Single-Batch Expression of an Experimental Recombinant Snakebite Antivenom Based on an Oligoclonal Mixture of Human Monoclonal Antibodies

Abstract

Oligoclonal antibodies, which are carefully defined mixtures of monoclonal antibodies, are valuable for the treatment of complex diseases, such as infectionss and cancer. In addition to these areas of medicine, they could be utilized for the treatment of snakebite envenoming, where recombinantly produced monoclonal human antibodies could overcome many of the drawbacks accompanying traditional antivenoms. However, producing multiple individual batches of monoclonal antibodies in an industrial setting is associated with significant costs. Therefore, it is attractive to produce oligoclonal antibodies by mixing multiple antibody-producing cell lines in a single batch to have only one upstream and downstream process. In this study, we selected four antibodies that target different toxins found in the venoms of various elapid snake species, such as mambas and cobras, and generated stable antibody-producing cell lines. Upon co-cultivation, we found the cell line ratios to be stable over 7 days. The purified oligoclonal antibody cocktail contained the anticipated antibody concentrations and bound to the target toxins as expected. These results thus provide a proof of concept for the strategy of mixing multiple cell lines in a single batch to manufacture tailored antivenoms recombinantly, which could be utilized for the treatment of snakebite envenoming and in other fields where oligoclonal antibody mixtures could find utility.

1. Introduction

Snakebite envenoming is a neglected tropical disease that causes an estimated 100,000 annual fatalities worldwide, including 33,000 deaths in sub-Saharan Africa [1]. Currently, the only specific treatment for snakebite envenoming is antivenom derived from venom-immunized large mammals, such as horses or sheep [2]. Following hyperimmunization, the whole immunoglobulin G (IgG) fraction is isolated from the plasma, often resulting in an antivenom with a low therapeutic content due to the presence of antibodies directed against nontoxic or unrelated antigens [3]. Additionally, antivenoms show batch-to-batch variations when derived from different horses or from different immunizations [4], making it difficult to control the quality and efficacy of antivenoms. Moreover, antivenoms can cause adverse reactions such as anaphylactic shock or serum sickness upon administration due to their foreign origin [5].

To overcome these issues, recombinant antivenoms could act as a promising alternative or additive treatment [6, 7]. Recombinant antivenoms contain a defined mixture of toxin-neutralizing monoclonal antibodies produced in mammalian cells, such as Chinese hamster ovary (CHO) cells, which have been used for antibody production for more than 25 years [8]. It is an increasingly common concept to mix several antibodies post-production to enable binding of several targets at once, such as in cancer therapy, treatment against Ebola virus, or allergy treatment [9–13]. However, this method is associated with significant costs, as each antibody production process requires validation, and with an increasing number of antibodies, costs increase [14].

When several antibodies are required for treatment, one strategy to lower manufacturing costs is to mix cell lines in a single-batch cultivation, thus reducing the validation costs, if the regulatory landscape allows for it. Symphogen (Ballerup, Denmark) pioneered this research field with their antibody cocktail Sym001 [15], which targeted the hemolytic disease of the newborn. By mixing 25 cell lines generated through targeted integration, researchers at Symphogen obtained a product with a reproducible antibody abundance and ratio, although the antibody production between the cell lines still varied 10-fold [16]. Using another approach that included six cell lines generated by random integration, the researchers chose two cell lines to mix in different ratios and showed that stable antibody ratios could be maintained over time [17]. No data are available on varying cell line ratios of more than two cell lines. Inspired by Symphogen’s approach, Laustsen et al. [6] and Jenkins and Laustsen [18] speculated that defined cell line mixtures could present a viable approach for antivenom production. On the basis of theoretical calculations, they concluded that a fed-batch or perfusion cultivation followed by continuous caprylic acid precipitation might enable the production of recombinant antivenoms that are cost-competitive compared to traditional plasma-derived antivenoms.

