Inhibition of Human Endogenous Retrovirus-K10 Protease in Cell-free and Cell-based Assays*

A full-length and C-terminally truncated version of human endogenous retrovirus (HERV)-K10 protease were expressed in Escherichia coli and purified to ho-mogeneity.Bothversionsoftheproteaseefficientlyproc-essedHERV-K10Gagpolyproteinsubstrate.HERV-K10 Gag was also cleaved by human immunodeficiency virus, type 1 (HIV-1) protease, although at different sites. To identify compounds that could inhibit protein processing dependent on the HERV-K10 protease, a series of cyclic ureas that had previously been shown to inhibit HIV-1 protease was tested. Several symmetric bisamides acted as very potent inhibitors of both the truncated and full-length form of HERV-K10 protease, in subnanomolar or nanomolar range, respectively. One of the cyclic ureas, SD146, can inhibit the processing of in vitro translated HERV-K10 Gag polyprotein substrate by HERV-K10 protease. In addition, in virus-like particles isolated from the teratocarcinoma cell line NCCIT, there is significant accumulation of Gag and Gag-Pol precursors upon treatment with SD146, suggesting the compound efficiently blocks Life Sciences). The reaction mix was initially denatured at 94 °C for 5 min and then subjected to 30 cycles of denaturation at 94 °C, annealing at 50 °C, and extension at 72 °C. An aliquot of PCR reaction was used directly in DNA sequencing, with either PRT-A or PRT-B as a primer. Molecular Modeling of HERV-K10 Protease— The three-dimensional homology model of the truncated version of HERV-K10 protease was constructed using coordinates of HIV-1 protease complexed with SD146 as a template (Protein Data bank file 1QBT.pdb; Ref. 36) with the program Molecular Operating Environment (Chemical Computing Group Inc.). The sequence was aligned initially maximizing the homology but later adjusted to accommodate the insertion at position 39 (HIV numbering) at the elbow of the flap and the insertion at position 80 (HIV numbering) at the active site, mimicking the three-dimensional structure of feline immunodeficiency virus protease. The homology al-gorithm of the Molecular Operating Environment software created 10 models, each of which was generated by making a series of Boltzmann-weighted choices of side chain rotamers and loop conformations from a set of protein fragments of high resolution protein structures. An aver-age model was potential energy-minimized using AMBER forcefield.

The human genome contains a large number of endogenous retroviral sequences that are virtually all highly defective because of multiple termination codons, deletions or the lack of a 5Ј long terminal repeat (1,2). It is assumed that at some time during the course of human evolution, exogenous progenitors of human endogenous retroviruses (HERVs) 1 integrated into the cells of the germ line and thereby obtained the ability to be inherited by offspring of the host as a mendelian trait (3).
HERVs are grouped into at least a dozen single and multiple copy number families and are classified according to the tRNA that they use as primer for reverse transcription (1,4). The retroviral element that carries a primer binding site complementary to the 3Ј end of a lysine tRNA is called HERV-K. HERV type K represents the biologically most active form of a variety of retroviral elements present in the human genome (1,5). Although the HERV-K group comes closest of all known HERVs to containing infectious virus, no corresponding replication-competent virus has so far been described (1,3). Although humans harbor several dozen proviral copies of HERV type K per haploid genome (4,6,7), some of which code for the characteristic retroviral proteins Gag, Pol, and Env (8,9), recent studies raised a suggestion that no complete proviral copy of HERV-K exists (10,11); the issue remains to be clarified. In terms of infectious virion production, HERV-K could be defective at multiple levels, including the observed arrest during budding, inefficient RT enzyme activity, and incomplete Env expression and processing (1).
HERV-K elements exhibit restricted cell type expression, observed mainly in germ cell tumors (including testicular teratocarcinoma cell lines) and their testicular precursor lesions (8,12,13). Typically the coding regions of HERV-K elements are far less disrupted by mutations than other HERV families, and protein synthesis has been observed for all the main retroviral genes. The HERV-K Gag precursors are cleaved into major core, matrix, and nucleocapsid components (14 -16), presumably by HERV-K protease, because functional activity has been demonstrated for this enzyme (15,17).
Detailed electron microscopic surveys have revealed the existence of retrovirus-like particles in breast carcinoma and teratocarcinoma cell lines (18 -20). The phenotype of human teratocarcinoma-derived retrovirus particles has been correlated with complex mRNA expression of HERV-K sequences in those cells, reminiscent of the mRNA expression pattern observed after exogenous retrovirus infection with, for example, lenti-or spumavirus strains (8,9).
Several hypotheses have so far been proposed about possible implication of HERV expression in certain pathogeneses, including autoimmune diseases such as insulin-dependent diabetes mellitus (21), tumor development, and even cardiovascular disease (22). In addition, numerous possible roles have been proposed for HERVs in reproductive physiopathology (reviewed in Ref. 23). In the study published by Sauter et al. (16), authors reported that HIV-1-infected patients and especially patients with seminomas exhibit elevated titers of anti-HERV-K10 Gag antibodies. Towler et al. (25) reported that HERV-K10 protease is highly resistant to a number of clinically used HIV-1 protease inhibitors, including ritonavir, indinavir, and saquinavir.
