Proteinase inhibitor 6 cannot be secreted, which suggests it is a new type of cellular serpin.

We have recently described a new serine proteinase inhibitor, proteinase inhibitor 6 (PI-6). This serpin has features that suggest it may function intracellularly, but its close resemblance to ovalbumin serpins like plasminogen activator inhibitor 2 (PAI-2) raises the possibility that it is secreted to regulate an extracellular proteinase. To determine whether PI-6 is secreted, we have examined its cellular distribution by immunohistochemistry and have attempted to induce its release from platelets and from cultured cells. We find that PI-6 is present in endothelial and epithelial cells, but it is apparently cytoplasmic and it is not released from cells in response to phorbol ester, dibutyryl cAMP or tumor necrosis factor α treatment. It is also not released from activated platelets. The addition of a conventional signal peptide to the amino terminus of PI-6 directed its translocation into the endoplasmic reticulum (ER), resulting in glycosylation but not secretion of the molecule. By contrast, the addition of the same signal peptide to PAI-2 markedly enhanced its translocation and secretion. Glycosylated PI-6 was sequestered in the ER and was incapable of interacting with thrombin. The failure of PI-6 to move along the secretory pathway, and the loss of inhibitory function of ER-localized PI-6, demonstrates that unlike PAI-2, PI-6 is not naturally secreted. Taken together, these results suggest that PI-6 has evolved to fulfil an intracellular role and that it represents a new type of cellular serpin.

We have recently described a new serine proteinase inhibitor, proteinase inhibitor 6 (PI-6). This serpin has features that suggest it may function intracellularly, but its close resemblance to ovalbumin serpins like plasminogen activator inhibitor 2 (PAI-2) raises the possibility that it is secreted to regulate an extracellular proteinase. To determine whether PI-6 is secreted, we have examined its cellular distribution by immunohistochemistry and have attempted to induce its release from platelets and from cultured cells. We find that PI-6 is present in endothelial and epithelial cells, but it is apparently cytoplasmic and it is not released from cells in response to phorbol ester, dibutyryl cAMP or tumor necrosis factor ␣ treatment. It is also not released from activated platelets. The addition of a conventional signal peptide to the amino terminus of PI-6 directed its translocation into the endoplasmic reticulum (ER), resulting in glycosylation but not secretion of the molecule. By contrast, the addition of the same signal peptide to PAI-2 markedly enhanced its translocation and secretion. Glycosylated PI-6 was sequestered in the ER and was incapable of interacting with thrombin. The failure of PI-6 to move along the secretory pathway, and the loss of inhibitory function of ER-localized PI-6, demonstrates that unlike PAI-2, PI-6 is not naturally secreted. Taken together, these results suggest that PI-6 has evolved to fulfil an intracellular role and that it represents a new type of cellular serpin.
Serine proteinase inhibitors (serpins) are a family of structurally related proteins that regulate the activity of serine proteinases involved in extracellular processes such as coagulation, fibrinolysis, complement fixation, and embryo implantation. Several members of the family have lost proteinase inhibitory function and have evolved extracellular functions such as serving as lipophilic molecule transporters and peptide hormone precursors (1).
Recently, we and others have identified a new Arg-serpin known as proteinase inhibitor 6 (PI-6), 1 the placental thrombin inhibitor, or the cytoplasmic antiproteinase (2,3). Although PI-6 efficiently inhibits the extracellular proteinases plasmin, trypsin, thrombin, and urokinase in vitro (4), it is unusual because it is present in cytosolic extracts, it is not found in the medium of cultured cells, it lacks a conventional signal sequence, and it is sensitive to oxidation (2,5). These properties suggest that PI-6 may have an intracellular function. At present, the only serpin with a clearly defined intracellular role is the viral protein crmA, which is an inhibitor of granzyme B and the interleukin-1␤-converting enzyme (6,7). PI-6 closely resembles the ovalbumin serpins. This group of proteins includes ovalbumin, plasminogen activator inhibitor 2 (PAI-2), the squamous cell carcinoma antigens (SCCA-1 and SCCA-2), maspin, and the monocyte neutrophil elastase inhibitor (8,9). All of the ovalbumin serpins lack conventional signal sequences, yet they are found as extracellular glycoproteins. At least two of these ovalbumin serpins, PAI-2 and SCCA, appear to exist mainly as cytosolic proteins but are efficiently secreted and glycosylated in response to specific stimuli. For example, glycosylated PAI-2 is released from monocytes in response to tumor necrosis factor ␣ and phorbol ester treatment (10), and SCCA is released from transformed squamous epithelial cells (11). Thus it cannot be inferred from the lack of a conventional signal sequence and an apparent cytosolic location that PI-6 is confined intracellularly or that it has an intracellular function.
