Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Aug 21.
Published in final edited form as: Metab Eng. 2025 Jul 23;92:174–184. doi: 10.1016/j.ymben.2025.07.006

Improving Recombinant Antibody Production Using FcBAR: An In Situ Approach to Detect and Amplify Protein-Protein Interactions

Mina Ying Min Wu 1,3,*, Frances Rocamora 3,*, Mojtaba Samoudi 3, Caressa M Robinson 1,3, Chih-Chung Kuo 1,3, Nuša Pristovšek 4, Lise Marie Grav 4,5, Helene Faustrup Kildegaard 4, Gyun Min Lee 6, Alexandre Rosa Campos 7, Nathan E Lewis 1,2,3,5,8,§
PMCID: PMC12366646  NIHMSID: NIHMS2099777  PMID: 40712833

Abstract

Recombinant proteins, in particular monoclonal antibodies and related molecules, have become dominant therapeutics. As they are produced in mammalian cells, they require the concerted function of hundreds of host cell proteins in the protein secretion pathway. However, the comprehensive set of host cell machinery involved remains unclear. Thus, it is often unknown why some recombinant proteins fail to express well. Here we present and deploy an approach called Fc-targeting Biotinylation by Antibody Recognition (FcBAR), which allows for the in situ detection of protein-protein interactions for any recombinant protein with Fc domain. Briefly, cells are permeabilized and incubated with an anti-Fc antibody, conjugated with horseradish peroxidase. All proteins interacting with Fc-bearing proteins are then biotinylated, pulled down and identified via mass spectrometry. We applied this method on a panel of rituximab-producing CHO-S clones with a range of productivity levels. Through analysis of FcBAR protein-protein interactions and RNA-Seq, we identified protein interactions positively correlated with rituximab secretion, and tested 7 of these targets. We found overexpression of AGPAT4, EPHX1, and NSDHL significantly increased rituximab production. Thus, FcBAR provides an unbiased approach to measure PPIs supporting recombinant antibody production in situ, and can guide efforts to boost production of biotherapeutics and biosimilars by addressing production bottlenecks.

Introduction

The most high-demand biopharmaceutical products are monoclonal antibodies (mAbs), which accounted for more than half of all new pharmaceutical product approvals from 2015–2022 and dominated global sales, bringing in more than 60% of the sales in 2021 (Walsh & Walsh, 2022). The demand for mAbs and related biologics will continue to grow to address their need for treating complex illnesses. Although biosimilars are making mAbs more affordable, the cost remains high. Moreover, some therapeutic proteins are difficult to express, sometimes due to bottlenecks in the secretory pathway, which limits their accessibility. Here, we use rituximab, a well-characterized mAb, as a model protein to investigate proteomic interactions that influence its secretion, with the goal of identifying strategies to enhance its production. Rituximab is a glycosylated murine/human chimeric IgG1κ mAb that targets the surface marker CD20 on malignant B-cells to treat non-Hodgkin’s lymphoma, chronic lymphocytic leukemia, and rheumatoid arthritis (Baldo, 2016; Selewski et al., 2010). Biosimilars of rituximab, such as Ruxience, have shown a 30% reduction in price compared to Rituxan (Kvien et al., 2022). Although biosimilars are making rituximab and other mAb treatments more affordable and accessible, the cost remains high.

One approach to reduce the cost of rituximab is to increase its production yield. Rituximab and its US-approved biosimilars are produced in Chinese Hamster Ovaries (CHO) cells. Optimizing the fed-batch bioprocess for rituximab-expressing CHO cells can increase in volumetric productivity (Mellahi et al., 2019). Building on these promising bioprocess results, our focus is on optimizing the secretory pathway machinery (SecM) at the cellular level to enhance rituximab productivity, as transgene mRNA and protein levels do not always correlate with high titers. This is because secretory pathway inefficiencies often act as bottlenecks in recombinant protein production.

The SecM involves over 1,127 genes associated with post-translational modifications (PTMs), protein folding, quality control (ER-associated degradation pathway, Unfolded Protein Response), and trafficking between the endoplasmic reticulum (ER) and Golgi apparatus in the mammalian system (Masson et al., 2024b). Given that many of the SecM proteins will physically interact with a secreted protein of interest to facilitate its folding, modification and transport through the pathway, co-expressing certain SecM components with difficult-to-express proteins in CHO cells has improved secretion (Malm et al., 2022). For mAbs, the bottleneck is primarily at the folding and assembly stages (Dinnis & James, 2005; Kaneyoshi et al., 2019). Overexpression of vesicle-associated proteins such as SNAP-23 and VAMP, which facilitate vesicle docking and fusion, has led to increased production of recombinant proteins including anti-CD18 monoclonal antibodies (IgG1) (Peng et al., 2011). However, the host cell machinery required for specific secreted proteins remains unclear, so it is difficult to know what should be augmented for any given cell line or product.

To address this issue, here we present FcBAR, an unbiased approach adapted from the Biotinylation by Antibody Recognition (BAR) method (Bar et al., 2018). Fc-BAR allows one to quantify protein-protein interactions (PPIs) between Fc-bearing proteins (e.g., mAbs, Fc-fusion proteins, etc.) and the host-cell proteome. Briefly, in FcBAR, the cell line of interest is fixed, permeabilized, and incubated with an anti-Fc antibody conjugated with horseradish peroxidase (HRP). All interacting proteins are then biotinylated by treatment with H2O2 and tyramide-biotin; biotinylated proteins are then pulled down and analyzed by mass spectrometry (MS). Unlike other methods, such as BioID, which requires the addition of a biotin ligase domain to the protein of interest (POI), our FcBAR approach can be done on any cell line without modifying the host or POI (Kim et al., 2016; Pfeiffer et al., 2022; Sears et al., 2019). We hypothesize that FcBAR can help identify known and novel protein interactors in the secretory pathway, which represent SecM proteins that support mAb production. This approach is particularly relevant given that approximately 40% of heavy chains and 65% of light chains are retained in the ER and Golgi apparatus (Kaneyoshi et al., 2019).