They reported the protease to be a homodimer with a pH optimum at 4.5 and with a higher enzymatic activity and stability at elevated ionic strengths. The authors raised an interesting speculation that HERV-K protease might somehow complement HIV-1 protease under conditions where the latter activity is impaired because of either the presence of drug resistance mutations or the presence of potent HIV-1 protease inhibitor.
The aim of this study was to identify potent inhibitors of HERV-K10 protease and to demonstrate their action in virusproducing cells. The results shown in this report indicate that some members of the cyclic urea class can act as very potent inhibitors of this protease in a nanomolar range and are capable of blocking processing of HERV Gag in vitro as well as in the teratocarcinoma cell line NCCIT.

EXPERIMENTAL PROCEDURES
Cloning of Truncated Version of HERV-K10 Protease-Genomic DNA was extracted from the buffy coat fraction of fresh human blood (24). DNA coding for core region of HERV-K10 protease (25) was then amplified by polymerase chain reaction with Taq DNA polymerase (PerkinElmer Life Sciences). Oligonucleotides 5Ј-CTAGGAAGCTTCA-TATGGACTATAAAGGCGAAATTCAA-3Ј (PRT-A) and 5Ј-GCTGTGG-ATCCTTACTACATGGTGATTTCCGCACC-3Ј (PRT-B) were used as sense and antisense primer, respectively. PCR product was cloned into mammalian expression vector pcDNA3.1(ϩ) (Invitrogen) via HindIII and BamHI restriction sites. DNA sequencing of several clones revealed presence of substantial polyporphism. The clone with the DNA sequence identical to that published by Ono et al. (6) was chosen for further experiments. This clone was subjected to another round of PCR amplification, this time with oligonucleotides 5Ј-AGACTGGATCCGA-CTATAAAGGCGAAATTCAA-3Ј and 5Ј-ACAGATCTCGAGCATGGTG-ATTTCCGCACC-3Ј. The amplification product was cloned into Escherichia coli expression plasmid pET21a(ϩ) (Novagen) via BamHI and XhoI restriction sites.
Cloning of the Full-length Version of HERV-K10 Protease-The cloning of the full-length version of HERV-K10 protease into pET21a(ϩ) and its site-directed mutagenesis was described previously (25).
Expression and Purification of 13-kDa Form of HERV-K10 Protease-Luria-Bertani broth (1 L) supplemented with ampicillin (200 g/ ml) was inoculated with 5 ml of overnight culture of E. coli BL21(DE3) expression strain (Novagen) harboring pET21a(ϩ)/HERV-K10 protease construct. When an A 600 value of 0.6 was reached, the expression of HERV-K10 protease was induced by addition of isopropyl-1-thio-␤-Dgalactopyranoside (Sigma) to a final concentration of 0.4 mM. After 3 h at 37°C the bacterial cells were pelleted by centrifugation at 6000 ϫ g for 10 min. The cells were resuspended in 50 ml of 5ϫ TE buffer (0.1 M Tris/HCl, 5 mM EDTA, pH 7.5) and subjected to sonication (6 ϫ 30 s, 40 W, microtip). The soluble fraction was discarded. Inclusion bodies were washed twice with 20 ml of 5ϫ TE buffer and then dissolved in 100 ml of 8 M urea, 0.1 M Tris/HCl, pH 7.5, 1 mM DTT. Refolding of HERV-K10 protease was achieved by dialyzing the solution against 4 liters of 20 mM PIPES, pH 6.5, 1 M NaCl, 1 mM DTT, at 4°C for 3 h and then against 4 liters of fresh buffer overnight. During the renaturation procedure the precursor form of HERV-K10 protease (20 kDa) completely autoprocessed to give rise to the mature, catalytically active 13-kDa form. The solution was centrifuged for 10 min to eliminate the precipitated proteins and then further clarified by filtration through a 0.45-m membrane. The solution was then mixed 1:1 with buffer A (50 mM PIPES, pH 6.5, 1 M NaCl, 1 mM EDTA, 1 mM NaK tartrate, 10% glycerol). Pepstatin A-agarose suspension (Sigma) was then added, and the flask was left overnight at 4°C. An Amersham Pharmacia Biotech column was packed with the slurry and then connected to a fast protein liquid chromatography system (Ä KTA, Amersham Pharmacia Biotech). The column was washed with 5 column volumes of buffer A at 1 ml/min. The bound proteins were eluted with buffer B (0.1 M Tris/HCl, pH 8.0, 1 mM NaK tartrate, 10% glycerol, 5% ethylene glycol). The protease containing fractions were pooled and concentrated with Amicon stir cell over YM3 membrane to about 2 ml. Protease concentration was determined with UV spectrophotometry (27). Calculated molar absorption coefficient of 29850 M Ϫ1 cm Ϫ1 was used. The protein solution was aliquoted and stored at Ϫ80°C.