In this study, we applied our established targeted integration system (isoCHO) to generate cell lines for antibody production. Previously, this methodology has been used to generate cell lines with comparable growth behavior and transcriptomes while producing etanercept, erythropoietin, growth differentiation factor 5, and C1 esterase inhibitor [19]. We expressed four antibodies from four different cell lines, with the individual genes integrated into the same genomic site (“isogenic” cell lines). We mixed the cell lines and performed batch cultivations to characterize a cell line coculture while producing an antibody cocktail with a predefined composition. The cell lines, as well as the produced antibodies, are designated A, B, C, and D (original names can be found in Table 1). The antibodies included in the mixture bind to toxins from three different protein families: Kunitz-type protease inhibitors, three-finger toxins, and phospholipase A₂s (PLA₂s). These toxins are medically important in snakebite envenomings caused by elapid snake species, such as mambas and cobras. The toxins targeted by antibodies A and B are dendrotoxins from the Kunitz-type protease inhibitor family. Dendrotoxins are frequently found in the venom of the black mamba, Dendroaspis polylepis, and exert their toxicity by blocking potassium channels, thus causing involuntary muscle contractions [20, 21]. Antibody C targets α-cobratoxin, which is a long-chain α-neurotoxin from the three-finger toxin family found in the venom of the monocled cobra, Naja kaouthia. This toxin binds to nicotinic acetylcholine receptors and thereby blocks neuromuscular transmissions, which induces flaccid paralysis and, in severe cases, fatal respiratory failure [22, 23]. Finally, antibody D, which targets a PLA₂ toxin from the venom of the black-necked spitting cobra, Naja nigricollis, was included. This toxin has an anticoagulant effect and may also interact synergistically with cytotoxins in the venom and cause tissue damage [24–26]. By combining monoclonal antibodies that neutralize toxins from the venoms of multiple snake species, we demonstrate that our experimental recombinant antivenom serves as an example of how oligoclonal mixtures of defined antibodies might be produced to treat snakebite envenoming. Our data showcase how growing the generated cell lines in a single mixed-batch cultivation resulted in acceptably stable and predictable cell line ratios, and antibody titers that matched expectations and that the antibodies retained their target binding properties as expected. The findings demonstrate how the use of cell line mixtures can be beneficial for the production of oligoclonal antibodies to be used as treatments against complex multitarget diseases, such as snakebite envenoming and potentially other diseases.

Table 1. Antibodies included in the study.

mAb simplified name	mAb original name	Citation of previous work	Target
A	363_01_F07	Laustsen et al. [27]	Dp6 fraction of black mamba (dendrotoxin)
B	367_01_H01	Laustsen et al. [27]	Dp8 fraction of black mamba (dendrotoxin)
C	2551_01_A12	Ledsgaard et al. [28]	Long-chain α-neurotoxins from cobra and black mamba
D	TPL0004_01_A11	Moore et al. [29]	PLA2 from cobra

Note: Antibodies were renamed for simplicity. Dp6/8, Dendroaspis polylepis venom fraction 6/8; PLA₂, phospholipase A₂.

2. Material and Methods

2.1. Antibodies and Cell Lines

The antibodies produced were discovered as single-chain variable fragments (scFvs) using phage display technology and naïve human antibody libraries (IONTAS kappa and lambda libraries constructed from 43 human donors) [30]. Table 1 provides an overview of the included antibodies, their designated names used in this article, their targets, and discovery method. Antibody C was discovered through a campaign that included light chain shuffling followed by cross-panning as described by Ledsgaard et al. [28] The genes encoding the variable heavy (V_H) and light (V_L) chains were cloned into vectors to be expressed in full-length human IgG format (Section 1 of the Supporting Information, Figure S1A). In the IgG format, the constant regions from the antibody rituximab were used with a few modifications [31]. The fragment crystallizable (Fc) domain harbored the LALA (Leu234Ala/Leu235Ala) mutation for reduced binding to Fcγ receptors I, II, and III to decrease antibody-dependent cellular cytotoxicity (ADCC) [32]. Additionally, the Fc domain harbored the YTE (Met252Tyr/Ser254Thr/Thr256Glu) mutation for enhanced neonatal Fc receptor (FcRn) binding and recycling of antibodies for increased half-life in circulation [33]. Moreover, the Fc domain contained an unintentional mutation A98V due to a database error.