We applied FcBAR to 10 rituximab-expressing clones with a range of productivity levels. We detected >1,100 unique proteins that correspond to rituximab-expressing clones. Furthermore, through a multiomics analysis of the FcBAR proteomics and RNA-Seq data, we compiled a list of protein interactions correlated with high productivity. Our results demonstrate that FcBAR is an effective tool for identifying proteins in the secretory pathway that enhances production yield. This multi-omics approach shows promise for addressing production challenges for rituximab and potentially other difficult-to-express proteins.

Results

Generating Stable Cell Lines Expressing Rituximab

A plasmid containing genes for the heavy chain (HC) and light chain (LC) of rituximab was transfected into suspension CHO-S cells through random integration, and the cells were subsequently cloned by single cell sorting. The productivity (pg/cell/day) of rituximab secretion were quantified for each clone through the Octet RED96 system, a bio-layer interferometry instrument that uses surface plasmon resonance (SPR) technology to measure protein interaction due to its sensitivity and suitability for measuring both low-abundance and highly abundant proteins. We identified 10 clones with similar mRNA levels but varying rituximab production levels (Supplementary Figure S1A). The clones were sorted into low or high producers based on the amount of rituximab they produced. Three low producers—B2, E9, and G12—had productivity ranging from 6.50 to 7.20 pg/cell/day, while seven high producers—A4, A6, B5, E6, E7, E10, and F1—showed productivity ranging from 8.87 to 14.46 pg/cell/day (Figure 1B). Additionally, western blot analysis confirmed successful secretion of rituximab in each clone (Supplementary Figure S1B).

Figure 1. Utilizing biotinylation by antibody recognition targeting the Fc Region (FcBAR) to study protein-protein interactions in rituximab clones.

Figure 1.

Open in a new tab

(A) Surface Plasmon Resonance (SPR) identified three low-producing rituximab clones (G12, E9, and B2) with productivity ranging from 6.50 to 7.20 pg/cell/day, and seven high-producing rituximab clones (B5, E10, E6, A4, F1, A6, and E7) with productivity ranging from 8.87 to 14.46 pg/cell/day. (B) An illustrated depiction of the FcBAR method applied to rituximab-expressing clones. Following fixation and permeabilization, cells were stained with an HRP-conjugated anti-human Fc antibody. Biotinylation was performed for 5 minutes, labeling proteins in proximity to rituximab. (C) Biotinylation was confirmed via western blot using streptavidin-HRP staining. Each clone included a negative control (B-) in which biotin was not added. A CHO-S wild-type not expressing rituximab served as an additional control. Clones expressing rituximab and treated with biotin contained more biotinylated proteins, indicating successful biotinylation of proximate proteins associated with rituximab.

Targeting the Fc Region of Rituximab using FcBAR

To investigate the PPIs of rituximab, FcBAR was optimized to target the Fc region of rituximab. FcBAR and bulk RNA-Seq were performed on the rituximab-expressing clones, harvesting the cells during the mid-exponential phase. FcBAR, the cells were fixed, permeabilized, and stained with an anti-human Fc antibody conjugated to HRP (Figure 1B). The cells were then biotinylated for 5 minutes, after which cell lysates were harvested. The biotinylated proteins were captured using streptavidin beads, eluted, and trypsinized for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis.

Western blot analysis confirmed biotinylation of the samples, revealing the biotin profile by staining with streptavidin (Figure 1C). Each clone included a negative control (B-), where biotin was not added. The western blot results showed that the rituximab-producing clones contained more biotinylated proteins compared to both the WT control and the negative control, confirming the specificity of the endogenous biotin labeling for interacting proteins. Additionally, the cells were fixed, permeabilized, and stained with an anti-human Fc-650 antibody targeting rituximab, DAPI for nuclear staining, and streptavidin-594 for biotinylated proteins. Confocal microscopy verified the colocalization of rituximab detected in each rituximab producing clone in the far-red channel (650 nm; shown in red) with biotinylated proximal proteins detected in the red channel (594 nm; shown in green) (Figure 2A). This was quantitatively confirmed with Pearson’s coefficient of 0.87, indicating strong colocalization between the rituximab and the biotinylated proximate proteins (Figure 2B).

Figure 2: Colocalization of rituximab with biotinylated proteins confirmed by confocal microscopy after performing BAR.

Figure 2:

Open in a new tab

(A) Confocal microscopy of the BAR samples for each rituximab clone demonstrated colocalization of rituximab (far red channel, anti-human-650 nm shown in red) with biotinylated proximate proteins (red channel, streptavidin-594 nm shown in green). The scale bar corresponds to 12 um. (B) Intensity plot between the 650 (rituximab) and 594 (streptavidin) channels showed colocalization with a Pearson's coefficient of 0.87, indicating strong colocalization between the rituximab and the biotinylated proximate proteins. Cells were fixed, permeabilized, and stained with DAPI for nuclear staining, anti-human Fc-650 nm antibody targeting rituximab, and streptavidin-594 nm to detect biotinylated proteins. The scale bar corresponds to 6 um.

