosylation derivatives were significantly increased compared with wt aPC, with each derivative having an anticoagulant activity of 2-3 times that of human plasma-derived aPC. We did not attempt to determine kinetic parameters in the crude clotting assay. Alternatively, we determined the kinetic parameters for wt and each of the glycosylation derivatives using the synthetic tripeptide substrate (Table  111). While the increases in activity of Q313 and Q248 were primarily due to increased kc.,, Q329 displayed a slightly higher affinity (decrease in K",). T h e overall enzyme efficiency ( k J

ytic and anticoagulant actitdim of HPC' glycosylation sit? drrivatiurs
Functional activities were determined as described under "Experimental Procedures" using the material purified from the ion-exchange calcium fraction. The first numher in parentheses is the fold increase in activity over plasma-derived HPC. The number of intlependent samples ( n ) determined in duplicate or triplicate is indicated.

TABLE Ill Kinetic pararnetrrs for u1t and derivative HPCs using synthrtic tripeptide (;lu-Pro-Arg-p-nitroanalidP IS-2.766) substratr
Results are the mean of two experiments with 20 data points used for nonlinear regression analysis. The error values were determined from the best fit curve using the EnzFitter program. K,) of each heavy chain derivative was approximately twice that of wt HPC.

Inhibition of the Activated Gl.vcosylation Dcriuatiucs--nl-A T is a major physiologic inhihitor of aPC at the active site.
To determine if the glycosylation derivatives had an altered interaction with nl-AT, varying concentrations of purified n l -A T were incubated with each HPC and the inhihitory activitv was determined. The rate of inhibition of wt aPC and each of the derivatives was identical at all concentrations of n,-AT tested; the pseudo-first-order plots for inhihition of wt a P C and Q313 are shown in Fig. 5 . The association rate constants ( k , ) and half-lives were determined for wt aPC and each derivative as described under "Methods" and are summarized in Table IV. The data clearly demonstrated that there were no differences in the kinetics of inhibition hy n,-AT. The association rate constants of -10 s" and T,, values of

in normal human plasma and by cui-antitrypsin
Values were determined and calculated as described under "Experimental Procedures." Results are the average of three independent determinations each performed in duplicate. s" and 26 min, respectively, obtained for wt aPC by Heeb et al. (30). The half-lives for wt aPC and each derivative in freshly prepared human plasma also were found to be identical (Table IV). Therefore, even though the functional activities of the heavy chain glycosylation derivatives were increased, none were inhibited any more rapidly either in plasma or by active-site inhibition with a,-AT. Actiuation by Thrombin-The efficient activation of HPC to aPC by the thrombomodulin-thrombin complex requires a calcium-dependent conformational change in the activation region (33)(34)(35)(36). T o determine if the removal of glycosylation affected this calcium-dependent activation, we examined the activation rates for wt HPC and each glycosylation derivative using bovine thrombin in complex with either detergentsolubilized rabbit thrombomodulin or recombinant soluble human thrombomodulin TMD1-105 kDa (29). The rates of activation by thrombin were identical with either source of thrombomodulin. As shown in Fig. 6A, the rates of activation of wt HPC, Q097, Q248, and Q329 were identical. However, the rate of Q313 activation by the TM/T complex was approximately 3-fold higher. We determined the rate of activation for Q313 by the TM/T complex at various calcium ion concentrations (Fig. 6B). There was no significant difference in calcium dependency; activation was inhibited at low concentrations of calcium, and both w t aPC and Q313 had halfmaximal stimulatory calcium concentrations of -0.35 mM.
T o determine whether the increase in rate of activation of Q313 by the TM/T complex was due to an increase in affinity for the enzyme-cofactor complex or an increase in turnover, kinetic analyses were performed at 3 mM calcium. As shown in Table V, the kc,, for Q313 was slightly less (-20%) than that of the wt HPC, however, the K,,, for $313 was substantially reduced. The K,,, obtained for wt HPC is in good agreement to the value of 5 FM previously reported for plasmaderived and recombinant human HPC (33,36). The overall efficiency (k,,,/K,) with Q313 as substrate was 2.5-fold greater than that observed with wt HPC. We also compared the rates of activation of Q313 and wt HPC by thrombin alone. Because the activation of HPC by thrombin alone is inhibited at physiological levels of calcium, we compared rates both at 3 mM calcium and in the presence of EDTA. Under both conditions, the rate of activation of Q313 was 2.5-3-fold higher than the wt HPC, similar to that observed above for the TM/ T complex. Due to the very high K,,, for activation of HPC by thrombin alone (33), we did not attempt to establish whether these increased rates were due to K,,, or kc,, differences.
However, our overall data suggest that the removal of glycosylation a t Q313 increases the affinity for thrombin in a calcium-independent manner.  The results are the average of three independent determinations. The values were determined by nonlinear regression analysis as described under "Experimental Procedures."

