Glycation and Post-translational Processing of Human Interferon-γ Expressed in Escherichia coli*

Until recently, nonenzymatic glycosylation (glycation) was thought to affect the proteins of long living eukaryotes only. However, in a recent study (Mironova, R., Niwa, T., Hayashi, H., Dimitrova, R., and Ivanov, I. (2001) Mol. Microbiol. 39, 1061-1068), we have shown that glycation takes place in Escherichia coli as well. In the present study, we demonstrate that the post-translational processing (proteolysis and covalent dimerization) observed with cysteineless recombinant human interferon-γ (rhIFN-γ) is tightly associated with its in vivo glycation. Our results show that, at the time of isolation, rhIFN-γ contained early (but not advanced) glycation products. Using reverse phase high performance liquid chromatography in conjunction with fluorescence measurements, enzyme-linked immunosorbent assay, and mass spectrometry, we found that advanced glycation end products arose in rhIFN-γ during storage. The latter were identified mainly in the Arg/Lys-rich C terminus of the protein, which was also the main target of proteolysis. Mass spectral analysis and N-terminal sequencing revealed four major (Arg140↓Arg141, Phe137↓Arg138, Met135↓Leu136, and Lys131↓Arg132) and two minor (Lys109↓Ala110 and Arg90↓Asp91) cleavage sites in this region. Tryptic peptide mapping indicated that the covalent dimers of rhIFN-γ originating during storage were formed mainly by lateral cross-linking of the monomer subunits. Antiviral assay showed that proteolysis lowered the antiviral activity of rhIFN-γ, whereas covalent dimerization completely abolished it.

In the present study, we demonstrate that the post-translational processing (proteolysis and covalent dimerization) observed with cysteineless recombinant human interferon-␥ (rhIFN-␥) is tightly associated with its in vivo glycation. Our results show that, at the time of isolation, rhIFN-␥ contained early (but not advanced) glycation products. Using reverse phase high performance liquid chromatography in conjunction with fluorescence measurements, enzyme-linked immunosorbent assay, and mass spectrometry, we found that advanced glycation end products arose in rhIFN-␥ during storage. The latter were identified mainly in the Arg/Lys-rich C terminus of the protein, which was also the main target of proteolysis. Mass spectral analysis and N-terminal sequencing revealed four major (Arg 140 2Arg 141 , Phe 137 2Arg 138 , Met 135 2Leu 136 , and Lys 131 2Arg 132 ) and two minor (Lys 109 2Ala 110 and Arg 90 2Asp 91 ) cleavage sites in this region. Tryptic peptide mapping indicated that the covalent dimers of rhIFN-␥ originating during storage were formed mainly by lateral cross-linking of the monomer subunits. Antiviral assay showed that proteolysis lowered the antiviral activity of rhIFN-␥, whereas covalent dimerization completely abolished it.
Natural human interferon-␥ (hIFN-␥) 1 is a glycoprotein with a carbohydrate moiety dispensable for its biological activity (1)(2)(3). Glycosylation of the protein has been extensively studied (4 -6) and shown to be a source of substantial heterogeneity in natural (7,8) and recombinant hIFN-␥ (rhIFN-␥) preparations (9 -13). Partial proteolysis yielding at least three truncated species further contributes to protein microheterogeneity (14 -16). Extensive experimental evidence indicates that cleavage of rhIFN-␥ occurs at its C terminus adjacent to basic amino acids (17)(18)(19). Proteolysis of rhIFN-␥ is usually attributed to specific (trypsin-like) proteases; however, it is observed also in purified protein samples either stored in solution (18,20) or lyophilized (21). Another striking observation concerns the formation of covalent rhIFN-␥ dimers (7,22). Because the mature form of hIFN-␥ represents a rare case of a cysteineless protein (23), its covalent dimers could not be a result of intermolecular disulfide bond formation. Lauren et al. (18) reported that rhIFN-␥ dimers are formed by terminal cross-linking between a native N and a truncated C terminus. The authors suggested that proteolysis and dimerization are associated reactions, but whether they are enzyme-catalyzed or not remained unclear.