Preprocessing and Quality Control Analysis of FcBAR Proteomic Data

The FcBAR proteomic mass spectra were analyzed using MaxQuant software (Tyanova et al., 2016), and the MS/MS spectra were searched against the Cricetulus griseus Uniprot protein sequence database. To analyze the proteomic data, we utilized the Differential Enrichment analysis of Proteomics data (DEP) package (Zhang et al., 2018) in RStudio, first performing quality control (QC) by plotting the total protein abundance for each sample. We expected the total amount of biotinylated proteins to be similar across replicates for each sample. Based on the plot of biotinylated proteins detected per sample, one replicate each from cell lines B5 and E7 showed significantly lower numbers of detected biotinylated proteins and thus failed quality control; these replicates were excluded from further analysis (Figure 3A).

Figure 3. Quality control of FcBAR and comparative proteomics analysis: linear regression of protein abundance versus productivity and differential expression analysis between high- and low-productivity clones.

Figure 3.

Open in a new tab

(A) The rituximab FcBAR samples were plotted against the number of biotinylated proteins detected per sample. During LC-MS/MS analysis, one replicate from each of clones B5 and E7 showed inconsistent numbers of captured biotinylated proteins compared to their respective replicates and was therefore excluded from the analysis. (B) PCA on the normalized proteomic dataset revealed distinct clustering, with non-producer cells (control) forming a separate cluster from the rituximab producer cell lines, as captured in PCA1, explaining 72% of the variance. (C) Linear regression analysis was performed on the protein abundance vs productivity (pg/cell/day) to identify proteins that show a positive association with high productivity (slope > 0, p < 0.05, and R2 > 0.50). (D) Welch's t-test was performed to compare high versus low producer cell lines, identifying > 60 (red dots) and >80 (blue dots) proximate proteins as differentially expressed in the high and low producing cell lines, respectively (p < 0.05) as visualized in the volcano plot.

We normalized the data using variance stabilizing normalization (VSN) (Huber et al., 2002; Karp et al., 2010), which transforms the data so that the variance of protein abundance values is stabilized across the measurement range. In raw proteomic data, low-abundance proteins typically exhibit low variance, while high-abundance proteins tend to show higher variance—indicating that variance depends on abundance. This dependence complicates statistical comparisons (e.g., t-tests) and can introduce bias in clustering analyses. Applying VSN stabilizes variance across all expression levels, enabling more reliable statistical comparisons and unbiased data interpretation. Next, the proteomic data was imputed using a left-shifted Gaussian distribution based on the assumption that missingness is not at random (MNAR), which means that missing values are likely to represent low-abundance proteins (Pereira et al., 2024). This approach avoids false zeros by acknowledging that missing proteins are likely just below the limit of detection. It ensures that imputed values are realistic and improves the reliability of statistical tests for differential expressions. We then analyzed the variance and clustering of the rituximab clones of the normalized data using Principal Component Analysis (PCA) of the normalized data, which transforms high-dimensional data into a lower-dimensional subspace while preserving the variance. The PCA plot clearly revealed that the nonproducers clustered separately from the producer cell lines, with PC1 capturing 72% of the total variance, indicating a clear difference between proteins captured in producer and nonproducer clones (Figure 3B).

Identifying Protein Interactors with Proteomic and Transcriptomic Analysis

To identify proteins associated with productivity levels, we performed linear regression analysis between the mean abundance of each protein against the productivity (pg/cell/day) of each clone (Figure 3C). We identified >20 proteins with a significant positive association with productivity (p < 0.05, slope > 0, R2 > 0.50) (Supplementary Tables S1), some of which are related to protein transport (Atp11c, Rab8a, Myo18a, Emc1, Atp13a1) and autophagy (Lamp2, Snx4, Pgrmc1), suggesting these proteins may enhance rituximab secretion. In contrast, >15 proteins showed a significant negative association with productivity (p < 0.05, slope < 0, R2 > 0.50) (Supplementary Tables S1), indicating potential inhibitory effects on productivity, some of which are involved in protein degradation (Psmd11, Herc4), protein folding (Sacs, Ahsa2p, Dnajc7), and pro-apoptotic signals (Stk24, Crlf1).

In addition, we identified differentially expressed proteins by employing Welch's t-test on the FcBAR data comparing high and low producer cell lines. We identified >1,100 potential interactors with rituximab, revealing >60 interactors significantly expressed (p < 0.05) and unique in high producers and >80 interactors significantly expressed (p < 0.05) and unique in low producers (Figure 3D). To complement the proteomic analysis, bulk RNA-Seq analysis was performed to determine if any of the identified PPIs overlapped with the transcriptomic data. Using Welch's t-test, approximately 1,300 genes were significantly expressed in high producers (p < 0.05) and >900 genes were significantly expressed in low producers (p < 0.05) at the mRNA level (Figure 4A). To identify enriched biological processes, Gene Ontology over-representation analysis revealed that high-producing cells were significantly enriched in secretory pathway-related functions, including protein degradation, modification, and transport pathways. These findings suggest that high protein production requires mobilization of an extensive secretory machinery. (Figure 4B). In contrast, significantly enriched genes in low producers were primarily associated with RNA splicing, processing, modification, and localization, as well as mitochondrial function, translation, and protein localization. This suggests that low-producing cells may prioritize RNA regulatory mechanisms and energy-intensive metabolic processes (Figure 4C).

Figure 4. Bulk RNA-Seq transcriptomics in high- and low-producing rituximab cell lines.

Figure 4.

Open in a new tab

(A) Bulk RNA-Seq analysis was conducted to compare high- and low-producing cell lines. The analysis identified approximately >1,300 differentially expressed genes in high-producing cell lines (red dots) and >900 differentially expressed genes in low-producing cell lines (blue dots) (Welch's t-test, p < 0.05) as visualized in the volcano plot. (B) Gene enrichment analysis focused on biological processes revealed significant pathways associated with the secretory pathway in high-producing cell lines. These pathways included protein secretion, ERAD pathway, vesicle budding, and related processes. (C) In contrast, low-producing cell lines showed gene enrichment primarily in RNA processing, splicing, modifications, and mitochondrial functions.

