Bacterial expression and secretion of various single-chain Fv genes encoding proteins specific for a Salmonella serotype B O-antigen.

Active single-chain Fv molecules encoded by synthetic genes have been expressed and secreted to the periplasm of Escherichia coli using the ompA secretory signal. Four different constructs were developed to investigate the effects of peptide linker design and VL-VH orientation on expression, secretion, and binding to a Salmonella O-polysaccharide antigen. Peptide linker sequences derived from the elbow regions of the Fab molecule were used alone or in combination with the flexible (GGGGS)2 sequence. VL and VH domain order in the single chain molecules had a profound effect on the level of secretion but hardly influenced total expression levels, which were approximately 50 mg/liter, chiefly in the form of inclusion bodies. With VL in the NH2-terminal position, the amount of secreted product obtained was 2.4 mg/liter, but when VH occupied this position the yield was less than 5% of this value. Enzyme immunoassays of the four products showed domain order and linker sequence affected antigen binding by less than an order of magnitude. Attempts to express active Fv from dicistronic DNA were unsuccessful, but active Fv was obtained from single-chain Fv by enzymic cleavage at a site in the elbow linker peptide. The thermodynamic binding parameters of intact and cleaved single-chain Fvs determined by titration microcalorimetry were similar to those of bacterially produced Fab and mouse IgG.

digestion of the intact antibody are rarely successful.
Recently, the production of antigen-binding proteins by recombinant DNA technology has opened up new avenues to build novel antibody-based molecules useful for the treatment of human diseases (1). The most interesting of such molecules produced thus far are single-chain Fvs (sFvs) in which the COOH terminus of one variable domain is joined to the NH2 terminus of the second variable domain by a linker of appropriate length and flexibility. Linking the domains in this way results in a stable, active molecule and overcomes possible problems with VH-VL dissociation at low concentration (2). Following the first reports of the production of these molecules with either VL or VH in the amino-terminal position (3,4), sFvs blocking rhinovirus infection ( 5 ) and sFvs linked to a toxin to form immunotoxins directed against ovarian cancer cells (6) have been described. Single-chain Fvs have also been useful in studying structure-function relationships in the VL and VH domains (2, 7) and in expressing antibody variable domains as fusion proteins on bacteriophage surfaces for the purpose of screening for antigen binding (8).
Typically, linkers designed from simple amino acid sequences such as (GGGGS), have been used, but in one report ( 5 ) linkers incorporating an interdomain sequence from the VL-CL junction in the Fab molecule were employed with some success. However, many aspects of linker design including "natural" versus "synthetic" have yet to be examined. In this report, we describe the production of several sFv constructs and the attempted production of Fv in the periplasm of Escherichia coli by expressing synthetic genes encoding these molecules. The effects of linker sequence and the orientation of the VH and VL domains on expression levels, secretion, and antigen binding properties were examined. Two types of linkers were used (i) sequences derived solely from the elbow regions of the corresponding Fd or light chains and (ii) sequences incorporating the previously used GGGGS sequence (3,4). The antibody we selected was Se155-4, specific for a trisaccharide epitope of Salmonella serogroup B O-polysaccharide (9). Previously, Se155-4 Fab has been produced in E. coli using synthetic dicistronic DNA and has been shown to be as active as mouse Fab in antigen binding and competitive immunoassays (10,11).

MATERIALS AND METHODS
Enzymes, Oligomers, and General Techniques-Restriction enland Biolabs, GIBCO-BRL and Boehringer Mannheim. Deoxyribo-zymes and DNA modifying enzymes were purchased from New Engnucleotides were synthesized using an automatic DNA synthesizer model 380A (Applied Biosystems Inc.). Plasmid pBtac2 was purchased from Boehringer Mannheim. Goat anti-mouse X antibody conjugated to alkaline phosphatase and anti-mouse X/biotin conjugates were purchased from Caltag. Other reagents used in EIA were purchased from Kirkegaard and Perry Laboratories, GIBCO-BRL, or Bio-Rad. Mouse Fab, E. coli-produced Fab and Se155-4 IgG were obtained by affinity chromatography as described previously (11).