Escherichia coli lacks enzyme activities for protein glycosylation of the type occurring in eukaryotes. Thus, hIFN-␥ synthesized in this bacterium is devoid of the native glycosylation pattern. However, proteins may become glycoproteins in an alternative type of glycosylation called glycation (to be distinguished from the enzymatic type of glycosylation). Glycation does not require enzyme catalysis and therefore is thought to proceed slowly under physiological conditions. The first step of glycation includes the formation of reversible Schiff bases (aldimines) at free amino groups in proteins, which are then converted to more stable Amadori products. Subsequent chemical rearrangements lead to the formation of complex terminal adducts, which remain tightly bound to proteins and are designated as advanced glycation end products (AGEs). After the initial attachment of glucose, the next rearrangements may occur under sugar-free conditions (24). Glycation was extensively studied in humans because of its clinical importance (25) and relation to aging (26). However, no one expected that glycation could occur in bacteria because of their short life span and intense protein turnover. This could explain why bacterial glycoproteins are scarcely studied. Such proteins are speculated to account for the existence of the enzymatic rather than nonenzymatic type of glycosylation in prokaryotes (27). The only bacterial glycoprotein considered so far as being a result of glycation is the Bacillus thuringiensis crystal protein (28,29). However, because this protein is secreted in the culture medium, it was concluded that the "sugars detected were the product of fermentation conditions rather than bacterial genetics" (30).
Recently, we reported that nonenzymatic glycosylation occurs also in E. coli under physiological conditions (20). To the best of our knowledge, this was the first report providing evidence for endogenous glycation in bacteria. We also demonstrated that 4-month-old rhIFN-␥ preparations isolated from E. coli were modified with AGEs. The purified protein was also prone to progressive proteolysis and covalent dimerization during storage. It is well known that chemical rearrangements in the late glycation stages cause peptide bond cleavage (31,32) and covalent cross-linking at lysine and arginine (but not cysteine) residues (33,34). That is why we concluded that glycation might provide a rational explanation for both the proteolysis and dimerization of purified rhIFN-␥ over time. Here, we present new data in favor of the link between glycation and post-translational processing of rhIFN-␥.

EXPERIMENTAL PROCEDURES
Purification of rhIFN-␥-A gene coding for cysteineless hIFN-␥ was expressed constitutively (35) in E. coli XL1-Blue (supE44 hsdR17 recA1 endA1 gyrA46 thi relA1 lac Ϫ FЈ (proAB ϩ lacI q lacZ⌬M15 Tn10 (tet r ))). Transformed cells were cultured overnight at 37°C in 1-liter flasks containing 250 ml of LB medium supplemented with 10 g/ml tetracycline and 0.1% glucose. Bacteria were harvested by centrifugation; suspended in buffer containing 1 M urea, 0.4 M guanidine hydrochloride (GdnHCl), and 20 mM Tris-HCl (pH 8.8); and disrupted by sonication. Cell lysate was centrifuged for 20 min at 10,000 ϫ g. The crude pellet containing rhIFN-␥ in the form of inclusion bodies was suspended in 7 M GdnHCl (pH 7), slowly diluted to 1 M GdnHCl with ice-cold water, and clarified by centrifugation. The supernatant was loaded onto an octyl-Sepharose CL-4B column ( Concentration Measurements and SDS-PAGE-The concentration of rhIFN-␥ was measured either by the method of Bradford (36) or spectrophotometrically. Standards in the Bradford assay were prepared with bovine serum albumin (BSA) at 10 -100 g/ml. In the spectrophotometric measurements at 280 nm, an extinction coefficient of 0.6 for 0.1% protein solution was used. 18% SDS-polyacrylamide gels were run according to Laemmli (37).