We identified >50 proteins that exhibited significantly different abundances in PPIs and significantly different mRNA expression levels when comparing high-producing to low-producing clones (p < 0.05) (Supplementary Tables S2 and S3). These proteins fell into several functional categories of interest. For example, some significant targets of particular interest included those with transport (Rab8a, Ocr1, Atp11c, Myo18a, Ergic1), protein degradation and membrane trafficking (Cul4b, Sacm1l, Ergic1, Pgrmc1, Psme2), cell adhesion and extracellular matrix components (Bgn, Col6a1/4a1, Thbs1/s3, Emilin1, Adam9, Tspan6, Itgb5/av, Arhgef6/12), metabolism (G6pd, Prps1, Nsdhl), and DNA/RNA processing (Fmr1, Sun2, Babam2, Dtymk).

From these significant genes, we selected seven of them to test to see if their overexpression could increase rituximab production: AGPAT4, ACSL4, NSDHL, PSME2, LOC100774522, PGRMC1, and LOC100758590 (EPHX1). These were of particular interest since they are anticipated to serve distinct but complementary cellular functions. AGPAT4 regulates membrane curvature (Karagiota et al., 2022), while ACSL4 catalyzes the activation of fatty acids to form acyl-CoA, a crucial step in lipid biosynthesis and fatty acid degradation, and is localized in the ER (Radif et al., 2018). Similarly, NSDHL is associated with lipid droplets and cholesterol biosynthesis (Caldas & Herman, 2003; Ohashi et al., 2003). In contrast, PSME2 (PA28β) functions in protein quality control as a subunit of the PA28 proteasome activator complex, which interacts with the 20S proteasome to degrade misfolded proteins (Rechsteiner & Hill, 2005). Together with PA28α, PSME2 forms a hexameric structure that aids in refolding of misfolded luciferase bound to Hsp90 along with Hsc70 and Hsp40 (Minami et al., 2000). LOC100774522, a pseudogene related to UBE2N, which has shown to maintain protein homeostasis (Ishikawa et al., 2025). Additionally, PGRMC1 is an ER transmembrane protein that participates in ER-phagy by acting as a size-selective cargo receptor, targeting smaller, misfolded prohormones for degradation through its interaction with reticulon-3 (Chen et al., 2021). Finally, EPHX1 is an ER-localized enzyme involved in hydrolyzing endogenous epoxy fatty acids (Gautheron & Jéru, 2021; Václavíková et al., 2015). Collectively, these proteins were selected based on their essential roles in lipid metabolism and protein degradation—processes critical for efficient protein secretion.

The three proteins, LOC100774522, PGRMC1, and EPHX1 not only showed increased mRNA abundance and PPIs in the high producers, but they further showed significant positive associations with productivity (p < 0.05, R2 > 0.60). Thus, we tested the ability for overexpression of these 7 proteins to increase rituximab production.

Overexpression of Identified Protein Interactors Increased Rituximab Expression

We tested selected targets by overexpressing the genes using polyethylenimine (PEI) Max transient transfection in the low-producer rituximab cell line, E9 with each test conducted in triplicate. RNA was extracted on day 2 post-transfection for quantitative PCR (qPCR) confirmation of overexpression. Gene expression was analyzed and normalized to the empty vector backbone (control), which was set to 1.0. Relative fold change calculations confirmed that all genes were successfully overexpressed, with Ephx1 showing particularly high expression levels, indicating that this protein is natively expressed at low levels in the E9 cell line (Figure 5A).

Figure 5. Validation of candidate genes identified from proteomic and transcriptomic datasets using overexpression and quantifying with Fc-ELISA.

Figure 5.

Open in a new tab

(A) RNA was extracted two days post-transfection and reverse transcribed into cDNA for qPCR analysis using the 2^(-ΔΔCT) method for quantification. Gene expression levels were normalized to the control (empty vector), which was set to 1, confirming successful overexpression of all tested genes in the transfected cells.(B) Cell counts and supernatant were collected on day four post-transfection. Indirect Fc-ELISA, depicted in the schematic was performed on the supernatant, revealing that three candidate genes—Agpat4, Ephx1, and Nsdhl—significantly enhanced rituximab secretion in the low-producing cell line, E9, compared to the control (empty vector backbone; paired t-test, p < 0.05).

On day 4, we performed cell counts to determine viability and harvested supernatant for rituximab quantification by indirect Fc-ELISA. The obtained concentrations were normalized to viable cell numbers, resulting in their yield (pg/cell) (Figure 5B). Compared to the control (empty vector backbone), overexpression of Agpat4, Ephx1, and Nsdhl significantly improved rituximab secretion in cell line E9 (p < 0.05, paired t-test) with fold changes of 1.57, 1.98, and 1.25 respectively. These three proteins were detected only in rituximab-producing cells in the FcBAR proteomic data and not in the control cell line indicating that these protein interactors are rituximab specific. In addition, paired t-test shows that there was no significant difference (p > 0.05, paired t-test) in viable cell count (cells/ml) and cell viability percentage between the overexpression of these validated genes and the control (Supplementary Figure S1C). These proteins all localize to the ER, suggesting that ER-related processes represent a key bottleneck in rituximab secretion in these cells.

