Evaluation of “Difficult‐to‐Express” Monoclonal Antibodies in a CHO‐Based Hybrid Site‐Specific Integration System Under Industrially Relevant Conditions

ABSTRACT

Variation in the primary sequence of monoclonal antibodies (mAbs) can negatively affect their behavior in biopharmaceutical manufacturing platforms, and efforts to identify mAbs with poor “developability” characteristics lack robust methods for assessing mAb expression from an industrially relevant platform. Recent advancements in site‐specific integration‐based (SSI) platforms in Chinese hamster ovary (CHO) cells can mitigate the high transcriptional variation observed with random integration and the low industrial relevance of transient expression by providing a flexible platform for mAb expression from a consistent clonal background. This work applies a novel SSI‐based expression system capable of generating isogenic cell pools in less than 1 month to systematically compare the expression of ten sequence variants of two therapeutically relevant mAbs from two genomic loci under industrially relevant culture conditions. Eight single amino acid mutations in trastuzumab resulted in reduced productivity compared to the wild‐type mAb in batch cultures, and three mutations maintained a low‐expressing phenotype in fed‐batch cultures. The mutations resulted in variant‐specific patterns of decreased domain stability and increased ER stress. The application of industrially relevant SSI systems in developability workflows could strengthen the understanding of the sequence determinants of mAb expression to improve mAb design, candidate selection, and process development decisions.

Graphical Abstract and Lay Summary

Improved understanding of the primary sequence features characteristic of “difficult‐to‐express” mAbs can improve mAb design, candidate selection, and process development decisions. This work uses a flexible recombinase‐mediated cassette exchange system for rapid comparison of low‐expressing mAb sequence variants at two genomic loci for two clinically‐relevant mAbs. Characterization of titer and intracellular bottlenecks for batch and fed‐batch cultures showed variant‐specific patterns of reduced expression, decreased domain stability, and increased ER stress.

1. Introduction

Monoclonal antibodies (mAbs) continue to dominate the biopharmaceutical market, with almost 1200 antibody‐derived therapies entering the clinic as of November 2022, driving efforts to increase the speed at which mAbs can enter this competitive clinical landscape [1, 2]. Reducing development timelines has necessitated the establishment of better platforms for evaluating molecule “developability” characteristics, which can determine the ease with which a mAb can be produced, purified, and formulated through today's manufacturing processes [3]. While expression has been recognized as a key developability metric, it has remained largely unstudied in industrially relevant systems [4]. Previous work has relied on transient expression as a rapid, high‐throughput platform for large developability assessments to avoid the low throughput and high clonal variability of random integration (RI) cell lines in Chinese hamster ovary (CHO) cells, which remain the industry platform for mAb production [5, 6].

Recent developments with site‐specific integration (SSI) aim to reduce the clonal heterogeneity observed with RI, where the inherent genomic instability of CHO cells and epigenetic factors determined by gene integration location can lead to clone‐specific patterns of production instability, defined by decreases in transgene expression with increasing cell age [7]. As a result, global regulatory agencies require lengthy clone stability studies with RI cell lines to ensure that the selected clone is capable of high mAb production over time. SSI systems based on hybrid mechanisms allow for the insertion of a genetic “landing pad” (LP) construct at a pre‐selected genomic locus with proven capabilities for stable and high transcription that can then be rapidly exchanged for a mAb expression cassette using recombinase‐mediated cassette exchange (RMCE) [8]. mAb‐expressing cell pools generated via RMCE typically show improved transcriptional and expression homogeneity at the clone level, reducing the size of clone screening workflows required for the identification of a cell line suitable for large‐scale manufacturing [9, 10, 11]. While LP/RMCE systems may not eliminate the need for cell line stability testing, the features of these systems allow for stability testing to be performed “off the critical path,” especially if the chosen genomic locus has a proven history of stability, providing significant time savings for cell line development [12].

These RMCE‐based SSI systems provide a convenient option for rapidly comparing the expression of mAb constructs and expression‐enhancing genetic elements in a similar format to those typically used for clinical and commercial production. This feature makes them an attractive alternative to both transient expression and RI by simultaneously preserving industrial relevance and increasing experimental throughput. Previous work focused on characterizing “difficult‐to‐express” (DTE) mAbs using stable expression systems relied on cell lines generated through RI; however, the transcriptional heterogeneity associated with RI made it difficult to deconvolute transcriptional patterns from expression and often involved lengthy single cell cloning workflows to characterize clonal distributions, limiting the throughput of these studies [13, 14, 15]. Alternatively, work relying on transient expression has characterized intracellular protein production bottlenecks at the mechanistic level, but cannot assess these bottlenecks under industrially relevant fed‐batch conditions due to the short‐lived nature of transient expression [16, 17]. The utility of SSI systems for determining the underlying mechanisms of a DTE mAb has recently been demonstrated with a targeted integration‐based inducible cell line; however, the details of that system, including the genomic integration location and mAb primary amino acid sequences, were not disclosed [18].