Cell lines producing each of the four antibodies were generated by targeted integration into a parental cell line derived from CHO-S cells harboring a landing pad (isoCHO) using our recombinase-mediated cassette exchange system, as described previously [19]. The cell lines were generated, validated, and banked in different campaigns as detailed in the Supporting Information S1 (see Section 1, Table S1–S4, and Figures S1, S2).

2.2. Batch Cultures of Cell Lines to Compare Cell Growth and Metabolic Profiles

CHO cells were thawed and passaged three times a week. The cell lines were cultivated at 37^◦C, 7.5% CO₂, and 81% humidity at 130 rpm (25 mm shaking diameter) using 1× CD CHO medium (Gibco by Life Technologies, 10743029) containing 8 mM of glutamine (Gibco, 25030081) and 0.2% of an anticlumping agent (Gibco, 01-0057AE) in 125 mL shake flasks (Corning) with a working volume of 30 mL in a Kuhner SHAKERX (Kuhner). The cultivations were continued for 8 days or until cell viability dropped below 75%. Viable cell density (VCD) and cell viability were assessed with a Vi-CELL BLU cell viability analyzer (Beckman Coulter). Metabolites were measured with the Bioprofile Flex2 (Nova Biomedical). Total seeding density was 0.3 × 10⁶ cells mL⁻¹. For the first mixture (mix1), cells were mixed in equal numbers at inoculation (25% of each cell line; 1:1:1:1). For the second mixture (mix2), 24.2% of cell line A, 3.4% of cell line B, 22.1% of cell line C, and 50.3% of cell line D were mixed (1:0.14:0.91:2.08). The mixtures, as well as separate single cell lines (SCLs), were cultivated in triplicate. Antibody titers were measured using 1:20 diluted supernatants by surface plasmon resonance in a Biacore 8K+ instrument using a protein A sensor chip (29127557, Cytiva), according to the manufacturer’s description.

Material for the enzyme-linked immunosorbent assay (ELISA) was produced in a 150 mL production batch where the shake flasks were incubated at 8% CO₂, 75% humidity, and 125 rpm as fixed parameters in a Multitron shaker (INFORS HT).

2.3. Amplicon Sequencing of gDNA From the Cell Mixtures

One milliliter of cell culture was centrifuged, and the pelleted cells (maximum 4 million cells) were used for genomic DNA (gDNA) extraction using the GeneJET Genomic DNA Purification Kit (Thermo Fisher Scientific, K0722). The gDNA concentration was adjusted to 10 ng µL⁻¹, and 2 µL were used in polymerase chain reaction (PCR) with the Platinum SuperFi II PCR Master Mix (Thermo Fisher Scientific 12368010) by using a C1000 Thermal Cycler (Bio-Rad). For the PCR, primers enclosing the V_L were designed with an overhang for amplicon indexing with Nextera XT indexes (Table S5). PCR products were run on a 1% agarose gel to confirm successful amplification before samples were handed over to the Australian Genome Research Facility (Melbourne facility) for bead clean-up, library preparation (amplicon indexing PCR), and sequencing using Illumina 250 bp Paired End MiSeq (total amount of data was 3.7 Gb for 192 pooled samples, total paired reads were 7,360,841; Flowcell ID: 000000000-KYDCW). The raw sequences were trimmed with Trimmomatic (Usadel Lab, Forschungszentrum Jülich, Germany), head cropping the first 33 bases and cropping 66 bases in total (Section 2 of the Supporting Information). Further, sequencing reads were aligned to the four V_L gene sequences using kallisto (Pachter Lab, California Institute of Technology, USA) [34].

2.4. Purification of Antibodies

Supernatants from 150 mL batch cultures in 500 mL shake flasks (vented cap, Corning) were thawed overnight at 10^◦C and centrifuged at 4500 × g for 15 min to remove any cell debris. Antibodies in the clarified supernatants were purified by protein A chromatography as described previously [28].