Discussion

Here we presented and used FcBAR to study a secreted protein, rituximab. We target the Fc regions of the rituximab rather than other domains such that the Fc region is highly conserved across antibodies, enabling this method to be extended to other mAbs or Fc-fusion proteins without modification, Fc-directed labeling avoids interference with the antigen-binding site (Fab), reserving antibody function, and Fc-specific detection antibodies are more widely available. We hypothesized that our Fc-targeted approach should capture proximal interactors across the entire antibody structure, as rituximab has a maximum linear dimension of 197 Angstrom (19.7 nm) (Belviso et al., 2022), while the biotinylation radius spans 10 to >200 nm (Rega et al., 2025). Since the labeling radius comfortably exceeds the antibody's dimensions, we would expect to detect primary Fc interactors (direct binders) and proteins associated with other antibody regions (Fab, hinge) or proximate to the Fc regions. To maximize the detection of the diverse protein interactions, cells are harvested during the mid-exponential phase for FcBAR analysis, as this ensures that the cells are actively growing and maintaining high levels of protein synthesis to capture protein interactions with rituximab before secretion. In contrast, if cells are harvested during the lag phase, the cells are still adapting to their environment and preparing to divide, with lower protein production. During the stationary phase, protein production reaches its peak of secretion, but cell growth slows down, waste products accumulate, and the cells enter survival mode (González-Hernández & Perré, 2024; Templeton et al., 2013). Future studies could optimize protocols to assay interactions during different production phases to gain further insights into the host cell machinery driving production.

Thus, FcBAR allowed us to identify PPIs that enhance secretion through the secretory pathway. We expressed rituximab in suspension CHO-S cells and identified a range of clones with different productivity levels. To identify factors affecting productivity, we employed FcBAR to study the PPIs of rituximab, along with RNA-Seq, integrating both approaches to identify proteins that interact with rituximab in the secretory pathway. We then performed linear regression analysis of the FcBAR data to identify proteins that are associated with high productivity and identified proteins that are differentially expressed in high producers in both the proteomic and transcriptomic datasets. For validation through overexpression, we selected proteins from both analyses that are involved in lipid metabolism and protein degradation.

Our validation experiments revealed that overexpression of AGPAT4, NSDHL, and EPHX1 significantly increased rituximab secretion. Notably, two of these proteins are directly involved in lipid metabolism. AGPAT4 (LPATTδ) belongs to the acylglycerophosphate acyltransferase (AGPAT) enzyme family, which catalyzes the synthesis of phosphatidic acid (PA) from lysophosphatidic acid (LPA) and acyl-CoA, participating in phospholipid and triacylglycerol biosynthesis (Karagiota et al., 2022). In addition, PA is involved in membrane trafficking, membrane curvature, and protein localization to membranes (Zhukovsky et al., 2019). AGPAT4 is shown to be located in various cellular compartments, including the ER, Golgi apparatus, and mitochondria (Karagiota et al., 2022). It interacts with the protein, BARS (Brefeldin A ADP-Ribosylated Substrate), a member of the C-terminal binding protein (CtBP) family, which plays a role in membrane fission and post-Golgi carrier formation (Pagliuso et al., 2016). Through this interaction, AGPAT4 aids in Golgi membrane fission by converting LPA to PA, which induces membrane curvature or recruits proteins that bind to PA (Karagiota et al., 2022; Pagliuso et al., 2016).

The second lipid-related protein we identified, NSDHL, is a NAD(P)H steroid dehydrogenase-like enzyme involved in cholesterol biosynthesis (Caldas & Herman, 2003; Ohashi et al., 2003). Located in the ER and localized with lipid droplets, NSDHL plays a role in cholesterol biosynthesis and lipid metabolism (Caldas & Herman, 2003; Ohashi et al., 2003). To properly localize to the ER membrane, NSDHL requires transport through the Golgi apparatus (Caldas & Herman, 2003). The third validated protein, EPHX1, indirectly interacts with fatty acids. EPHX1 is a microsomal epoxide hydrolase located in the microsomal fraction of the ER, where it converts highly reactive epoxides into less reactive diols, such as trans-dihydrodiols (Gautheron & Jéru, 2021; Václavíková et al., 2015). EPHX1 plays a role in hydrolyzing endogenous epoxy fatty acids (EpFAs) derived from polyunsaturated fatty acids, such as epoxyeicosatrienoic acids (EETs) to diols, which are produced by CYP450 enzymes in the ER (Edin et al., 2018; Gautheron et al., 2021). In fibroblasts, EPHX1 knockout prevents adipocyte differentiation, increases oxidative stress, and leads to cellular senescence (Gautheron et al., 2021).

Through this work, we demonstrated that by studying PPIs and RNA-Seq data, we identified novel genes that, when overexpressed, increased rituximab production. This enhancement may be attributed to the fact that both AGPAT4 and NSDHL are lipid-associated proteins whose overexpression might improve lipid and cholesterol metabolism, while EPHX1 indirectly interacts with fatty acids. The expected impacts on membrane composition and fluidity could impact protein trafficking, folding, and secretion, ultimately boosting rituximab secretion. Additionally, overexpression of these genes might activate other pathways that indirectly enhance rituximab production. It would be of interest to determine whether expressing all three proteins or in dual combinations would produce a significant increase in rituximab production compared to individual overexpression, since these proteins have different functions.

Our approach here is a simplified version of BAR, tailored to proteins with an Fc domain. BAR is a versatile protein labeling method that requires only specific binding antibodies for POI, avoiding the laborious process of introducing foreign DNA into the host cell, (Sears et al., 2019). Additionally, BAR can be done on fixed cells and it provides a temporal snapshot of proximal proteins at a specific time point, thus avoiding prolonged labeling times that could affect protein function. However, BAR does have limitations. The method requires highly specific binding antibodies for the POI, which may not be readily available depending on the organism. Furthermore, BAR does not distinguish between direct and indirect interactions and may miss transient PPIs due to the temporal nature of the labeling process.

