Biosynthesis and Enzymatic Characterization of Human SKI-1/S1P and the Processing of Its Inhibitory Prosegment*

Biochemical and enzymatic characterization of the novel human subtilase hSKI-1 was carried out in various cell lines. Within the endoplasmic reticulum of LoVo cells, proSKI-1 is converted to SKI-1 by processing of its prosegment into 26-, 24-, 14-, 10-, and 8-kDa products, some of which remain tightly associated with the enzyme. N-terminal sequencing and mass spectrometric analysis were used to map the cleavage sites of the most abundant fragments, which were confirmed by synthetic peptide processing. To characterize its in vitro enzymatic properties, we generated a secreted form of SKI-1. Our data demonstrate that SKI-1 is a Ca2+-dependent proteinase exhibiting optimal cleavage at pH 6.5. We present evidence that SKI-1 processes peptides mimicking the cleavage sites of the SKI-1 prosegment, pro-brain-derived neurotrophic factor, and the sterol regulatory element-binding protein SREBP-2. Among the candidate peptides encompassing sections of the SKI-1 prosegment, the RSLK 137- andRRLL 186-containing peptides were best cleaved by this enzyme. Mutagenesis of the latter peptide allowed us to develop an efficiently processed SKI-1 substrate and to assess the importance of several P and P′ residues. Finally, we demonstrate that, in vitro, recombinant prosegments of SKI-1 inhibit its activity with apparent inhibitor constants of 100–200 nm.

Over the last 30 years (1, 2), our understanding of the complex cellular processing by limited proteolysis of inactive secretory precursors into active polypeptides and proteins has greatly expanded. It is now becoming clear that, following removal of the signal peptide, precursor cleavage can occur intracellularly, at the cell surface or within the extracellular milieu. The sites of cleavage are composed of either (i) single or pairs of basic residues (Lys or Arg) within the general motif (R/K)-(X) n -(K/R)2, where n ϭ 0, 2, 4, or 6 and X is any amino acid (aa) 1 except Cys, or (ii) hydrophobic (e.g. Leu, Phe, Val, or Met) and small aa such as Ala, Thr or Ser (3). The former cleavage type occurs in many growth factors and their receptors, most polypeptide hormones and neuropeptide precursors, surface glycoproteins (including adhesion and viral envelope glycoproteins), as well as a host of other secretory proteins (1,2). The latter type of cellular processing has been implicated in the generation of bioactive peptides (4 -6), proteins (7), and transcription factors (8).
Efforts to identify the proteinases responsible for the intracellular processing of precursors at hydrophobic or small aa have led to the recent cloning of a new subtilase called SKI-1 (10) or S1P (11), whose aa sequence is highly conserved among human and rodent species. According to Siezen and Leunissen's classification (9), this enzyme belongs to the pyrolysin branch of subtilases (compared with PCs, which are within the kexin branch). Tissue distribution analyses by both Northern blots and in situ hybridization reveal that SKI-1 mRNA is widely expressed (10,12). We reported previously that human SKI-1 (hSKI-1) produces a 28-kDa product from the 32-kDa brain-derived neurotrophic factor precursor (proBDNF) via selective cleavage within the sequence RGLT2SL (10). Inde-pendently, Sakai et al. demonstrated that hamster SKI-1/S1P is responsible for the site 1 cleavage of sterol regulatory element-binding proteins (SREBPs) (11), highlighting the critical role of SKI-1/S1P in the regulation of the synthesis and metabolism of cholesterol and fatty acids. In their model, SKI-1/S1P cleaves SREBP-2 at an RSVL2SF sequence within the lumen of the endoplasmic reticulum (ER). Mutational analyses demonstrated that the presence of Arg at the P4 substrate position is critical for cleavage, whereas the P1 Leu could be replaced by a number of other aa (8).
In this work, we first present data regarding the cellular biosynthesis of membrane-bound hSKI-1 and its zymogen processing. Then, based on our previous discovery of a secreted (shed) form of hSKI-1 (10), we produced a vaccinia virus (VV) recombinant of a soluble form of this enzyme, which, by analogy to rPC7 (13), we called before the transmembrane domain SKI-1 (BTMD-SKI-1). This isoform, collected from cell media, was used to study the in vitro cleavage properties of this enzyme on a number of synthetic substrates. In addition, we present data on the in vitro inhibitory character of three prosegment constructs of SKI-1, which we obtained as bacterial recombinant proteins. Moreover, we examined the processing of hSKI-1 in LoVo cells infected with a VV recombinant as well as in a stable transfectant of HK293 cells (10).

EXPERIMENTAL PROCEDURES
Vaccinia Virus Recombinant of BTMD-SKI-1-The preparation of a soluble form of hSKI-1 involved the initial amplification by polymerase chain reaction (PCR) of a 1250-base pair product encompassing nucleotides 491-1740 of the hSKI-1 cDNA (12), which includes the initiator methionine. The sense (s) and antisense (as) oligonucleotides were 5Ј-GTGACCATGAAGCTTGTCAACATCTGG-3Ј and 5Ј-ACACTGGTCCC-TGAGAGGGCCCGGCA-3Ј, respectively. This completely sequenced fragment, which had been inserted into the PCR2.1 TA cloning vector (Invitrogen), was first digested with NotI and AccI. It was then ligated with the similarly digested full-length hSKI-1 cDNA 3.5-kilobase pair product, resulting in a product called 5Ј-hSKI-1-FL. In order to obtain a soluble form of hSKI-1 with a hexa-His sequence just before the stop codon, PCR amplification was carried out using the sense and antisense oligonucleotides: 5Ј-ATTGACCTGGACAAGGTGGTG-3Ј and 5Ј-GGAT-CCTCTAGATCAGTGGTGGTGGTGGTGGTGGTGCTCCTGGTTGTA-GCGGCCAGG-3Ј. This resulted in a 165-base pair fragment encoding the C-terminal sequence PGRYNQE 997 -(H 6 )* (10). Following digestion with 5Ј EcoNI and 3Ј XbaI, the product was ligated to the aforementioned and similarly digested 5Ј hSKI-1-FL. This cDNA, coding for BTMD-SKI-1 ending with a hexa-His sequence, was then transferred to the BamHI/XbaI site of the (VV) transfer vector PMJ601. A recombinant was then isolated as previously reported (13). The VV recombinant of full-length hSKI-1 has been described (10).
