Development of an E. coli strain for cell‐free ADC manufacturing

Abstract Recent advances in cell‐free protein synthesis have enabled the folding and assembly of full‐length antibodies at high titers with extracts from prokaryotic cells. Coupled with the facile engineering of the Escherichia coli translation machinery, E. coli based in vitro protein synthesis reactions have emerged as a leading source of IgG molecules with nonnatural amino acids incorporated at specific locations for producing homogeneous antibody–drug conjugates (ADCs). While this has been demonstrated with extract produced in batch fermentation mode, continuous extract fermentation would facilitate supplying material for large‐scale manufacturing of protein therapeutics. To accomplish this, the IgG‐folding chaperones DsbC and FkpA, and orthogonal tRNA for nonnatural amino acid production were integrated onto the chromosome with high strength constitutive promoters. This enabled co‐expression of all three factors at a consistently high level in the extract strain for the duration of a 5‐day continuous fermentation. Cell‐free protein synthesis reactions with extract produced from cells grown continuously yielded titers of IgG containing nonnatural amino acids above those from extract produced in batch fermentations. In addition, the quality of the synthesized IgGs and the potency of ADC produced with continuously fermented extract were indistinguishable from those produced with the batch extract. These experiments demonstrate that continuous fermentation of E. coli to produce extract for cell‐free protein synthesis is feasible and helps unlock the potential for cell‐free protein synthesis as a platform for biopharmaceutical production.