Expression and Purification of Full-length Forms of HERV-K10 Protease-E. coli BL21(DE3) strain was transformed with expression plasmids containing either wild type form or active site mutant (D26N) of the 18-kDa version of HERV-K10 protease. Overnight culture was diluted 1:50 into 1 liter of LB broth. At an A 600 value of 0.6, 1 mM isopropyl-1-thio-␤-D-galactopyranoside was added, and the culture was then incubated in a 37°C shaker for 1 h. Cells were spun down and washed with 50 mM Tris/HCl, pH 8.0, 5 mM EDTA. Cells were then resuspended in 25 ml of lysis/wash buffer (40 mM phosphate buffer, pH 7.0, 0.3 M NaCl, 20 mM imidazole). Lysozyme was added to 0.2 g/ml, and the suspension was incubated on ice for 30 min. Cells were then sonicated (6 ϫ 30 s) and then centrifuged at 10,000 ϫ g for 30 min. Supernatant was filtered through 0.45-m syringe filter, and the pellet was discarded. One ml of Qiagen nickel-nitrilotriacetic acid Superflow resin was put in a 10-ml Bio-Rad disposable column and then equilibrated with 10 ml of lysis/wash buffer. The soluble fraction was applied to the column and allowed to enter by gravity flow. 20 ml of lysis/wash buffer were used to wash the resin. The protease was eluted with 6 ml of elution buffer (40 mM phosphate buffer, pH 7.0, 0.3 M NaCl, 300 mM imidazole) and then further purified by ion exchange chromatography. Nickel-nitrilotriacetic acid purified material was dialyzed against 1 liter of buffer C (20 mM sodium acetate, 2 mM EDTA, 2 mM DTT, pH 5.0) and then applied to an Amersham Pharmacia Biotech MonoS HR 5/5 column. The resin was washed with buffer C, and the protease was eluted with a linear NaCl gradient from 0 -1 M NaCl. The fractions containing protease were pooled. Protein concentration was determined with UV spectrophotometry. Molar absorption coefficients of 33,690 and 34970 M Ϫ1 cm Ϫ1 were used for wild type and active site mutant forms, respectively.
Expression and Purification of HIV-1 Protease-HIV-1 protease was expressed in E. coli and then renatured from inclusion bodies as described previously (28).
N-terminal Amino Acid Sequence Analysis-The N-terminal sequence was determined using the Hewlett Packard G1005A protein sequencing system with on-line PTH analysis. All methods, reagents, and consumables used were those recommended by the manufacturer.
Mass Spectrometry-Matrix-assisted laser desorption ionization mass spectrometry data were obtained on a PerSeptive Biosystems Voyager DE-Pro mass spectrometer. The spectra were acquired in the linear mode with delayed extraction. External calibration was performed using calibrant 3 supplied by the manufacturer. The sample was diluted 1:10 in sinapinic acid matrix solution. The matrix was prepared by dissolving 10 mg/ml sinapinic acid in aqueous 30% acetonitrile containing 0.3% trifluoroacetic acid.
Generation of Anti-HERV-K10 Protease Antiserum-1 mg of truncated version of HERV-K10 protease was loaded on SDS-PAGE, and the band was excised from the gel. The gel slice was covered with phosphate-buffered saline and emulsified with a syringe through a 23-gauge needle. The emulsion was then used directly to immunize rabbits with 100 g/dose.
Enzyme Assay-To measure the inhibitory potency of compounds, the discontinuous HPLC method described in Erickson-Viitanen et al. (34) was used. The synthetic fluorescent cationic peptide substrate 2-aminobenzoyl-Ala-Thr-His-Gln-Val-Tyr-Phe(NO 2 )-Val-Arg-Lys-Ala (28) was incubated with truncated or full-length HERV-K10 at 25°C in an assay buffer containing 50 mM MES, pH 5.0, 1 M NaCl, 20% glycerol, 1 mM EDTA. The synthesis of the substrate has been described elsewhere (28). The enzymatic reaction was terminated with 0.2 M ammonium hydroxide. Enzymatic hydrolysis of the substrate yielded the fluorescent anionic product, (2-aminobenzoyl)-ATHQVY. The extent of hydrolysis was determined using anion-exchange HPLC. An Amersham Pharmacia Biotech HR5/5 MonoQ column eluted at 1.0 ml/min with 0 -70% buffer B for 10 min was used to separate the fluorescent cleavage product from the fluorescent substrate. The mobile phase buffer A contained 20 mM Tris/HCl, 0.02% sodium azide, and 10% acetonitrile at pH 9.0, whereas buffer B consisted of buffer A plus 0.5 M ammonium formate at pH 9.0. The column was washed with 100% buffer B for 5 min and then stepped down to 0% buffer B to recycle the gradient for the next injection. The cleavage product was measured at an emission wavelength of 430 nm and excitation wavelength of 330 nm. Linearity of enzymatic activity with time was first established, and based on the results, reactions involving the truncated or full-length HERV-K10 protease were quenched after 20 in or 40 min, respectively (Fig. 1).
The K m values were determined with fixed enzyme concentration (0.5 nM) and substrate concentrations of 0.5-50 M; the data were fitted directly to Michaelis-Menten equation with GraFit software version 4.0.10 (Erithacus Software Ltd.). In the next step, potent inhibitors were identified as described below. The active site concentrations of the proteases were determined by titrating the enzymes with different concentrations of SD146. This data then enabled us to convert v max values into those for k cat . Inhibition Kinetics-Samples of the HIV-1 protease inhibitors indinavir (MK-639), saquinavir (Ro 31-8959), and ritonavir (ABT-538) were synthesized at DuPont Pharmaceuticals. Pepstatin A was purchased from Sigma-Aldrich. The cyclic ureas were prepared as described elsewhere (29 -32). All inhibitors were dissolved in dimethyl sulfoxide and stored at Ϫ20°C. Their chemical structures are shown in Fig. 2. The activity of the proteases was measured in the absence and presence of seven different concentrations of inhibitor at a fixed concentration of both enzyme and substrate. The proteases were preincubated 5 min at 25°C with inhibitors. Substrate was then added to the final concentration of 2 M, and the assay was carried out as described above. Fractional activities ranging from 0.2 to 0.8 relative to uninhibited control were fitted directly to the following Morrison equation (33).