This work presents the application of a previously established hybrid LP/RMCE system in a CHO‐K1 host for the study of DTE mAb variants in a stable expression format at an unprecedented scale [19]. The RMCE‐based system allows for rapid, efficient, and reproducible integration of mAb variants at known genomic loci using only chemical selection methods in less than 1 month, which increases throughput and reduces timelines compared to RI or de novo CRISPR/Cas9 gene integration and subsequent clone selection. This approach allows mAb variants to be expressed from a consistent clonal and transcriptional background using industrially relevant culture conditions to study the post‐transcriptional bottlenecks and cellular responses to expression characteristic of cell lines used for mAb manufacturing. Through this work, ten single amino acid mutations of two IgG₁κ mAbs were expressed from two different genomic loci, resulting in the comparison of 44 distinct expression conditions. Eight variants of trastuzumab showed lower expression as compared to the wild‐type (WT) mAb at both loci, and three of these variants displayed low productivity even under fed‐batch conditions. Evaluation of intracellular protein assembly products and ER stress markers in trastuzumab fed‐batch samples revealed unique patterns in cellular responses that may also be characteristic of bottlenecks in large‐scale producer cell lines. These findings show that LP/RMCE systems represent a valuable tool for developability assessment workflows, as these systems can reveal valuable expression patterns for supporting candidate selection and de‐risking process development decisions.

2. Materials and Methods

Kabat numbering for antibody sequences is used throughout the text.

2.1. LP/RMCE SSI System

The development of the LP/RMCE system and generation of two LP host cell lines, R26‐C9 and D1b‐P1C11, has been described previously [19]. Briefly, both cell lines contain a genetic landing pad construct containing a thymidine kinase (TK) gene and an enhanced monomeric near‐infrared fluorescent protein 670 (emiRFP670, RFP) marker fused to a blasticidin S deamidase (BSD^R) gene. The landing pad contains orthogonal recombinase recognition sites to provide RMCE capability using the Bxb1 serine recombinase. Landing pads were inserted using CRISPR/Cas9‐mediated insertion at the Rosa26 locus and the Dop1b locus. Cells were selected with Blasticidin S HCl (Gibco) then single‐cell cloned to identify host cell lines (R26‐C9 and D1b‐P1C11 for LP insertions at Rosa26 and Dop1b, respectively) with single‐copy on‐target integrations and high levels of RFP expression.

2.2. Molecular Cloning of RMCE Cargo Cassettes

RMCE cargo plasmids expressing mAb sequence variants for trastuzumab or adalimumab were generated as previously described using standard molecular cloning techniques [20]. Gene fragments containing the mutations of interest were purchased from Integrated DNA Technologies (IDT). Plasmids were transformed into E. coli TOP10 cells (Thermo Fisher Scientific) and prepared using a ZymoPURE Plasmid Miniprep Kit or a ZymoPURE Plasmid Midiprep kit (Zymo Research). Prior to use, all plasmids were sequence confirmed using Sanger sequencing at the University of Delaware Sequencing and Genotyping Center.

2.3. Cell Culture

CHO‐K1 cell cultures were maintained in ActiPro medium (Cytiva) supplemented with 6 mM L‐glutamine (Fisher). Shaken cultures were grown at 37°C, 5% CO₂, 80% humidity (Infors MultiTron, orbital diameter = 25.4 mm). Cultures in 24‐deep well plates (square well, v‐bottom, Axygen) were maintained at a culture volume of 2 mL and shaken at 200 rpm. Cultures in 125 mL flasks (Corning) were maintained at 25 mL and shaken at 135 rpm. Cultures were routinely passaged every 2–3 days in 50 mL mini bioreactor tubes (Chemglass NEST) with a working volume of 10 mL and shaken at 200 rpm. Cell counts were obtained using a Vi‐CELL XR cell viability analyzer (Beckman Coulter). For all transfections, the appropriate host cells were passaged to 1 × 10⁶ cells/mL with a complete media exchange on the day prior to transfection.

2.4. RMCE Pool Generation and Batch Analysis

Stable RMCE pools were generated by the co‐transfection of the pCAG‐NLS‐HA‐Bxb1 plasmid (a gift from Pawel Pelczar, Addgene plasmid #51271) and the appropriate RMCE cargo plasmid at a Bxb1:cargo molar ratio of 1:6. Transfections were performed using 2 × 10⁶ cells per transfection and 250 fmol total DNA per million cells. All conditions were transfected in biological triplicate. Cells were passaged every 3–4 days to approximately 1 × 10⁶ cells/mL in 24‐deep well plates in the presence of 500 µg/mL geneticin (Gibco) or 500 µg/mL geneticin and 1 µM ganciclovir (Sigma) (Figure S1).

Pools selected on geneticin only were passaged to 0.5 × 10⁶ cells/mL in 24‐deep well plates in the absence of selection on day 14 post‐transfection for batch analysis. On Day 4 of batch culture, supernatant samples were taken for titer analysis, mRNA, and genomic DNA. mRNA was extracted using a Quick RNA‐96 kit (Zymo Research) with on‐column DNAse I digestion, and genomic DNA was extracted using an in‐house protocol (Supplemental Methods). The productivity (qP, pg/cell‐day) of each culture was calculated by dividing the final Day 4 titer by the integral of viable cell density (IVCD), with IVCD determined using a trapezoidal approximation from Day 0 and Day 4 cell counts. All productivities were then normalized by setting the average qP of the WT cultures expressed from the same genomic locus to 1.