To determine if PI-6 is released to function in the extracellular milieu, we have examined its cellular distribution using immunohistochemistry and have attempted to induce its secretion from cultured cells and platelets. Furthermore, we have provided it with a conventional signal sequence to assess whether it can be efficiently glycosylated and released if directed into the secretory pathway. We find that PI-6 is located in endothelial cells, in platelets, and in a subset of epithelial cells but that it is not released from activated platelets nor from cultured cells in response to tumor necrosis factor ␣, phorbol ester, or cAMP analogues. PI-6 directed into the secretory pathway is glycosylated but loses inhibitory activity and is retained in the endoplasmic reticulum. On the basis of these studies, we conclude that PI-6 is not naturally secreted and that it is a true intracellular serpin.

EXPERIMENTAL PROCEDURES
Cell Culture-COS-7 and K562 cells were maintained as described previously (5). U937 cells were maintained as for K562 cells. COS-7 cells were transfected using the DEAE-dextran/chloroquine method as described (12).
Antibodies and Histology-Rabbit anti-aminoglycoside 3Ј-phosphotransferase (NEO) antibodies were purchased from 5 Prime-3 Prime, Inc. The anti-PAI-2 monoclonal antibody was from American Diagnostics. The anti-PI-6 antiserum used in the secretion, pulse-chase and immunofluorescence experiments has been described previously (2). For the histological experiments, new rabbit anti-PI-6 antibodies were prepared as before, except that the antigen was recombinant PI-6 produced in a yeast expression system (4). Preparation of 4-m paraffin-embedded tissue sections and immunohistochemical staining was as described previously (13), except that 3-amino-9-ethlycarbazole (stock solution 0.4% (w/v) in formamide) was used as the developing reagent (DAKO). Briefly, sections were dewaxed and blocked in 3% (v/v) hydrogen peroxide followed by 10% (v/v) horse serum in phosphate-buffered saline (PBS). Sections were incubated for 1 h in an empirically determined dilution of the primary antibody (typically 1:200), washed in PBS, and then incubated for 20 min in a 1:100 dilution of biotinylated swine anti-rabbit immunoglobulins (DAKO E353). Following a further wash in PBS, streptavidin-horseradish peroxidase (DAKO K377) was added for 30 min. Sections were washed again in PBS and developed using 3-amino-9-ethlycarbazole freshly diluted to 6% (v/v) from the formamide stock solution into 0.1 M acetate buffer (pH 5.2). Slides were counterstained and mounted in Crystal/Mount (Biomeda Corp.). As controls, serial sections were incubated with a similar dilution of nonimmune rabbit serum as primary antibody.
Thrombin-Thrombin was prepared from prothrombin purified from human plasma (14). Iodinations, and estimations of its concentration and activity were performed exactly as described previously (2).
Isolation and Activation of Platelets-Human platelets from 40 ml of blood were prepared and washed as described previously (16). 5 ml of a platelet suspension containing 4 ϫ 10 8 cells/ml was divided into three parts and treated as follows. (i) 250 l were activated by the addition of iodinated thrombin (5 nM) for 10 min at 37°C. (ii) 2.5 ml of the suspension were pelleted, resuspended in 2.5 ml Tyrode's buffer (140 mM NaCl, 1.3 mM KCl, 0.2 mM MgCl 2 , 24 mM NaHCO 3 , 5 mM Hepes, 5.5 mM glucose, pH 7.5), and freeze-thawed 3 times using liquid nitrogen. The disrupted platelets were centrifuged at 100,000 ϫ g for 1 h, and 250 l of the supernatant were incubated with 5 nM iodinated thrombin for 10 min at 37°C. (iii) 2.25 ml of platelets were treated as in (ii) but were first activated with thrombin (5 nM) for 10 min at 37°C and then treated with 1 mM diisopropyl fluorophosphate (Sigma). The samples were either subjected directly to analysis by reducing SDS-PAGE, or were first immunoprecipitated using anti-PI-6 antibodies.
