Systematic screening of antibody and antibody fragment production in the cytoplasm of E. coli

Antibodies contain a large number of disulfide bonds in their native state e.g. an IgG₁ contains 19 disulfide bonds. However, they generally have a regular pattern with one intra-molecular disulfide per domain plus inter-molecular disulfide bonds which link the heavy and light chains together and the two heavy chains together. Exceptions to this pattern exist, for example in IgA and IgE there is an additional disulfide bond in the first constant domain of the heavy chain (Fig. 1). The low number of disulfide bonds suggests that it should be possible to make natively folded scFv (two disulfides) and Fab antibody fragments (five disulfides, except IgA and IgE which have six disulfides) using CyDisCo in the cytoplasm of E. coli. These are especially attractive targets for production in the cytoplasm as they lack the consensus N-glycosylation sites found in the Fc region of antibodies.

We therefore decided to systematically screen the production of scFv and Fab antibody-fragments in the cytoplasm of E.coli using the CyDisCo system. Eleven antibodies were chosen for this screen, three human IgG₁ (Humira; PDB structures 3WD5, 4NYL [12, 13]; Maa48 [14] and K2 [15]), two humanized IgG₁ (Avastin; PDB structures 1BJ1, 1CZ8 [16, 17] and Herceptin; PDB structures 1FVC, 1N8Z, 4HKZ [18–20]), a mouse IgG₁ (3211; An anti-BNP antibody, Veijola, Vuolteenaho and Takkinen, unpublished observations), a mouse IgG₂ (PDB structure 1IGT [21]), a humanized IgG₄ (Tysabri; PDB structure 4IRZ [22]), a human IgA₁ (PDB structure: 3M8O [23]), a human IgE (PDB structure: 2R56 [24]) and a human IgM (PDB structure: 1QLR [25]). Each of the 11 scFv and Fab fragments derived from these antibodies were expressed from otherwise identical vectors and expressed and purified under identical conditions i.e. not optimized for individual proteins, such that antibody specific differences could be observed.

Previously it has been reported that some antibody fragments can be expressed in ΔtrxB/Δgor strains of E. coli (for examples [26, 27]) in which the reducing pathways have been removed, but which have no active pathways for disulfide bond formation [28]. This suggests that some antibody fragments may require little or no assistance from catalysts of disulfide bond formation to reach a soluble state. To examine this possible effect all 22 antibody constructs were expressed with and without CyDisCo components in 24 deep well plates using the KEIO collection parental K12 E. coli strain. This strain has the cytoplasmic disulfide bond reducing pathways intact. All expression tests were conducted at least in quadruplicate.

Examination of the soluble lysates by SDS-PAGE followed by Coomassie staining allowed the direct visualization of antibody fragment production in high yields for many of the constructs with CyDisCo present (data not shown). Since densitometric analysis of yields from lysates is prone to error e.g. due to co-migrating E.coli proteins, the 22 constructs expressed in the presence and absence of CyDisCo components were purified by IMAC from 3 mls of culture grown in 24 deep well plates. Since disulfide bond isomerization can occur in SDS if a free thiol group is present, the purified proteins were treated with N-ethylmalemide (NEM) to block free thiols before being analysed by reducing and non-reducing SDS–PAGE. The results (Fig. 2; Table 1) indicate that 10 out of the 11 scFv and 10 out of the 11 Fab were expressed in sufficient yield to be visible in Coomassie stained gels (limit of detection circa 3 mg/L yield) when CyDisCo components were present. In contrast only two scFv, Herceptin and Tysabri, showed high-level CyDisCo independent production and no Fab fragments were purified in the absence of CyDisCo.

Antibody name	Type	scFv (mg/L)	Fab (mg/L)
Humira	IgG1 human	33	5
Maa48	IgG1 human	4	43
K2	IgG1 human	11	26
Avastin	IgG1 humanized	–	3
Herceptin	IgG1 humanized	271	50
Tysabri	IgG4 humanized	139	25
3211	IgG1 mouse	19	9
Mab123	IgG2 mouse	6	42
3M8O	IgA1 human	86	17
2R56	IgE human	32	–
1QLR	IgM human	26	9

Yields were determined by densitometric analysis using yields of purified scFv and Fab from shake flasks as calibration controls

All of the scFv produced were monomeric, with the exception of a very faint band of disulfide linked dimer for the Herceptin scFv present in some replicates. In contrast the Fab were predominantly a disulfide linked dimer of the heavy and light chains, though some monomeric light and heavy chains i.e. not disulfide linked, could be observed e.g. for Herceptin and Tysabri. When small amounts of monomeric purified heavy and light chains were observed in the non-reducing gel analysis the apparent ratio was always 1:1 suggesting that dimers of the heavy and light chain were purified, but that some of these lacked the inter-chain disulfide bond and so ran as monomers on non-reducing SDS–PAGE.