Previous applications of BAR have been successfully applied to both non-secreted and secreted proteins. For non-secreted proteins, BAR has been used to study the mitochondrial matrix protein Lamin A/C in HeLa cells, primary cell cultures, and primary human muscle and adipose tissues, successfully identifying putative interactors (Bar et al., 2018). For secreted proteins, BAR has been applied to fetuin-B and able to identify unique interactome corresponding to fetuin-B (Masson et al., 2024a). In this study, we modified BAR to simplify the protocol for Fc-bearing proteins. We demonstrated it with a model secreted mAb, showing it can identify potential protein interactors of rituximab within the ER and Golgi apparatus. This specificity is possible because HRP is only active in oxidizing environments, such as the lumen of the ER and Golgi apparatus (Bosch et al., 2021).

Taken together, the identification of AGPAT4, EPHX1, and NSDHL as key interacting proteins provides confidence in the utility of FcBAR for gaining valuable insights into optimizing protein secretion in CHO cells. These findings highlight the importance of lipid metabolism and ER function in antibody secretion. Our approach offers a promising strategy for optimizing the manufacturing of complex biologic therapies, potentially reducing production costs and improving patient access. Future studies could explore whether these protein overexpression strategies can be applied to other therapeutic antibodies and difficult-to-express proteins.

Materials and Methods

Cell Line Engineering and Protein Expression

The rituximab vector was constructed using the pcDNA3.1(+) backbone with uracil-containing primers and Phusion Hot-Start Flex polymerase (New England Biolabs, Ipswich, MA, USA). The rituximab vector consists of CMV-HC-BGHpA-EF1α-LC-BGHpA with selection marker SV40-NeoR-SV40pA. CHO-S suspension cells (Life Technologies, Carlsbad, CA, USA) were transfected with the rituximab plasmid using FuGENE HD transfection reagent (Promega) as described previously in the supplement (Pristovšek et al., 2019). Cells were selected with 0.500 μg/ml Geneticin Selective Antibiotic (Gibco, USA) to establish a stable polyclonal population. Fluorescence-activated cell sorting (FACS) was used to isolate rituximab-expressing clones, which were subsequently expanded as previously described (Pristovšek et al., 2018). The expanded clones were cultured in 125 mL Erlenmeyer shaker flasks with a baffled bottom (Fisher Scientific) in the presence of 0.25 mg/ml Geneticin.

Quantifying Rituximab Productivity Through Surface Plasmon Resonance

Rituximab titer was determined for each clone using an Octet RED96 (Pall, Menlo Park, CA, USA, a biolayer interferometry instrument that uses surface plasmon resonance (SPR) to measure protein binding as described previously (Grav et al., 2015), with cell counts taken daily. Following equilibration in sample diluent buffer (PBS with 0.1% BSA and 0.02% Tween-20), Protein A biosensors measured rituximab concentrations for 120 seconds at 30°C. Absolute concentrations were determined by comparing measurements to a calibration curve prepared from human IgG dilution series (GENSA01006, VWR, Dublin, Ireland). Biosensor tip regeneration was performed between measurements using 10 mM glycine (pH 1.7). Specific productivity (pg/cell/day) for rituximab clones was calculated based on rituximab titer and the integral viable cell density measurements as previously described (Pybus et al., 2014) to determine high and low rituximab producing clones.

Cell Culture

A panel of ten cell CHO-S clones expressing rituximab were obtained. A Laminin-alpha1 knock-out CHO-S clone line was obtained to use as a control. CHO-S cells were grown in 125ml Erlenmeyer shaker flasks with baffled bottom (Fisher Scientific, USA) in a humidified 37°C incubator with shaking at 130 RPM at 5% CO2 with CD CHO Medium (Gibco, USA) supplemented with 8mM L-Glutamine (Gibco) and 1x anti-clumping agent (Gibco). The rituximab producing clones were grown in media containing 250 ug/ml Geneticin Selective Antibiotic (G418 sulfate; Gibco) for selection.

Western blot

Western blots were performed to analyze rituximab using a precast polyacrylamide gel, 4–15% Mini-Protean TGX Gels (Bio-Rad, Hercules, CA, USA). The samples were run under reducing conditions by mixing with loading dye (4x Laemli + 10% β-Mercaptoethanol) (Bio-Rad) and boiling at 95°C for 5 minutes. The gel was run on a Mini-PROTEAN Tetra Vertical Electrophoresis Cell (Bio-Rad) at 160V for 40 minutes with the Precision Plus Protein standards ladder (Bio-Rad). The gel was transblotted onto 0.2 μm nitrocellulose membranes (Bio-Rad) with the Trans-Blot Turbo Transfer System (Bio-Rad). The membranes were blocked with Intercept Blocking Buffer (LI-COR, Lincoln, NE, USA) for 1 hr. Rituximab was detected with IRDye 800CW Goat anti-human IgG Secondary Antibody (LI-COR) at 1:15,000 in Intercept Antibody Diluent (LI-COR) for 1.5 hours at room temperature. The membranes were washed 3x for 10 minutes with 0.1% TBST and imaged using the LI-COR 9120 Odyssey Infrared Imaging System.

For staining of intracellular biotinylated proteins, 20 μg of total protein from labeled cells was loaded and transblotted. The membrane was blocked by 3% BSA in 0.1% TBST for 1 hour and probed with HRP-conjugated streptavidin (Abcam, Cambridge, UK) diluted in blocking buffer at 1:2000 for 40 minutes. For visualizing the proteins' bands, the Clarity Western ECL Substrate (Bio-Rad) was used.