Plasmids-The PI-6 expression vector pSVTfPTI/P is described in Coughlin et al. (5). The plasmids pSVHA/NEO and pSVmHA/NEO are described in Ref. 17. The plasmid pSVHA/PI-6 was constructed as follows. A mutagenic oligonucleotide 5Ј-GCCATCATAGATCT-TCTCGC-3Ј was synthesized that removes the initiation codon of PI-6 to form a BglII site and substitutes Val 3 with Leu (Bresatec, Australia). 20 pmol of this oligonucleotide and 20 pmol of a T3 primer (Promega) were used in a PCR, which also included 5 ng of a PI-6 cDNA template (PTI/P cDNA (5) cloned into Bluescript II KS Ϫ (Stratgene)). Amplification was performed using the proof-reading Vent polymerase (1 unit) under its specified reaction conditions (New England Biolabs). 30 cycles of 95°C for 90 s, 45°C for 60 s, 70°C for 180 s were performed. The amplified fragment was cloned into pCR TM II (Invitrogen) and sequenced completely to verify the presence of the desired alteration and to rule out second site mutations. A BglII-XbaI fragment containing the modified PI-6 cDNA was then separated from the pCR TM II vector fragment and ligated to pSHT (18) that had been digested with BamHI and SpeI. The resulting plasmid was sequenced to verify an in-frame fusion between the pSHT HA signal sequence and the PI-6 cDNA formed by ligation of the compatible BamHI and BglII ends.
The PAI-2 expression plasmid pEUKPAI-2 (a gift of Dr. R. Medcalf) consists of the human PAI-2 cDNA cloned into pEUK-C1 (Clontech). A PAI-2 derivative containing the HA signal sequence was constructed in a similar manner to pSVHA/PI-6. PCR primers 5Ј-ATGGAGGATCCT-TGTGTG-3Ј (sense) and 5Ј-GGACTAGTTAGGGTGAGCAAAATCT-3Ј (antisense) were designed to amplify the coding sequences of PAI-2. The sense primer inserts a BamHI site near the initiation codon and substitutes Leu 4 with Pro. The antisense primer inserts an SpeI site just after the termination codon. Following amplification with Vent polymerase, the fragment was cloned into pCR TM II for verification, released by BamHI-SpeI digestion and ligated to pSHT cleaved with BamHI and SpeI.
Pulse-Chase Experiments-At 48 h posttransfection, 2 ϫ 10 6 COS cells were washed once in PBS and placed in warm serum-free RPMI 1640 medium lacking methionine. After 30 min, the medium was replaced with warm serum-and methionine-free RPMI 1640 medium containing 100 Ci of [ 35 S]methionine and cysteine (Expre 35 S 35 S protein labeling mix, DuPont NEN). The labeling was terminated after 30 min or 1 h, depending on the experiment, by either collecting the medium and lysing the cells or by replacing the labeling medium with warm Dulbecco's modified Eagle's medium containing 10% (v/v) Nu-Serum (Collaborative Research Inc.). In the latter case, the incubation (chase period) was continued for a specified time then terminated by collecting the medium and lysing the cells. Medium and cell extracts were prepared and immunoprecipitated using the appropriate antiserum and protein A-Sepharose as described above. 1 g of thrombin was added to some samples immediately prior to immunoprecipitation. Immunoprecipitates were analyzed by reducing SDS-PAGE. Gels were enhanced in Amplify (Amersham Corp.), and the samples were visualized by fluorography.
Endoglycosidase and Tunicamycin Treatments-Immunoprecipitates from transfected, labeled COS cells were resuspended in 34 l of 0.5% (w/v) SDS, 1% (v/v) ␤-mercaptoethanol and boiled for 10 min. The sample was split in two, 2 l of 0.5 M sodium citrate pH 5.5 was added to each portion followed by 1 l (1000 units) of endoglycosidase H f (New England Biolabs) to one portion only. After 1 h at 37°C, the samples were analyzed by SDS-PAGE and fluorography.