| INTRODUCTION
Antibodies have desirable properties that make them particularly well suited for therapeutic applications. They are large and have an affinity for the neonatal Fc receptor which can result in serum half-lives greater than 15 days (Booth et al., 2018). They have excellent affinities for their target, which can bind with picomolar K D (equilibrium dissociation constant) (Kamat et al., 2020). Finally, they are human proteins with exquisite specificity, reducing immunogenicity and other off-target effects (Irani et al., 2015).
Cell-free protein synthesis (CFPS) has emerged as a powerful method for the rapid production of a variety of immunoglobulin formats including IgG, scFv, Fab, and bispecific antibodies (Yin et al., 2012;Xu et al., 2014). Previous studies inspired by the IgG folding pathway in the mammalian endoplasmic reticulum identified two key protein factors (DsbC and FkpA) for expressing immunoglobulins at high yield with in vitro protein synthesis (Groff et al., 2014). Disulfide bond isomerization is a key process in immunoglobulin maturation that allows the correct conformation of disulfide bonds to occur by allowing disulfide shuffling until the protein reaches its mature and most stable form. In prokaryotic expression systems, this process can be efficiently catalyzed by the disulfide isomerase DsbC (Yin et al., 2012;Frey et al., 2008). The other biochemical activity key for folding IgG constructs is peptidyl-prolyl isomerase (PPI), which interconverts the thermodynamically favored trans proline isomer to cis. There is a key proline residue in the CH1 domain of the heavy chain that must be converted to cis before the mature IgG fold is reached (Feige et al., 2010). The periplasmic E. coli PPI, FkpA can facilitate this isomerization and has been shown to greatly enhance IgG folding and assembly in CFPS reactions (Groff et al., 2014).
Earlier methods for antibody conjugation utilize the reactivity of natural amino acid functional groups such as the thiol group on cysteine or the primary amine in the lysine side chain and modify a fraction of these amino acids. This leads to heterogeneity in the sites and numbers of residues modified for the IgG bioconjugate. There is a growing desire and market for antibodies with bioconjugation handles at clearly defined locations (J. Walsh et al., 2021). The production of antibodies with site-specific conjugation handles can facilitate the production of fluorophore or metal labeled antibodies for in vivo imaging applications (Adumeau et al., 2016), precision attachment of radioisotopes for directed radiation treatment (Kitson et al., 2013), tethering of boron for boron neutron capture therapy (Wu et al., 2004), or attachment of cytotoxins to produce antibody-drug conjugates (ADCs) for oncology treatments (Kline et al., 2015). The use of site-specific conjugation techniques also facilitates the production of homogenous antibody conjugates with desired activity and pharmacokinetic profiles (Yin et al., 2017).
There are a variety of different technologies for producing antibodies with precisely defined conjugation sites including the use of an engineered cysteine (Bhakta et al., 2013), sugar modification (Pergolizzi et al., 2016), or introduction of small tags for enzymemediated conjugation including transglutaminase or aldehyde tags (Falck & Müller, 2018). Nonnatural amino acid (nnAA) mutagenesis offers a particularly advantageous method for site-specific bioconjugation because it can utilize biorthogonal chemistries, eliminating any off-target conjugation with naturally occurring amino acids (Axup et al., 2012). This technique requires mutation of only a single amino acid residue, allowing the conjugation site to be placed at any residue in the protein which does not impact folding, effector functions, or substrate interactions while preserving biological activity that could be affected by the introduction of a larger tag.
nnAA mutagenesis utilizes an amber suppressor orthogonal tRNA-AAtRNA (aminoacyl tRNA) synthetase system which incorporates a nnAA co-translationally in response to the TAG "amber" stop codon. In mammalian cell lines, clinical scale antibody production has been demonstrated with this system using stable cell lines with chromosomally integrated tRNA and synthetase (Axup et al., 2012). However, nnAA incorporation at TAG codons competes with translational termination.
This limits the efficiency of nnAA incorporation, precludes nnAA mutagenesis at difficult-to-suppress sites, and typically restricts the number of nnAA incorporated to one per protein chain (Axup et al., 2012). As an alternative to mammalian cell production, CFPS of nnAA containing IgGs offers some important advantages. First, the open nature of CFPS allows components of the nnAA incorporation machinery to be directly added to the reaction. This is particularly important for the nnAA itself which would otherwise need to cross the cell membrane in a cell-based expression system (Zimmerman et al., 2014). The translational machinery of E. coli is also easier to modify than mammalian cells, which facilitates knockout or engineering of factors responsible for translational termination, enabling the incorporation of multiple nnAAs and incorporation at challenging sites (Johnson et al., 2011;Yin et al., 2017). Finally, in CFPS, growth is decoupled from protein synthesis and proteins essential for growth such as RF1, which terminates translation at TAG codons, can be removed through proteolysis before in vitro protein synthesis. IgG expression in extract lacking RF1 allows efficient production of IgGs with up to 8 nnAAs (Yin et al., 2017).
One of the most important challenges with CFPS of nnAA IgGs at larger scales is access to sufficient quantities of extract. Extract production conventionally requires batch or fed-batch cell fermentations where cells are grown to a high OD in a bioreactor and the biomass is harvested in a single pass. Batch extract production is highly successful at smaller scales, however, it becomes limiting at the scales needed for commercial production of therapeutic antibodies in a CFPS system.
One way to address this challenge is to grow the extract strain chemostatically at high cell density and rates of division. This mode of fermentation has the potential to increase the E. coli cell extract output for a bioreactor 20 times (Kopp et al., 2019 and Scheme 1a) While in continuous fermentation mode, cells are grown much longer than typical batch or fed-batch fermentation and go through many more rounds of division. This necessitates strains with high genetic stability which is complicated by the fact that our previously engineered extract strains overexpress chaperones from a plasmid with antibiotic selection for plasmid maintenance (Groff et al., 2014). Antibiotic selection is less effective at high cell density and continuous fermentation would generate a large volume of antibiotic contaminated waste (Velur Selvamani et al., 2014). To address these issues, an auxotrophic GROFF ET AL. | 163 selection scheme was devised, based on glutamine auxotrophy, that enabled the retention of a plasmid for at least a week of continuous fermentation. However, over the course of this fermentation, instability within the expression cassette resulted in decreasing levels of FkpA. This problem was solved by integrating expression cassettes for both chaperones onto the chromosome, resulting in stable expression of both DsbC and FkpA at g/L quantities in the extract.
To facilitate site-specific ADC production with nnAAs it would also be desirable to produce the amber suppressor tRNA during extract fermentation. Our previous chaperone studies suggested that stable, highlevel expression of the amber suppressor tRNA should be achievable with genes transcribed from the chromosome (Groff et al., 2014). Initial studies with a synthetic orthogonal tRNA (o-tRNA) operon demonstrated that tRNA activity correlated very well with promoter strength. Unfortunately, it also became clear that even with the strongest promoter tested, a single integrated o-tRNA operon could not achieve tRNA levels sufficient for maximal amber suppression during cell-free nnAA-IgG synthesis. This is a common challenge in moving from a multicopy plasmid to a single copy chromosomal integrant. One elegant solution for dealing with this challenge called Chemically Induced Chromosomal Evolution utilizes recA to form up to 40 concatemerized gene or pathway duplications (Tyo et al., 2009). Implementing this approach offered a 10-fold stability enhancement relative to plasmid expression of the same pathway. However, over time the copy numbers were reduced by homologous recombination. From initial inoculum growth and week-long continuous fermentation, the extract strain may be growing for as long as 9 days, therefore necessitating even greater stability. This was achieved through the integration of the o-tRNA operon onto the chromosome at multiple defined sites which allowed us to spatially separate these genes and precisely control and optimize o-tRNA levels within the extract strain. This also demonstrates the utility in expressing factors for CFPS from the chromosome. Further improvements can be prototyped through plasmidbased overexpression before genes are moved onto the chromosome in an iterative process.
The final ADC-extract strain was capable of constitutive expression of the chaperones DsbC and FkpA and o-tRNA from genes integrated into the chromosome. All three were expressed at quantities sufficient for high-level nnAA-IgG synthesis in CFPS, simplifying the reaction setup and eliminating one reagent fermentation and lysate production needed for ADC production (Scheme 1b). Extracts produced from these cells showed consistent chaperone levels, IgG synthesis, and nnAA incorporation on the first and final days of continuous fermentation, indicating that these chromosomally integrated genes were stable. In addition, extract produced from continuously grown cells had cellfree performance comparable to batch-produced extract demonstrating that continuously growing cells for extract production is a viable route for increasing yields of cell-free extract-a key requirement for enabling commercial production of biotherapeutics by CFPS. For plasmid selection, media was supplemented with 50 μg/ml kanamycin and 100 μg/ml carbenicillin where appropriate. Complex media and antibiotics were purchased from Teknova. All plasmid construction was performed using the Choo-Choo homology cloning kit from McLab, unless noted otherwise. All polymerase chain reactions (PCRs) were performed using Phusion DNA polymerase from New England Biolabs. Strains utilized in this study can be seen in Table 1.