2.5. Fed‐Batch Expression Studies

RMCE pools selected with geneticin and ganciclovir were passaged without selection for at least one passage before being seeded at 0.8 × 10⁶ cells/mL in 25 mL of ActiPro for fed‐batch studies in shake flasks. Cells were counted daily starting on Day 2. Starting on Day 3, cells were fed using Cell Boost 7a and 7b feeds (Cytiva) and a pyramid feeding schedule as previously described for the VRC01 cell line [21]. Additional glucose from a concentrated solution (Sigma) was fed to a total of 6 g/L on Days 3, 4, and 5, then increased to 9 g/L from Day 6 onwards [21]. Glucose, lactate, glutamine, and glutamate concentrations were determined daily using a YSI 2900 metabolite analyzer starting on Day 3. Culture temperature was decreased to 32°C starting post‐feeding on Day 4 [22, 23]. On Days 4, 6, 8, and 10, mRNA from 5 × 10⁶ cells was extracted using a Quick‐RNA Miniprep Kit (Zymo Research) with on‐column DNAse I digestion. Intracellular proteins from 10 × 10⁶ cells were extracted using RIPA buffer (Thermo Fisher Scientific) supplemented with HALT protease and phosphatase inhibitor (Thermo Fisher Scientific) and 1 mM EDTA. Pellets were incubated with prepared RIPA buffer (100 µL buffer per million cells) at −20°C for at least 15 min, then centrifuged at >18,000 × g for 10 min to remove cellular debris. Fed‐batch productivities for each culture were calculated by determining the slope of a linear regression model fit to IVCD versus titer data.

2.6. Titer Determination

Antibody titers were determined on an Octet 96e (Sartorius) using Protein A biosensors (Sartorius). Supernatant samples were diluted 2× (batch samples) or 100× (fed‐batch samples) in Dilution Buffer (Dulbecco's PBS with 0.1% BSA). A standard curve between 50 and 0.391 µg/mL was generated by diluting a 700 µg/mL Protein A standard (Sartorius) to 50 µg/mL in Dilution Buffer with 50% ActiPro (batch samples) or 1% ActiPro (fed‐batch samples) to match the sample buffer matrix, then making serial 2× dilutions. Samples were loaded in technical triplicates into black, flat‐bottom polypropylene 96‐well plates (Corning or Gator Bio). In between samples, biosensors were regenerated in 10 mM glycine, pH 1.5, and neutralized in Dilution Buffer with ActiPro at the same percentage as the samples. Binding rate values were obtained using the pre‐built high‐sensitivity assay (5 min read time, 1000 rpm shaking, 30°C), and the standards were fit to a “Dose Response – 4PL Unweighted” equation for concentration determination.

2.7. Droplet Digital PCR Analysis for Gene Expression Analysis

Droplet Digital PCR (ddPCR) analyses were performed on a QX ONE ddPCR system (Bio‐Rad) using custom‐designed, multiplexed assays (IDT, Table S1). Expression of light chain and heavy chain mRNA was determined using the One‐Step RT ddPCR Advanced Kit for Probes (Bio‐Rad) according to the kit protocol. Approximately 0.25–0.5 ng total mRNA was targeted per ddPCR reaction. Antibody gene expression was normalized to the expression of the housekeeping gene Rab10. All primer/probe assay sets were purchased pre‐mixed at a 3.6:1 ratio (900 nM primers to 250 nM probes, IDT).

2.8. Characterization of Intracellular Bottlenecks and Thermal Stability

Trastuzumab fed‐batch samples were subject to confocal microscopy and colocalization analysis, Western blotting, XBP1 gene expression analysis with qPCR, and differential scanning fluorimetry (DSF) using in‐house derived methods provided in the Supporting Information.

2.9. Statistical Analyses

All statistical analyses were performed and plots generated in R (v 4.0.4) using the Tidyverse family of packages (v 1.3.0). Unless otherwise noted, error bars represent one standard deviation from the mean of n biological replicate cultures, with n indicated in figure captions. The statistical tests used are provided in the figure captions or main text, with p < 0.05 considered significant for all tests.

3. Results

3.1. LP/RMCE System Can Be Used to Study the Expression of DTE Trastuzumab Variants

The establishment of an optimized mAb production platform based on RMCE allows for the expression of mAb variants to be studied under a consistent clonal background, without the position‐dependent transcriptional effects present in RI cell lines and without the lengthy timelines and extensive clone screening that would be required from de novo Cas9‐mediated gene insertions. Ten trastuzumab variants that were previously identified as low‐expressing in a transient expression system were produced using an optimized RMCE cargo cassette containing the CpG island from the Azin1 locus and the Piggybac transposase 5’ inverted terminal repeat in between the LC and HC genes [19, 20]. Plasmids containing the variants of interest (Table 1) were recombined into two landing pad‐containing host cell lines, R26‐C9 and D1b‐P1C11, which contain a RMCE‐capable landing pad integrated at the Rosa26 and Dop1b loci, respectively, as described previously [19]. mAb expression was measured from a four‐day batch process following 10 days of geneticin selection.

TABLE 1.

Equivalent single amino acid mutations tested in trastuzumab and adalimumab, in the format [wild‐type amino acid] [Kabat position] [mutated amino acid].

Construct	Trastuzumab mutation	Adalimumab mutation
LC1	F62S
LC2	V33K	L33K
LC3	P80I
LC4	G68F
LC5	F71Y
HC1	F67S
HC2	I34K	M34K
HC3	L11A
HC4	I51A
HC5	Y59S

Overall cellular productivities (qP, pg/cell‐day) were lower from the D1b‐P1C11‐derived pools as compared to the R26‐C9 pools, as previously observed, but the rank order of variant expression was maintained across sites (Figure 1A). Productivities at each site were normalized to the expression of the WT mAb, with nine variants at Rosa26 and eight variants at Dop1b identified as low‐expressing (Figure 1B,C). At both sites, HC2 was the lowest expressing variant, showing a 3‐fold reduction in qP at Rosa26 and a 2‐fold reduction in qP at Dop1b as compared to WT. Most other significant variants had productivities between 60% and 80% of WT, with HC3 identified as the only variant at both sites to show no significant difference in qP. Analysis of LC and HC mRNA expression for all pools showed that differences in LC or HC transcription alone could not account for the 3‐fold range of qP observed across the dataset (Figure 1D,E). In general, LC mRNA expression showed no correlation with qP. HC mRNA expression was weakly correlated with qP, with the highest productivity variants also showing the highest levels of HC transcription, however, the variants with the lowest HC transcription levels showed qP values differing by up to 2‐fold. HC mRNA expression was more variable than LC mRNA expression, especially amongst variants containing mutations in the HC. Additional mRNA expression measurements combined with a larger variant panel would help verify this apparent trend.