Biosynthetic Analyses-Seventeen hours following infection with 2 plaque-forming units each of VV:SKI-1 and VV:BTMD-SKI-1 recombinants, human LoVo cells (3 ϫ 10 6 ) were radiolabeled with 500 Ci of [ 3 H]Leu for 2 h or pulsed for 15 min followed by a chase of 2 h, in the presence or absence of 5 g/ml fungal metabolite brefeldin A (BFA) as described (10,14). Media and cell lysates were immunoprecipitated with SKI-1 antiserum directed against either aa 634 -651, or the prosegment comprising aa 18 -188 (10). Immune complexes were resolved by SDS-PAGE on an 8% or 14% polyacrylamide/Tricine gel (10) and the dried gels autoradiographed (10,14). All biosynthesis experiments were performed at least twice.
Isolation and Purification of Recombinant hSKI-1 Prosegments-Three N-terminal fragments of hSKI-1 were isolated by PCR using a common (s) oligonucleotide (5Ј-GGATCCGAAGAAACATCTGGGCGA-CAGA-3Ј) and one of three (as) oligonucleotides (5Ј-CTCGAGGGAGAG-GCTGGCTCTTCG-3Ј, 5Ј-CTCGAGGGCTCTCAGCCGTGTGCT-3Ј, or 5Ј-CTCGAGTGTCTGGGCAACCTGGCGCGGG-3Ј). These prosegment fragments, ending at aa 169, 188, and 196 (10), were cloned in the PCR 2.1 TA cloning vector for sequencing. They were then transferred into the BamHI/XhoI sites of the bacterial expression vector pET 24b (Novagen). These recombinants were transformed into the Escherichia coli strain BL21. Protein expression was induced with 1 mM isopropyl-␤-Dthiogalactoside, and the cultures were grown for 3 h at 37°C. The cell pellets were sonicated on ice in a binding buffer containing 6 M guanidine-HCl (Novagen) until a clear solution was obtained. The clarified and filtered solution was then applied to a nickel affinity column (Novagen) and eluted with 500 mM imidazole. The eluates were dialyzed overnight at 4°C against 50 mM sodium acetate (pH 7). The protein precipitate was solubilized with glacial acetic acid, filtered through a 0.45-m disc, and further purified on a 5-m C4 column (0.94 ϫ 25 cm; Vydac) by reverse-phase high performance liquid chromatography (RP-HPLC). The purity was assessed by Coomassie staining and the identity of the products verified by mass spectrometry on a matrix-assisted laser desorption/time of flight (MALDI-TOF) Voyageur DE-Pro instrument (PE PerSeptive Biosystems). The amounts of prosegments were determined by quantitative amino acid analysis (13).
Expression and Purification of Recombinant BTMD-SKI-1-Following infection of BSC40 cells (75 ϫ 10 6 cells) with 2 plaque-forming units/cell of recombinant VV:BTMD-SKI-1, the cells were washed and incubated at 37°C for 18 h in a serum-free minimal essential medium (Life Technologies, Inc.). Media (45 ml) were then dialyzed, concentrated 20-fold to 2.2 ml on Centriprep-30 (Amicon) and stored at Ϫ20°C in 40% glycerol. For purification, 2 the concentrated media were applied to a Ni 2ϩ affinity resin (Novagen) or a Co 2ϩ affinity resin (CLONTECH) as described by the manufacturer. After two washes with 5 mM imidazole, the protein was eluted with 200 mM imidazole and tested for enzymatic activity and immunoreactivity by Western blot (see below).
Synthesis of Peptide Substrates-All Fmoc amino acid derivatives (L-form), the coupling reagents, and the solvents for peptide synthesis were purchased from PE Biosystems Inc. (XI) Q-hSKI-1(134 -142), Abz-RSLKYAESD-Y(NO 2 )-A. Except for the first two peptides, which were purchased from the Sheldon Biotechnology Institute (McGill University, Quebec, Canada), all other peptides were synthesized with the C terminus in the amide form. Peptides III-XI were prepared on a solid phase peptide synthesizer (Pioneer model, PE Biosystems) using either 2-(1H-benzotriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate/N-hydroxybenzotriazole or O-[7-azabenzotriazol-1-yl]-N,N,NЈ,NЈ-2 Although we managed to produce limited quantities of partially purified SKI-1 using metal chelating resins, there was insufficient enzyme to carry out full kinetic analyses. However, since the medium of WT virus-expressing (or control vector-expressing) cells produced no significant peptide hydrolysis (with the exception of peptides VIII and IX), we mainly used the concentrated media of BSC40 cells infected with VV:BTMD-SKI-1. Thus, the metal chelation-purified enzyme served mainly to verify that the enzyme from concentrated media behaved similarly to this form. We therefore confirmed all of the peptide cleavage sites, the SREBP-2 pH optimum, and the Ca 2ϩ requirement presented below. tetramethyluronium hexafluorophosphate/diisopropyl ethyl amine-mediated Fmoc chemistry with polyamide-linker-polyethyleneglycol (PAL-PEG) unloaded resin and the standard side chain protecting groups (16). For the incorporation of the two unnatural amino acids (Abz and Y(NO 2 )), an extended coupling cycle was used instead of either the standard or fast cycles.