2.5. Antibody Binding to Toxins in ELISA

Both the individual IgG antibodies and the oligoclonal mixtures were assessed for binding to their cognate antigens by ELISA (n = 2). White high-binding ELISA plates (Greiner, 655074) were coated overnight at 4C^◦, with streptavidin (Thermo Fisher Scientific, 21135) diluted in phosphate buffered saline (PBS) to 10 µg mL⁻¹. After the plates were washed three times with PBS containing 0.05% Tween and blocked for 1 h with 3% skimmed milk in PBS, biotinylated α-cobratoxin (Latoxan, L8114), Dendroaspis polylepis venom fraction 6 and 8 (Dp6 and Dp8), and Naja nigricollis venom fraction 19b (Nn19b) containing PLA₂ were added at 5 µg mL⁻¹ in PBS with 3% skimmed milk. Purified IgG antibodies were diluted to 10 µg mL⁻¹ in PBS with 3% skimmed milk, titrated 1:3, and left to bind the washed plate for 1 h at room temperature. After washing, bound IgG antibodies were detected with donkey anti-human-horseradish peroxidase (Jackson Immunoresearch, 709-035-149) diluted 1:5000 in PBS with 3% skimmed milk, followed by addition of substrate (Pierce SuperSignal, 37070) diluted 1:10 in TRIS buffer. Luminescence was read in a VICTOR Nivo plate reader after incubation for 10 min at room temperature.

Data were plotted and sigmoidal curves were fitted using a concatenated logistic fit by Origin 2019b (OriginLab) software. One outlier was removed from the mix1 Dp6 ELISA (41.2 µg mL⁻¹ IgG, n = 1).

2.6. Statistical Analysis

Maximum growth rates (µ_max), growth, consumption, and by-product formation were compared by using analysis of variance (ANOVA) and a Tukey’s honestly significant difference (HSD) test in JMP Pro 15.0.0 (390308).

Cell line ratios were analyzed using the Multivariate Response Generalized Linear Models (MGLM) package in R. For each sample, we assume that the amplicon reads are distributed according to a discrete multivariate distribution. The simplest multivariate distribution is the multinomial (MN) distribution with parameters n equal to the number of reads for the sample and the probability vector p = (p_A, p_B, p_C, p_D), ∑p_i = 1. The MN distribution does not allow for overdispersion, which is common in real data; hence, the Dirichlet multinomial (DM) distribution is a common alternative. The DM distribution is a compound MN distribution in which the probability vector is drawn from a Dirichlet distribution with parameter vector, a = (a_A,α_B,a_C,a_D). The Dirichlet distribution makes a certain assumption about the covariance matrix, which can be relaxed by having the probability vector drawn from a generalized Dirichlet distribution (GDM) at the cost of introducing additional parameters, here a = (a_A, a_B, a_C), b = (b_A, b_B, b_C). We are interested in how the distribution parameters change over time, t, in culture. We link parameters in the distributions to the covariate, x = (0,t), through the inverse link functions:

$$\begin{array}{c}\text{MN}:p_j=\frac{e^{x^T{\beta }_j}}{{\displaystyle \sum }{e}^{x^T{\beta }_j}},j=A,B,C,D\text{and}{\beta }_D=0\hfill \\ \text{DM}:a_j=e^{x^T{\beta }_j},j=A,B,C,D\hfill \\ \text{GDM}:a_j=e^{x^T{\alpha }_j},b_j=e^{x{\beta }_j},j=A,B,C\hfill \end{array}$$

All three models were fitted to the data for each of the two cell line mixes. The DM model had significantly better Bayesian information criterion and Akaike information criterion scores than the MN model and slightly better scores than the more complex GDM model. The full DM model (with an intercept and a slope) was compared to the simpler model (no slope) using a likelihood ratio test ${\lambda }_{\text{LR}}=-2\left(l\left({\theta }_0\right)-l(\widehat{\theta })\right)\le {\chi }_{4.95\text{\%}}^2$, where l(…) is the log-likelihood function and θ₀ is the parameter subset with no slope. Other special tests were formulated as Wald tests.