Thiobarbituric Acid Colorimetric Method-After the second chromatography step (see above), rhIFN-␥ fractions with different retention times were precipitated with trichloroacetic acid at a final concentration of 20%. The trichloroacetic acid pellet was dissolved in Tris-buffered saline, and the protein concentration was adjusted to 5 mg/ml based on the absorbance at 280 nm. The Amadori product content in rhIFN-␥ was determined (as nmol of Fru eq) by the thiobarbituric acid method (38) using a standard curve prepared with fructose at 10 -100 nmol/ml.
Reverse Phase High Performance Liquid Chromatography (RP-HPLC)-Purified rhIFN-␥ samples were analyzed on a TSKgel phenyl 5PW column (4.6 ϫ 75 mm) at a flow rate 0.5 ml/min at ambient temperature. Solvent A was 0.8% trifluoroacetic acid in water, and solvent B was 1% trifluoroacetic acid in acetonitrile. Elution was carried out isocratically for 5 min at 15% solvent B and then with a linear gradient to 90% solvent B for 30 min. The HPLC system (Japan Spectroscopic) was equipped with a Model 875-UV detector and a Model 820-FP fluorescence detector. For preparative isolation of glycated peptides, 100 g of rhIFN-␥ was loaded onto the RP-HPLC column, and the fluorescent fraction (designated as F in Fig. 2) was collected manually, evaporated to dryness, and dissolved in 10 l of distilled water. The peptide fraction was loaded onto an ODS HG-5 column (2.0 ϫ 150 mm) and eluted at a flow rate 0.3 ml/min isocratically for 5 min with 5% solvent B (2.0% trifluoroacetic acid in acetonitrile), followed by raising its concentration to 70% in 40 min. Solvent A was 1.6% trifluoroacetic acid in water.
Competitive Enzyme-linked Immunosorbent Assay (ELISA)-BSA was dissolved in 0.5 M D-glucose at 10 mg/ml, filter-sterilized, and incubated at 37°C for 3 months to obtain AGE-modified BSA for use in the assay. Glycated BSA was dialyzed against phosphate-buffered saline. Then, 96-well microtiter plates were coated with 50 l/well AGE/ BSA solution at 0.5 mg/ml. Plates were blocked overnight at 4°C and then washed three times with 0.05% Tween 20 in phosphate-buffered saline. Samples or standards (25 l/well) dissolved in phosphate-buffered saline were mixed with 50 l of assay buffer (0.25% BSA in phosphate-buffered saline) and 50 l of horseradish peroxidase-labeled anti-AGE antibody (diluted 1:20 in assay buffer). The two monoclonal anti-AGE antibodies used (AG-1 and AG-10) were specific for imidazolone and N ⑀ -(carboxymethyl)lysine (CML), respectively (39 -41).
Competition reaction was carried out for 40 min at room temperature. After three washing steps, color was developed with 100 l/well ophenylenediamine dissolved in 0.2 M Na 2 HPO 4 citrate buffer (pH 5.2) and 0.02% H 2 O 2 at a concentration of 1 mg/ml. The reaction was stopped with 100 l/well 0.8 M H 2 SO 4 , and the absorption at 492 nm was read with a microplate reader. AGE/BSA was used as a standard at 0.05-0.6 mg/ml. The concentration of AGE-modified rhIFN-␥ was expressed as mg/ml AGE/BSA eq. Liquid Chromatography/Electrospray Ionization Mass Spectrometry (LC/ESI-MS)-Mass spectra of rhIFN-␥ were obtained on a Finnigan MAT triple stage quadrupole TSQ 7000 mass spectrometer with an atmospheric pressure ionization ion source operating in the positive ion mode. The electrospray was created at a potential difference of 4.5 kV. The nitrogen sheath gas was set to 70 p.s.i. (1 p.s.i. ϭ 6894.76 pascals). Nitrogen heated to 250°C was used as a drying gas and was introduced into the capillary region at a flow rate of 25 liters/min. The system was run in an automated LC-MS mode. The mass spectrometer was equipped with a Hewlett-Packard HPLC Series 1050 system. Protein samples were applied to the RP-HPLC column at an injection speed of 2 l/s. The HPLC method employed was the same as the RP-HPLC method described above except for acetic acid, which was substituted for trifluoroacetic acid in solvents A and B. Experiments were performed at a mass spectrometric resolution of 1000 using a step size of 0.1 atomic mass unit. External calibration of the mass scale was achieved with multiple charge ions by injection of human ␤ 2 -microglobulin (average molecular mass of 11,732.2 Da; Sigma). Mass spectra were acquired in the m/z range of 600 -1400 atomic mass units by scanning the magnetic field in 2 s and analyzed in the BIOMASS TM Deconvolution and BIOMASS TM Calculation modules of the BioWorks ICIS window (Version 8.2).