Purification, Analysis, and Digestion of Peptide Substrates-The crude peptides were purified by RP-HPLC using a semi-preparative Chromatographic Sciences Co. Inc. (CSC)-Exsil C18 column (0.94 ϫ 25 cm). Monitoring at 210 nm, the peptides were eluted with a 1%/min linear gradient (5% to 60%) of aqueous 0.1% trifluoroacetic acid/CH 3 CN at 2 ml/min. The peptide purity and concentration were determined by quantitative amino acid analysis (16). The identity of each purified peptide was confirmed by MALDI-TOF spectrometry using the matrix ␣-cyano-4-hydroxycinnamic acid (Aldrich).
For digestions, each peptide was typically reacted at 37°C with 10 l of the concentrated enzyme preparation in a buffer consisting of 50 mM HEPES (ICN Biomedicals Inc), 50 mM MES (Sigma), and 3 mM Ca 2ϩacetate (pH 6.5). The digestion products were separated by RP-HPLC on a Beckman 5-m Ultrasphere C18 column (0.46 ϫ 25 cm) and eluted with a 1%/min linear gradient of aqueous 0.1% trifluoroacetic acid/ CH 3 CN (5-45%) at a flow rate of 1 ml/min. The collected peptides were characterized by mass spectrometry and amino acid composition, which was also used to quantitate the amount of various substrates and products. The digestions of the quenched fluorogenic peptides were analyzed by RP-HPLC using a dual UV (210 nm) and fluorescence (excitation and emission wavelengths of 320 and 420 nm, respectively) detector (Rainin).
pH Optimum, Calcium Dependence, and Inhibitor Profile-The protocols used were essentially the same as reported previously (13). Stocks of the buffer described above were adjusted to pH 5.0 -8.5 at 0.5-unit increments by addition of either acetic acid or sodium hydroxide. In order to investigate the calcium requirement of SKI-1, increasing concentrations of Ca 2ϩ -acetate were used ranging from 0 to 10 mM. For inhibition studies, the enzyme in the reaction buffer was preincubated with the desired agents for 30 min prior to addition of peptide II.
K m(app) , V max(app) , and K i(app) Determinations-Following digestion reactions with increasing substrate concentrations, the products were separated by RP-HPLC. The rate of substrate hydrolysis was obtained from the integrated peak areas of the chromatograms. K m(app) and V max(app) values were estimated using nonlinear regression analysis (Enzfitter software; Elsevier Biosoft, Cambridge, United Kingdom) of plots of the hydrolysis rate versus the substrate concentration. For apparent inhibitor constant (K i(app) ) determinations, variable inhibitor concentrations within the range of 15-70% inhibition were used at three concentrations of peptide IV ranging from 0.6 to 3.5 times the K m(app) value. The K i(app) values were estimated from Dixon plots as described (16). For the two quenched peptides, kinetic parameters were determined as described (17).

SKI-1 Overexpression, Purification, Biosynthesis, and Prosegment Processing-
We have previously shown that overexpression of full-length SKI-1 (FL-SKI-1) in HK293 cells results in shedding of a 98-kDa form (sSKI-1) of this enzyme into the medium (10). Based on this finding, we engineered a soluble form of SKI-1 (BTMD-SKI-1), ending at residue 997, to which we added a hexa-His sequence at the C terminus (Fig. 1A). In a comparative biosynthetic analysis, shown in Fig. 1B, LoVo cells were infected with the SKI-1 virus constructs VV:FL-SKI-1, VV:BTMD-SKI-1, and wild type virus (VV:WT). After labeling the cells for 3 h with [ 35 S]Cys, proteins in the media were immunoprecipitated with an antiserum directed against either the prosegment of SKI-1 (Ab:P) or an internal SKI-1 sequence (Ab:S). In both cases, a protein of ϳ14 kDa co-immunoprecipitated with the 98-kDa sSKI-1 or the 100-kDa BTMD-SKI-1 (bSKI-1, Fig. 1B) that was not seen with VV:WT infections. Since Ab:P was raised against a recombinant SKI-1 prosegment peptide and has been shown previously to detect the SKI-1 zymogen (10), we concluded that the ϳ14-kDa peptide is most likely derived from the cleaved prosegment (the full-length prosegment is ϳ24 kDa; see below). The fact that it co-immunoprecipitated with the enzyme under denaturing conditions suggests a strong interaction between SKI-1 and this FIG. 1. A, schematic representation of the structure of FL-SKI-1 and its truncation mutant BTMD-SKI-1. The various SKI-1 domains depicted are, respectively, the signal peptide, pro-segment, catalytic domain, and the C-terminal region comprising a cytokine receptor/growth factor motif, a transmembrane domain, and a cytosolic tail. The positions of polypeptides used to produce SKI-1-specific antisera (Ab: P, N, and S) are also displayed. B, biosynthetic analysis of SKI-1. VV:FL-SKI-1, BTMD-SKI-1 (bSKI-1) or control VV:WT-infected LoVo cells were pulse-labeled with [ 35 S]Cys for 3 h. Media were immunoprecipitated with either Ab:S or Ab:P and then resolved by SDS-PAGE on an 8% gel, followed by autoradiography. Arrows point to the migration positions of the 100-kDa BTMD-SKI-1 (bSKI-1), the 98-kDa shed form (sSKI-1), as well as the 14-kDa prosegment product. C, Western blot analysis of the overexpressed BTMD-SKI-1. Samples from VV:WT-or BTMD-SKI-1-infected BSC40 cells (left and middle panel) were processed as described under "Experimental Procedures" and run on an 8% SDS-PAGE reducing gel. Following electrotransfer to polyvinylidene difluoride membranes, protein bands were visualized via ECL detection using primary rabbit antisera Ab:S or Ab:N. Purified BTMD-SKI-1 (right panel, *) was obtained from a Ni 2ϩ affinity resin as described under "Experimental Procedures," then processed as described above. A mixture of Ab:S and Ab:P were used as primary antisera. Elution buffer was used as a control (CTL). region of its prosegment. The actual stoichiometry of enzymeto-prosegment is not clear from this experiment, since it was carried out using two different antisera and denaturing conditions. We also observed that some of the 100-kDa BTMD-SKI-1 is cleaved into a 98-kDa form similar to that found with FL-SKI-1 (Fig. 1B). This conversion is presumably carried out by endogenous "shedding enzymes" (10, 18) that can act on both forms of SKI-1, although C-terminal sequencing would be needed to confirm this hypothesis.