3. Results

3.1. Targeted Integration Yielded Genetically Similar Cell Lines

For the production of the four toxin-targeting IgG antibodies (A, B, C, and D), we constructed one cell line per antibody as described in Section 1 of the Supporting Information and Figures S1 and S2. The first approach for cell line generation utilizing the initial donor plasmids yielded several clonally stable cell lines for antibody B and one cell line for antibody D, but none for antibodies A and C. Stable cell lines for producing antibodies A and C were generated by changing the heavy chain promoter in the plasmid for antibody C and including an antibiotic selection step to generate a cell pool for antibodies A and C prior to single cell sorting. Thus, distinct strategies might be required to generate cell lines that express different antibodies. Upon integration of the antibody genes into the cell genome, sequencing of the insert junctions revealed that a recombination error had resulted in the heavy chain of antibody D losing its polyA signal. Digital PCR revealed 1.7 copies of the gene encoding the light chain in cell line C. Therefore, the generated cell lines are not isogenic but genetically similar, and one cell line per antibody (where applicable) was selected and characterized for growth rate, nutrient consumption, by-product formation, and antibody expression levels.

3.2. Cell Lines Display Similar Growth and Metabolite Profiles

The four cell lines were cultured individually in a shake flask batch experiment and showed similar courses of growth over time (Figure 1A). Batches for cell line A were terminated 1 day earlier than the batches for the other cell lines to avoid excessive drops in viability. The four cell lines showed similar glutamine and glucose consumption profiles, as well as lactate and ammonia production profiles (Figure 1A). However, statistical tests revealed small but significant differences between the cell line maximum growth rates (µ_max). The µ_max for cell lines A, B, and C were significantly different from each other (Figure 1B); the µ_max of cell line D (4 × 10⁻² ± 4.5 × 10⁻⁴ h⁻¹) fell between the values of cell line A (4.1 × 10⁻² ± 8.0 × 10⁻⁴ h⁻¹) and cell line B (3.9 × 10⁻² ± 2.5 × 10⁻⁴ h⁻¹) and was distinct from that of cell line C (3.7 × 10⁻² ± 6.5 × 10⁻⁴ h⁻¹). Additionally, the antibody titer (Figure 1C) and the specific antibody productivity (data not shown) were significantly different between the cell lines. Although the µ_max values were significantly different between cell lines, the VCDs for all cell lines were constant after inoculum and during the last days of cultivation (days 7 and 8) (Figure S3, Table S6). After having characterized each cell line separately, we mixed the cell lines to assess how they behave in a mixture.

Figure 1. Single cell lines producing antibodies A, B, C, D, and mixtures thereof.

(A) Batch cultivation in triplicate. Column 1: From top to bottom: growth curves (MMcells = million cells), glutamine consumption, ammonia production. Column 2: Cell viability, glucose consumption, and lactate production. (B) Maximum growth rates (µ_max) of the single cell lines. Significant differences between growth rates were assessed by analysis of variance (ANOVA) and Tukey’s honestly significant difference (HSD) test using the JMP Pro 15 (SAS) software and are depicted with letters above the bars in the figure and details can be found in Table S6 (n = 3). (C) Antibody titers in cell line supernatants measured with surface plasmon resonance (SPR) in Biacore. Red triangle, A; green diamond, B; blue square, C; yellow circle, D; dark blue cross, mix1; purple star, mix2. Mix1 represents 25% of each cell line. Mix2 represents 24.2% of cell line A, 3.4% of cell line B, 22.1% of cell line C, and 50.3% of cell line D. Cultivation was performed in single-use shake flasks at 37^◦C, 7.5% CO₂, 130 rpm, 25 mm throw diameter, and 81% humidity.

3.3. Mixed Cell Cultures Maintain a Stable Cell Line Ratio and Produce Expected Antibody Titers

We blended the cell lines for the inoculum of the mixed cultures in two fashions: (i) equal abundance (mix1: 25% of each cell line) and (ii) adjusted to the antibody titer of each cell line (mix2: 24.2% of cell line A, 3.4% of cell line B, 22.1% of cell line C, and 50.3% of cell line D). The growth, nutrient consumption, and by-product formation profiles of the mixed cultures present an average of the SCLs (Figure 1A). The growth rates of the mixtures (mix1: 4 × 10⁻² ± 1.1 × 10⁻³ h⁻¹ and mix2: 4 × 10⁻² ± 5.8 × 10⁻⁴ h⁻¹) are not significantly different from those of any of the SCLs, except SCL C (Figure 1B), and the mixtures present antibody titers that lie between the SCL titers (Figure 1C).