Tunicamycin (10 g/ml, Boehringer Mannheim) was added to the medium of transfected COS cells 18 h before labeling commenced and was included throughout the labeling procedure.
Indirect Immunofluorescence-Transfected COS cells were prepared for analysis by indirect immunofluorescence as described previously (19).

RESULTS
Tissue Distribution of PI-6 -Our previous studies of human and mouse tissues have shown that PI-6 mRNA is present in many embryonic and adult organs (5,20). To identify cells that synthesize PI-6, we carried out an immunohistochemical survey of human adult tissues. Affinity-purified anti-PI-6 antibodies were used with standard methods to probe sections of a variety of tissues including skin, breast, uterus, placenta, testes, skeletal muscle, bone marrow, lung, bowel, and liver. From this analysis, it appeared that PI-6 is synthesized predominantly in capillary endothelial cells and in epithelial cells such as those forming the spinous layer of the epidermis, forming hair follicles, sweat gland secretory ducts, endometrial glands, mammary intralobular ducts, testicular seminiferous tubules, and liver bile ducts. It was also observed in the syncytial trophoblast of placenta. In all of these cells, PI-6 staining appeared to be cytoplasmic, with no staining of membranes or intercellular bridges. The pattern of PI-6 expression is illustrated in Fig. 1, which shows a section of human dermis. Here PI-6 is evident in the small blood vessels and in the differentiated epithelial cells of the sweat gland ducts but not in the gland itself.
The demonstration of PI-6 staining in endothelial and epithelial cells accounts for the wide distribution of PI-6 previously observed by RNA analysis. The presence of PI-6 in these cells is also consistent with a model for PI-6 function in which it is released by epithelial or endothelial cells to participate in the regulation of extracellular proteinases. In this respect, it might resemble the closely related serpin, PAI-2, which is released to regulate urokinase (10). To test if PI-6 is normally released following synthesis or on stimulation of particular cells, we examined its production in a number of systems in which regulated or constitutive release might occur.
PI-6 Is Present in Platelets But Is Not Released on Activa-tion-We have previously noted the presence of PI-6 mRNA in the megakaryoblastic cell line, MEG-01 (5), and we have shown that these cells contain an SDS-resistant thrombin-complexing activity that is immunoprecipitable with anti-PI-6 antibodies. 2 The presence of PI-6 in MEG-01 cells suggests that it may also be present in platelets. To test this, we lysed human platelets by freeze-thawing, incubated aliquots of the lysate with iodinated thrombin, and tested for the presence of an SDS-resistant complex by reducing SDS-PAGE and autoradiography. As shown in Fig. 2 (lane 2), platelet lysate contains two SDSresistant thrombin complexes, the smaller of which is immunoprecipitable by anti-PI-6 antibodies (Fig. 2, lane 5). The larger complex was not immunoprecipitable (Fig. 2, lane 4), and almost certainly consists of thrombin bound to protease nexin I, which is a well characterized and potent thrombin inhibitor contained in platelet ␣-granules (21)(22)(23).
To test whether PI-6 is released on platelet activation, we stimulated platelets with iodinated thrombin to cause release of the granule contents and then separated the platelets from the releasate. The activated platelets were then subjected to lysis by freeze-thawing to prepare cytosolic extracts, which were incubated with a fresh aliquot of iodinated thrombin. All of the samples were then reduced and analyzed by SDS-PAGE and autoradiography. As shown in Fig. 2 (lane 1), the releasate from activated platelets contained the larger complex, which was not immunoprecipitable using anti-PI-6 antibodies (lane 4). By contrast, the cytosol of the activated platelets contained the smaller complex (lane 3), which was immunoprecipitable (lane 6). These results demonstrate that PI-6 is present in platelet cytosol but that its secretion is not induced on platelet activation.
PI-6 Is Not Released by Resting or Stimulated Cultured Cells-We have examined a number of cultured cell lines by RNA analysis, indirect immunofluorescence, or thrombin-com-2 P. B. Coughlin and L. Cerruti, unpublished data. plexing assays for the presence of PI-6. These include primary human umbilical vein endothelial cells; the human lines HeLa, HepG2, HT1080, K562, U937, and THP1; the simian line, COS-7; and the murine lines SP2, Balb/c 3T3, F9, E14, and STO. With the exception of THP1, all of these cells produce PI-6 (data not shown). We have tested conditioned media from most of these cells for PI-6 activity by the thrombin complexing assay or for PI-6 antigen by immunoblotting and have not detected any evidence for PI-6 release into the medium.