Western blot analyses of media now obtained from BSC40 cells infected with VV:BTMD-SKI-1 also revealed a secreted ϳ100-kDa immunoreactive band (Fig. 1C). The same band was detected using either an antiserum against the N-terminal region of the SKI-1 catalytic domain (Ab:N) or one against a more C-terminal region (Ab:S). When Ab:P was mixed together with Ab:S and used to probe the metal affinity column-purified SKI-1 preparation (indicated by the asterisk in Fig. 1C), we were able to again detect the ϳ14-kDa prosegment fragment, further supporting our hypothesis that it forms a strong association with the enzyme. It should be noted that, although a mixture of Ab:S and Ab:P was used in order to detect both proSKI-1 and BTMD-SKI-1 simultaneously, when either Ab:N or Ab:S were used alone, only the 100-kDa or 14-kDa species were observed, respectively (data not shown).
In order to evaluate the rate of zymogen processing and the fate of the prosegment, LoVo cells overexpressing VV:FL-SKI-1 were pulse-labeled with [ 3 H]Leu for 15 min and then chased for 2 h. Fig. 2 shows an SDS-PAGE analysis of the cell lysates immunoprecipitated with Ab:P (left panel). At least five immunoreactive polypeptides (molecular masses of ϳ26, 24, 14, 10, and 8 kDa), which were not present in controls infected with VV:WT, were detected. In order to further define in which organelle(s) this processing occurred, LoVo cells infected with VV:FL-SKI-1-were pulse-labeled with [ 3 H]Leu for 2 h in the presence or absence of BFA (Fig. 2, right panel). In both cases, the same five major, intracellular, immunoreactive prosegment forms could still be detected. Since the fungal metabolite BFA is known to disassemble the Golgi complex and cause the ER to fuse with the cis-, medial-, and trans-Golgi (but not the trans-Golgi network, TGN) (19), this result strongly implies that the initial zymogen processing of proSKI-1 occurs early along the secretory pathway. Possible locations include the ER or cis-Golgi, as was previously reported (10). Moreover, further processing of the prosegment into yet smaller fragments also occurs in these organelles.
To further characterize the prosegment of SKI-1, we took advantage of a stable transfectant of FL-SKI-1 in human HK293 cells that we had made previously (10). This system has the added advantage that the possibility of VV overexpression artifacts influencing the processing of the prosegment is eliminated. Concentrated culture medium from these cells (serumfree) was purified via RP-HPLC using first a semi-preparative C4 column (data not shown) followed by an analytical C4 column (Fig. 3A). The eluted fractions were analyzed by Western blot using Ab:P (Fig. 3B). Immunoreactive peptides ranging from ϳ4.5 to 24 kDa were apparent. N-terminal sequencing of the very abundant ϳ14-kDa protein in fraction IV (Fig. 3C) revealed a major sequence starting at Gly 18 of pre-proSKI-1 (10,12). This clearly defines the signal peptidase cleavage site as LVVLLC 17 2GKKHLG, which is 1 aa before that predicted by signal peptidase cleavage site algorithms (10,11). The Nterminal sequence of the ϳ4.5-kDa polypeptide (Fig. 3D) revealed that it starts at Pro 143 , indicating a cleavage at the sequence KYAESD 142 2PTVPCNETRWSQK. This fragment is most likely the product of cleavage between Asp and Pro that may be caused by the acidic conditions encountered in either RP-HPLC, Edman sequencing (20), or sample preparation for SDS-PAGE analysis (21). An unexpected benefit of this cleavage was our finding that phenylthiohydantoin (PTH)-Asn 148 , which occurs in the putative N-glycosylation site Asn-Glu-Thr was readily detected in this sequence. Thus, the predicted N-glycosylation site Asn 148 within the prosegment of SKI-1 is not employed, at least in this expression system. This conclusion was also supported by the prosegment's resistance to endo H and endo F digestion (data not shown). Of the two eukaryotic subtilases known to contain a potential N-glycosylation Asn-Glu-Thr site, i.e. kexin (22) and SKI-1 (10), it appears that at least the latter's prosegment is not N-glycosylated. Finally, the separation of the above prosegment fragments from mature SKI-1 using RP-HPLC (Fig. 3, A and B) and non-reducing SDS-PAGE (data not shown), suggests that none of the Cys residues in the prosegment (10) are linked by disulfide bridges to the rest of the enzyme.