Quantifying HC and LC with qPCR

Total RNA was extracted from 1–2 × 106 cells using the RNeasy Plus Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. RNA concentration was measured with Qubit fluorometric analysis (Life Technologies,Carlsbad, California, USA). cDNA was synthesized from 1.5–3 μg of total RNA using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR with dsDNAse treatment (Thermo Fisher Scientific). RT-qPCR analyses were performed on the QuantStudio 5 Real-Time PCR System using TaqMan™ Multiplex Master Mix (Thermo Fisher Scientific) in triplexes (gene of interest and two normalization genes) using the following amplification conditions: 50°C for 2 min, 95°C for 10 min; 40x: 95°C for 15 s, 60°C for 1 min. Custom-made TaqMan assays were used for rituximab light chain and heavy chain, as well as normalization genes Gnb1 and Fkbp1a (Brown et al., 2018). All the primers and probes were validated by melting curve analysis and primer efficiency test. Using the ΔΔCt method, the relative expression levels of rituximab heavy and light chains were calculated by normalization to the geometric mean of expression levels of the two normalization genes. Each experiment included controls with no template and was performed using technical triplicates.

Quantification of PPIs of Rituximab in CHO-S cells using the FcBAR method

FcBAR was used to target rituximab in 10 CHO-S clones that express rituximab at different productivity levels, in order to identify PPIs in each clone. To accomplish this, CHO-S clones expressing rituximab and non-producing CHO-S control cells, each in technical triplicate, were harvested at mid-exponential phase, fixed in 4% paraformaldehyde (PFA) in PBS (Thermo Scientific), and permeabilized with 0.4% PBST. Endogenous peroxidases were inactivated with 0.4% H₂O₂ (Fisher Scientific, USA), and cells were blocked with 5% goat serum (Gibco) in 1% BSA-PBST. Cells were then incubated with goat anti-human IgG Fc secondary antibody conjugated to HRP (Invitrogen, USA) for 1 hour at room temperature with rotation. Proximal protein biotinylation was performed by treating cells with H₂O₂ and tyramide-biotin using the TSA Biotin Reagent Pack (SAT700001EA, Akoya Biosciences, Marlborough, MA, USA), resulting in tyramide-biotin radicalization and deposition onto proximal proteins. The reaction was quenched with 0.5 M sodium ascorbate in PBS (Sigma-Aldrich). Cell lysates were extracted using 1.5% SDS (G Biosciences, St. Louis, MO) and 1% sodium deoxycholate (bioWorld, Dublin, OH, USA) in 0.1% PBST, then heated at 99°C for 1 hour with mild shaking. Samples were centrifuged at maximum speed for 5 minutes, and the supernatant was collected for protein quantification using the BCA Protein Assay (Lambda Biotech, Inc., St. Louis, MO, USA). Samples were stored at −80°C until shipment to the Sanford Burnham Prebys Proteomics Core (San Diego, CA, USA) for LC-MS/MS analysis. Biotinylated proteins were enriched using streptavidin beads, digested with trypsin, and subjected to LC-MS/MS analysis.

Fluorescence microscopy

Colocalization of rituximab and biotinylated proteins in the endomembrane system was assessed by fluorescent staining and confocal super-resolution microscopy following the BAR protocol. For colocalization studies, a subsample of labeled cells was probed with goat anti-human-DyLight 650 conjugate (Thermo Fisher Scientific) and streptavidin-DyLight 594 conjugate (Thermo Fisher Scientific) to target rituximab and biotinylated proteins, respectively. These antibodies were diluted 1:300 and 1:1000 in the blocking buffer, respectively, and incubated with fixed cells for 30 minutes at room temperature. Cells were washed, counterstained with DAPI, and mounted on slides using antifade Vectashield mounting medium (VectorLabs, Newark, CA, USA). Images were acquired using a Leica SP8 confocal microscope with Lightning deconvolution (Wetzlar, Germany). Colocalization analysis was performed on deconvolved images using the Coloc_2 plugin in Fiji (ImageJ 1.52p) to assess pixel intensity overlap between the 650 nm channel (rituximab) and 594 nm channel (streptavidin) (Schindelin et al., 2015). This analysis generates Pearson's correlation coefficients (range: −1.0 to 1.0) to quantify the degree of colocalization between the two fluorescent signals. Background subtraction was applied using Fiji's rolling ball algorithm within defined regions of interest, and threshold values were automatically determined using Coloc_2's bisection method to optimize background correction.

LC/MS-MS on Biotinylated Samples

The BAR samples were sent to Sanford Burnham Prebys Proteomic Core for LC-MS/MS analysis. Biotinylated proteins were affinity-purified using the Bravo AssayMap platform (Agilent) with AssayMap streptavidin cartridges (Agilent). On-cartridge digestion of the bound proteins was performed using mass spectrometry-grade Trypsin/Lys-C Rapid digestion enzyme (Promega, Madison, WI) at 70°C for 2 hours. The resulting peptides were desalted on the Bravo platform using AssayMap C18 cartridges. Organic solvents were removed using a SpeedVac concentrator prior to LC-MS/MS analysis. The dried peptides were reconstituted in 2% acetonitrile with 0.1% formic acid and analyzed via LC-MS/MS using a Proxeon EASY nanoLC system (Thermo Fisher Scientific) coupled with a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific).