FIGURE 1.

Trastuzumab variants expressed in RMCE pools under batch conditions. (A) Average productivities from RMCE pools expressing wild‐type (WT) trastuzumab or trastuzumab variants in the D1b‐P1C11 or R26‐C9 cell lines show a consistent rank order across sites, despite lower expression at Dop1b. Points indicate the mean productivity of biological triplicate RMCE pools. Batch pool productivities were normalized to the average WT expression at (B) Rosa26 and (C) Dop1b, with eight of the ten variants showing significantly lower expression compared to WT at both sites (*p < 0.05, Dunnett's test). mRNA expression of trastuzumab LC and HC genes from (D) Rosa26 and (E) Dop1b in batch conditions show no correlation with productivity. Relative mRNA expression of each transcript to the housekeeping gene Rab10 (expression normalized to 1, dotted line) was determined using multiplexed ddPCR assays. Each pool was measured with technical duplicates. Primer/probe sequences are provided in Table S1. Error bars represent one standard deviation from the mean of biological triplicate pools.

Low‐productivity phenotypes for trastuzumab are also observed in fed‐batch cultures despite similar growth, metabolic, and transcriptional profiles. R26‐C9‐derived RMCE pools expressing trastuzumab variants that had undergone both geneticin and ganciclovir selection were expanded and cultured under fed‐batch conditions in shake flasks to determine if the productivity phenotypes observed in batch cultures also translated to a more industrially relevant and metabolically favorable culture format. Of the ten variants tested, six variants (LC1, LC2, LC3, HC1, HC2, and HC4) were selected for further characterization and grown alongside WT RMCE pools and pools expressing an empty cargo vector (Null) containing the geneticin selection gene only. Cells were grown for 11 days with daily feeding starting on Day 3 and a temperature shift to 32°C starting on Day 4. Cultures showed very similar growth profiles across biological replicate RMCE pools and across variants, with cultures reaching a maximum VCD of >40 × 10⁶ cells/mL (Figure S2A,B). The mAb‐expressing pools grew similarly to the Null cultures, showing that exogeneous protein expression did not affect growth and metabolite profiles. All measured metabolite concentrations (glucose, lactate, glutamine, and glutamate) were also similar across conditions, with the differences observed at the end of culture likely attributable to differences in culture viability on Day 11 (Figure S2C–F). Titers from these fed‐batch cultures ranged from approximately 600 mg/L for the lowest‐expressing variants to 1200 mg/L for the WT cultures (Figure 2A). Despite similar growth and metabolic profiles, three of the six variants (LC2, HC1, and HC2) still showed reduced productivities as compared to WT trastuzumab (Figure 2B). Once again, the HC2 variant was the lowest expressed, with a 2‐fold lower qP compared to WT.

FIGURE 2.

Trastuzumab variant fed‐batch results. (A) Fed‐batch titers were measured starting on Day 6 with Octet Protein A biosensors in technical triplicate. (B) Productivities for each culture were determined by fitting a linear regression model (R ² > 0.92 for all models) to the integral of viable cell density (IVCD) versus titer from Day 6 to Day 11. Three variants show lower expression compared to wild‐type (WT) trastuzumab (*p < 0.05, Dunnett's test). (C) mRNA expression of the LC and HC genes and (D) HC:LC transcript ratio over the course of fed‐batch culture was monitored starting on Day 4 using multiplexed ddPCR assays in technical duplicate, with samples normalized to the expression of Rab10. Error bars show one standard deviation from the mean of biological triplicate pools. Primer/probe sequences are provided in Table S1.

Analysis of LC and HC mRNA expression from samples taken on Day 4, 6, 8, and 10 of the fed batches showed a progressive increase in transcription over the course of the culture, with samples on Day 10 showing 25–35× higher LC expression compared to Rab10 and 50–80× higher HC expression compared to Rab10 (Figure 2C), but an approximately constant ratio of HC:LC transcripts throughout culture (Figure 2D). Differences in transcriptional profiles between variants alone could not account for the differences in productivity, and taken together with the batch experiment data, these results further demonstrate that the productivity differences observed are consistent with post‐transcriptional bottlenecks.

3.2. Bottlenecks to mAb Production are Post‐Translational

To further investigate the mechanisms contributing to the observed low‐expression phenotypes, samples from fed‐batch cultures were analyzed for mAb assembly products, thermal stability, ER stress responses, and colocalization of mAb polypeptides with subcellular organelles. Western blot analysis of the intracellular assembly products showed that the HC1 and HC2 pools had a distinct decrease in the amount of HC dimer present, with HC2 pools showing almost no detectable HC dimer on Day 8 (Figure 3A). An unknown heavy chain‐containing product was present in the WT and LC mutation conditions but was absent from all HC mutant pools. For all conditions, very little intracellular light chain species were observed.

FIGURE 3.