As a preliminary means of characterizing the SKI-1 prosegment fragments, MALDI-TOF analysis (Fig. 3E) of fraction IV from Fig. 3B was carried out. Three major molecular ions of masses 13,351, 13,518, and 13,685 Da were detected, with an expected error of Ϯ25 Da for this mass range. Combined with the previous N-terminal sequencing results of the ϳ14-kDa peptide (Fig. 3C), these mass values indicate that this peptide has heterogeneous C termini that are derived from cleavages near the sequence RKVFRSLK 137 , as indicated in Fig. 3E. In fact this region contains three potential SKI-1 cleavage sites (8) with an Arg or Lys at the P4 position and an Phe, Arg, or Lys at the P1 position. Although the calculated molecular masses of 13,339, 13,496, and 13,696 for the polypeptides G 17 KK-RKVF 133 , G 17 KK-RKVFR 134 and G 17 KK-RKVFRSL 136 , respectively, match within experimental error (Ϯ 22 Da) the observed masses in Fig. 3E, these assignments should only be taken as a first indication (see below). Moreover, the predicted G 17 KK-RKVFRSL 136 fragment does not correspond to the expected SKI-1 cleavage motif of a basic residue at the P4 position. Hence, this secreted peptide could result either from cleavage at G 17 KK-RKVFRSL 136 or, more likely, at G 17 KK-RKVFRSLK 137 , followed by basic carboxypeptidase cleavage of the C-terminal Lys (23). Since we were unable to obtain consistent mass spectra of the ϳ4.5-kDa polypeptide that was sequenced in Fig. 3D, we could not use this technique to approximate its C terminus, which presumably corresponds to the C terminus of the processed SKI-1 pro-segment. We therefore resorted to synthetic peptide cleavage as a tool to accurately define potential prosegment cleavage sites.
Analysis of Synthetic Prosegment-derived Peptide Cleavages-Based on our detection of ϳ26and 24-kDa SKI-1 prosegment products (Fig. 2), as well as on a mutagenesis study of SREBP-2 cleavage sites (8), we synthesized three SKI-1 prosegment peptides encompassing potential, C-terminal, autocatalytic cleavage sites (10,11). All contain Arg at P4 and either Leu or Ala at P1 (peptides III, VI, and VII shown in Table I). Of these peptides containing only native sequences, the only one with detectable cleavage by SKI-1-containing concentrated medium (from either VV:BTMD-SKI-1-infected BSC40 cells or SKI-1-transfected HK293 cells) was peptide III (WHATGRHSSRRLL 186 2RAIPR) (see Table I). No cleavages were observed when VV:WT-infected or empty vector-transfected media were used (data not shown). Metal chelation chromatography-purified enzyme further supported that this cleavage is effected by SKI-1 ( Fig. 4A; peptide IV), and the products were positively identified via mass spectrometry.
Similarly, based on the mass spectrometry data in Fig. 3E, we synthesized two peptides (VIII and IX) encompassing the putative internal processing site(s) of the SKI-1prosegment. Both were cleaved at multiple locations by SKI-1-containing concentrated medium from HK293 transfectants (data not shown). Further analysis revealed that one of these cleavages, corresponding to PQRKVF 133 2RSL, was as prevalent in empty vector-transfected HK293 medium as in SKI-1-transfected medium (see Table I, peptide VIII). In contrast, the PQRKVFRSLK 137 2YAESD cleavage was only seen in SKI-1containing medium. This cleavage was also confirmed using metal chelation chromatography-purified enzyme ( Fig. 4B;   FIG. 3. Purification and identification of secreted recombinant pro-SKI-1. A, media obtained from HK293 cells stably expressing FL-SKI-1 were concentrated and sequentially applied to C4 semi-preparative column (data not shown) followed by a C4 analytical RP-HPLC column, and then eluted by the indicated linear CH 3 CN gradient. B, the fractions labeled I-IV were collected and analyzed by Western blotting using the primary antiserum Ab:P. C and D, proteins contained in fraction IV were separated on a 10% SDS-PAGE reducing gel. Following electrotransfer, the proteins were stained with Ponceau Red. The immunoreactive 14-kDa (C) and non-immunoreactive but colored ϳ4.5-kDa (D) polypeptides were excised and submitted to N-terminal sequencing (X represents an undefined residue). E, mass spectrometric analysis by MALDI-TOF spectrometry of fraction IV. The C-terminal residue sites believed to corresponding to the three ϳ14-kDa polypeptides are underlined, whereas the expected (potential) cleavage sites are indicated by dashed arrows.
peptide IX) and mass spectrometry to identify the products. However, also clearly visible are the PQRKVF 133 2RSLKYA-ESD cleavage products. We acknowledge that there could be residual contaminating proteases in our purified SKI-1 preparations (minor bands were visible on colloidal gold-stained membranes of SKI-1 preparations). Thus, while we are confident that SKI-1 cleaves its prosegment at the C-terminal WHATGRHSSRRLL 186 2RAIPR site and at the internal PQRKVFRSLK 137 2YAESD site, our data do not allow us to rule out SKI-1-mediated cleavage at the PQRKVF 133 2-RSLKYAESD site.
Comparing the simple cleavage rates of the SKI-1 prosegment internal and C-terminal sites, we observed that the former was vastly superior to the latter (data not shown). We also noticed that the peptides best processed by SKI-1 contain an acidic residue at the P3Ј or P4Ј substrate site, whereas those that did not, appeared to be cleaved poorly or not at all (Table  I). Moreover, we had previously established that SKI-1 does not cleave the fluorogenic peptides RGLT-MCA, RGLTTT-MCA, and RSVL-MCA (10), which lack PЈ residues. Based on these observations, we asked if replacing the Ile and Pro residues at P3Ј and P4Ј of the C-terminal prosegment processing site would significantly improve the SKI-1-mediated cleavage of peptide III. Thus, we synthesized two mutants of this peptide (peptides IV and V, the latter truncated by 8 aa at the N terminus) in which the Ile and Pro residues at P3Ј and P4Ј were replaced by Leu and Glu, respectively. As shown in Table II, this change significantly improved the processing of these peptides, such that we were able to determine V max(app) /K m(app) values. The approximately 2-fold difference in these values for peptides IV and V further suggests that determinants N-terminal to the P4 position may also play a role in substrate specificity. The SKI-1 specificity of these peptide cleavages was also verified using metal chelation chromatography-purified enzyme (when VV: WT-infected or empty vector-transfected media were used, no peptide processing was observed).
