Optimizing assembly and production of native bispecific antibodies by codon de-optimization

Generation of K11L7–2 variants and characterization in transient transfections

We first generated a panel of 6 variants of K11L7–2 containing different combinations of antibody heavy (VHCH) and light (VKCK K11 and VLCL L7–2) chains encoded by sequences optimized for expression in mammalian cells (Table 1; constructs 1 to 6). All DNA sequences encoded the same amino acid sequence. These constructs were used for transient transfections in PEAK cells, and the content of the culture supernatant was then purified and characterized. While the total IgG productivity remained unaltered (Fig. 2D), increase in light chain abundance correlated with sequence optimization (Figs. 2B and C). Indeed, the over-representation of the lambda light chain was further increased in the K11L7–2 variants 1 and 3 containing an optimized sequence for the lambda light chain. The situation remained unchanged when both light chains were optimized as in variants 2 and 4. Finally, optimization of the kappa light chain sequence in variants 5 and 6 lead to an increase of assembled kappa chain. However, even in variants 5 and 6 in which the kappa chain was optimized, the IgGλλ remained the major form secreted (Fig. 2C).

To improve this situation, a second strategy aimed at reducing the expression of the highly expressed lambda light chain was explored. Three new variants of K11L7–2 were generated, each containing the same optimized heavy and kappa chain sequences as in variant 5 combined with genes encoding the lambda light chain that were de-optimized to various degrees. The de-optimization was achieved by including codons less frequently used in CHO cells. The variants, i.e., constructs 7, 8 and 9, respectively, contained increasing numbers of non-optimal codons (Table 1; see materials and methods and Fig. S1-4 for details). The different IgG forms isolated from the supernatant of PEAK cells transiently transfected with K11L7–2 and the variants 4, 5 and 7 to 9 were analyzed by IEF (Fig. 3A). A reduction of IgGλλ was observed with increasing levels of lambda light chain codon de-optimization. Inversely, IgGκκ was increasingly more abundant in variants 7, 8 and 9, respectively. The bands corresponding to the bispecific IgGκλ were also more intense for variants 7, 8 and 9 compared with the original K11L7–2 construct or the optimized variants 4 and 5 (Fig. 3A). The overall productivity was unaffected, with total IgG concentrations ranging between 20 and 30 µg/mL (Fig. 3B).

To better quantify the relative abundance of the three forms, hydrophobic interaction chromatography (HIC) was performed on these samples (Fig. 3C). The integration of the three distinct peaks indicated that the bispecific content was significantly increased, from 22% in the initial K11L7–2 to 43% for variant 8 (Fig. 3D and Table 2). This data confirmed the IEF analysis indicating that codon de-optimization of the lambda light chain sequence had an effect on the distribution of the three IgG forms secreted in the culture supernatant. Interestingly, although the IgGκκ/IgGλλ ratio varied significantly in variants 7, 8 and 9 (0.38, 1.07 and 2.27, respectively), the proportion or IgGκλ remained comparatively stable (38%, 43% and 40%, respectively). This observation is in line with the fact that when one of the light chains is less expressed, the assembly of the monospecific IgG form incorporating two copies of the less available light chain is more affected than the bispecific form incorporating a single copy of the less abundant chain. Thus, within a certain range, unbalanced expression of the two light chains has a low impact on the assembly of the IgGκλ. This observation supports the fact that for most κλ body candidates expressed, the proportion of the bispecific form ranges between 35% and 50% without any codon optimization.