DISCUSSION
In recent years, carbohydrate side chains have begun to receive attention as being integral to the functional properties of glycoproteins. It has become clear that N-linked side chains can control a wide variety of functions, including plasma clearance, signal transduction, receptor activation, intracellular folding, and activity (reviewed in Refs. [37][38][39][40][41][42]. In this paper, we demonstrate that glycosylation a t each of the sites on HPC has an effect(s) on the properties of the protein, including dramatic increases in functional anticoagulant activity.
A striking effect of HPC glycosylation appears to be to attenuate the functional activity of the protein; elimination of each glycosylation site in the heavy chain resulted in an increase in both functional amidolytic and anticoagulant activity (Table 11). Previous studies (15,43) have demonstrated that the degree of glycosylation of HPC may affect functional activity, as removal of sialic acid residues by neuraminidase treatment results in a 2-3-fold increase in anticoagulant activity. Sialic acid residues themselves can play a role in the function of a protein as patients with oversialylated fibrinogen have extended thrombin times and abnormal fibrin monomer aggregation (44), and desialylation of factor IX results in a loss of coagulant activity (45). It is not clear whether or not the increase in activity of the heavy chain glycosylation derivatives is entirely due to removal of the charged sialic acid residues upon removal of the entire side chain, or if it is simply due to removing steric hindrance of the active site to protein substrates or cofactors. With the synthetic substrate, the increase in activity following removal of glycosylation at 248 and 313 was primarily due to an increase in enzyme kcat.
This would suggest that the conformation of the active site is slightly altered in these variants. Of interest, the increased activity of 329 was at least in part due to a decreased K , for substrate. Although the kinetic parameters with synthetic substrate may not reflect the reason for the increased anticoagulant activities, the data nevertheless suggest that the naturally occurring @-form of HPC is more active. The data further suggest that changes in the ratios of the a-and pglycoforms in certain patients (see Ref. 46) may have clinical significance.
In the circulation, aPC is inhibited primarily by two proteins, al-AT and protein C inhibitor, both of which form a complex via acylation of the active site. As shown in Table  IV, there were no differences in the rates of inhibition by al-AT among any of the HPCs. Additionally, the half-lives of each HPC in plasma were the same, indicating that the interaction with protein C inhibitor was also unaltered by eliminating glycosylation sites. As indicated above, the increase in activity of Q329 appeared in part to be due to an increased affinity. If this increase in active-site affinity was due to removal of steric hindrance, one might have speculated a corresponding increase in the rate of inhibition would have occurred. Most likely, inhibitor docking is the rate-limiting step controlling the rate of inhibition, not active-site availability. Furthermore, our data indicate that the carbohydrate side chains in HPC are not affecting the interaction with either inhibitor in the wt protein.
We previously demonstrated that treatment of an HPCsecreting 293 cell line with tunicamycin, which inhibits Nlinked glycosylation, resulted in a 5-10-fold drop in secretion. We now demonstrate that glycosylation at position 97 in the EGF domain of the light chain appears to be the site critically required for the efficient secretion of the protein, as its removal results in a 70-75% drop in secretion (Fig. 2). The effect of N-linked glycosylation on the folding and transport of proteins in the endoplasmic reticulum has been recently reviewed (40), and it is believed that the presence of side chains in strategic positions on the polypeptide are required for correct folding and subsequent efficient transport. Removal of Asn-97 had no effect on the proteolytic processing or y-carboxylation of the secreted material, however, the ratio of a to p heavy chain increased. It would appear that, in addition to transport, the correct conformation of the EGF domain in the endoplasmic reticulum also affects the conformation of the serine protease domain, thereby altering the accessibility of the glycosylation site a t position 329. In support of this, we also have observed an increase in the amount of a-form HPC in two individual mutants of the site of phydroxylation in the EGF domain.' Thus, the EGF domain of HPC appears to play a role in the overall structural properties of the molecule during secretion.