Construction of F u Expression Plusmid-The plasmid pSal VL was derived from pSal-L (10) and was joined at its 3' Sac1 terminus with the VH gene at its 5'-EcoRI site by a synthetic 24-nucleotide spacer duplex (11). This spacer duplex contained a ribosomal binding site eight nucleotides from the ATG start codon of the ompA signal peptide preceding the VH gene. The resultant gene construct is shown in Fig. 1A. Both variable domains are preceded by the ompA signal peptide allowing their secretion into the periplasm of E. coli. For expression studies, the EcoRI-Hind111 cassette containing the cistronic genes was cloned into the corresponding sites of the pBtac2 plasmid and designated ptFv.
Construction of sFu Expression Plasmids-Four sFv genes in both orientations, VL-VH or VH-VL, carrying both elbow (el) and flexible (fl) linkers were constructed from plasmids carrying the V L , VH or the dicistronic Fab genes (10,11). As in the Fv construct, suitable restriction fragments carrying the desired sequences were bridged by synthetic oligonucleotides. The organization of all four genes and the linker sequences incorporated into the four designs are shown in Fig.  1B. The 5'-end of each gene was preceded by the ompA secretory signal. The four constructs were cloned into the pBtac2 vector for the purpose of expression. The recombinant pBtac2-derived plasmids were designated ptsFvLH(el), ptsFvLH(fl), ptsFvHL(el), and pts-FvHL(fl).
Expression of Fu and sFu-The plasmid ptFv harboring the cistronic Fv gene was transformed into competent E. coli TG1 cells and grown in shake flasks at 30 'C in M-9 minimal medium supplemented with 0.4% casamino acids and 100 pg/ml of ampicillin. At 48 h, the cultures were induced with supplementary nutrients (12 g of tryptone, 24 g of yeast extract, 4 ml of glycerol/liter) and 2 mM isopropylthiogalactopyranoside. The cultures were harvested 24 h later and periplasmic extracts prepared as described earlier (11). Periplasmic extracts were checked for activity by indirect EIA and also analyzed by SDS-PAGE and Western blotting. SDS-PAGE (12.5% acrylamide) was performed using the buffer system described by Laemmli (12), and gels were stained with Coomassie Brilliant Blue. In Western blotting, proteins were transferred to Immobilon-P (Millipore) with 25 mM Tris-glycine, pH 8.2, containing 15% methanol as the transfer buffer. The detection reagent was goat anti-mouse X/alkaline phosphatase. After extensive dialysis against 50 mM Tris-HC1, pH 8.0, buffer containing 0.15 M NaC1, the extracts were applied to an antigen-based affinity column as described earlier (11).
Plasmids harboring the sFv genes were transformed into E. coli TG1 cells. Cultures were grown as described above, and the single chain products were isolated from periplasmic extracts by affinity chromatography in a similar manner.
Quantitation of SFU Expression and Secretion-Cultures (100 ml) harboring ptsFvLH(el), ptsFvLH(fl), ptsFvHL(el), or ptsFvHL(fl) were grown and induced as described above. The periplasmic extracts were prepared from 50 ml of each culture. The two supernatants obtained by sucrose and shock treatment, respectively, were combined (total volume, 5 ml). The cells from the remaining 50 ml of each culture were suspended in 5 ml of water. Dilution series of cell suspensions and periplasmic fractions were analysed by SDS-PAGE/ Western blotting and the amounts of single-chain product estimated by comparison with a dilution series of purified product.
sFu Cleauage-Purified VL-VH(el) sFv was incubated at 30 "C in PBS for up to 7 days at an antibody concentration of 300 pg/ml. To test for inhibition of cleavage by protease inhibitors, incubations were set up with each of the following inhibitors: 1 mM phenylmethylsulfonyl fluoride, 1 mM EDTA, aprotinin (10 pg/ml), E-64 (1 mg/ml), and pepstatin (0.7 pg/ml). Aliquots were taken at appropriate intervals for analysis by SDS-PAGE, Western blotting, and EIA.