Assessment of intracellular assembly products and ER stress responses in trastuzumab fed‐batch cultures samples on Day 8. Variant and pool replicate identifiers are included above the blots. (A) Intracellular protein lysates were subjected to non‐reduced SDS‐PAGE and Western blotting to detect distributions of intracellular antibody assembly products. Lysate from approximately 6666 cells was loaded in each lane. An unidentified HC‐containing species was observed in four of the conditions (*). β‐actin was used as a loading control (mouse anti‐β‐actin = 2500×, donkey anti‐mouse IgG = 5000×). (B) Activation of UPR markers and protein chaperones was measured by Western blot analysis using antibodies specific for the indicated markers. Intracellular protein samples were subjected to reducing conditions, and lysate from approximately 66,666 cells was loaded in each lane, with β‐actin used as a loading control (mouse anti‐β‐actin = 10,000×, donkey anti‐mouse IgG = 20,000×). qPCR analysis to detect (C) spliced and (D) unspliced forms of XBP1 mRNA. RNA samples were extracted from 5 × 10⁶ cells on Day 8. Relative expression was calculated using the ΔΔC^T method with Rab10 as a reference gene and the Null‐1 pool for normalization (dashed line) [61]. Gray points indicate the pool averages from four technical replicates, with horizontal lines and error bars representing the average and standard deviation of the biological triplicate pools. Average expression values that differ significantly from the WT pools (two‐tailed Student's t‐test) are indicated (*p < 0.05). Primer/probe sequences are provided in Table S1, and detection antibodies are provided in Table S2.

Cell lysates and mRNA samples isolated on Day 8 of fed‐batch culture were assayed for various downstream ER stress response markers across the three branches of the unfolded protein response (UPR), with PERK signaling detected with ATF4 and CHOP expression, ATF6 activation monitored with PDI, GRP94, and ERp57, and IRE1 activation measured with qPCR analysis of XBP1 mRNA splicing (Table S2) [24, 25, 26]. BiP (GRP78) is a well‐known chaperone for promoting the folding of immunoglobulin chains and is believed to be the primary signaling protein for activating the UPR [27]. Through the coordinated regulation of these proteins, which act as folding chaperones (PDI, GRP94, ERp57, BiP) or transcription factors (ATF4, CHOP, XBP1), cells can either expand its secretory capacity, activate pro‐apoptotic responses, or modulate metabolic processes to restore homeostasis in the presence of accumulated misfolded proteins in the ER [28].

Western blotting showed similar levels of BiP across mAb‐expressing conditions, as might be expected from cultures producing high levels of exogeneous protein (Figure 3B). Comparison of PDI showed elevated protein expression in cultures expressing LC3, HC1, and HC2, and ERp57, another member of the PDI family, showed very low expression in most pools and no expression in Null cultures. HC2 showed the strongest ERp57 signal across replicates. Another folding chaperone, GRP94, was also expressed in all cultures, with HC2 once again showing the highest expression compared to both WT and Null cultures. Low levels of ATF4 activation were detected in all mAb‐expressing cultures, with almost no ATF4 expressed in Null conditions, suggesting no preferential activation in response to any variant. Despite high culture viability on Day 8, CHOP expression was observed in all cultures and showed the inverse trend to PDI, with LC3, HC1, and HC2 showing the lowest CHOP expression across biological triplicates. The strongest CHOP signals were observed in the Null and WT conditions. qPCR analysis of spliced XBP1 mRNA (XBP1s) and unspliced XBP1 (XBP1us) showed increased expression of XBP1s across all cultures relative to the Null conditions, and upregulation in the HC1 pools compared to WT pools (Figure 3C). Relative expression of XBP1us remained similar across pools, with slightly reduced expression of XBP1us in HC2 pools (Figure 3D).

Subcellular organelles including the ER, Golgi, ribosomes, secretory vesicles, lysosomes, and endosomes were stained with organelle‐specific markers in combination with either the LC or HC to visualize the colocalization of antibody chains within the cell using confocal microscopy [29]. Staining patterns of the ER showed no abnormal structures that would be indicative of Russell body formation, as has been previously observed (Figures S3 and S4) [16, 17, 29]. As expected, the most significant colocalization signal was observed between the mAb chains and the ER, due to ER‐localized mAb folding and assembly reactions. Interestingly, the distribution of chains within each cell varied between variants and chain type. Light chain staining intensity was uniform across cells and across variants; however, heavy chain staining intensity varied more from cell‐to‐cell, with some cells showing almost no detectable HC signal, especially in the LC2, LC3, and HC4 variants (red arrows, Figure S4). Staining for the other organelles showed no significant or preferential colocalization with the mAb chains and no structural abnormalities (data not shown).

Thermal stability as determined by DSF revealed three distinct transition patterns across the variants (Figure 4, Table S3). The first pattern corresponded to variant LC3 and WT trastuzumab, with two clear peaks in the fluorescence profile at similar temperatures. For both conditions, transition temperatures at 68.7°C–68.9°C and 79.8°C–79.9°C were observed, similar to the melting temperatures previously reported for trastuzumab from DSC analysis [30]. The second pattern, including variants HC1 and HC4, was characterized by a shifted main peak with a leading shoulder, where the transition temperature of the shoulder peak corresponded to T₁, and the main peak possibly corresponding to a lower T₂. In this pattern, the T₁ values were both within 1°C of the WT value but the T₂ transition showed a distinct shift of 4.8°C–5.6°C below the WT T₂. The third pattern, including variants LC1, LC2, and HC2, shows a single main peak with a long tail and only one identifiable transition temperature, likely corresponding to T₁. LC1 showed a similar T₁ to WT (68.3°C), but LC2 and HC2 showed values 1°C–2°C less than WT, with LC2 showing the lowest T₁ (66.5°C) of all variants. The presence of a single transition temperature suggests a significant decrease in T₂ in these samples.