In this equation, [I] is inhibitor concentration, [E t ] is the concentration of active enzyme
, v i is the activity at a particular inhibitor concentration, v o is activity of uninhibited enzyme, v i /v o is fractional activity, and K i (app) is the estimated apparent inhibition constant. On the basis of previous studies with HIV-1 protease (34), the mode of inhibition was assumed to be competitive. To verify this assumption, the dose response data were obtained for SD146 as a representative compound at four substrate concentrations; IC 50 values increased linearly with increasing substrate concentration, indicating the competitive nature of inhibition (35).
Effect of SD146 on the Cleavage of in Vitro Translated HERV Gag Polyprotein by HERV-K10 Protease-Plasmid pcDNA3.1(ϩ)/HERV-K10 gag was used as template in TnT® Quick Coupled Transcription/Translation (Promega) reactions to produce [ 35 S]methionine-labeled HERV Gag polyprotein that then served as a substrate for HERV-K10 protease. The in vitro translation product was incubated together with 0.54 M HERV-K10 protease (truncated form) and various concentrations of SD146 (0 -1 M) in 20 mM PIPES, pH 6.5, 0.1 M NaCl, 1 mM DTT, 10% glycerol, for 1 h at 37°C. The substrate and cleavage products were separated on NuPage SDS-polyacrylamide gel (Novex) and autoradiographed. Subsequently, the dried gel was scanned for radioactivity with a Bio-Rad Molecular Imager FX, and the HERV Gag polyprotein bands were quantitated using QuantityOne software (Bio-Rad).
Mammalian Cell Cultures and Collection of Particulate Material-Human teratocarcinoma cell lines NCCIT, PA-1, and NTERA-2, as well as the embryonic kidney line 293 (all purchased from American Type Culture Collection) were cultured in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) supplemented with 10% fetal calf serum, 2 mM glutamine, and antibiotics (100 units/ml penicillin, 100 g/ml streptomycin). Cell cultures were subcultured routinely twice per week. NCCIT cell line was treated with several concentrations of SD146 (up to 2 M) or left untreated, and aliquots of culture supernatants were taken at time 0 and after 1 day. HERV-K particles were recovered by ultracentrifugation. After a 10-min centrifugation in a Sorvall RT6000B table top centrifuge at 1500 rpm to remove unbroken cells and large cell debris, the samples were centrifuged for 3 h in a Sorvall RC80 ultracentrifuge at 78,000 ϫ g at 8°C. Medium was discarded. The virus pellets were resuspended in a minimal volume of 1% SDS, 1% mecaptoethanol, and 7% glycerol, heated at 56°C for 1 min, and loaded onto 10% polyacryamide gels.
Immunoblotting-Protein samples were separated with SDS-PAGE and transferred to Immobilon-P polyvinylidene diflouride membranes (Millipore) using semidry method. The membranes were probed with either anti-HERV-K10 protease antiserum at a dilution of 1:250 or polyclonal anti-HERV-K Gag antiserum (15) at a dilution of 1:10,000. Blots were stained indirectly by using horseradish peroxidase-conjugated donkey anti-rabbit antibodies and subsequent chemiluminescence detection (PerkinElmer Life Sciences).
Viral RNA Isolation and RT-PCR-RNA was isolated from concentrated virus particles with QIAamp Viral RNA Mini Kit (Qiagen). Eluted RNA was treated with RNase-free DNase I to digest any contaminating cell genomic DNA and repurified with the same kit. RNA was eluted in 60 l of diethylpyrocarbonate-treated water. Reverse transcription was performed in a volume of 20 l containing 5 l of viral RNA, 0.5 mM dNTP mix, 10 units of RNAsin, 100 ng of primer PRT-B, and 4 units of Omniscript reverse transcriptase (Qiagen). The reaction was carried out for 1 h at 37°C. 5 l of RT reaction was used in PCR amplification, together with 0.1 M primers PRT-A and PRT-B, 1.5 mM MgCl 2 , 0.2 mM dNTP mix, and 1 unit of Taq DNA polymerase (PerkinElmer Life Sciences). The reaction mix was initially denatured at 94°C for 5 min and then subjected to 30 cycles of denaturation at 94°C, annealing at 50°C, and extension at 72°C. An aliquot of PCR reaction was used directly in DNA sequencing, with either PRT-A or PRT-B as a primer.