Enzyme Immunoassay-Indirect EIA was carried out using microtiter plates coated with BSA-0-polysaccharide conjugate2 at a concentration of 10 pg/ml. Serial dilutions of antibody fragments were added to the plates, and bound materials were detected with an antimouse h/biotin conjugate and streptavidin-horseradish peroxidase using 3,3',5,5'-tetrarnethy1benzidine/H2O2 as substrate. Indirect EIA was also carried out with Salmonella essen lipopolysaccharide-coated plates, as previously described (lo), using a goat anti-mouse X chain antibody alkaline phosphatase conjugate.
Competition measurements were conducted using a direct EIA system (13). Protein was absorbed to plates, and the procedure was modified slightly in that the BSA blocking step was omitted. All incubations were done at 37 "C, and all buffers contained 0.05% E. Altman, unpublished results.
Tween 20. The binding of biotinylated 0-polysaccharide to antibody fragments in the wells was inhibited by various concentrations of 0polysaccharide (14). In addition, a competitive indirect EIA was developed for purposes of comparing the four sFvs and E. coliproduced Fab. In this assay microtiter plates were coated with a BSA-0-polysaccharide conjugate (as above). Antibody fragments were used at concentrations giving 70% of the maximum signal in indirect EIA with BSA-0-polysaccharide conjugate. The fragments were mixed with various concentrations of 0-polysaccharide prior to addition to the wells. Bound antibody fragments were detected with an antimouse X/biotin conjugate and streptavidin/horseradish peroxidase using TMB/H202 as the substrate.
Titration Microcalorimetry-Microcalorimetry was carried out on an OMEGA titration microcalorimeter from Microcal Inc. (Northampton, MA). This instrument has been described in detail by Wiseman et al. (15). Approximately 6-7 mg of protein in 1.3 ml of PBS, pH 7.0, was placed in the calorimeter cell and was stirred at 400 rpm at 25 "C. The sample was titrated with a 2 mM solution of the synthetic trisaccharide methyl 2-O-(a-D-galactopyranosyl)-3-O-(-3,6-dideoxy-xylo-ol-~-hexopyranosyl)-a-~-mannopyranoside (the Salmonella serogroup B epitope) in PBS using 20 aliquots of 5 pl added at 3-min intervals from a 100-p1 syringe. The reference cell was filled with water, and the instrument was calibrated by standard electrical pulses. The non-linear least squares analysis was performed as described elsewhere (9).

RESULTS
Expression of F u Genes-The previously successful strategy of secreting the X and Fd chains of Se155-4 Fab to the E. coli periplasm (10,11) where correctly folded and active molecules were formed was applied to the VL and VH domains of Fv.
Significant amounts of a protein of M, 13,000 were detected in the periplasmic fractions of cultures harboring the ptFv plasmid by SDS-PAGE/Western blotting with anti-X chain antibody, indicating VL expression. However, no Fv product was detectable by affinity chromatography suggesting problems with VH expression, VH secretion, or a lack of VH-VL association in the periplasm.
Assembly of sFu Genes-Four sFv genes were constructed, two with VL in the amino-terminal position and two with VH in the amino-terminal position (Fig. 1B). Each orientation was constructed using flexible and elbow linkers. The linker designs incorporated sequences from the VL-CL elbow, the VH-CHI elbow, and the flexible sequence (GGGGS), (Fig. 1B).
An additional feature of the VL-VH elbow linker was the incorporation of a chemical cleavage site. This linker was composed of 12 amino acids from the NH, terminus of the CL domain (ending with SerlZ5) followed by Asn-Gly, a dipeptide sequence which is sensitive to hydroxylamine cleavage. The objective was to produce a "pro-protein'' that could be cleaved in the folded state. The VH-VL elbow linker version of the molecule did not contain the hydroxylamine cleavage site and was composed solely of a 16 amino acid sequence from the CHI domain (Fig. 1B).