To assess the development of the cell line ratios in the mixtures over time, we amplified the gDNA from the mixtures at different time points and sequenced a fragment including the V_L chain, which is different in all cell lines. For the analysis, we excluded the final time point (day 8), where cell viability fell below 80% (Figure 1A), which was reflected in a change in the composition (Figure 2). We fitted a DM model to the read counts,

$$Y_{j,\eta }~DM\left(n_{j,\eta },a\right),\log a_i={\alpha }_i+{\beta }_i{t}_j,i=A,B,C,D$$ (1)

where Y_j,η is the read count vector (#A, #B, #C, and #D) in the j^th time point for the n^th biological replicate. Similarly, n_j,η is the total read count for the j^th time point for the n^th biological replicate.

Figure 2. Cell line percentages in mix1 and mix2. Percentages were assessed by amplicon sequencing of the variable light chain region.

(A) In mix1, 25% of each cell line was attempted. (B) In mix2, 24.2% of cell line A, 3.4% of cell line B, 22.1% of cell line C, and 50.3% of cell line D were attempted.

For mix1, the aim was to have an equal abundance of the four cell lines. However, cell line B was clearly more abundant than the others, possibly due to a cell counting device bias related to cell size that was observed consistently (data not shown) before the mixing (Wald test = 645.9 on 3 degree of freedom [DoF], p = 0), while cell lines A, C, and D were equally abundant (Wald test = 1.26 on 2 DoF, p = 0.532). There remained a small but significant negative drift in the Dirichlet parameters (Wald test = 15.8 on 3 DoF, p = 0.0033), indicating that the dispersion was increasing over time; that is, the biological replicates vary more from each other the longer they stay in culture, but the average ratios stay nearly constant. For mix2, the overdispersion relative to the MN is greater than that for mix1, and there is a larger variation in ratios between the cell lines over time. Given the higher noise, there is no evidence of drift of the parameters (likelihood ratio test = 3.95 on 4 DoF, p = 0.41).

Using the values from the sequencing and the titers from the SCLs, we calculated the theoretical titer of the mixtures according to Equation (2):

$$\begin{array}{l}\text{Titer}\left(\text{theoretical}\right)\\ =\underset{\text{SC}L=A}{\overset{E}{{\displaystyle \Sigma }}}\text{titer}(\text{day}n)\ast \frac{{\text{percentage}}_{\text{SCL}}\left(\text{day}n\right)}{100}\end{array}$$ (2)

The theoretical antibody titer for mix1 is consistent with the observed antibody titer, while the mix2 titer slightly diverges from the theoretically calculated titer (Figure 3). Having characterized the mixtures on a cultivation level, the next step was to purify the antibodies and confirm that the expressed antibodies bind to their respective toxin as expected.

Figure 3.

Theoretical versus actual antibody titers for mix1 and mix2. Theoretical values are calculated from daily single cell line antibody titers and the actual cell line percentage from the sequencing result (Equation 2). The actual antibody titers were determined with surface plasmon resonance (SPR) in Biacore using a protein A chip. Dark blue cross, mix1; purple star, mix2 (n = 3); light blue cross, mix1 theoretical titer (n = 1); light purple star, mix2 theoretical titer (n = 1).

3.4. Antibody Binding Properties Remain Unchanged in the Oligoclonal Mixture

The purified individual IgG antibodies and the mixture contained mainly monomeric components as determined by size exclusion chromatography (Figure S4), and ELISA experiments on the purified antibodies from the SCLs and mix1 showed binding of the antibodies to their respective target (A, dendrotoxin-containing fraction Dp6; B, dendrotoxin-containing fraction Dp8; C, α-cobratoxin; and D, PLA₂-containing fraction Nn19b). For the mixture, we calculated the expected percentages of the different antibodies from the SCL titers on day 7 in the batch and applied these approximate numbers to determine the concentration of the individual IgG antibodies used in the ELISA (Figure 4, Table S7). There is a small difference in the EC₅₀ values between the oligoclonal mixture and the individual IgG antibodies, which is likely due to the approximated concentrations of the individual IgG antibodies in the mixture. Furthermore, a relatively weak binding to Nn19b is observed for both the individual IgG antibodies and the oligoclonal mixture, which makes the calculation of an EC₅₀ value uncertain. Overall, the results demonstrate that the oligoclonal antibodies bind their cognate antigens as expected and that the antibody concentrations in the oligoclonal mixture align with the expected values.