By analogy to the situation for PAI-2 or SCCA, we considered the possibility that PI-6 is only released into the medium in response to a specific signal. To test this, we treated K562 cells, U937 cells, or COS-7 cells with inducers of the protein kinase C signal transduction pathway (phorbol 12-myristate 13-acetate) and the protein kinase A pathway (dibutyryl cAMP). We also treated the cells with tumor necrosis factor ␣, which is a potent inducer of PAI-2 (24). Following the treatments, the medium was removed and cytosolic extracts were prepared. Iodinated thrombin was added to both media and extracts, and PI-6 antiserum was used to immunoprecipitate any complexes formed. As shown in Fig. 3, thrombin⅐PI-6 complexes were detected in cytosolic extracts of both untreated and treated cells, but not in any of the media samples. There was no evidence of any increase in the amount of intracellular complexing forming activity in response to any of the treatments, suggesting that PI-6 biosynthesis is not stimulated by these agents. This was supported by RNA analysis, which showed that PI-6 mRNA levels did not increase in the treated cells (data not shown).
Given that PI-6 is inactivated in oxidizing conditions (5), it is possible that PI-6 is released from cultured cells but that functional assays fail to detect it because it is rapidly inactivated. To test this possibility, we analyzed selected media samples for PI-6 antigen by immunoblotting, but we were unable to detect PI-6 protein (data not shown). In addition, we carried out pulse-chase experiments in COS cells transfected with a PI-6 expression vector. PI-6-producing cells were starved in media lacking methionine and then labeled for 30 min with [ 35 S]methionine. After the labeling (pulse) period, complete media were added and the cells were incubated for specified times (chase). At each time point, the medium was collected and the cells were lysed. Both media and lysates were immunoprecipitated with PI-6 antiserum and the immune complexes analyzed by SDS-PAGE. Lactate dehydrogenase assays carried out on the samples showed that a negligible degree of nonspecific cell lysis occurred during the experiment.
As shown in Fig. 4, the predominant protein immunoprecipi-tated from the cell lysates immediately after the labeling period was a 42-kDa species, as expected for PI-6 (5). Preimmune serum did not recognize this protein (data not shown). The amount of PI-6 present in the cell extracts did not decrease markedly over 10 h, and no release into the medium was detected, suggesting that PI-6 is reasonably stable in the cytosol and that it is not secreted under these conditions. Taken with the experiments on release of PI-6 activity from cultured cells, these results suggest that PI-6 is not normally secreted and may have evolved to function intracellularly. Addition of a Signal Peptide to PI-6 -Although a simple interpretation of our results is that PI-6 is not a secreted protein, its similarity to PAI-2 and SCCA leaves open the possibility that it is secreted under certain (perhaps rare) circumstances. If this is true, a simple prediction can be made that if directed to the endoplasmic reticulum (ER), PI-6 should be able to travel through the secretory pathway, and that glycosylated PI-6 should retain proteinase inhibitory function. To test this, we decided to efficiently direct PI-6 into the secretory pathway by providing it with a conventional signal sequence.
As shown in Fig. 5, a derivative of PI-6 (HA/PI-6) containing the influenza virus HA signal sequence fused to the amino terminus of PI-6 was constructed by PCR-mediated mutagenesis of PI-6 and in-frame cloning into the expression vector, pSHT (18). This vector provides the SV40 early promoter followed by the HA signal sequence, cloning sites, and termination codons. A similar derivative of PAI-2 (HA/PAI-2) was constructed as a control (Fig. 5). (Although it is predominantly cytosolic, PAI-2 is known to be capable of travelling through the  conventional secretory pathway (25,26), and the efficiency with which it enters the ER can be enhanced by attaching a heterologous signal sequence (27).) The HA/PI-6, PAI-2, and HA/PAI-2 expression plasmids were transfected into COS cells and subjected to pulse-chase analysis as described above. Entry into the ER and travel through the secretory pathway was predicted to result in an apparent increase in the size of both proteins and release into the medium. Since PI-6 (42 kDa) and PAI-2 (47 kDa) each have three potential N-linked glycosylation sites, increases in size of at least 10 -12 kDa were expected for both molecules. As shown in Fig. 5A, proteins approximately 42, 45, 47, and 50 kDa in size were immunoprecipitated from extracts of COS cells expressing HA/PI-6. These probably represent HA/PI-6 glycosylated at 0, 1, 2, or 3 sites, respectively. The number and sizes of these proteins did not alter during a 3-h chase period, and none were detected in the media, suggesting that HA/PI-6 cannot exit the secretory pathway.