Molecular Modeling of HERV-K10 Protease-The three-dimensional homology model of the truncated version of HERV-K10 protease was constructed using coordinates of HIV-1 protease complexed with SD146 as a template (Protein Data bank file 1QBT.pdb; Ref. 36) with the program Molecular Operating Environment (Chemical Computing Group Inc.). The sequence was aligned initially maximizing the homology but later adjusted to accommodate the insertion at position 39 (HIV numbering) at the elbow of the flap and the insertion at position 80 (HIV numbering) at the active site, mimicking the three-dimensional structure of feline immunodeficiency virus protease. The homology algorithm of the Molecular Operating Environment software created 10 models, each of which was generated by making a series of Boltzmannweighted choices of side chain rotamers and loop conformations from a set of protein fragments of high resolution protein structures. An average model was potential energy-minimized using AMBER forcefield.

Expression and Purification of HERV-K10
Proteases.-Two versions of HERV-K10 protease were expressed in E. coli. The amino acid sequences of polypeptide chains that were expressed are shown in Fig. 3. The C-terminal boundary of the truncated version was chosen on the basis of sequence homology with the mature HIV-1 protease (25). An additional 58 amino acid residues included at the N-terminal end of the protein were expected to be cleaved off in an autocatalytic manner. This N-terminal flanking portion was expressed to allow us to readily monitor autoprocessing activity (28). The nucleotide sequence of the clone that was chosen for E. coli expression was in complete agreement with the cDNA sequence of HERV-K10 protease ORF published in Ono et al. (6). The truncated, "core" protease was expressed as a 185-amino acid precursor at a high level in form of insoluble cytoplasmic inclusion bodies. During the renaturation step with dialysis all of the precursor (20 kDa) was autocatalytically processed to give rise to mature, enzymatically active 13-kDa form. The site of N-terminal autoprocessing was determined with N-terminal amino acid sequencing. It was shown to be GKAAY-WASQ with the dash designating the scissile bond and was in agreement with previous findings (25). When analyzed with mass spectroscopy, the protein showed a molecular mass that was in agreement with the expected size and that also suggested that no C-terminal autoprocessing occurred (data not shown). In addition to the monomer, a peak representing the mass of a dimerized protease was present. Affinity chromatography with pepstatin A as an immobilized ligand was used efficiently to purify the 13-kDa form of HERV-K10 protease (37). The method described by Wondrak et al. (37) for HIV-1 protease purification was adjusted to ensure the best yield of HERV-K10 protease. Several NaCl and (NH 4 ) 2 SO 4 concentrations in pepstatin A binding buffer were tested. The majority of HERV-K10 protease was bound to pepstatin A in the presence of 0.5 M NaCl and complete absence of (NH 4 ) 2 SO 4 . Bound protease was eluted with no salt buffer and appeared to be homogenous as assessed with SDS-PAGE, isoelectric focusing, and native PAGE (data not shown).
The expression plasmid for full-length HERV-K10 protease was constructed so that only five additional amino acids (in addition to T7 tag) were present at N terminus because the presence of longer flanking region was observed not to be necessary for proper autoprocessing. At the C terminus the protease extends all the way to the termination codon that is present in full-length HERV-K10 provirus (6), which accounts for additional 50 amino acid residues not present in the truncated, core protease version. Nucleotide and deduced amino acid sequence of the clone that coded for full-length protease differ from that published by Ono et al. (6) as described previously (25). The full-length version differs from the truncated form also in the residue at position 65 (mature HERV protease numbering; Fig. 3); this residue is not positioned close to the active site or in the flaps and is believed not to be important for substrate or inhibitor binding. Metal chelation chromatography and subsequent ion exchange chromatography were used to purify both mature wild type full-length HERV-K10 protease and its active site mutant (D26N). Soluble fraction of E. coli cells was applied to nickel resin, and His tag-containing protease was bound. After elution with high imidazole buffer, the protease was further purified to homogeneity with cationic ion exchange chromatography on MonoS column. The mature wild type enzyme had a molecular mass of 18.2 kDa (including His tag), whereas active site mutant showed a molecular mass of 20 kDa because of the presence of T7 tag and remaining N-terminal pentapeptide that was not cleaved off because of lack of enzymatic activity of the protein.
Enzymatic Activity of HERV-K10 Proteases-Enzymatic activity of the enzymes was quantitatively assessed by determining kinetic constants for the hydrolysis of 2-aminobenzoyl-Ala-Thr-His-Gln-Val-Tyr-Phe(NO 2 )-Val-Arg-Lys-Ala. First, the K m values were determined. After identifying compounds with potent inhibitory activity, the active sites of the protein preparations were titrated, and the k cat values were then calculated from v max . As can be seen in Table I, the K m value for the truncated version of HERV-K10 protease was about 20 times lower than that of the full-length counterpart. Similar ratio was observed previously for the hydrolysis of a different peptide substrate and under slightly different reaction conditions; in that report the K m for the truncated version was about 10 times lower than that of the full-length enzyme (25). The turnover capacity (k cat ) of the full-length protease was about 10 times higher than that of the truncated form, resulting in a catalytic efficiency that was twice higher for the 13-kDa protein than what one could observe with the 18-kDa form. The ratio of k cat values of both protease forms differs from that in a previous report (25); the difference is probably to be attributed to different substrates and to slightly different reaction conditions that were used in the assays.