| Auxotrophic selection
Starting strains were made auxotrophic for amino acids by disrupting glnA, cysE, or argA, which is required for the biosynthesis of glutamine, cysteine, or arginine respectively from the genome of the cellfree optimized E. coli host. Gene knockouts were performed using homologous recombination with a selection marker as described previously (Datsenko & Wanner, 2000).
The complementing plasmids used in this study are based upon the medium copy ACYC origin of replication. This auxotrophic plasmid system has been used for the expression of FkpA and an o-tRNA for nnAA incorporation. An example of this type of plasmid is shown in Figure 1a. For cloning this variant of restoring plasmid, miniprepped pACYC-PC plasmid was cut at a single BsrBI site. glnA, cysE, or argA was amplified out of WT E. coli strain SBJY001 with PCR primers that had homology to sequence 3ʹ and 5ʹ of the BsrBI restriction site.
These PCR primers also introduced a strong constitutive promoter, CP9, or medium strength constitutive promoter, CP42, upstream of each gene, responsible for its transcription. This fragment was then cloned into pACYC-PC using choo-choo cloning (McLab).
To prepare strains with auxotrophic selection, mutants SBDG098, SBDG099, SBDG100 lacking glnA, cysE, or argA respectively, were made electrocompetent and then transformed with a complementing plasmid. The plasmid could be maintained in media with kanamycin or media lacking the specific amino acid.
Plasmid retention was measured with a plate assay. Stationary phase cultures were diluted 1 × 10 7 into LB broth and plated onto LB agar plates with and without kanamycin. The retention rate was calculated by dividing the number of colonies on the kan plates by the number of colonies that grew without antibiotics.
Growth assays were performed by picking colonies off LB-Agar plates, inoculating in MOPS minimal media with 0.1% glucose, and incubating at 37°C degrees overnight in a shaking incubator at 250 RPM. Samples that grew to an OD595 > 0.25 were considered positive for growth.

| Integration of FkpA and tRNA strain expression cassettes
A bacterial strain SBDG112 derived from SBHS016 was used as the initial source strain. The FkpA and tRNA expression strains were constructed using previously described methods (Groff et al., 2012).
To insert the tandem copies of FkpA into strain SBDG112, the FkpA

| Cell banking
A single colony is picked to inoculate 25 mL LB media in a Corning 125 ml Erlenmeyer flask incubated in 37°C, 250 rpm shaker. During the exponential phase (OD 595~1 ), 6.25 ml of 80% glycerol is added to the flask culture and 2 ml aliquots of the cell stock are flash-frozen in liquid nitrogen.

| Cell extract prepared from 200 ml chemostat fermentations
Cell extracts were prepared using derivatives of Sutro extract strain SBHS016, which has a protease-sensitive RF1 that is degraded during extract production. Strains were cultured to an OD 595 approximately 50, grown using glucose and an amino acid fed-batch continuous   isopropanol, pH 6.5. Two reactions were set up with extract produced continuously, and one reaction was set up using extract produced with a batch fermentation for comparison. Titers of all three reactions were measured as described for 1 ml scale in duplicate for each bioreactor.