LC/ESI-MS Peptide Mapping-Tryptic digestion was performed by the gel matrix method of Rosenfeld et al. (42) with some modifications. SDS-polyacrylamide gels were stained with a fresh solution of 0.2% Coomassie Brilliant Blue R-250 in 10% acetic acid and 25% methanol and destained with 30% methanol until bands became visible. The latter were excised from the gel, washed twice with 0.5 ml of 50% acetonitrile in 0.2 M NH 4 HCO 3 (pH 8.9) for 20 min at 30°C and once with 0.5 ml of 0.2 M NH 4 HCO 3 (pH 8.9), and evaporated to a semidry state in a rotary evaporator. Gel slices were rehydrated in 20 l of enzyme solution containing 0.5 mg of modified trypsin (Promega) in 0.2 M NH 4 HCO 3 (pH 8.9). After absorption of the protease solution, aliquots of 5 l of ammonium carbonate buffer were added until the slices regained their original size and became completely immersed in the reaction solution. The slices were incubated for 4 h at 37°C, and the resulting peptides were extracted twice with 60% acetonitrile for 20 min at 30°C and once with 60% acetonitrile and 0.02% Tween 20. The extracts were combined and concentrated to 20 l in a rotary evaporator.
The peptides were introduced into the ion source of the triple stage quadrupole mass spectrometer via the RP-HPLC method described for peptide isolation. Acetic acid was used instead of trifluoroacetic acid in the mobile phase solutions. Peptide samples were injected automatically at a speed 1 l/s, and the ion source was operated in the positive ion mode. Drying nitrogen was heated to 275°C, and mass spectra were acquired by scanning the magnetic field for 3 s in the m/z range of 300 -1300 atomic mass units. Data were analyzed in the PEPMAP TM module of the BioWorks ICIS window (Version 8.2) at a mass tolerance of 1.0 atomic mass unit.
Gel Filtration Chromatography-Gel filtration was performed using an LKB Bromma HPLC system equipped with an LKB/UltroPac TSK-G3000SW column (7.5 ϫ 300 mm) according to the method described by Lauren et al. (18).
Antiviral Assay-The antiviral assay was based on the protective effect of hIFN-␥ against the cytopathic action of the vesicular stomatitis virus on the human amnionic cell line WISH as described (43).

Purified Preparations of rhIFN-␥ Undergo Proteolysis and
Dimerization over Time-rhIFN-␥ was purified from E. coli inclusion bodies by two consecutive chromatography steps. The second step included separation of refolded proteins on CM-Sepharose based on their net charge. Refolded rhIFN-␥ was usually eluted from the second column as a broad peak covering the salt concentration range of 0.4 -0.5 M NaCl (data not shown). Eluted rhIFN-␥ was pooled into three fractions, desig-  Fig. 1A shows the SDS-PAGE pattern of these three fractions immediately after purification, whereas Fig. 1B represents the same fractions after filter sterilization and storage at 4°C for 2 months. As shown in Fig. 1A (lanes 1-3), at the time of isolation, the three fractions contained apparently homogeneous rhIFN-␥ . However, 2 months later, Fraction 1 was completely converted into lower molecular mass form(s) (Fig.  1B, lane 1). rhIFN-␥ in Fraction 2 underwent partial cleavage concomitant with the appearance of higher molecular mass forms (Fig. 1B, lane 2). The electrophoretic mobility of the latter was indicative of covalent dimerization affecting both the full-length and truncated rhIFN-␥ species. Fraction 3 turned out to be unaffected during storage (Fig. 1B, lane 3). Truncation and dimerization of rhIFN-␥ in Fractions 1 and 2 were observed even upon storage at Ϫ20°C (data not shown).