The sequence (GGGGS), was used to create flexible linker versions of each orientation (Fig. 1B). The location of suitable restriction sites dictated the incorporation of six constant domain amino acids into the VL-VH design and 5 constant domain residues into the VH-VL design. Hydroxylamine-cleavage sites were not included in either construct.
Expression and Secretion Levels-It was observed that the levels of secreted product with the VL-VH(el) construct were approximately 20-fold higher when the gene was under the control of the tac promoter, as compared to the lac promoter (data not shown). The effect of domain orientation and linker sequence on expression and secretion was studied using the tac series of plasmids. Periplasmic extracts from cells harboring the VL-VH constructs were shown by SDS-PAGE/Western blotting to contain significantly more single-chain product than periplasmic extracts from cells harboring the VH-VL   to an equal volume of SDS sample buffer containing 5% 8-mercaptoethanol and boiled for 5 min prior to application to the gels. plasmids (Fig. 2). Only the sFv regions of the Western blots are shown in Fig. 2, but a complete and representative Western blot of a periplasmic extract is shown in lane 6 of Fig. 3.
Levels of secreted product for the V,,-VH(el), VI,-VH(fl), VH-V,,(eI), and V~-Vr<(fl) constructs were estimated to be 2400, 2400, 80, and 160 pg/liter of culture, respectively. With all constructs, a relatively small fraction of the total product was secreted. The levels of expression were similar in each instance, approximately 50 mg/liter.
Single-chain F v Purification-The products of the sFv genes were isolated from periplasmic extracts in a single step by affinity chromatography (Fig. 3). This also provided a more accurate means of confirming the secretion levels estimated by SDS-PAGE/Western blotting (Fig. 2). The values obtained for each construct by the two procedures were in excellent agreement. The affinity-purified products gave single Coomassie-staining bands on both non-reducing and re- ducing SDS-PAGE gels (Fig. 3). Two minor, higher molecular weight species were detected by Western blotting (Fig. 3). These may be the products that are cross-linked through the free cysteine at position 94 of the light chain, since these bands disappeared on reduction. NH,-terminal amino acid sequence analysis of the VI,-VH(el) product showed correct processing of the leader sequence and confirmed the NH,terminal sequence up to residue 20.
Single-chain F u Cleavage-Use of the Asn-Gly site for the specific cleavage of the linker region by hydroxylamine was made unnecessary by the observation that there was spontaneous cleavage of the VI.-VF,(el) linker upon prolonged storage a t 4 "C. Incubation of the sFv at 30 "C gave virtually complete cleavage after 4 days (Fig. 3). Complete binding to the antigen column of the cleaved product a t a protein concentration of 4.5 mg/ml indicated that it existed in a fully active form under these conditions. NH2-terminal amino acid sequence analysis indicated that peptide cleavage had occurred between Ser'"" mM EDTA suggesting that the hydrolysis was mediated by a metalloprotease. Hydrolysis was unaffected by the other pro-and Ser125 . The cleavage could be completely inhibited by 1 tease inhibitors that were tested. Since Fv itself could not be obtained (see above), this cleaved sFv was used instead for immunological and microcalorimetric comparisons with intact sFv.
Indirect EZA-The antigen-binding activities of the sFv products were compared by indirect EIA with Fab produced in E. coli (Fig. 4). Similar patterns were obtained with lipopolysaccharide or BSA-0-polysaccharide as the antigen, but the results were more consistent and differences more pronounced with the BSA conjugate and the anti-mouse X-biotin detection system. If the amount of antibody required to give 50% maximum activity is used as an indication of affinity, the assay showed that the sFvs and the Fab were generally quite similar on a mass basis. The V H -V L ( f l ) construct consistently displayed the highest binding, which was approximately 10-fold higher than the least active VL-VH(fl) construct. The activity profiles of intact and cleaved Vt-Vdel) were similar.