| ADC conjugation and characterization
Polished aCD74 IgG was diluted to 1 mg/ml in PBS and incubated at room temperature overnight with a three-fold molar excess (40 μM

| Auxotrophic selection
Overexpression of the chaperones DsbC and FkpA in extract is required for the folding and assembly of IgGs during CFPS. It was previously demonstrated that more than 1 g/L IgG titers were achievable with an extract strain in which DsbC was overexpressed from the chromosome and FkpA was overexpressed from a medium copy plasmid with a p15A origin (Groff et al., 2014). Though we were initially interested in adapting this system for compatibility with continuous fermentation, antibiotic selection is not desirable in continuous fermentation because the high cell density reduces antibiotic concentration and decreases selection pressure for plasmid maintenance (Kopp et al., 2019). In addition, the large culture volumes associated with continuous fermentation produce a large volume of antibiotic-contaminated spent media. To address these issues, we have opted to maintain the plasmids in our continuous fermentation strains with auxotrophic selection by knocking out an essential gene required for amino acid biosynthesis and rescuing with gene expression from a plasmid (Figure 1a).
The media for continuous fermentation is a defined media with mineral salts, glucose, vitamins, and 13 amino acids. That leaves 7 amino acids that could potentially be targeted by auxotrophic selection. Of these remaining amino acids, we identified three whose biosynthesis could be disrupted by deletion of a single gene.
These three genes were cysE for cysteine biosynthesis, glnA for glutamine biosynthesis, and argA for arginine biosynthesis. Initially, we verified that mutants with these single-gene knockouts were unable to grow in our continuous media (data not shown). To test the ability of each auxotrophic system to work for plasmid selection, strains with disrupted amino acid biosynthesis pathways were transformed with the complementing plasmid and grown in minimal media. In Figure 1b, the auxotrophic selection system maintained the plasmid in nearly 100% of cells for all three amino acid systems, comparable to antibiotic selection with both promoters, (CP9 and CP42) indicating the selection was robust with respect to GlnA expression levels. In the absence of selection, kanamycin-resistant cells were not detectable indicating the plasmid was lost from these cells.
Previous research indicates plasmid selected by auxotrophy will also be unstable if they don't confer a growth advantage (Dong et al., 2010). Figure 1c and S2 shows each of the strains with restored amino acid biosynthesis had growth rates similar to the prototrophic parental strain in minimal media, while the untransformed strains did not grow (only the Gln auxotroph is shown for simplicity). We were also interested in testing whether this auxotrophic selection system would function in complex media which would contain all 20 amino acids. Figure 1d shows the growth of the untransformed auxotrophic FkpA in these cells is expressed from a p15A plasmid which should have a copy number 10-15 times higher than the chromosome.
However, during the course of the continuous fermentation, the FkpA levels for SBDG150 were relatively stable at around 2 g/L or about 50% of the FkpA titers from the plasmid-based FkpA production at its highest time point. After 4 days of continuous fermentation, extract made from these cells had higher FkpA concentrations than plasmid expressed FkpA in SBDG161. Overexpressing both chaperones off the chromosome allowed continuous cell growth for extract production and stable, high-level production (>1 g/L) of two different chaperones for at least 5 days.