FIGURE 4.

Characterization of trastuzumab variants by differential scanning fluorimetry (DSF). Antibodies produced through fed‐batch culture were purified, buffer exchanged, and diluted to 1.5 mg/mL in a 10 mM histidine pH 6.5 buffer for DSF analysis with SYPRO Orange in technical duplicate. Representative baseline‐subtracted fluorescence values (RFU) are shown, calculated by subtracting the signal at each temperature from a buffer‐only condition. The transition profile for wild‐type trastuzumab is overlaid in gray in each plot.

3.3. Transcriptional Bottlenecks in Adalimumab Pools Likely Limit Productivity

The equivalent single amino acid mutations were introduced into the coding sequences of the adalimumab LC and HC genes to study how these mutations affect productivity across IgG₁κ mAbs (Table 1). These variants were expressed using the same optimized cargo plasmid design in both cell lines and measured for productivity after four days in batch cultures (Figure S5A). Similar to previous work, adalimumab productivities were observed to be at least 3‐fold lower than trastuzumab productivities, with these differences driven by lower titers, not changes in cell growth [19].

Interestingly, most of the variants studied showed either no difference in productivity as compared to WT adalimumab or nominal increases in productivity, with no consistent pattern of expression observed across cell lines (Figure S5B,C). Gene expression analysis revealed similarly low HC and LC transcription as compared to previous observations, which may contribute to low productivity (Figure S5D,E). The HC2 variant was the only variant to show significantly lower expression compared to WT adalimumab. Fed‐batch cultures of six variants (LC1, LC2, LC3, HC1, HC2, HC4) in adalimumab expressed in the R26‐C9 cell line confirmed the observations seen in the batch experiment, where despite similar growth and metabolic profiles (Figure S6), no differences or nominal increases in productivity were observed for all variants tested (Figure S7A,B). The LC2 and HC2 adalimumab variants had the highest productivities, in contrast to the trastuzumab results, where these two variants had some of the lowest productivities measured in batch and fed‐batch experiments. Analysis of mRNA expression of adalimumab LC and HC genes over the course of fed‐batch culture showed 20–30× expression of both genes relative to Rab10; however, HC transcription continued to increase between Day 8 and Day 10, whereas relative LC transcription remained approximately constant between these days (Figure S7C). As a result, the HC:LC transcript ratio increased across the course of adalimumab fed‐batch cultures, in contrast to the trastuzumab fed‐batch cultures, where HC:LC transcript ratio remained constant (Figure S7D). This result suggests that HC transcripts may be limiting in adalimumab cultures. It is hypothesized that the overall low productivity of adalimumab in both cell lines may impact the ability to identify productivity differences in the tested variants.

4. Discussion

The emergence of SSI systems for the industrial production of mAbs has been driven by reduced clonal heterogeneity and improved production stability as compared to RI cell lines [9, 10, 31]. In this work, an optimized LP/RMCE system enabled the characterization of expression variability, intracellular production bottlenecks, and biophysical properties of DTE mAbs. Rapid generation of isogenic RMCE pools using only chemical selection enables higher throughput experimentation with reduced timelines as compared to past studies using RI workflows, and R26‐C9‐derived pools expressing trastuzumab reached an average titer of 1.2 g/L under fed‐batch conditions using commercially available media and feeds, which is on the same order of magnitude as the best‐performing industry SSI cell lines [32, 33].

With this system, ten single amino acid mutations in trastuzumab and adalimumab were expressed from two genomic loci for a total of 44 distinct expression conditions. The variants chosen for evaluation reflect four broad classes of sidechain chemistry (aromatic, charged, hydrophobic, and polar) at partially exposed and buried residues, as well as two pairs (LC1/HC1 and LC2/HC2) of structurally equivalent mutations across the chains. These pairs reflect polar and charged mutations at inaccessible residues, both of which represent molecular contexts strongly associated with low expression [20]. Transcriptional profiles of trastuzumab LC and HC gene expression could not account for the differences observed in productivity in batch or fed‐batch cultures, suggesting that expression changes are due to post‐transcriptional bottlenecks. Antibody expression bottlenecks are commonly linked to the limited secretory capacity and activation of the UPR, a conserved set of three signaling pathways (PERK, IRE1, and ATF6) that can activate many downstream signaling cascades to restore ER homeostasis in the presence of accumulated unfolded and misfolded proteins in the ER [14, 16, 28, 34, 35]. Therefore, we evaluated a panel of downstream UPR response proteins and folding chaperones in trastuzumab fed‐batch culture samples to investigate the mechanisms driving low expression of the selected variants.