Fresh rhIFN-␥ Preparations Contain Early (but Not Late) Glycation Products-We tested fresh rhIFN-␥ Fractions 1-3 for modifications with the early glycation marker fructosamine (or the Amadori product). As shown in Table I, the Amadori product was detected only in Fractions 1 and 2, but not in Fraction 3. Additionally, the content of the Amadori product in Fraction 1 was about six times higher than that in Fraction 2. The protein content in the three fractions was measured by two independent methods: Coomassie Brilliant Blue G staining (Bradford assay) and absorption at 280 nm (A 280 ). The concentrations for rhIFN-␥ in Fractions 1 and 2, determined by the Bradford assay, were five and two times lower, respectively, compared with the values obtained by A 280 (Table I). Fraction 3 gave comparable concentrations independent of the method used. Bearing in mind that Coomassie Brilliant Blue dye interacts with positively charged groups in proteins, which are also glycation targets (44 -46), we are tempted to explain the lower concentrations obtained by the Bradford assay as the result of glycation of rhIFN-␥ in Fractions 1 and 2. As glycation does not affect tryptophan residues, A 280 values remain unchanged. Fresh rhIFN-␥ Fractions 1-3 were also tested for the presence of AGEs by ELISA using CML-and imidazolonespecific monoclonal antibodies. Neither of these two compounds was detected in the fresh rhIFN-␥ preparations.
Processed rhIFN-␥ Fractions Contain AGEs-Two months later, the above three rhIFN-␥ fractions were analyzed by RP-HPLC. The eluate was monitored for both absorption at 280 nm and AGE-specific fluorescence (Ex 370 nm /Em 420 nm ). ELISA was also applied to test the peak fractions for the presence of AGEs. Our results show that the truncated protein form(s) in Fraction 1, which eluted at a retention time of 22.6 min, was neither fluorescent nor reactive to either of the anti-AGE antibodies ( Fig. 2A). However, this fraction contained some material with a low retention time (5.4 min), which was AGE-positive as determined by both fluorescence analysis and ELISA. Notably, this fraction (designated as fraction F) did not absorb at 280 nm ( Fig. 2A, lower panel) and demonstrated strong reactivity to the anti-imidazolone antibody (upper panel). Fraction 2 showed two peaks with close retention times (22.2 and 22.6 min) in the RP-HPLC chromatogram (Fig. 2B, lower panel). The earlier peak (retention time of 22.2 min) showed poor fluorescence and was not immunoreactive, and its SDS-PAGE mobility corresponded to that of full-length rhIFN-␥ (data not shown). The later (shoulder) peak (retention time of 22.6 min) was highly fluorescent and represented a mixture of dimer and truncated rhIFN-␥ species. Because truncated rhIFN-␥ species were not fluorescent ( Fig. 2A, lower panel), dimers should account for the high AGE-specific florescence in this mixed fraction. In the RP-HPLC chromatogram of Fraction 2, one can also see the same AGE-positive fraction F (retention time of 5.4 min) observed in Fraction 1. rhIFN-␥ in Fraction 3, which eluted at a retention time of 22.2 min, was ELISA-negative and non-fluorescent and was composed of full-length protein only (Fig. 2C). These results clearly demonstrate that, after 2 months of storage, glycation of rhIFN-␥ in Fractions 1 and 2 had come to its terminal stages, leading to accumulation of typical AGEs.