By contrast, three forms of PAI-2 were detected in extracts of COS cells immediately after labeling (Fig. 5B). The smallest, most abundant form represents cytosolic, unglycosylated PAI-2 (47 kDa), which is not released into the medium. Three larger forms were present in extracts in much lower amounts and represent glycosylated PAI-2 (50 -55 kDa). Slight but increasing amounts of these larger forms were detected in media samples during the chase period (Fig. 5B). This pattern of expression is consistent with the inefficient secretion of PAI-2 that has been described previously (26). Addition of the HA signal sequence to PAI-2 significantly altered the pattern of expression (Fig. 5C). In this case, far less 47-kDa PAI-2 was observed, and significant quantities of the larger forms were present in the cell extracts and were secreted into the medium. This confirmed that the HA signal can markedly increase the efficiency of PAI-2 entry into the ER, leading to a substantial increase in the amount of PAI-2 that exits the secretory pathway.
HA/PI-6 Is Glycosylated and Retained in the ER-To confirm that the forms of HA/PI-6 observed in transfected COS cells are glycoproteins, the effect of tunicamycin on HA/PI-6 biosynthesis was examined. Tunicamycin is an inhibitor of N-linked glycosylation that effectively prevents the transfer of precursor oligosaccharides to nascent polypeptides in the ER (28). COS cells producing either PI-6 or HA/PI-6 were labeled in the presence or absence of tunicamycin, and extracts were prepared and immunoprecipitated with anti-PI-6 antiserum. As shown in Fig. 6, treatment with tunicamycin had no effect on the production or size of normal PI-6, demonstrating that the molecule is not usually glycosylated. By contrast, tunicamycin abolished the production of the 45-, 47-, and 50-kDa forms of HA/PI-6, showing that these species are glycoproteins and confirming that HA/PI-6 can enter the secretory pathway.
The failure to detect secretion of the HA/PI-6 glycoforms (Fig. 6) suggested that they are trapped somewhere along the secretory pathway. To assess where this block occurs, indirect immunofluorescence experiments were carried out. COS cells producing either PI-6, HA/PI-6, PAI-2, or HA/PAI-2 were fixed, permeabilized, and probed with either PI-6 or PAI-2 antibodies. After detection with FITC-conjugated secondary antibodies, cells were examined by fluorescence microscopy (Fig. 7). Cells producing PI-6 and PAI-2 showed the diffuse, intracellular pattern of staining expected for cytosolic proteins, whereas cells producing HA/PAI-2 showed the characteristic Golgi staining observed for secreted glycoproteins. By contrast, cells containing HA/PI-6 showed a reticular pattern of staining usually associated with proteins located in the ER.
To confirm its apparent ER localization, the pattern of HA/ PI-6 staining was compared with that seen in COS cells producing HA/NEO, which is a chimeric protein consisting of the HA signal fused to the bacterial enzyme neomycin 3Ј-phosphotransferase (HA/NEO). It has previously been shown that the HA signal can direct the NEO polypeptide into the ER where it is trapped, whereas mutation of the HA signal sequence results in a protein (mHA/NEO) that is cytosolic (17). The expression patterns in cells producing HA/NEO and mHA/NEO resembled those of HA/PI-6 and PI-6, respectively (Fig. 7), supporting the proposition that HA/PI-6 is sequestered in the ER.