Competitive EZA-It was observed that while a blocking step with bovine serum albumin did not significantly affect the signal obtained with mouse Fab in direct EIA, it reduced that obtained with E. coli-produced Fab and sFv by over 90%.
Fortunately, a blocking step was found not to be necessary in this assay. The underlying cause of this difference in response was not investigated.
Bacterially produced Se155-4 Fab was compared with the VL-VH(el) sFv using this assay (Fig. 5). The concentration of 0-polysaccharide antigen required to inhibit binding of biotinlabeled 0-polysaccharide was identical for both preparations, indicating similar affinity for antigen.
The competitive indirect assay was the method of choice for comparing all four sFv constructs because the VH-VL constructs could not be detected in the direct assay. In this system, affinity for antigen is inversely proportional to inhibitor concentration at 50% inhibition (13). Results obtained with this assay (Fig. 6) were in good agreement with those described above for indirect EIA which show that the VH-VL(fl) has a 10-fold high affinity compared to VL

Comparison of the bindingproperties of IgG, E. coli-produced Fab, sFv, and cleaved sFu by titration microcalorimetm
M' x I @ kJ mol" J mot' K" to mouse IgG, E. coli-produced Fab, single-chain VL-VH(el), and cleaved single-chain VL-VH(el) were determined by microcalorimetry (Table I). Association constants were also determined from the same data and showed little significant difference for the four species. There was a corresponding agreement in the enthalpy of association. The results for the Single-chain Fus cleaved Fv were less accurate than other measurements since the experiment was performed at a somewhat lower protein concentration (Table I). The observed association constants in the order of lo6 M" lay in the range typical of antibodycarbohydrate binding. As reported previously (9), the binding of the trisaccharide epitope is mostly enthalpy driven but with a significant contribution from the entropy term (-TASo).

DISCUSSION
Single-chain antibody fragments are one of the most novel developments in the rapidly evolving area of antibody engineering (1,16). Although bacterial leader sequences have been used to target Fab (17) and Fv (18) products to the periplasm, most sFvs so far constructed were expressed as inclusion bodies in E. coli (3,4). While inclusion body isolation followed by denaturation and refolding steps can give acceptable yields of active sFv, the procedure is tedious and plagued by solubility problems. Our present study reports on four aspects of sFv production: (i) the targeting of sFv to the E. coli periplasm, (ii) the effect of domain orientation on secretion, (iii) the feasibility of using linker sequences derived solely or partially from the elbow regions of the corresponding Fab molecule, and (iv) the effect of linker sequence and domain orientation on antigen binding properties. We also observed the serendipitous proteolytic cleavage of our sFv within the peptide linker sequence, thereby yielding an active Fv.
A major observation reported in the present study was that the uniformly high expression levels of products from all four constructs made with the Se155-4 V L and V H domains did not translate into correspondingly high yields of secreted protein.
Less than 5% of the expressed sFv was secreted, and this secretion was significantly dependent on domain orientation in the construct. Nevertheless, there are obvious advantages in avoiding the harsh denaturing conditions required for purification of sFvs from inclusion bodies, and the yields of secreted products, while acceptable for many purposes, could probably be improved by changing factors such as culture conditions, bacterial strain, or leader sequence.
With a VH-VL orientation, the secreted yield was 20-fold less than that obtained with a VL-VH orientation. Although some contribution to this effect by minor differences in linker sequence is possible, the major factor of domain orientation is thought to be primarily responsible. While there are many examples of heterologous protein secretion to the E. coli periplasm, the molecular mechanisms involved in the process are not completely understood. Attachment of a signal sequence does not, in itself, always lead to secretion (19), implying that information contained in the mature protein sequence is also important (20). It has been suggested that entry to the prokaryotic secretory pathway is contingent on a protein assuming a soluble and suitably folded conformation (21). Domain orientation might significantly affect the degree to which sFv molecules meet such a requirement. Some findings relating to heavy chain secretion in mammalian cells may also ap,ply in bacterial systems. Polypeptide chain-binding proteins implicated in protein folding can be grouped into families that span the prokaryote/eukaryote boundary (31). In mammalian cells, secretion of immunoglobulin molecules is mediated by the light chain and is impeded when light chain is limiting (22). Also, while free light chain is readily secreted (23, 24) free heavy chain is generally not secreted (25,26,27,28). In sFv molecules, VH at the NH2-terminal position may limit secretion because of its folding taking place in the absence of VL, resulting in inclusion body formation.