As previously reported, approximately 50% of the nonglycosylated rHPC secreted from tunicamycin-treated cells was single chain (47). We show in Fig. 3 that removal of a glycosylation site at position 248 resulted in a substantial * B. Gerlitz and B. W. Grinnell, unpublished observation. increase in the amount of single chain protein, suggesting that the carbohydrate at this site affects the efficiency of the processing of the KR dipeptide and, like position 97, plays some role in the structural conformation of the molecule during secretion. It is interesting to note that several mammalian cell lines, including CHO, C127, and BHK-21, are not capable of efficient processing of the KR dipeptide (14,48,49). Although this cell line dependence for processing may be due to differences in the efficiency of dibasic cleavage, the effect could also be due to the specific carbohydrate side chain processed at Asn-248. As indicated above, the increase in enzymatic activity of Q248 was primarily due to an increase in active site turnover, suggesting that glycosylation at this site also affects the conformation of the fully processed and secreted protein.
Recently, Miletich and Broze (46) demonstrated that an antibody to the region containing Asn-329 did not recognize the a-form but would recognize the p-form, suggesting that the p-form of HPC represents protein not glycosylated at position 329. Our data by site-directed mutagenesis directly confirm this finding. We and others (11,13,46) have previously demonstrated that plasma-derived HPC contains from 70 to 80% a-form heavy chain. In contrast, the rHPC secreted from several cell lines typically contains from 50 to 60% aform. We also have reported that the functional activity of rHPC has a 1.3-1.4-fold higher specific activity when compared with plasma-derived HPC having essentially the same Gla content (13,15). Possibly, this higher than expected activity can now be explained by the fact that the recombinant material has a higher percentage of high activity p-form protein.
Based on an examination of HPC from warfarin-treated patients, Miletich and Broze (46) proposed that the partial glycosylation a t position 329, which has the unusual consensus sequence Asn-X-Cys, is controlled by the rate of translation and thus temporal availability of the site. This hypothesis may not be valid in cell culture with rHPC, as we have not observed a difference in the ratio of a-to P-form HPC in recombinant lines secreting from 50 ng/ml to 30 pg/ml HPC, nor have we observed a difference in the rate of HPC translation upon inhibition of y-carboxylation using warfarin or Chloro K (18). Alternatively, these authors have suggested that under-carboxylated molecules are structurally (rather than temporally) more accessible for glycosylation a t Asn-329. In support of this, we have found that HPC containing only 7 of the 9 Gla residues (15) possesses an increased proportion of a-chain HPC.3 Therefore, it would appear that the degree of glycosylation at Asn-329 is influenced by the correct post-translational processing of the Gla domain. Coupled with our observations on the EGF domain, our data indicate that the correct modification of the light chain as a whole influences the conformation of the heavy chain and the accessibility of Asn-329 during secretion into the lumen of the endoplasmic reticulum.
We have demonstrated that the removal of glycosylation a t Asn-313 results in an increase in the rate at which thrombin can cleave the activation peptide, primarily due to an increase in affinity of thrombin for this HPC substrate. Lack of glycosylation at Asn-313 did not alter the calcium dependency of activation (Fig. 6). Therefore, the removal of this glycosylation site does not affect the calcium-induced conformational change(s) thought to occur in the activation region. The carbohydrate at 313 likely is near the activation region in the three-dimensional structure and to some extent sterically

12190
Through site-directed mutagenesis, we have begun to un-