Glycosylation of nascent proteins is an ordered process that commences in the ER and continues in the Golgi apparatus. Proteins remaining in the ER normally have different oligosaccharide structures compared with those that have travelled to the Golgi and can be distinguished by the effect of endoglycosidase H (endo H). Resident ER proteins or nascent secretory proteins that have not left the ER contain "high mannose" oligosaccharides that can be removed by endo H. Proteins that have entered the Golgi apparatus have their N-linked carbohydrates modified and are resistant to endo H. On this basis, it was predicted that HA/PI-6 proteins trapped in the ER would be sensitive to endo H. COS cells producing HA/PI-6 or HA/ NEO were metabolically labeled as described above, chased for 0 or 2 h, lysed, and immunoprecipitated using the appropriate antibodies. Immune complexes were split and treated or not treated with endo H prior to SDS-PAGE analysis.
As shown in Fig. 8, endo H treatment of HA/PI-6 immunoprecipitates completely removed the HA/PI-6 glycoforms, and no endo H-resistant proteins were observed 2 h after the labeling was terminated. Similar results were obtained with immunoprecipitates from cells containing the ER-resident HA/NEO protein. These results support the notion that HA/PI-6 is sequestered in the ER, and suggest that little movement of HA/ PI-6 from ER to Golgi occurs.
ER-localized HA/PI-6 Is Nonfunctional-Cytosolic and glycosylated forms of PAI-2 do not differ in proteinase inhibitory activity (25,29). To test whether HA/PI-6 retains inhibitory function, thrombin was added to labeled extracts of mock transfected COS cells, and to those producing PI-6 or HA/PI-6. As described above, the thrombin⅐PI-6 interaction results in an SDS-resistant complex that can be immunoprecipiated using anti-PI-6 antibodies. It was therefore expected that normal PI-6 bound to thrombin would give rise to a 67-kDa complex (2,30), whereas HA/PI-6 and thrombin would give rise to a larger complex due to glycosylation of PI-6. Following immunoprecipitation, SDS-resistant complexes were observed in the mock samples (due to low level production of endogenous PI-6 by COS cells (30)) and in those from cells producing normal PI-6 ( Fig. 9). By contrast, a larger complex between HA/PI-6 and thrombin did not form, although a species corresponding to thrombin complexed with simian PI-6 was evident in these samples. These results suggested that HA/PI-6 in the ER has lost inhibitory function.
Loss of complex forming ability could be due to steric hindrance mediated by the carbohydrate side chains on HA/PI-6 or to malfolding of the molecule in the ER. Treatment of transfected COS cells with tunicamycin did not result in the formation of thrombin⅐HA/PI-6 complexes (data not shown), suggesting that the loss of inhibitory function is not caused by glycosylation of HA/PI-6.

DISCUSSION
Our previous studies have shown that PI-6 is an Arg-serpin that is produced in many tissues and most closely resembles a group of proteins collectively known as the ovalbumin serpins (2,4,30). Two of these ovalbumin serpins, PAI-2 and SCCA, are predominantly cytosolic but can be secreted under certain circumstances (8). This, coupled with the fact that PI-6 efficiently inhibits extracellular proteinases such as plasmin, thrombin, and urokinase, suggested that PI-6 might function outside the cell. Our observation that PI-6 is synthesized by endothelial and epithelial cells is consistent with this idea. However, as discussed below, our failure to detect release of PI-6 under a number of conditions and our demonstration that PI-6 directed into the ER is nonfunctional and not secreted, strongly suggests that PI-6 has an intracellular role.
Most serpins that function extracellularly possess aminoterminal signal peptides that serve to direct entry of the nascent protein into the ER. The ovalbumin serpins are unusual in that secretion of these molecules occurs in the absence of conventional signal sequences. The nature of the signal(s) that direct ovalbumin serpin secretion is poorly understood, but it is thought to comprise sequences in the first and second helices (near the amino terminus) (10). Although PI-6 resembles the ovalbumin serpins in this region, it is not possible to predict from sequence information alone whether PI-6 is secreted.