In the present study, the use of elbow region sequences as linkers in sFvs resulted in molecules that were indistinguish-able from the corresponding Fab molecule with respect to antigen binding thermodynamics as measured by EIA (Figs. 4-6) and titration microcalorimetry (Table I). This contrasts with the report by Condra et al. (5) in which an sFv containing an elbow linker had less than 10% of the activity of the Fab molecule. However, these authors presented evidence suggesting that the elbow sequence they used was too short. If suitable restriction sites are present, the use of elbow sequences can provide a straight-forward means of generating sFv genes from the Fab genes. In addition, a natural linker may have lower immunogenicity for in uiuo applications. Here we also incorporated a chemical cleavage site in one linker sequence so that the sFv product could serve as a "pro-protein'' for the Fv product. It was found that chemical cleavage was unnecessary, the sFv being successfully cleaved by an endogenous enzyme to yield a fully functional Fv. Although the enzymic nature of the cleavage was established, the origin of the enzyme and its fortuitous copurification with the sFv were not investigated. This enzymic cleavage was particularly useful for characterization purposes as we were unable to isolate an Fv molecule by expression of dicistronic DNA constructs, possibly because of poor VH expression or secretion. Field et al. (29) found that VL was expressed far better than VH for an anti-lysozyme F,, also suggesting unfavorable VH properties.
The EIA data using both indirect and direct assay formats suggested that the various constructs had comparable binding constants with the 50% inhibition points all falling within an order of magnitude. For the Se155-4 constructs, linker attachment through Gly as in the VL-VH(el) sFv gave a little better affinity than attachment through Ser as in the VL-VH(fl), and the v~-VL(fl) was better than either of these. Longer range effects of other sequence differences in the linkers may also be important. These differences may be related to the effects that changes in the NH2-terminal residues of the VH can have on affinity for antigen, as shown with antibodies and sFv specific for digoxin (3). The effects may be explained in terms of a H-bond network connecting the NHz-terminal residues to residues in the binding site (30).
The similarity of Fab and sFv binding constants was confirmed by titration microcalorimetry for sFv VL-VH(e1). Where quantities of purified protein permit, titration microcalorimetry provides an accurate and powerful method for the elaboration of a complete thermodynamic description of ligandprotein interaction. In the present study, the binding thermodynamics for the trisaccharide hapten to mouse IgG and the three cloned products E. coli Fab, single-chain VL-VH(el), and cleaved single-chain VL-VH(el), were shown to be indistinguishable. The favorable contribution of entropy to the free energy of binding at 25 "C, observed with this mouse IgG, and also consistently observed for the bacterially expressed fragments, is an unusual trait in protein-carbohydrate interactions. It is interesting to note that there is no significant change in the thermodynamic functions between the sFv and the cleaved sFv. Presumably, the concentration of the protein in the microcalorimeter (approximately 50 p M ) is high enough that there is a very low concentration of the dissociated form of the Fv. Alternatively the binding of the hapten might increase the association of the VL and VH chains (2).
In conclusion, the experiments reported here show that successful design of genes for sFv products with good affinity and secretion levels will involve a balance of two opposing factors. These are the benefits of placing the VL domain in the NH2-terminal position to improve secretion, and the negative effects that attachment of the linker sequence to the NHz-terminal residue of the VH domain can have on the affinity of the sFv. Hence, for any parent antibody, optimal design of the sFv for good affinity and secretion level cannot be done to a formula, but instead will require evaluation of the two factors by experiments of the kind we have conducted with Se155-4.