Figure 4.

ELISA antigen binding of antibodies derived from cultivation of individual single cell lines and a cell line mixture. Transformed x-axis values refer to expected antibody concentrations in the mixture, which were calculated on the basis of the single cell line antibody titers at day 7 (n = 2). (A) Binding to Dendroaspis polylepis venom fraction 6 (Dp6) containing dendrotoxin. One outlier at 41.15 µg mL⁻¹ of immunoglobulin G (IgG) was removed. (B) Binding to Dp8 containing dendrotoxin. (C) Binding to α-cobratoxin. (D) Binding to Naja nigricollis venom fraction 19b (Nn19b) containing phospholipase A₂ (PLA₂). Sigmoidal curves were fitted with Origin 2019b (OriginLab) using a concatenated logistic fit. Statistical parameters can be found in Table S7.

4. Discussion

Defined mixtures of monoclonal antibodies can be used to target multiple different antigens, achieving effects that a single monoclonal antibody alone cannot provide [35]. This is particularly important when several different antigens are to be targeted simultaneously, which includes applications such as infectious diseases, where multiple epitopes are involved, and envenomings by venomous animals, such as snakes, where multiple toxins are involved [36–38]. The venom from a single snake species typically contains dozens of different toxins that can be divided into several (sub)families, and many of these toxins need to be neutralized for an antivenom to be effective [39–41]. Currently, neutralization is achieved by using polyclonal antibodies derived from the plasma of immunized animals (typically horses). However, the use of oligoclonal mixtures containing defined recombinantly expressed antibodies (i.e., recombinant antivenoms) could potentially be beneficial in regard to improved safety and efficacy compared to conventional therapies [7, 27, 42]. Mammalian cells would be the production host of choice, as they have been historically exploited to produce recombinant monoclonal antibodies exhibiting close-to-human glycosylation patterns [43]. However, when several monoclonal antibodies are produced to be mixed in the same final product, it can be cumbersome and costly to manufacture these antibodies separately in parallel or in subsequent bioreactor runs, as multiple productions require more time, possibly higher equipment costs, multiple downstream operations, and several process optimizations. In addition, the cost of cell banking and upstream and downstream processing becomes significant [14].

For snakebite envenoming, as well as for many other neglected tropical diseases, reducing manufacturing costs is of high importance, as this may help drive down the cost of treatment and make it more accessible to victims from tropical low- and middle-income countries, where most snakebites occur [18, 44, 45]. To reduce manufacturing costs for oligoclonal antibody mixtures, one strategy involves coculturing stable cell lines in a single bioreactor and then purifying the oligoclonal antibody mixture in one downstream process, as exemplified by Rozrolimupab, which is composed of 25 different monoclonal IgG antibodies against the Rhesus D antigen that are produced in a single batch [16, 46]. Another strategy for recombinant production of multiple IgG antibodies in one batch is transfecting a plasmid mixture into cells, generating a pool that expresses different antibodies, as exemplified by GIGA-2050, which is a mixture of approximately 12,500 antibodies (analyzed at the transcript level) developed for the treatment of COVID [47–49]. However, the use of large plasmid mixtures for antibody production results in an undefined composition of cell lines and antibodies, including loss of antibodies [47], which may come with limitations when the exact composition of the product is important and/or needs to be documented.