In Vitro Kinetic Properties of SKI-1: Comparative Analysis of Synthetic Peptide Cleavages-In a previous report (10), sSKI-1 was shown to cleave the 32-kDa proBDNF into a 28-kDa form at the RGLT2SL sequence in vitro with a pH optimum close to neutrality. Similar to PCs (1-3), we suggested that SKI-1 might be a Ca 2ϩ -dependent enzyme since the calcium ionophore A23187 inhibited the ex vivo cleavage of proBDNF (10). In order to obtain kinetic analyses of defined SKI-1 substrates, we examined a 14-aa peptide spanning the hproBDNF processing site (10), K 50 AGSRGLT2SLADTF 63 (peptide I) and a 27 aa hSREBP-2-related peptide (8), G 504 GAHDSDQHPHSGSG-RSVL2SFESGSGG 530 (peptide II). Concentrated SKI-1-containing medium (from either VV:BTMD-SKI-1-infected BSC40   were digested for 18 h with metal chelation chromatography-purified BTMD-SKI-1. The cleavage products were separated by RP-HPLC using a 5-m analytical Ultrasphere C18 column (Beckman) as described under "Experimental Procedures." The peptides contained in all but two peaks were identified by mass spectrometry. The unidentified peaks are attributable to contaminating activities seen in WT/empty vector controls. cells or SKI-1-transfected HK293 cells) was reacted with these peptides at pH 6.5, followed by MALDI-TOF mass spectrometric analysis of the RP-HPLC-purified products. The expected cleavages were confirmed and did not occur using WT/empty vector-derived media (Fig. 5). Again, the metal chelation chromatography-purified enzyme generated the same products as the concentrated media (data not shown). We then demonstrated that the optimal pH and calcium concentrations for efficient cleavage of the hSREBP-2 peptide (II) are pH 6.5 and 2 mM Ca 2ϩ , respectively (Fig. 6). Interestingly, the pH optimum observed with the proBDNF peptide (I) is sharper than that obtained with peptide II. In the former case, the enzyme still retains about 30% of its activity at pH 5.0 and 55% of its activity at pH 8.5 (Fig. 6A). Similar results for the pH optimum of peptide II cleavage were obtained with metal chelationpurified BTMD-SKI-1 (data not shown). In contrast, however, the pH optimum of peptide IX with the purified enzyme was 8.0, with no activity detectable below pH 5.5.
A summary of the kinetic analyses of the synthetic proBDNF (peptide I) and SREBP-2 (peptide II) cleavages by SKI-1 is shown in Table II. Both peptides are cleaved at comparable kinetic efficiencies with V max(app) /K m(app) values of 0.002 and 0.004 h Ϫ1 , respectively. In comparison, the V max(app) /K m(app) value estimated with peptide IV is 5-10-fold higher than those obtained with peptides I and II (Table II). The N-terminal truncation of peptide IV from 17 to 9 aa (peptide V, Table I) caused a 2-fold reduction in catalytic efficiency (Table II). Table III shows the inhibitor profile of SKI-1, in which it is clear that this enzyme is quite sensitive to metal chelators such as EDTA and to the calcium chelator EGTA. In addition, the transition metals Cu 2ϩ and Zn 2ϩ , but not Ni 2ϩ or Co 2ϩ , inhibit the enzyme at mM concentrations. As reported using the 32-kDa proBDNF (10), assays with the synthetic SREBP-2 peptide demonstrated that the metal chelator o-phenanthroline becomes inhibitory at concentrations above 1 mM. The other nonchelator inhibitors tested had minimal or no effects on SKI-1 activity.
In order to develop a convenient in vitro assay for SKI-1, we designed a number of internally quenched fluorogenic substrates and tested their cleavage efficacy by SKI-1. The two best peptides encompassed the processing site RSLK2 within the hSKI-1 prosegment (peptides X and XI, Table I). Mass spectrometric analysis confirmed that both peptides were cleaved at the RSLK2 site by shed SKI-1 derived from HK293 cell transfectants, but not by medium obtained from HK293 empty vector transfectants. This processing generated the fluorescent N-terminal peptides Abz-VFRSLK or Abz-RSLK, and a non-fluorescent C-terminal peptide YAESDY(NO 2 )-A (data not shown). Measurements of kinetic parameters demonstrated that peptides X and XI are about 20-and 200-fold better substrates than the C-terminal prosegment peptide IV (Tables  II and IV), suggesting that the shorter peptide XI may be the best SKI-1 substrate tested to date. This cleavage was completely abolished in the presence of 10 mM EDTA, in agreement with the Ca 2ϩ dependence of SKI-1 activity (Fig. 6B).
SKI-1 Inhibition by Its Prosegment-One important question remaining is whether the SKI-1 prosegment functions as an inhibitor of its enzymatic activity, analogous to the prosegments of other subtilases (3). We thus prepared prosegment constructs, designated ending near the proposed C-terminal processing site RRLL 186 (Fig. 4A): PS1, extending to Leu 169 ; PS2, extending to Ala 188 ; and PS3, extending to Leu 197 (Fig. 7). To each C terminus we coupled a hexa-His tag. These prosegment constructs were expressed in bacteria and purified by Ni 2ϩ -chelation chromatography followed by RP-HPLC (see "Experimental Procedures"). The purity of these prosegments was confirmed by SDS-PAGE/Coomassie staining (Fig. 7B) and aa analysis (data not shown). A summary of the inhibitory potency of each prosegment using peptide IV as a substrate is shown in Table V. Kinetic analysis using Dixon plots (15) indicated a competitive inhibition mechanism (data not shown). Although PS2 exhibits the best apparent inhibitory constant (K i(app) ϭ 97 nM), PS3 (K i(app) ϭ 127 nM) and PS1 (K i(app) ϭ 182 nM) are similarly potent SKI-1 inhibitors. When PS2 was digested with carboxypeptidase B to eliminate the His tag, its inhibitory potency was not affected (data not shown), confirming that this tag is not responsible for the observed inhibition. We also tested the inhibitory activity of the RP-  5. Processing of proBDNF and SREBP-2 peptides by BTMD-SKI-1. The 14-aa peptide I (A) and 27-aa peptide II (B) were digested with BTMD-SKI-1 for 150 and 60 min, respectively. The cleavage products were separated by RP-HPLC using a 5-m analytical Ultrasphere C18 column (Beckman) as described under "Experimental Procedures." The peptides contained in the major peaks were identified by mass spectrometry and amino acid analysis (data not shown).