Enzymatic activities of both versions of HERV-K10 protease were then tested against polyprotein substrate. Radioactively labeled HERV-K10 Gag polypeptide was shown to be successfully cleaved by both versions of the protease, as well as by recombinant HIV-1 protease (Fig. 4A). The specificities of fulllength and truncated forms seemed to be identical as suggested by similar cleavage patterns. HIV-1 protease, however, cleaved HERV-K10 Gag polypeptide at different sites, suggesting different substrate specificity under the reaction conditions that we used in our assay. Active site mutant of full-length HERV-K10 protease was not enzymatically active, as expected. These results seem to be consistent with differential cleavage of HIV-1 Gag and Pol precursors by HERV-K10 protease in the context of chimeric virions, where the HERV enzyme cleaved HIV-1 polyproteins at both apparently authentic as well as nonauthentic sites (17,38).
Identification of Potent HERV-K10 Inhibitors-To evaluate the capacity of potent HIV-1 protease inhibitors to inhibit HERV-K10 protease, K i (app) values for a series of P2,P2Ј-substituted cyclic ureas were determined. In addition, pepstatin and three Food and Drug Administration-approved HIV-1 protease inhibitors, ritonavir, saquinavir, and indinavir, were tested. The apparent inhibition constants for both versions of HERV-K10 protease are shown in Table II, together with previously reported values for inhibition of wild type HIV-1 protease (39). Although potent inhibitors of wild type HIV-1 protease activity, the three Food and Drug Administration-approved compounds turned out to be weak inhibitors of both versions of HERV-K10 protease. The linear peptidyl mimetic inhibitors had K i (app) values ranging from 0.6 to 5.7 M.
A series of 13 compounds of the cyclic urea class was tested, all of them being P2,P2Ј-substituted. The symmetric substituted cyclic ureas in general fared better in inhibition assay than the five asymmetric compounds tested. From the latter, compound Q8467 exhibited the weakest activity, with the apparent inhibition constants being 16 and 61 nM for truncated and full-length HERV-K10 protease, respectively. The remaining asymmetric ureas (SD152, SD145, XW805, and XV651) did not differ significantly from each other, their K i (app) values being in the range of about 3-8 nM for 13-kDa protein and about 30 -40 nM for 18-kDa form. Among cyclic C 2 symmetric ureas the compound with the smallest, cyclopropyl side groups, XK234, fared the worst, the K i (app) values being about 0.7 and 1.9 M. This compound had also turned out to be less efficient in inhibiting HIV-1 protease than the bulkier members of this group. XM412, also known as DMP450, containing m-aminomethylbenzyl groups, exhibited more inhibitory potency toward HERV-K10 proteases, although with apparent inhibition constants of about 90 and 400 nM, it was still much less potent than the remaining six cyclic ureas. XV643, XV644, SD146, XV648, and XV652 were capable of inhibiting 13-kDa protease in subnanomolar range, with the K i (app) values ranging from 0.10 nM for XV648 to 0.52 nM for XV652. The group of these five compounds inhibited the 18-kDa enzyme in nanomolar range; apparent inhibition constants were 2.3-4.3 nM. In general, K i (app) values for the full-length form of HERV-K10 protease were about 3-20 times higher than those for the truncated counterpart; however, the compounds that acted as weak inhibitors with one version of the protease were also weak with the other and vice versa. The differences in K i (app) values between both versions of the protease were consistent with the lower K m value obtained for the 13-kDa form and were ob-  served also in a previous report where compounds KNI-227 and KNI-272 were measured (25). The differences are very likely to be attributed to 50-amino acid C-terminal extension present in the full-length enzyme; however, in the absence of x-ray data it is not possible to provide a more detailed explanation. Inhibition of HERV-K10 Gag Processing-SD146 was previously reported (40) to have potent activity in cells to block HIV-1 Gag processing by a variety of HIV-1 protease mutants. Because of this and its excellent potency against HERV-K10 proteases (Table II), it was chosen for detailed studies of HERV-K10 Gag processing. To estimate the range of concentration at which which SD146 inhibits processing of HERV-K Gag polyprotein, we first tested the system with recombinant HERV-K10 protease and in vitro translated HERV-K10 Gag polyprotein substrate. SD146 inhibited the processing of HERV Gag, with the dose response data shown in Fig. 4B. On the basis of the quantitation of substrate disappearance, the IC 50 was estimated to be 0.35 M.
Among the cell lines we examined, the only one in which we could detect synthesis/release of HERV Gag polypeptides was NCCIT (Fig. 5), although PA-1 appeared to express small quantities of complete and partially processed intracellular HERV Gag (not shown). NCCIT cells released HERV-K Gag polypeptides that were detected mainly at 30 kDa, although also observed were varying amounts of larger polypeptides of 39 and 76 kDa (Fig. 5A) (1, 16). When 1 or 2 M SD146 was added to the cells, and cells were incubated for 1 day or more, the pattern changed drastically, with the released particles containing little or no 30-kDa polypeptide and correspondingly greater amounts of the 76-kDa full-length Gag precursor (Fig.  5A). Also seen in inhibitor-treated NCCIT cells and virus particles were forms of Gag-related antigens larger than 76 kDa, suggesting possible accumulation of Gag-Pol precursors. This was concurrent with disappearance of the processed 30-kDa core polypeptide. The difference in cell lysates was present but much less obvious than that in virion samples, mainly because only a small quantity of processing of HERV Gag is intracellular. In cells treated with 1 M SD146 we could observe the disappearance of p30 (Fig. 5B). Dose response data were obtained for the inhibition of HERV Gag and HERV protease processing (Fig. 5, C and D, respectively). The size of the mature HERV-K protease seemed to be slightly higher than 18 kDa (Fig. 5D). On the basis of the quantitation of product appearance, we estimated the IC 50 to be 0.37 M in the case of  Gag processing and 0.42 M in the case of protease maturation. Taken together, the results in Fig. 5 show that the HIV-1 protease inhibitor SD146 is able to effectively block HERV-K10 Gag processing, both in a teratocarcinoma cell line and in the released particles, as predicted from our enzyme inhibition cell-free results (Table II and Fig. 4B).