AGE-modified C-terminal Fragments Are Cut Off from the rhIFN-␥ Molecules-The fluorescent fraction F was collected and reloaded onto a narrow-bore RP-HPLC column. As shown in Fig.  3, in addition to the bulky early eluting material, two peaks, designated as P1 (retention time of 24.8 min) and P2 (retention time of 26.9 min), were clearly distinguished. The eluate corresponding to these two peaks was collected, concentrated, and subjected to both N-terminal sequencing and LC/ESI-MS analysis. Sequence analysis revealed sequence DDFEK for peptide P1 and sequence AIHEL for peptide P2, accounting for proteolysis in the rhIFN-␥ molecule between Arg 90 2Asp 91 and Lys 109 2Ala 110 , respectively. The absence of tryptophan in rhIFN-␥ following Arg 90 (23) explains the lack of absorption at 280 nm in fraction F (see above). The two cleavage sites were further confirmed by LC/ESI-MS analysis ( Fig. 4 and Table II). The m/z values observed in this experiment were matched against the monoisotopic m/z values expected for peptides P1 (Asp 91 -Gln 144 ) and P2 (Ala 110 -Gln 144 ). In addition, we expanded the set of calculated m/z values with those corresponding to putative modifications of the two peptides with either CML or imidazolone. We were motivated to do this by the observed reactivity of fraction F with both the anti-CML and anti-imidazolone antibodies. Thus, for peptide P1, we detected four multiple charge ion series (Fig. 4A and Table II) corresponding to the non-modified peptide (series D) or to modifications of this peptide with one carboxymethyl residue (series C) and either one (series I) or two (series II) imidazolone moieties. Peptide P2 produced only one ion series corresponding to the non-modified peptide (series A in Fig. 4B and Table II).

TABLE I Amadori products and protein content in fresh rhIFN-␥ preparations
The concentration of fresh rhIFN-␥ in Fractions 1-3 was measured by either the Bradford assay (BA) or A 280 . Following trichloroacetic acid precipitation, the concentration of rhIFN-␥ in the three fractions was adjusted to 5 mg/ml based on A 280 . Presented are the data for the content of Amadori products (AP) in 5 mg (ϳ300) nmol of rhIFN-␥. ND, not detected. Truncation of rhIFN-␥ Is Due to Cleavage at the C Terminus Adjacent to Either Lysine or Arginine Residues-To reveal additional cleavage sites in rhIFN-␥, 2-month-old Fractions 1-3 were subjected to LC/ESI-MS analysis. As expected, the most homogeneous and stable rhIFN-␥, which was in Fraction 3, yielded multiple charge ions corresponding to the full-length protein (Table III, series M). The experimentally determined mass of full-length rhIFN-␥ (16,905.4 Da) was close to that calculated for the N-terminally methionylated protein (16,907.4 Da) and deviated significantly from the mass expected for the demethionylated form (16,776.2 Da). These data indicate that the full-length rhIFN-␥ carried an N-terminal methionine and was devoid of any chemical modifications (remember that this fraction proved to be AGE-negative in all foregoing analyses). In addition to the ion series M, rhIFN-␥ in Fraction 2 produced two ion series designated as T 1 and T 2 (Tables III and IV). Ion series T 1 resulted from cleavage of rhIFN-␥ between Arg 140 and Arg 141 , whereas ion series T 2 corresponded to a cleavage site between Phe 137 and Arg 138 . Fraction 1 was composed of truncated rhIFN-␥ species only. This fraction contained two shorter rhIFN-␥ derivatives, designated as T 3 and T 4 (Tables III and IV), indicative of proteolysis at Met 135 2Leu 136 and Lys 131 2Arg 132 , respectively. It is noteworthy that five of a total of six cleavage sites in rhIFN-␥ were adjacent to either Lys or Arg residues, known to be preferable targets for glycation (34, 44 -47). The observed molecular masses for all truncated rhIFN-␥ species were several Da higher than the theoretically predicted masses (Table IV), which may reflect some unknown chemical rearrangement taking place at the sites of proteolysis.