| Chromosomal expression of o-tRNA
Site-specific ADC production with CFPS requires co-translational incorporation of nnAAs (Yin et al., 2017). This requires three additional components in the cell-free reaction: nnAA with an orthogonal bioconjugation handle, an amber suppressor tRNA not recognized by endogenous E. coli synthetases, and an aminoacyl tRNA synthetase capable of specifically recognizing the nnAA and charging it onto the amber suppressor tRNA (Wang et al., 2001).
Producing either of the macromolecular components for nnAA incorporation in the extract strain would eliminate a reagent, simplify the reaction, and reduce the costs of producing ADCs with CFPS.
First, we wanted to focus on the amber suppressor tRNA because it is the most resource-intensive reagent. Endogenous tRNA and amber suppressor tRNAs are typically expressed from tRNA operons with a tRNA promoter and terminator (Wang et al., 2001). For our first set of experiments, we wanted to test whether one could transcribe an nnAA tRNA with an alternative promoter during extract fermentation. A small library was created with the natural proK tRNA promoter, or strong synthetic promoters including the MTL promoter, PL6 promoter, and a mutant MTL promoter with a 1 bp deletion in the −10 site, see Table S1 for sequences and details, which were used to drive GFP production off a plasmid with the pheV terminator. The results from the GFP expression experiment can be seen in Figure 3a. copies of genes onto the chromosome to achieve plasmid-like copy numbers, but these methods suffer from a lack of control over the exact copy number of integrated genes (Tyo et al., 2009). Excess o-tRNA levels had been observed to be deleterious for cell-free reaction titers, so we were interested in a precise o-tRNA titration.
This was achieved by sequentially integrating single o-tRNA expression cassettes in defined locations within the chromosome as single, or tandem tRNA constructs using the MTL promoter to drive transcription. After four more integrations with the cassette containing one or two copies of tRNA, mutants with exact tRNA copy numbers ranging from 1 to 9 were produced. Each of these cell lines was grown in a chemostat then used to prepare extract and used for cell-free reactions producing aCD74 F404pAMF IgG. From Figure 3c, it can be seen that strains with three or fewer integrated tRNAs resulted in nnAA containing IgG titers which were lower than the control extract, e224, which had tRNA expression driven from a plasmid. For these strains, suboptimal o-tRNA activity was limiting amber suppression. Strains with 4-6 integrated tRNAs had optimal aCD74 expression and nnAA IgG titers at or above strain S224 with plasmid-based o-tRNA transcription. Additional copies of tRNA beyond this leads to statistically significant decreases in titer. This may be a result of having too much o-tRNA which has been previously reported to negatively affect the titers of proteins in CFPS (Zimmerman et al., 2014). Strain SBDG299 with four individual integrations of the pAzPhe tRNA was chosen for further work because this o-tRNA configuration maximized titer, amber suppression, and specific growth rate (µ > 0.6 h −1 ) with the fewest chromosomal alterations and o-tRNAs (Figure 3c).

| Comparability of continuous extract quality
Strain SBDG299 was capable of stably co-expressing DsbC, FkpA, and o-tRNA during a multiday continuous fermentation. As a final test, we wanted to assess whether extract made from this strain with continuous fermentation produced ADCs with titers and quality similar to extracts produced in batch fermentation. Extract from SBDG299 was made using either batch or continuous fermentation and then used to produce aCD74 F404pAMF IgG. Table 3 shows that extract generated with continuous fermentation can produce nnAA-IgG with titers at or above the levels of extract made with batch fermentation. The nnAA-IgG made in these reactions was purified with protein A affinity chromatography, polished with SEC chromatography, and then conjugated with a cytotoxic maytansine warhead (Figure 4a). The assembly status of the purified IgG was monitored at each step using SEC and the conjugation efficiency was measured with MS. The percentage of fully assembled IgG, partially assembled fragments and aggregate were similar in the proA capture and polished material with both sources of extract.
Each extract produced good quality ADCs with 96% of the final product as fully assembled IgG and DAR > 1.90 (Table 3,

| DISCUSSION
The manufacture of protein biologics with CFPS would be facilitated by having access to the large volumes of extract enabled with continuous fermentation. This study demonstrates that it is possible to stably grow cells that have been engineered for cell-free synthesis of nnAA IgGs in a chemostat. For the production of biologics, day-to-day consistency is important. We discovered that while it is possible to achieve very high titers of chaperone expression from a plasmid, this system cannot achieve stable expression over the course of a five-day continuous fermentation.
On the other hand, the chromosome is a privileged location for the overexpression of these genes with much-improved stability. We have demonstrated that it is possible to achieve chaperone co-expression at levels more than 1 g/L during continuous fermentation over a period of 5 days. This level of chaperone expression is sufficient to support high-level IgG expression and assembly in CFPS (Groff et al., 2014).
T A B L E 3 Titer, product quality, and conjugation of aCD74 ADC produced in batch and continuous extract  Of crucial importance, the synthesis and performance of nnAA IgGs and corresponding ADCs appear identical with cellfree reactions using either continuous extract or batch extract.
SBDG299 extract produced with continuous fermentation leads to cell-free titers at least as good as extracts made using batch fermentation. The quality of the IgGs produced in terms of fully assembled IgG, aggregate, and partially assembled IgG is very similar with both extract production methods. The conjugation and performance in in vitro cell killing assays for ADCs made with either method are indistinguishable. Continuous fermentation can produce high-quality extract over extended periods of time with stable co-expression of three different macromolecular components for nnAA IgG synthesis from the chromosome. Access to large volumes of high-quality extract will be critical for enabling economical CFPS of ADCs at manufacturing scales.

CONFLICT OF INTERESTS
The authors declare that there are no conflict of interests. Carlos, Jeffrey A. Hanson, Xiaofan Li, and Gang Yin wrote the manuscript. All authors helped to revise the manuscripts.

DATA AVAILABILITY STATEMENT
All data are available upon request.