BiP acts across all three branches of the UPR, and similar BiP levels were observed across all mAb‐producing cultures with slightly reduced expression observed in Null cultures, suggesting that consistent upregulation of UPR pathways under mild stress may result in similar steady‐state levels, especially due to the relative stability of BiP mRNA and its high protein half‐life [36]. PERK signaling is considered an early UPR response due to rapid protein‐level modulation of eIF2α through phosphorylation to attenuate global protein synthesis and initiate transcription of downstream effector genes such as ATF4 and CHOP. The presence of ATF4 across all mAb‐producing pools at a low but constant level coupled with more varied expression of CHOP suggests the consistent activation of PERK signaling in response to mAb expression. While CHOP has classically been considered a pro‐apoptotic factor of the UPR, it has been hypothesized that CHOP may adopt a binary mechanism through multiple steps and additional protein‐protein interactions that require progressively higher stimuli for activation to prevent excessive cell death in the presence of mild stress [25, 36]. The high cell viabilities observed on Day 8 of fed‐batch culture suggest that the CHOP levels observed may not be sufficient to propagate a pro‐apoptotic signal. The combination of low ATF4 but high CHOP expression in Null pools shows that these cultures were also subject to PERK activation, but the stress has now been resolved, as CHOP downregulation has been observed to lag behind ATF4 due to the sequential activation of these factors [37]. It is important to note that eIF2α phosphorylation is also a key mechanism in the integrated stress response (ISR), which can adapt to other cellular stresses such as metabolite deprivation and hypoxia, so the activation of this pathway can result from stresses not directly related to protein folding [38]. Additional work to optimize fed‐batch conditions and monitor canonical markers of the ISR may reveal the origins of PERK activation in these pools.

Canonical downstream targets of ATF6 signaling were found to be preferentially upregulated in some cultures expressing DTE variants. The transcriptional targets of ATF6 are considered more specific for the alleviation of ER stress, as they are predominantly ER resident proteins and folding chaperones, such as PDI, ERp57, and GRP94 [39, 40]. Both PDI and ERp57 are members of the broader protein disulfide isomerase family and are responsible for catalyzing the formation of disulfide bonds, with ERp57 known to specifically interact with glycoproteins [41]. GRP94's role in protein folding within mammalian cells remains elusive but it has been found to interact with maturing antibodies following BiP association, suggesting a sequential mechanism of action and a distinct binding motif compared to BiP [27, 42]. The increased expression of both PDI and GRP94 in HC2 pools implies that expression of the HC2 variant results in persistent activation of ATF6 signaling and requires additional folding chaperones to possibly mitigate poor folding and assembly in the ER [40]. This conclusion is further supported by the decrease in assembled HC dimer in the HC2 pools. Assembly of IgGs is believed to proceed from dimerization of two HCs followed by subsequent association of LCs, so this observation suggests that the HC2 mutation affects HC dimerization or reduces the amount of HC available for mAb assembly [43]. A similar depletion of HC dimer was observed for HC1, suggesting a similar bottleneck.

Splicing of XBP1 mRNA was also measured as a marker of IRE1 activation. The upregulation of XBP1s across all mAb‐producing cultures as compared to Null cultures suggests activation of this pathway due to exogeneous protein expression, similar to the results observed with ATF4. The detectable quantities of XBP1s in Null cultures also supports the conclusion that Null pools experience mild stress. On average, the HC1 variant showed significant upregulation of XBP1s compared to WT pools, suggesting persistent activation of IRE1 signaling. The minimal change in XBP1us expression (less than 2‐fold) as compared to XBP1s expression (up to 8‐fold) across pools is supported by the transcriptional activation of XBP1 by ATF6, which can restore XBP1us to normal levels and aid the cell in returning to equilibrium [44].

Because of the 3‐week pool generation process and the constitutive expression of the mAb genes throughout that time, the responses observed here are anticipated to be more characteristic of chronic ER stress, and the cells in the RMCE pools likely represent subpopulations that have sufficiently adapted to constitutive expression of exogeneous mAb. Many of the extensive UPR characterizations performed previously capture the dynamics of acute UPR responses induced by transient expression of mAbs or chemical induction of ER stress [16, 17, 36]. Therefore, limited understanding exists as to how the UPR can distinguish between its adaptive and pro‐apoptotic mechanisms in the presence of milder, chronic ER stress, as may be observed in the mAb‐producing cultures studied here [25]. Our assessment of UPR activation in response to mAb expression provides unique insight into how this signaling pathway manifests over longer culture durations (∼ 4 weeks) and may reflect the adaptive mechanisms adopted by industrially relevant production host cell lines. Improved understanding of these mechanisms for accommodating high protein production may inform secretion‐ and UPR‐based engineering approaches for CHO cell line development [45].

The strongest intracellular responses generally correlated with the lowest‐expressing variants, and a similar trend was observed with thermal stability as measured by DSF. The observed decrease in the second transition temperature for five trastuzumab variants supports the conclusion that introduction of these mutations into the variable domains can result in broader destabilization of the Fab domain. The lowest‐expressing mutations (HC1, HC2, and LC2) all show significantly different thermal profiles compared to WT, with both LC2 and HC2 showing the same decrease in T₁ and complete elimination of the T₂ transition, and HC1 showing a decrease in T₂ by almost 5°C. The relationship between thermal stability and expression has previously been identified in work studying yield differences between human antibody germline sequences in E. coli, where differences in these naturally occurring frameworks dramatically impacted domain expression [46]. These trends are especially pronounced in bacterial systems, where the absence of regulated and compartmentalized protein folding pathways results in a strong dependence between thermodynamic stability, folding capacity, and expression. The trend was recently extended to mammalian systems, most notably in a recent developability assessment that found a correlation between mAb melting temperature and transient expression titers in HEK293 [5]. Studies specifically evaluating thermal stability in the context of DTE mAbs also found that single amino acid mutations could result in changes in thermal stability alongside decreases in expression [13, 47]. However, because DSF struggles to resolve transitions at higher temperatures, additional experiments with more sensitive methods would be required to conclusively attribute the altered thermal profiles observed here to Fab destabilization.