HPLC-fractionated native prosegment (see Fig. 3). Only the material from fraction IV, which included the full-length ϳ24-kDa prosegment, was inhibitory, whereas that of the others, including the ϳ14-kDa peptide alone or in combination with smaller fragments, were not inhibitory (data not shown).

DISCUSSION
Limited proteolysis of inactive precursor proteins at sites marked by paired or multiple basic residues is a widespread process (1,2). Less common is the recent finding that bioactive peptides or proteins can also be generated by limited proteolysis after either hydrophobic or small residues (3). SKI-1 represents the first mammalian member of subtilisin-like processing enzymes with such substrate specificity (10,11). It is a widely expressed enzyme (10) that may play a crucial role in cholesterol and fatty acid metabolism (11). Due to its very recent discovery, information regarding its enzymatic properties, substrate specificity, and the function of its proregion have only begun to be addressed.
Many peptidyl hydrolases, including subtilases, possess a prodomain that acts both as an intramolecular chaperone and a highly potent inhibitor of its associated protease (24,25). Activation of the enzyme typically requires release of the prosegment in an organelle-specific manner. For furin (26) the release occurs in the TGN, whereas for PC1 and PC2 (27) it occurs in immature secretory granules. The data presented in this report demonstrate that SKI-1 is unique among the mammalian subtilases, since both the C-terminal and internal cleavages of its prosegment occur in the ER. Hence, this enzyme does not appear to require an acidic environment for activation, assuming, by analogy with other subtilases (3), that prosegment release is the crucial step leading to zymogen activation. We propose the following sequence of events presumably leading to SKI-1 activation. 1) The signal peptide is removed in the ER by a signal peptidase cleavage at LVVLLC 17 2GKKHLG (Fig. 3C). 2) The prosegment is processed into a non-N-glycosylated polypeptide with an apparent molecular mass of ϳ24 -26-kDa (Fig. 2). 3) This prosegment is further processed into 14-, 10-, and 8-kDa intermediates (Fig.  2). While these multiple cleavages may be catalyzed by SKI-1 itself, the participation of other proteases cannot be excluded. The major cleavages leading to the formation of the ϳ24and ϳ14-kDa products occur within 10 min, and the other secondary ones within 30 min (data not shown). Since treatment of cells with BFA did not significantly alter these processing events, they most likely occur in the ER (Fig. 2). It is possible that the generation of prosegment fragments from the ϳ24 -26-kDa pro-form leads to a loss of inhibition in a fashion similar to that of subtilisin E (24,25). Indeed, our results demonstrate that, while the full-length prosegment is inhibitory, its ϳ14-kDa product is not. Surprisingly, some pro-region-derived polypeptides are found associated with SKI-1 in cell culture media. Thus, in contrast to furin (26), the low pH and high Ca 2ϩ concentrations prevailing in the TGN do not lead to propeptide dissociation. High ionic concentrations (up to 1 M NaCl) such as those used in immunoprecipitation (Fig. 1B) and metal chelation protein purification (Fig. 1C) also do not dis-   rupt the complex. It is only during RP-HPLC purification (Fig.  3A), in the presence of strong acids and organic solvents, that the prosegment peptides dissociate from SKI-1. These data suggest that hydrophobic interactions may be critical, as is the case for subtilisin (24,25).
To distinguish the SKI-1 prosegment autoprocessing sites (C-terminal and internal) from several closely situated candidate sites, we employed a combination of mass spectrometry and synthetic peptide digestion. In the case of the C-terminal site, only one of three candidate peptides (III) was processed by SKI-1 (Table I), indicating that RRLL 186 2RAIP is the most likely autoprocessing site. For the internal site, preliminary mass spectrometric data suggested three distinct cleavages occurring within the sequence PQRKVFRSLKYAESD 142 (Fig.  3E). Two of the three possible sites (PQRKVF 133 2RSLKYA-ESD and PQRKVFR 134 2SLKYAESD) appeared to satisfy the proposed SKI-1 recognition motif requiring a P4 basic residue (8). The third possibility (PQRKVFRSL 136 2KYAESD) could be considered by assuming the cleavage actually occurred at PQRKVFRSLK 137 2YAESD, followed by endogenous, basic carboxypeptidase removal of the C-terminal Lys residue (23). Assays carried out in vitro with synthetic peptides corresponding to this region of the prosegment (peptides VIII and IX) produced the same cleavage products (data not shown), but only the PQRKVFRSLK 137 2YAESD cleavage was unique to SKI-1. Thus, we propose that the aforementioned site is the most likely internal autoprocessing site, with the qualification that PQRKVF 133 2RSLKYAESD may occur to a lesser extent (see "Results" and Fig. 4B).