The efficiency of these unconventional signals varies markedly, ranging from the ovalbumin signal that directs complete secretion of the molecule, to the one on PAI-2 that does not appear to function until stimulation of PAI-2 biosynthesis greatly increases its intracellular concentration. This variation in efficiency can be explained by Rapoport's model for the interaction of a signal sequence with the signal recognition particle (SRP) (31), in which this interaction is postulated as an equilibrium between unbound SRP on one hand and the SRPsignal complex on the other. Thus SRP can have different binding affinities for different signals, and in the case of a poor signal, binding to the SRP might not occur until a significant increase in the signal concentration kinetically favors the for-mation of the SRP-signal complex. Consequently, if PI-6 possesses a weak signal sequence, it can be predicted that increased PI-6 transcription and the biosynthesis of large quantities of PI-6 might be accompanied by constitutive secretion of the molecule. This is certainly the case for PAI-2 produced in phorbol ester-treated U937 cells; PAI-2 transcription increases markedly and is paralleled by secretion of up to 70% of nascent PAI-2 (25). In this study, we were unable to identify a treatment that increases expression of endogenous PI-6 mRNA or that leads to the release of PI-6 protein. Furthermore, overexpression of human PI-6 in COS cells did not lead to secretion.
An alternative pathway for PI-6 release might be through regulated secretion, in which the molecule is stored in an intracellular compartment and released in response to a specific signal. Although our histological and immunofluorescence experiments provide no evidence for such a compartment, we used platelets to model this situation because they contain PI-6, the regulated release of platelet contents is well-characterized, and they are known to release protease nexin 1 (another serpin) on activation. In addition, we treated several PI-6-producing cell lines with agents designed to activate intracellular signaling pathways likely to trigger regulated secretion. PI-6 was not released from activated platelets, nor was it released from stimulated cell lines, suggesting that regulated secretion of PI-6 does not occur. Another argument against intracellular storage and regulated secretion of PI-6 is that entry of proteins into storage compartments usually occurs via the secretory pathway after movement through the Golgi. Since PI-6 cannot move past the ER, it is unlikely to be stored in a conventional secretory granule.
A number of studies have been performed in which normally cytosolic or nuclear proteins have been introduced into the ER by attaching a heterologous conventional signal sequence (17,32,33). In all cases, the proteins successfully entered the ER and were glycosylated but did not move along the secretory pathway. The reason for this is thought to be a failure to fold correctly due to oxidation and formation of inappropriate disulfide bonds. Malfolded proteins in the ER are retained and degraded by a mechanism that remains obscure (34). By contrast, heterologous signal sequences added to normally secreted proteins do not impair processing and secretion (17,27). On the basis of such studies, we predicted that if PI-6 is a intrinsic cytosolic protein, attachment of the HA signal would result in incorrect folding and failure to exit the ER. On the other hand, if PI-6 can be glycosylated and secreted under certain circumstances, attachment of the HA signal should simply enhance the amount appearing in the medium. Our studies clearly support the first prediction and argue strongly that PI-6 is a cytosolic serpin that has evolved to meet an intracellular function. Given that PI-6 is an inhibitory serpin, it is likely that this involves the regulation of an intracellular proteinase.
Taken with our previous work demonstrating differences between PI-6 and the ovalbumin serpins in gene localization and structure (20,35), the results of this study show that PI-6 can now be distinguished from the ovalbumin serpins by three criteria: gene structure, gene localization, and the failure to exit the secretory pathway. The recent finding that the MNEI gene co-localizes with PI-6 on human chromosome 6p25 (36,37) indicates that MNEI may not belong to the ovalbumin serpins as suggested previously (8). If this is the case, it is conceivable that MNEI will have a gene structure similar to PI-6, and will prove to be nonsecreted. Thus PI-6 may be the prototype of a new class of intracellular serpins.
Acknowledgments-We thank Dr. J. Sun for assistance with the FIG. 9. Assessment of the thrombin complexing ability of HA/ PI-6. COS cells were transfected with pSVTf (Vector), pSVTfPTI/P (PI-6), or pSVtfHA/PI-6 (HA/PI-6) DNA. At 48 h posttransfection, cells were starved for 30 min in media lacking methionine and labeled for 4 h in media containing 100 Ci of [ 35 S]methionine. Cell extracts were prepared and analyzed without (Ϫ) or with (ϩ) the addition of thrombin prior to immunoprecipitation with PI-6 antibodies. Immune complexes were collected, reduced, and analyzed by 10% SDS-PAGE and fluorography. Arrow indicates the thrombin⅐PI-6 complex.
construction of HA/PI-6 and Dr. R. Medcalf for donating the anti-PAI-2 monoclonal antibody and providing the expression vector, pEUKPAI-2.