Taken together, the low expression of trastuzumab mutations HC1 (F67S), LC2 (V33K), and HC2 (I34K) in fed‐batch culture, their decreased thermal stability, and the presence of upregulated ER stress responses in these pools suggests significant structural disruption of the antibody chains due to the introduced mutations. The HC1 variant, HC‐F67S, is located on the exposed surface of the HC within a loop connecting the framework regions between CDR2 and CDR3. The buried orientation of the aromatic ring in the WT phenylalanine is accommodated by the positions of the methionine and leucine side chains at positions 82 and 18, respectively, and the sidechain participates in an extended interaction network throughout the surrounding residues by forming key hydrogen and nonpolar interactions [48, 49]. Previous work has shown that changes in the sidechain at position 67 to smaller, hydrophobic residues can have far‐reaching effects by impacting the conformation of positions 82, 18, and 9, and has even suggested that changes at position 67 could impact residues important for CDR2 conformation [50]. The introduction of a serine mutation at this position could therefore result in a significant change in sidechain orientation and disrupt key hydrophobic interactions within the protein core that may manifest as a poorly folded HC.

The LC2/HC2 (LC‐V33K and HC‐I34K) mutation pair impacts a buried residue adjacent to CDR1 in both chains that is believed to play a role in determining the conformation of the CDR1 loop [51]. In the WT trastuzumab light chain, atoms in the valine residue backbone form hydrogen bonds with a threonine residue within the CDR1 loop as well as with an alanine in the CDR2 loop. The formation of these hydrogen bonds may be impactful for maintaining the orientations of these two loops, and the introduction of a large, charged lysine residue at this position may result in unfavorable protein folding accommodations in the backbone that disrupt the geometry necessary for hydrogen bond formation. The nonpolar interactions that this residue makes with a buried phenylalanine (LC‐71) may also be disrupted upon mutation [48]. Similarly, in the heavy chain, the WT isoleucine forms two hydrogen bonds with neighboring HC‐I51, a position which has also been implicated as a key residue for mAb expression, with previous work demonstrating reduced antibody secretion with mutations to arginine and lysine [52]. HC‐I34 and HC‐I51 are physically adjacent residues with buried sidechains, and it was hypothesized that the decrease in antibody expression from mutations at HC‐I51 may result from enhanced BiP binding to this region, preventing HC maturation. BiP binding sites remain difficult to predict, but the distinct decrease in HC dimer present in the HC2 pools and subsequent increase in protein binding chaperones suggests a mechanism where heavy chains containing the I34K mutation cannot participate in the necessary interchain interactions to drive high mAb expression. The equivalent HC2 mutation in adalimumab (HC‐M34K) also demonstrated reduced expression in batch cultures and is known to form two similar hydrogen bonds with HC‐I51, suggesting a common structural basis for low expression of this variant in IgG₁ heavy chains [48].

The results presented here demonstrate that LP/RMCE systems are valuable tools for systematic comparisons of mAb expression under industrially relevant conditions at unprecedented scale, with ten mutations studied across two mAbs at two genomic loci for a total of 44 distinct experimental conditions. The use of RI or transient transfection methods in previous work on DTE mAbs has limited the throughput of these studies to eight variants at most, and in the case of transient transfection, has focused on short‐term intracellular responses to mAb expression, which may not be characteristic of industrial stable expression systems [35]. The low throughput of this previous work has resulted in a limited picture of the mechanistic relationship between mAb primary amino acid sequence and expression, with low‐expressing mAbs requiring cell line‐ and product‐specific optimization efforts tailored to the candidate's unique behavior [43]. The increased throughput, reduced clonal variation, and expression resolution observed with LP/RMCE systems would allow for screening of larger mAb variant panels to distinguish between sequence features that either improve or reduce expression levels under industrially‐relevant culture conditions, as previously demonstrated with work using LP/RMCE to optimize expression cassettes or identify novel genetic elements [53, 54, 55, 56, 57].

Industrial implementation of LP/RMCE systems across drug discovery and process development groups has become a crucial link across project lifecycle phases, and their use in developability assessments has helped to characterize expression distributions of mAb candidates and make informed candidate selection decisions [58]. Using LP/RMCE systems at this stage to understand intracellular bottlenecks to mAb expression can contribute to rational and informed engineering solutions for DTE mAbs, reducing time and effort during process development and increasing the chances of successfully manufacturing these molecules. While LP/RMCE systems are increasingly prevalent in industrial settings, additional work to identify stable, high‐expressing integration locations, appropriate selection and reporter mechanisms for generating LP host cells, and optimal mAb expression cassettes across loci can improve the expression levels of LP/RMCE systems to routinely meet the benchmarks set by RI workflows [9, 19, 32, 59, 60]. Applying LP/RMCE systems to better understand the sequence‐based determinants of expression in industrially relevant host cell lines can help bring more novel antibody‐based medicines into the clinic that may be therapeutically beneficial but challenging to manufacture.

Author Contributions

AS conceptualized and executed the study, analyzed data, and prepared the first manuscript draft and edited the manuscript. KHL analyzed data, supervised the study, acquired funding for the study, edited and reviewed the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: biot70102‐sup‐0001‐SuppMat.docx

Szkodny A. C. and Lee K. H., “Evaluation of “Difficult‐to‐Express” Monoclonal Antibodies in a CHO‐Based Hybrid Site‐Specific Integration System Under Industrially Relevant Conditions.” Biotechnology Journal 20, no. 8 (2025): 20, e70102. 10.1002/biot.70102

Data Availability Statement

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