Other information regarding the substrate preferences of SKI-1 was obtained by replacing the P3Ј and P4Ј Ile and Pro residues of the C-terminal cleavage site peptide (III) by Leu and Glu (peptides IV and V) to create a very well processed SKI-1 substrate. While it would appear that the presence of an acidic residue at P4Ј significantly enhances the rate of substrate hydrolysis, it is also possible that the presence of Pro at P4Ј hinders efficient substrate processing. The pres-ence of similar acidic residues at the P3Ј or P4Ј position of the two confirmed substrates of SKI-1 (peptides I and II) as well as in the prosegment internal cleavage site RSLK 137 2YAES (Table I) lends support to the first argument. In addition to these residues, others also appear to play a role in SKI-1 substrate cleavage catalysis. The peptide pairs IV/V and X/XI both point to influences of positions N-terminal to the P4 residue. Interestingly, the efficiency of the truncated C-terminal peptide V is lower than that of peptide IV, whereas that of the truncated internal (quenched) peptide XI is higher. Taken together, these data indicate the importance of aa at both the P and PЈ positions in SKI-1-mediated substrate hydrolysis.
The data presented in Fig. 6 indicate that SKI-1 functions most efficiently near neutral pH and at 2-3 mM Ca 2ϩ . This is in general agreement with the conditions that reportedly prevail in the ER (28,29). However, closer examination of the data reveal that the pH optimum of SREBP-2 cleavage (peptide II, Fig. 6A) is actually 6.5, an observation that we confirmed using our purified SKI-1 preparation (data not shown). This suggests that the processing of SREBP might occur outside of the ER, perhaps in the Golgi where pH values of ϳ6.5 have recently been reported (30,31). Indeed, there is now cellular evidence suggesting that SREBP cleavage may occur in the Golgi rather than in the ER (32,33). The pH optimum of SKI-1 appears to be dependent on the substrate employed; proBDNF (10) and its related peptide (I), appear to be well cleaved even at pH 5.5, suggesting that it could cleave this (and possibly other substrates) in acidic endosome-like compartments where it was previously localized (10). On the other hand, cleavage of the internal, autocatalytic, prosegment processing site PQRKVFRSLK 137 2YAESD (Fig. 4B) is optimal at pH 8 (data not shown), implying that this event, as we concluded from our biosynthesis assays, takes place most effectively in the ER. Overall, the pH and Ca 2ϩ profiles of SKI-1 resemble those of the constitutively secreted PCs (1,13). The inhibitor profile of SKI-1 (Ref. 10, Table III), showing that enzymatic activity is significantly inhibited by EDTA, EGTA, and only high concentrations of o-phenanthroline, tend to discount the likelihood that SKI-1 is a transition metal-dependent proteinase. In fact, SKI-1 activity is inhibited by low concentrations of certain transition metals, such as Cu 2ϩ and Zn 2ϩ .
Directed by the observation that peptides containing the primary processing site of the prosegment of PC1 are potent inhibitors of its activity, and that the C-terminal basic residues of furin and PC7 are essential for enzyme inhibition (34,35), we assessed the inhibitory potency of three SKI-1 recombinant propeptides (Fig. 7A). All of these end at sequences near the RRLL 186 RA cleavage site. Interestingly, the three FIG. 7. Pro-SKI-1 constructs. Three putative primary activation sites, together with the constructs made, are schematically represented (A). The number above each aa corresponds to the candidate processing site. A hexahistidine tag is directly fused to each prosegment. B, the prosegments were purified via Ni 2ϩ chelation column chromatography and RP-HPLC as described under "Experimental Procedures" and analyzed by Western blotting using Ab:P as the primary antibody (left panel). The purity was assessed by Coomassie staining (right panel).

TABLE V
Effect of pro-segment peptide constructs on BTMD-hSKI-1 activity Digestion reactions using BTMD-SKI-1 medium plus peptide IV were carried out as described under "Experimental Procedures." The prosegment peptides were preincubated with the enzyme for 30 min. Values were deduced from the Dixon plots obtained from three separate experiments. prosegments displayed comparable inhibitory potencies (Table V). Compared with proPC1 (34), pro-furin and proPC7 (35), the K i(app) values (Table V) are up to 250-fold higher. This suggests that the prosegment of SKI-1, although potentially inhibitory in vivo, may function more as a chaperone, catalyzing the productive folding of SKI-1. Indeed, since SKI-1 may be active in the ER (10,11), whereas the PCs are not (13,26), the lower inhibitory potency of the prosegment of SKI-1 may be adapted to the conditions prevailing in this cellular compartment. In the case of PCs, highly effective inhibition by the prosegment may be needed in order to ensure that these enzymes are activated only when they reach the TGN or secretory granules (1-3). The 14-kDa fragment, which represents the major secreted form of the prosegment, is tightly associated with SKI-1 (Fig. 1C), yet it is not inhibitory (data not shown). Accordingly, this segment may serve a chaperonin-like function similar to that reported for the N-terminal 150 aa of 7B2 toward proPC2 (36,37).
In conclusion, the present work firmly establishes that SKI-1 is a Ca 2ϩ -dependent subtilase with a reasonably neutral pH optimum, depending on the substrate employed. We also demonstrate that SKI-1 can cleave substrates C-terminal to Thr, Leu, and Lys residues, thus providing direct, in vitro evidence that it is a candidate converting enzyme responsible for the generation of 28-kDa proBDNF (10) and SREBP-2 processing at site 1 (11). For efficient cleavage, it appears that substrates should contain a basic residue at P4 and an aliphatic one at P2 (Table I). Furthermore, aa at the P3Ј and P4Ј positions seem to exert an important discriminatory effect. The best substrate tested so far is the quenched fluorogenic substrate Abz-RSLK 234 YAESDY(NO 2 )A, thereby providing a convenient and sensitive assay for SKI-1 activity. The present data demonstrate that only the full-length SKI-1 prosegment is inhibitory. Thus, overexpression of this prosegment in cell lines may provide a novel method for inhibiting the cellular activity of this enzyme in a fashion similar to that of over-expressed profurin and proPC7 (35). Finally, it is anticipated that precursor substrates other than the sterol-regulating SREBPs (8) and the neurotrophin proBDNF (10) will be identified, thereby extending the spectrum of activity of this unique and versatile enzyme.