Shake flask scale production, purification and analysis of antibody fragments

Small scale production in 24 deep well plates (DWP) allowed preliminary screening of the production of antibody and antibody fragments, but it did not produce sufficient protein for more detailed analysis. To examine in more detail the proteins produced using CyDisCo five scFv (Herceptin and Tysabri which showed CyDisCo independence, and 3211, 3M8O and 2R56 which showed CyDisCo dependence for production) and four Fab (Maa48, 3211, 3M8O and 1QLR) were chosen for production in shake flasks. Based on the results from DWP expression, Herceptin and Tysabri scFv were produced with and without CyDisCo, while the other seven antibody fragments were produced only with CyDisCo present to catalyse native disulfide bond formation. The proteins were purified by IMAC and the quality analysed by SDS–PAGE. Similar patterns were observed as observed in DWP expression (see Fig. 3a as example). The purified proteins were quantified using absorbance at 280 nm (Table 2). While the yields of scFv were in line with estimates from SDS–PAGE of lysates, the yields of the Fab fragments were on average two-fold lower than expected, suggesting that the standard IMAC protocol or the placement of the hexahistidine tag at the end of the heavy chain was non-optimal. The yields in mg/L obtained for the nine antibody fragments tested in shake flask scale were on average 103 % of those obtained in DWP (compare Table 1 with Table 2) demonstrating the ease of going from screening scale to a scale suitable for structural or functional studies using CyDisCo in EnPresso B media.

Fragment type	Antibody name	Purified yield
scFv	Herceptin (human IgG1)	251 mg/L (with CyDisCo)
		167 mg/L (without CyDisCo)
	Tysabri (human IgG4)	169 mg/L (with CyDisCo)
		141 mg/L (without CyDisCo)
	3211 (mouse IgG1)	11 mg/L
	3M8O (human IgA1)	68 mg/L
	2R56 (human IgE)	32 mg/L
Fab	Maa48 (human IgG1)	42 mg/L
	3211 (mouse IgG1)	13 mg/L
	3M8O (human IgA1)	19 mg/L
	1QLR (human IgM)	11 mg/L

Yields were determined by A280 measurements. Extinction coefficients used were calculated using ProtParam and assuming all of the cysteines present are in disulfide bonds. Except where stated all fragments were made in the presence of CyDisCo components

To confirm that the proteins obtained were correctly folded far UV circular dichroism (CD) was performed. All of the proteins showed a CD spectra with a minima around 217 nm (Fig. 3b), consistent with the β-sheet found in the Ig fold.

To further confirm that the antibody fragments obtained were natively folded, a selection of them were tested for their ability to bind to their reported antigens. The scFv and Fab of 3211 bound to a peptide fragment of BNP, the antigen it was raised against (Fig. 3c). Due to the availability of suitable substrates, the binding specificity of the Fab fragment of Maa48 was analysed in more detail. Maa48 Fab bound to malondialdehyde-acetaldehyde (MAA)-modified antigen, but not to malondialdehyde (MDA)-modified antigen or to the non-modified antigen (Fig. 3d, e). In addition, Maa48 Fab did not bind to copper oxidized LDL or carbamylated LDL. These results match the published specificity of Maa48 (also known as Fab-pre; [14]) and imply that antibody fragments produced using CyDisCo in the cytoplasm of E. coli retain biological activity and specificity.

Analysis of Herceptin and Tysabri scFv

Both Herceptin and Tysabri were unusual in that the scFv were solubly expressed in high yields in the absence of CyDisCo components i.e. under conditions in which disulfide bond formation should not occur. While the purified yields in the absence of CyDisCo were up to one-third lower of that obtained in the presence of CyDisCo components, the high level produced suggests either that the proteins are soluble and proteolytically stable in the absence of disulfide bond formation or that disufide bond formation occurs via an alternative route for these scFv. To examine the redox state of these the IMAC purified scFv produced in the presence and absence of CyDisCo were analysed.

Vector construction

Genes for the Erv1p, mature PDI along with the heavy and light chains of the antibodies tested (lacking the N-terminal signal sequence) were synthesized codon optimized for E .coli expression (GenScript; Additional file 1: Figure S3). IgE and IgM heavy chains were synthesized without the C-terminal region involved in oligomerization.