RT-PCR and DNA Sequencing of NCCIT-derived Virions-To verify that the particles derived from NCCIT cell line are indeed HERV-K encoded, viral RNA was isolated from the cell culture medium and its protease region RT-PCR amplified. A single product of expected size (ϳ500 base pairs) was obtained. Direct DNA sequencing of the PCR product resulted in a single sequence and revealed that this region differs from the HERV-K10 clone published by Ono et al. (6) in 2 nucleotides. Neither of the substitutions (T3545C and C3572T; numbering as in Ref. 6) lead to an amino acid change. When BLAST search was performed against all nucleotide sequences deposited in GenBank TM to that date, the amplified region of RNA of NC-CIT-derived HERV particles completely matched only the HERV protease region of a recently deposited Homo sapiens chromosome 5 clone CTB-69E10 (GenBank TM accession number AC016577). DISCUSSION Retroviral proteins are synthesized in the form of Gag or Gag/Pol precursors that are then processed by the action of a virus-encoded aspartic protease. The existence of a functional HERV-K protease was inferred from the presence of processed Gag proteins in teratocarcinoma cells (5). Direct evidence for a functional protease activity came from expression of different clones in E. coli (15,17,25,38).
Recently, a hypothesis has been proposed that HERV-K10 encoded aspartic protease might complement HIV-1 protease during infection and thereby interfere with clinical antiviral therapy because it is highly resistant to currently approved HIV-1 protease inhibitors (25). To identify low molecular weight compounds that could inhibit proteolytic activity of this enzyme, we first expressed two versions of this enzyme in an E. coli expression system. The N termini of both enzymes were the result of autocatalytic processing by the protease. The C terminus of smaller, core form was chosen on the basis of sequence homology with mature HIV-1 protease. The C-terminal boundary of full-length version corresponds to that found in prt-ORF of proviral DNA (6); this version has 50 additional amino acids on its C terminus. Whether it is the full-length enzyme that is biologically relevant or additional C-terminal processing occurs to give rise to smaller molecular species remains to be seen. Initial studies suggest that some limited cleavage of 13 amino acid residues at the C terminus occurs after prolonged incubation (25,38).
The DNA sequence of HERV-K10 protease ORF strongly suggests that this protease belongs to the group of aspartic proteases, because the ORF contains sequence motif LVDT-GAXX(T/S)(V/I). Furthermore, a sequence GLVGIG, a so-called "flap," is found downstream of the active center. In addition, the sequence GRDLL conserved in aspartic proteases, is found at nucleotide position 3723-3737 (Ref. 15; numbering as in Ref. 6). Schommer et al. (17) showed that presence of high concentration of HIV-1 protease inhibitor Ro 31-8959 (saquinavir) can inhibit autoprocessing of HERV-K10 protease in E. coli expression broth, suggesting a similarity between active sites of the two viral proteases. We therefore decided to test a series of our cyclic ureas, second generation HIV protease inhibitors (reviewed in Ref. 41), for their ability to inhibit HERV-K10 protease. Although as a whole the cyclic urea class has relatively poor pharmacokinetic properties, mostly because of low water and oil solubility (41), these compounds are extremely potent against HIV-1 protease in vitro, and some of them have very good resistance profiles. At least one of the cyclic ureas, DMP450 (42), is presently in human clinical trials versus HIV. Several symmetric bisamides exhibited high potency against both versions of HERV-K10 protease. In the absence of any available structural data, we built a homology three-dimensional model of the 13-kDa form of this enzyme to be able to understand the mode of action of the compounds.
The cyclic urea substituents at P1, P1Ј, P2, and P2Ј are optimized for good potency against HIV-1 protease. In this enzyme, P1 and P1Ј residues form van der Waals' contacts with Pro 81 , Val 82 , and Ile 84 , whereas P2 and P2Ј groups form contacts with Ile 47 , Ile 50 , and Ile 84 . Cyclic urea inhibitors with smaller P2 and P2Ј were shown to be less potent against HERV-K10 protease than HIV-1 protease (XK234, XM412). However, cyclic urea amides containing P3 and P3Ј groups are as potent against HERV-K10 protease as HIV-1 protease (e.g. XV652, XV643, XV644, SD146, and XV648). Most of the hydrogen bond contacts between the cyclic urea amide inhibitor and HIV-1 protease complexes are predicted to be maintained in the cyclic urea amide and HERV-K10 protease complexes (Fig.  6). The potency of cyclic ureas increase with the increasing potential of forming hydrogen bonds. For example, SD146 (HERV-K10 K i (app) ϭ 0.15 nM), which is capable of forming 12 hydrogen bonds, is ϳ4500 times more potent than XK234 (HERV-K10 K i (app) ϭ 670 nM). Besides the interaction with hydrogen bonds, the hydrophobic interaction is predicted to be important for the good potency of the cyclic urea amides. For instance, the substitution of Ile 47 in HIV-1 protease for Leu 52 in HERV-K10 protease is predicted to result in loss of van der Waals' interactions with P2 or P2Ј groups, but at the same time this change results in increased van der Waals' interactions between Leu 52 HERV and P3, P3Ј groups of cyclic urea amides. A similar effect caused by the hydrophobic interactions was observed previously in case of double mutant V82F/I84V of HIV-1 protease (40,42).