Expression in stably transfected CHO cells

We then investigated whether the results obtained using transient transfections would also apply to expression in stable CHO cells that represent a more relevant situation for large-scale production of therapeutic antibodies. K11L7–2 as well as variants 4, 5, 7, 8 and 9 were transfected into CHO cells. Between 4 and 13 stable pools were selected and expanded for each construct. Productivity and IgG species distribution were evaluated for fed-batch fermentation conditions. After total IgG purification by Protein A affinity chromatography, relative abundance of the three IgG forms was determined by HIC (Fig. 4A). Similar to transient expression conditions, the IgGκκ/IgGλλ ratio increased with the level of codon de-optimization of the lambda chain and the proportion of IgGκλ increased to reach a maximum of about 40% in variants 7 and 8 (Fig. 4A). Interestingly, the drastic reduction of lambda chain expression in variant 9 leads to a decrease in the abundance of IgGκλ. This observation indicates that the lambda chain expression became limiting for bispecific antibody assembly. The average total IgG titer measured in the supernatant increased in variants containing a de-optimized lambda chain (Fig. 4B). This increase in total IgG titer combined with a significantly larger proportion of IgGκλ lead to the generation of CHO cell pools with an increase in IgGκλ titer for variants 7 and 8 (Fig. 4B). Indeed, when considering the two best CHO stable cell pools for each construct, the yield of bispecific IgGκλ doubled in variant 8 (721 and 678 µg/mL) as compared with the K11L7–2 (348 and 365 µg/mL; Table 3). Purification of the bispecific IgGκλ was performed using the three sequential affinity chromatography steps (Fig. 5A). The purified bispecific IgGκλ had similar aggregate levels and equivalent binding to their targets (data not shown). Total IgG isolated after the Protein A step and purified κλ body were analyzed on an Agilent Bioanalyzer and by IEF (Fig. 5B and C). The intensity of the bands corresponding to the light chains (Fig. 5B) or the intact IgG species (Fig. 5C) confirmed the HIC data indicating that the assembly of the two light chains was more evenly distributed in variants 7 and 8, and lead to the highest level of IgGκλ secretion in the cell culture supernatant.

Codon optimization and de-optimization

Optimal and non-optimal codon sequences for mammalian cells were generated by GeneArt® (GeneOptimizer™ software) on the heavy chain, the kappa and the lambda light chains. Inhibitory motifs (such as possible splice sites) have been removed in all sequences. During the optimization process the following sequence motifs were avoided: internal TATA-boxes, chi-sites, ribosomal entry sites, AT-rich or GC-rich sequence stretches, RNA instability motifs, repeat sequences, RNA secondary structures, splice donor and acceptor sites. To reduce the lambda chain expression, three different levels of de-optimization were applied by progressive introduction of codons used less frequently. All sequences are shown in Figs. 1–3 and the codon quality distribution for each de-optimized sequences shown in Fig. S4 For variant 9, the parameters have been modified in such a way that the result leads to a nearly inverted codon usage table. For variants 7 and 8, the parameters were more relaxed.

Plasmid generation

The common heavy chain, the kappa and the lambda light chains previously generated were cloned into a single pNOVI expression vector containing three expression cassettes under the transcriptional control of the human cytomegalovirus promoter.

First, second and third promoters drove, respectively, the expression of the kappa light chain, which binds to hCD47, the common heavy chain, and the lambda light chain, which targets hCD19.

Different candidates were obtained by cloning the optimized and de-optimized chains in the wild type plasmid of K11L7–2 (Table 1).

Transient IgG expression in PEAK cells

Transformed Human Embryo Kidney monolayer epithelial cells (PEAK cells; Edge Bio) were maintained in 5% CO₂ at 37°C in a humidified atmosphere in DMEM (Thermo Fisher Scientific, Waltham, MA) containing 10% FCS (Sigma-Aldrich, St Louis, MO) and supplemented with 2mM glutamine (Sigma-Aldrich), referring to complete DMEM. One day before transfection, confluent cells were split 1:4 in T175 flasks (Nunc™ Cell Culture Treated Flasks with Filter Caps, Thermo Fisher Scientific) to have cells in the exponential growth phase. Transient transfections were performed using a mix containing 30 μg of DNA and 42 μL of Lipofectamine® 2000 transfection reagent (Thermo Fisher Scientific) in 1.4 mL of DMEM for 10⁷ cells per T175 flask in 50 mL of complete DMEM.

IgG expression was measured using the Octet RED96 with Protein A-coated biosensors (Pall ForteBio, Menlo Park, CA). According to antibody concentration, supernatants were harvested 7 to 10 d after transfection.

Stable pool generation in CHO-S cells

CHO-S (Thermo Fisher Scientific) cells were routinely cultured in suspension at 2×10⁵ cells/mL in CD CHO medium (Sigma-Aldrich) supplemented with 6 mM L-glutamine in Erlenmeyer flasks. Cell culture was performed at 37°C with 5% CO₂ and 85% relative humidity at 140 rpm. For stable transfection, CHO-S cells in exponential growth phase were resuspended at 1.43×10⁷ cells/mL in CD CHO medium without glutamine and mixed with 40 µg of linearized DNA in an electroporation cuvette of 0.4 cm (Bio-Rad, Hercules, CA).