Covalent Dimerization of rhIFN-␥ Results from Lateral Rather than Terminal Cross-linking of the Polypeptide
Chains-To shed light on the nature of the covalent rhIFN-␥ dimers, we performed peptide mapping experiments. The dimer band (Fig. 1B, lane 2) was excised from stained SDSpolyacrylamide gels and in-gel digested with trypsin. In parallel, the rhIFN-␥ monomer fraction (Fig. 1B, lane 3) was subjected to the same procedure and used as a control. As shown in Fig. 5 and Tables V and VI, most of the peaks in both reconstructed ion current (RIC) chromatograms could be assigned to certain tryptic peptides of rhIFN-␥. ⌻he peptide map covered ϳ90% of the hIFN-␥ sequence. In the RIC chromatogram of the dimer sample (Fig. 5B), we failed to detect ions corresponding to the internal sequence at positions 15-35 (i.e. peak 9 in Fig.  5A). Upon comparison of the two RIC chromatograms, one can see also that ions corresponding to internal peptides 70 -81 (peak 4) and 96 -108 (peak 6) were highly underrepresented in the dimer sample (Fig. 5). Therefore, we assumed that covalent rhIFN-␥ dimers were formed predominantly by internal (lateral) joining of the polypeptide chains. Because peptides with molecular masses exceeding 3 kDa were inefficiently extracted from the gel matrix, most of the putative cross-linked peptides were expected to have been lost during the extraction procedure. However, in the dimer RIC chromatogram, we detected two ions (peak L in Fig. 5B and Table VI) whose m/z values matched well those expected for the terminal cross-linked peptide 1-7/133-137 reported by Lauren et al. (18). Obviously, there is no particular mode of cross-linking in rhIFN-␥, and its covalent dimers should therefore be regarded as a heterogeneous population of molecules with variable rather than defined chemical structure.
Proteolysis of rhIFN-␥ Lowers Its Antiviral Activity, whereas Dimerization Completely Abolishes It-rhIFN-␥ Fractions 1-3 were tested for antiviral activity immediately after purification and after 2 months of storage at 4°C. The results presented in Table VII show that, at the time of isolation, there was no significant difference in the antiviral activity of the three fractions. Two months later, the most heterogeneous fraction (Fraction 2) was subjected to gel filtration (data not shown) to separate the rhIFN-␥ monomers from the dimers. Thus, the antiviral assay was carried out with four samples: total Fractions 1 and 3 together with the monomers and dimers isolated  Fig. 4. Uppercase letters denote rhIFN-␥ C-terminal peptides corresponding to peaks P1 (D, C, I, and II) and P2 (A) in Fig. 3, where D indicates non-modified peptide P1, C is CML-modified peptide P1, I is imidazolone-modified peptide P1, II is imidazolone-double-modified peptide P1, and A is non-modified peptide P2. NO, not observed; ND, not determined (out of the m/z scan range).

TABLE III m/z values for full-length rhIFN-␥ and its truncated forms
Shown are the observed (Obs.) and expected (Exp.) m/z values obtained in one of a total of five experiments performed to determine the average mass of the different rhIFN-␥ forms (see Table IV). Uppercase letters indicate full-length rhIFN-␥ (M) and truncated species (T; subscript numbers stand for different truncated species). ND, not determined (out of the m/z scan range); NO, not observed.   from Fraction 2. The results in Table VII demonstrate a 5-10fold drop in the antiviral activity of the monomers and total Fraction 1. The covalent rhIFN-␥ dimers were completely in-active, whereas Fraction 3, representing the most stable rhIFN-␥ fraction, maintained its antiviral activity during storage. DISCUSSION We have shown recently that nonenzymatic glycosylation (glycation) takes place in E. coli under physiological conditions (20). In the present study, rhIFN-␥ isolated from E. coli was analyzed for modifications with either early or late glycation products. Our results show that, at the time of isolation, the rhIFN-␥ preparations contained early glycation products, but not AGEs. This finding implies that glycation of rhIFN-␥ was initiated in the producing cells (note that we did not use any carbonyl compounds throughout the purification procedure). We observed that only rhIFN-␥ fractions containing early glycation products underwent proteolysis and dimerization during storage.