The expression vector used was a modified version of pET23 in which the T7 promoter was replaced with Ptac promoter from previously modified (SpeI site inserted) pMal-p2X [10] by digesting the pMal-p2X with MscI/SpeI and ligating the fragment into MscI/XbaI digested pET23. Synthetic multi-cloning sites for Fab fragments (EcoRV/XhoI) and scFv fragments (EcoRV/CelII) were synthesised (GenScript) and ligated into this vector backbone.

The variable domains of the light and heavy chain were amplified by PCR from the synthetic genes and cloned into a synthetic multi-cloning site using NdeI/KasI (light chain) and XhoI/BamHI (heavy chain) to generate a scFv with a C-terminal hexahistidine-tag (Fig. 4). Since the vector backbone, tag and linker region were constant any differences in scFv production comes from the variable regions.

The truncated heavy chain for Fab production was amplified by PCR from the synthetic gene and cloned XbaI/BamHI into a polycistronic vector with the light chain (cloned NdeI/Hind III). These polycistronic vectors include two ribosome binding sites (rbs) and make two proteins by co-expression from two translation initiation sites (Fig. 4). All heavy chain fragments included a C-terminal hexahistidine-tag. The tag was placed on the heavy chain rather than the light chain as preliminary studies on other Fab fragments had suggested that in some cases soluble light chain could be generated and purified in the absence of heavy chain co-expression whereas the opposite was not observed i.e. having the tag on the heavy chain was designed to increase the quality of the final product.

A polycistronic expression construct for codon optimized Erv1p and codon optimized mature PDI was made in modified pET23 as described previously [9]. The polycistronic fragment was transferred into the new vector with Ptac promoter by cloning XbaI/XhoI. From there the fragment containing Ptac promoter, codon optimized Erv1p and codon optimized PDI was cloned NsiI/AvrII into modified a pLysSBAD-vector described previously [10] to generate pMJS205 expression vector. A control construct identical to pMJS205, except lacking the genes for Erv1p and PDI was made by removing the genes by NdeI/SpeI digestion and ligating in short annealed primers with complementary sticky ends.

All plasmid purification was performed using the Gen-Elute HP Plasmid Miniprep Kit (Sigma Aldrich) and all purification from agarose gels was performed using the Gel/PCR DNA Fragments Extraction Kit GeneAid), both according to the manufacturers’ instructions.

All plasmids generated were sequenced to ensure there were no errors in the cloned genes.

Protein expression

For expression in EnPresso B media in 24 deep well plates, E. coli strains containing expression vectors were streaked out from glycerol stocks stored at −70 °C onto LB agar plates containing 5 g/L glucose and suitable antibiotics to allow for selection (100 μg/ml ampicillin for pET23 derivatives, 35 μg/ml chloramphenicol for pLysS derivatives) and the plates incubated at 37 °C overnight. The next day one–three colonies from these plates were used to inoculate 2 ml of LB media supplemented with 2 g/L glucose, containing suitable antibiotics, and the cultures grown at 30 °C, 200 rpm (2.5 cm radius of gyration) in 24 deep well plates covered with an oxygen permeable membrane for 6–8 h. These cultures were used to seed 24 deep well plates containing 3 mls of EnPresso B media (Biosilta Oy; as manufacturer’s instructions) per well containing suitable antibiotics and the cultures grown at 30 °C, 200 rpm (5 cm radius of gyration) in 24 deep well plates covered with an oxygen permeable membrane for approximately 16 h. The cultures were then boosted (as manufacturer’s instructions) and induced with 0.5 mM IPTG. Cultures were harvested after a further 24 h of growth. Final OD₆₀₀ values of the cultures were in the range 20–37. The cells were collected by centrifugation and resuspended in 3 mls of 50 mM sodium phosphate pH 7.4, 20 μg/ml DNase, 0.1 mg/ml egg white lysozyme. After 10 min incubation the resuspended cultures were frozen. Cells were lysed by freeze-thawing.

Protein expression in shake flasks was as per 24 deep well plates except the media volume was 25 mls (250 ml flask) of EnPresso B media and cultures were grown at 30 °C, 250 rpm (2.5 cm radius of gyration). Resuspension was done in the same volume as the initial culture.