In this study, we constructed four stable CHO cell lines and mixed them in two distinct ratios at inoculum in two different batches (n = 3) and monitored the cell line populations over 8 days. The two ratios were (i) equal cell numbers for all cell lines and (ii) cell numbers for each cell line adjusted to its production of respective IgG antibody, which differed among the cell lines (Figure 1C). Tracking of both cell line populations and antibody titers, combined with calculations of the expected antibody concentrations in the final mixtures, suggested that the cell line ratios stayed consistent over 7 days and that the oligoclonal antibody production batches yielded the expected concentrations of each individual antibody. The different antibody titers observed for the different cell lines were possibly a result of each cell line producing an antibody with different variable domains and of small variations in the genomic content of the cell lines, such as mutations resulting from genetic instability, production load, selection pressure (antibodies A and C), or epigenetic changes [50–52]. Although the clones were not fully isogenic, their growth and antibody production were similar between separate batches, which indicates that it might be feasible to mix and coculture these cell lines in bioreactors. The use of cell lines derived from targeted integration further facilitates the design of oligoclonal mixtures and may allow cell lines to be exchanged and added to the oligoclonal mixture if needed. The observed antibody titers of the cell lines were relatively low compared to industrial standards but sufficient to study cocultures. To increase the antibody titer in the future, multicopy cell lines (using several landing pads) [53] could be generated or different insertion loci could be explored [54, 55]. This could possibly be combined with sorting parental cells with a high mitochondrial membrane potential before the cell line generation, which has been shown to positively affect protein production [56]. Even though the maximum growth rates for the four cell lines had small but significant differences (Figure 1B), the ratios of the cell lines in the mixtures were stable over time (Figure 2). For the mixture adjusted to the antibody production (mix2), no significant change in the cell population could be detected over time, while for an equal mixture (mix1), cell line C showed a small but significant drift downward in abundance (Figure 2). Overall, the data demonstrate reproducible and predictable growth for cell line mixtures in batch cultivation and expected antibody titers. In addition, the binding of the antibodies to their cognate targets, which is required for the antibodies to be functional, was confirmed by ELISA.

Depending on the indication, the importance of a precisely defined ratio between the antibodies may be higher or lower. How important this ratio is for a therapeutic product may depend on whether the target is exogenous or endogenous and whether overdosing of one specific antibody could cause (severe) adverse reactions. For therapies against cancer and autoimmune diseases, tight control over precise dosing is often essential if the therapeutic window for the drug product is narrow, as drug targets are endogenous and often overexpressed in diseased tissue but also present in healthy tissue and/or cells [57]. However, we speculate that for snakebite and other animal envenomings, overdosing of human monoclonal antibodies against the toxins is less critical, since toxins are distinct from endogenous proteins. Nevertheless, it is still relevant to control the antibody composition fully in an antivenom product to optimize its therapeutic utility (including its neutralization capacity across multiple snake venoms).

While we demonstrate that our stable cell lines grew stably, predictably, and reproducibly in batch cultivations both individually and in defined mixtures, most industrial processes are run in fed-batch or perfusion mode, which is relevant to consider when designing manufacturing processes for recombinant antivenom products [6]. Showing cell line stability and reproducible antibody production in more complex cultivation modes may therefore be an obvious next step for future research efforts, as this could be key to ensure low-cost manufacturing, as such costs are currently an impediment to bringing recombinant antivenoms into the clinic [58]. However, the data presented here showcase how targeted integration and coculture of stable CHO cell lines in a single batch can be used to generate mixtures of defined recombinant human monoclonal antibodies, which are promising therapeutics for the treatment of bites and stings from venomous animals, such as snakes, but which may also be of utility for other human diseases where multitargeting of dissimilar antigens is required.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section.

Abbreviations

CHO

Chinese hamster ovary

Dp6

Dendroaspis polylepis venom fraction 6

Dp8

Dendroaspis polylepis venom fraction 8

ELISA

enzyme-linked immunosorbent assay

Fc

fragment crystallizable

gDNA

genomic DNA

IgG

immunoglobulin G

Nn19b

Naja nigricollis venom fraction 19b

PCR

polymerase chain reaction

PLA₂

phospholipase A₂

scFv

single-chain variable fragment

SCL

single cell line

VCD

viable cell density

V _H

variable heavy chain

V _L

variable light chain

Data Availability Statement

The data that support the findings of this study are available on a reasonable request from the corresponding author.