We identified in 2-month-old rhIFN-␥ preparations the markers for advanced glycation, imidazolone and CML, which were not found in fresh samples. These AGEs were detected mostly in peptides released from the rhIFN-␥ C terminus, but not in the core protein fraction. Human IFN-␥ is a basic protein (pI 9.7) containing 20 lysine and 8 arginine residues (23). X-ray crystallography revealed that the C terminus of rhIFN-␥ following Pro 123 is highly disordered and flexible (47). There are 8 basic amino acids in this region, which is a predisposition for both enzymatic and chemical rearrangements. The LC/ESI-MS analysis in this study proved that truncation of rhIFN-␥ was a result of C-terminal cleavage. Mass spectral analyses and Nterminal sequencing revealed four major cleavage sites after Pro 123 and two minor ones distal to the C terminus (Arg 90 2Asp 91 and Lys 109 2Ala 110 ). All cleavage sites except for one were found in the vicinity of either Lys or Arg residues, which are known to be efficient targets for glycation (34, 44 -46). Taking into consideration that glycation leads to sitespecific (at Lys and Arg) or random peptide bond cleavage in proteins (31,32,48), we assume that the proteolysis of rhIFN-␥ observed during storage is a consequence of glycation.
The antiviral assay demonstrated that the early glycation adducts did not interfere with antiviral activity, whereas proteolysis of rhIFN-␥ led to a partial loss of antiviral activity. It should be mentioned that there is a substantial disagreement in the literature dedicated to the functional significance of the hIFN-␥ C terminus. The conclusions vary from indispensable (49,50) to not important (51,52). However, the role of the sequence between Asn 79 and Phe 93 , called the "nuclear localization sequence," in the internalization of the hIFN-␥⅐receptor complex has been well established (53). We found that one of the two minor cleavage sites (Arg 90 2Asp 91 ) resides within this sequence and therefore might account for the observed decrease in the antiviral activity of the truncated rhIFN-␥ fractions.
The active form of rhIFN-␥ is a noncovalent homodimer composed of two antiparallel monomer subunits (54). In this and other studies (18,20), however, a covalent dimer form of   rhIFN-␥ was found. This observation raised the question of how covalent dimers are formed in the cysteineless protein and whether they are biologically active. Lauren et al. (18) reported that the covalent rhIFN-␥ dimers are the result of head-to-tail cross-linking involving tryptic peptides 1-7 and 133-137. We also detected two m/z values corresponding to this peptide in the dimer sample. However, most prominent in this sample was the lack or sharp decrease in the content of some internal tryptic peptides, which indicated that the covalent dimers were formed by lateral rather than terminal cross-linking of the polypeptide chains. We found also that the dimers were highly fluorescent under conditions typical for AGE-modified proteins.
Because some fluorescent AGEs such as pentosidine are crosslinkers (55), we assume that, in addition to the glycation of the C terminus (causing proteolysis), the glycation of internal amino acids leads to covalent dimerization of rhIFN-␥. The antiviral assay clearly demonstrated that the covalent dimers were devoid of antiviral activity. E. coli is the most widely used bacterial host for manufacturing of recombinant proteins. Our previous (20) and present studies provide substantial evidence that glycation is an intrinsic feature of bacterial metabolism. It affects both homologous as well as foreign proteins produced in E. coli. Consequently, recombinant proteins synthesized in this bacterium and used as drugs in medicine may suffer from instability or decrease in or loss of functionality as well as immunogenicity. Regulatory authorities should therefore be aware of all this and include in their demands analyses for glycation of recombinant proteins designed for medical use. Recently, Wiame et al. (56) reported the identification of a deglycating enzyme (amadoriase) in E. coli. In addition to the suggested role in fructosamine utilization in human intestine, this enzyme might be involved in the regulation of the glycation activity in E. coli. Presumably, it could be used also in recombinant DNA technology for stripping off glycation adducts to prepare stable and safe recombinant proteins.