Protein purification and analysis

Purification of hexa-histidine tagged proteins was performed by standard immobilized metal affinity chromatography using HisPur Cobalt Superflow Agarose (Thermos Scientific) resin under native conditions following clearance of the cell lysate by centrifugation (4000 rpm, 20 min, 4 °C) for 24 deep well plate. For 3 ml cultures from 24 deep well plates IMAC was performed using 0.5 ml resin in small gravity feed columns. The resin was washed with 2 × 5 mls of water, equilibrated with 2 × 5 mls of 50 mM phosphate buffer (pH 7.4). After loading the sample the column was equilibrated with 5 ml of 50 mM phosphate buffer (pH 7.4), washed with 4 × 5 mls of wash buffer (50 mM sodium phosphate, 5 mM imidazole, 0.3 M sodium chloride; pH 7.4) then 5 mls of 50 mM sodium phosphate (pH 7.4) before elution with 3 × 0.7 mls of 50 mM sodium phosphate, 150 mM imidazole (pH 7.4). For 25 ml cultures the same protocol was used with the following changes: 1.0 ml of resin; 6 × 5 mls of wash buffer; elution with 4 × 1 ml of buffer. Where needed, 2.5 ml of eluted sample was desalted into 50 mM sodium phosphate (pH 7.4) on PD-10 columns (GE Healthcare). Appropriate samples were treated with 20 mM NEM for 20 min at room temperature prior to making SDS–PAGE samples or mass spectrometry analysis.

Protein analysis

Far-UV circular dichroism spectra were recorded on a Chirascan- plus CD spectrophotometer. All scans were collected at 25 °C as an average of four scans, using a cell with a path length of 0.1 cm, scan speed 2 nm/s, step size 0.5 nm, a spectral band width of 1.0 nm. The maximal HT voltage was 750 V.

For determining binding of 3211 Fab and scFv to their ligand, 100 ng of recombinant NTproBNP1-76 [32] in 0.1 M sodium bicarbonate buffer pH 9.6 was coated per well on a shaking ELISA plate over-night at 4 °C. The wells were emptied and rinsed three times with 250 μl of 1xPBS (20 mM phosphate, 150 mM sodium chloride, pH 7.4) containing 0.05 %v/v tween 20 and then incubated with 250 μl of blocking buffer (0.2 % gelatin, 0.5 % BSA, 0.05 % v/v tween20 in 1xPBS, pH 7.4) for 20 min at room temperature. Then 100 μl samples containing 0–20 ng of scFv or Fab, diluted in the blocking buffer were incubated for 2 h at room temperature shaking. After removal of the sample and washing the wells six times with 300 μl of 1xPBS containing 0.05 % v/v tween 20, 100 μl of alkaline phosphatase labelled anti-HIS antibody (Sigma) diluted 1:10000 in blocking buffer was added and the reactions incubated for 1 h at room temperature. After removal of the detection antibody and washing the wells six times with 300 μl of 1xPBS containing 0.05 % tween 20, the 1 mg/ml pNPP-substrate solution (Sigma) in 0.2 M Tris, 5 mM MgCl₂ was added and incubated for 30 min at room temperature. The absorbance at 405 nm was measured with a Tecan Infinite M1000PRO multilabel reader.

For determining the specificity of Maa48 Fab binding, the ligand (MAA-LDL, MDA-LDL, copper oxidized-LDL, carbamylated-LDL, native LDL, MAA-BSA, MDA-BSA or native BSA; sources) at 0–20 μg/ml concentration in PBS was bound to an ELISA plate at 4 °C overnight. The antigens were prepared as described [15]. The plate was washed three times with 0.27 mM EDTA in PBS using an automated plate washer. Nonspecific binding was blocked with 0.5 % fish gelatin and 0.27 mM EDTA in PBS for 1 h at room temperature. Maa48 Fab (1 µg/ml) was incubated for 1 h at room temperature. Alkaline phosphatase-conjugated anti-human IgG (Fab) (Sigma) was used as a secondary antibody and LumiPhos 530 (Lumigen) as a substrate in the assay [14]. The chemiluminescence was measured as relative light units (RLU) with a Wallac Victor3 multilabel reader (Perkin Elmer).

Masses of purified and desalted proteins, treated and not treated with 20 mM NEM were measured by LCMS with an Aquity UPLC-system (Waters) connected to a Synapt G1 Q-ToF –type mass spectrometer. The analytical column was a BEH 300 C4, 2.1 ×100 mm (Waters) run at 0.4 ml/min using a gradient from 3 % acetonitrile in water/0.1 % formic acid to 70 % acetonitrile over 15 min. Samples were acidified with trifluoracetic acid to about 0.5 % v/v and 5 µl of the sample was injected. The mass spectrometer was operated in sensitivity mode with lock mass corrected 1 s scans in continuous mode for m/z 400–2000. Capillary voltage was 3.5 kV, cone voltage 30 V. Mass spectra were base line subtracted and deconvoluted with MaxEnt1.