The question of activity of the cyclic ureas in cells was addressed. In this study we demonstrated that HERV-K Gag processing in a cell environment can be blocked by synthetic protease inhibitors, as could be seen by substantially reduced proportion of HERV Gag precursor being cleaved to smaller polypeptides in NCCIT cell line treated with SD146 (Fig. 5A). To our knowledge, this is the first report of inhibition of HERV-K Gag maturation in cell milieu. Given the inability of cyclic ureas to inhibit cellular proteases (34), our results strongly support a model in which the aspartic protease of HERV-K10 processes homologous Gag polypeptides in human teratocarcinoma cells. Much of the HERV-K10 Gag within NCCIT cells is unprocessed. This is different from the case with HIV-1-infected cells, where a significant percent of HIV-1 Gag is cleaved. In contrast, processing of extracellular HERV Gag appears to be efficient, implying the HERV-K10 protease is inactive or unavailable except in maturing virions. HIV-1 protease is toxic to a variety of mammalian cells (43), but clearly human cells, including some teratocarcinoma cell lines, are not damaged by endogenous retroviral proteases. The question of whether the cells have evolved to resist the action of the endogenous viral proteases or the enzymes are sequestered/inactive until packaging/exit should be addressed.
The extracellular particle yield, as estimated by Western blotting of total viral proteins, was roughly the same in presence of the protease inhibitor, indicating that HERV-K Gag polypeptide processing is not a limiting step for particle release. Similar results obtained with viral RNA isolation from the particulate material of NCCIT cell culture medium and its subsequent quantification with RT-PCR amplification support this observation (data not shown). These data are consistent with the observation that HIV-1 protease inhibitors block the processing of Gag and Gag-Pol precursor polyproteins in HIV-1-infected cells but do not markedly alter either the number of particles released from the infected cells (44,45) or the amount of packaged viral RNA (46,47). In addition to using antigenspecific immunoblotting, we verified the identity of NCCIT released virions by checking the nucleotide sequence of packaged RNA. The sequence of the 500-nucleotide protease region that was RT-PCR amplified unequivocally shows that the virions belong to HERV-K family. However, additional regions would have to be sequenced for an exact clone number to be assigned, especially with regard to the fact that the recent estimates based on BLAST searches and phylogenetic analyses show that there could be as many as 170 HERV-K elements present in human genome (4). The protease amino acid sequence deduced from the obtained nucleotide sequence was identical to that of HERV-K10 clone (6).
Although cyclic ureas act as potent inhibitors of HIV-1 and HERV-K protease, they do not inhibit mammalian, nonretroviral cellular aspartic proteases (34). However, a question arises whether cell processes could be affected because of HERV-K protease inhibition. The fact that HERVs remain a constitutive part of the genome and the notion that ORFs for all major viral proteins exist and have retained coding capacity despite extensive deleterious effects normally associated with endogenization of retroviruses suggest that they may confer certain positive traits to the host (48). HERV encoded proteins, including HERV-K protease, might well be involved in normal cell physiology and pathophysiology. Our results in which the activity of HERV protease and inhibition of viral protein processing could be efficiently accomplished in teratocarcinoma cells may help to clarify the role of HERVs in cell physiology.
Acknowledgments-We acknowledge Beverly C. Cordova for providing us with the human lymphocyte fraction and Drs. Ralf Tönjes and Reinhard Kurth (Paul-Ehrlich-Institut, Langen, Germany) for kindly supplying pcG3gag clone. We are especially grateful to Ronald M. Klabe and Dr. James L. Meek for excellent technical advice with HPLC enzyme assay. We thank Leah A. Breth and Jennifer E. Kochie for FIG. 6. Schematic representation of hydrogen bonds between HIV-1/HERV-K10 protease and the SD146. Hydrogen bonds between HIV-1 protease and SD146 were determined by x-ray crystallography (36), and those for HERV-K10 protease were modeled. In the model of HERV-K10 protease complexed with SD146, all hydrogen bonds are predicted to be preserved except that between the side chain of Asp 30 and ring nitrogen atom of the inhibitor (thicker line), because Asp 30 is replaced with Val 31 in HERV-K10 protease. HERV residues are in parentheses and in bold type. All distances are in Å.
raising anti-HERV-K10 protease antiserum, Jeanne I. Corman for Nterminal amino acid sequencing and mass spectroscopy analysis, and Wilfred Saxe for help with modeling of HERV-K10 protease. Thanks also to Drs. Lee T. Bacheler and Robert A. Copeland for helpful discussions and to Dr. Susan K. Erickson-Viitanen for continuing support and useful suggestions.