After electroporation by single pulse of 300 V, 900 µF and infinite resistance using Gene Pulser Xcell™ Electroporation Systems (Bio-Rad), cells were immediately added in 50 mL of CD CHO medium without glutamine and distributed in two 6-well plates at 4 mL/well or in two T-75 flasks and placed in humidified 10% CO₂ incubator set at 37°C. The following day, L-methionine sulfoximine (MSX) (Sigma-Aldrich) was added to the culture at 50 µM final concentration for transfected cell selection. After 4 to 5 weeks of growth under selection pressure, pools were assessed for their IgG productivity, and transferred to Erlenmeyer flasks and amplified in suspension.

Productivity evaluation by fed-batch overgrow culture

The productivity of CHO pools was assessed by fed-batch overgrow culture evaluation. Cells were inoculated at 3×10⁵ cells/mL in 50 mL of CD CHO medium supplemented with 300 mg/L of L-cysteine, 120 mg/L of L-tyrosine, 50 µM MSX and fed with 15 mL CHO CD EfficientFeed™ B Liquid Nutrient Supplement (Thermo Fisher Scientific) at day 1. The glucose level was monitored using the GlucCell Glucose Monitoring System (CESCO Bioproduct), and adjusted when necessary. Cells were harvested at day 15 post inoculation or when viability had dropped below 75%. IgG quantitation were measured by Octet.

IgG purification

After 7 to 10 and 15 d of antibody production for PEAK cells and CHO cells, respectively, the supernatant was harvested, clarified by centrifugation 10 min at 2000 rpm and filtered on a 0.22 µm membrane (Merck Millipore, Darmstadt, Germany). Total IgGs were purified by one affinity chromatography step using the FcXL resin (Thermo Fisher Scientific) or the MabSelect™ SuRe™ resin (GE Healthcare, Chicago, IL) for PEAK and CHO supernatant, respectively. A serum-containing medium is used for production in PEAK cells; bovine IgGs do not bind to the FcXL resin, while they bind to the MabSelect™ SuRe™ resin. Then, two additional affinity chromatography steps were required to isolate the κλ body and eliminate the two monospecific mAbs: one purification on the KappaSelect resin (GE Healthcare) to eliminate the IgGλλ and one purification on the LambdaFabSelect resin (GE Healthcare) to get rid of IgGκκ.

An appropriate amount of MabSelect™ SuRe™ or FcXL resin was washed three times with phosphate-buffered saline (PBS) and resuspended in PBS (Sigma-Aldrich). The resin was added to the supernatant and the mix was incubated overnight at 4°C and 15 rpm. Samples were centrifuged at 2200 rpm for 10 minutes to collect the resin and the flow through was discarded. The resin was washed with PBS and transferred on SigmaPrep™ spin column (Sigma-Aldrich). Elution was performed with glycine 50 mM at pH 3.0. Following purification, the total IgG and the κλ body were formulated into 25 mM histidine, 125 mM NaCl at pH 6.0, by desalting on Amicon Ultra-4 centrifugal filters with membrane Ultracel 50 kDa (Merck Millipore) previously equilibrated with formulation buffer.

The final antibody concentration was evaluated by Nanodrop®.

Characterization of purified antibodies

Monospecific antibodies and κλ body distribution and integrity was assessed by electrophoresis, IEF, HIC-HPLC and SEC-HPLC.

Purified IgGs were analyzed by electrophoresis in denaturing and reducing conditions. The Agilent 2100 Bioanalyzer was used with the Protein 80 kit as described by the manufacturer (Agilent Technologies, Santa Clara, CA).

The distribution of the different formats of IgG (IgGλλ, IgGκκ and IgGκλ) was determined by isoelectric focusing (pH 7–11 IsoGel agarose plates, Cambrex, East Rutherford, NJ) and HIC-HPLC analysis using ProPac HIC-10 column (Dionex, Sunnyvale, CA). A gradient of mobile phase A (0.001 M phosphate buffer + 1 M ammonium sulfate, pH 3.5) from 85 to 35% and a growing gradient of mobile phase B (0.001 M phosphate buffer + acetonitrile 10%, pH 3.5) from 15% to 100% were applied. A blank was performed with mobile phase A, pH 7.0.

Aggregate and fragment levels were determined by SEC-HPLC with a Biosep-SEC-s3000 column (Phenomenex, Torrance, CA) using a 200 mM sodium phosphate, pH 7.0 mobile phase.