Peptide separation was conducted on an analytical C18 Aurora column (75μm × 250 mm, 1.6μm particles; IonOpticks) at a flow rate of 300 nL/min. The 80-minute gradient applied was: 0% to 6% buffer B over 0.5 minutes, 6% to 23% B over 50 minutes, 23% to 34% B over 29 minutes, and 34% to 48% B over 0.5 minutes (Buffer A: 0.1% FA; Buffer B: 80% ACN with 0.1% FA). The mass spectrometer operated in positive data-dependent acquisition mode. MS1 spectra were acquired at a resolution of 70,000, with an AGC target of 1e6 and a mass range of 350–1700 m/z. Up to 12 MS2 spectra per duty cycle were triggered, fragmented using HCD, and acquired at a resolution of 17,500, with an AGC target of 5e4, an isolation window of 1.6 m/z, and a normalized collision energy of 25. Dynamic exclusion was enabled with a duration of 20 seconds.

Mass spectra were analyzed using MaxQuant software (Tyanova, Temu, & Cox, 2016), version 1.5.5.1. MS/MS spectra were searched against the Cricetulus griseus Uniport (version August 2020) and GPM cRAP sequences (common protein contaminants). Precursor mass tolerance was set to 20 ppm for the initial search, which included mass recalibration, and 4.5 ppm for the main search. Product ions were searched with a mass tolerance of 0.5 Da. The maximum precursor ion charge state for the search was set to 7. Carbamidomethylation of cysteines was included as a fixed modification, while oxidation of methionines and N-terminal acetylation were considered variable modifications. The enzyme specificity was set to trypsin, allowing up to two missed cleavages. The target-decoy-based false discovery rate (FDR) filter for spectrum and protein identification was set to 1%.

FcBAR Analysis

The FcBAR proteomic mass spectrometry dataset was preprocessed and analyzed using the DEP package (Zhang et al., 2018) in RStudio to identify differentially expressed proteins. Proteins were assigned unique names, and samples were filtered to remove proteins with excessive missing values. Specifically, proteins were retained only if they were identified in at least 2 out of 3 replicates for any given condition. Samples were normalized using variance stabilizing normalization (VSN), and missing values were imputed based on data missing not at random (MNAR) using a manually defined left-shifted Gaussian distribution. The data were then visualized using Principal Component Analysis (PCA) of the normalized and imputed data. Proteins were filtered using the following criteria: detected in ⅔ replicates for each producer clone and present in <2 replicates in control, or present in ⅔ replicates in control but having p-adjusted < 0.05. These were marked as potential target proteins. Welch's t-test was performed comparing high producer cell lines versus low producer cell lines to identify proteins that were significantly expressed with p < 0.05 and fold change of >0 in the high producer cell lines. Linear regression analysis was performed in RStudio on the FcBAR proteomic data, examining the relationship between the mean abundance of each protein and the productivity level (pg/cell/day) from each rituximab clone.

RNA-Seq Prep and Analysis

The rituximab clones and its parental cell line were each cultured in triplicate. They were each harvested at the same time as the sample used for FcBAR. Total RNA was extracted from each sample using the RNeasy Kit (Qiagen), following the manufacturer’s instructions, and then quantified using nanodrop. Samples were sent to UCSD IGM Genomics Center for mRNA library preparation and sequencing. RNA quality was analyzed using the Agilent Tapestation 4200, and only samples with an RNA Integrity Number (RIN) exceeding 8.0 were selected for library construction. Libraries were prepared using the Illumina Stranded mRNA Sample Prep Kit with Illumina RNA UD Indexes (Illumina, San Diego, CA), following the manufacturer's guidelines. The finalized libraries were multiplexed and sequenced on an Illumina NovaSeq X Plus platform, generating 150 base pair (bp) paired-end reads (PE150) with an approximate depth of 25 million reads per sample. The samples were quality controlled using FastQC and the adapter sequences trimmed using Trimmomatic. The sequences were demultiplexed with bcl2fastq v2.20 Conversion Software. Reads were mapped to the CHO GCF_003668045.1_CriGri-PICR genome, aligned and quantified with Salmon to obtain TPM values. Statistical analysis was done in RStudio.

Generating Plasmids for Overexpression

The Acsl4, Agpat4, Ephx1, Loc100774522, Nsdhl, Psme2, Pgrmc1, and the empty vector backbone (pcDNA3.1(+)) plasmids were designed from Chinese amster (Cricetulus griseus) and purchased from GenScript (Piscataway, NJ, USA). The plasmids were transformed into DH5ɑ Competent Cells (Invitrogen) and expanded in Miller's LB Broth (Corning Inc., Corning, NY, USA) supplemented with 100 μg/ml of ampicillin (Sigma-Aldrich). The plasmids were extracted using QIAprep Spin Miniprep Kit (Qiagen).

Transient Transfection

PEI Max transient transfection was used for overexpression of the validation genes. PEI MAX-Transfection Grade Linear Polyethylene Hydrochloride (MW 40,000) (Kyfora Bio, Warrington, PA, USA) was dissolved at 1 mg/ml in distilled water and pH was adjusted to 7.0 using 1M of NaOH. The dissolved PEI was sterile filtered in 0.22 μm steritop (MilliporeSigma, Burlington, MA, USA). A day before transfection, cells were passaged at 0.5–0.8 × 106 cells/ml in media containing no anti-clumping. On the day of transfection, cell counts were done and were at least > 95% viable for transfection. Cells were harvested and resuspended at 1.0 × 106 cells/ml in media containing no anti-clumping. Cells were transfected with 1 μg plasmid and 7 μg of PEI Max solution per ml of cell culture (5% of total culture volume). The DNA/PEI mixture was incubated in OptiPRO SFM (Life Technologies) for 20 minutes at room temperature before adding to the cells in 6-well non-treated, polystyrene, flat bottom wells plate (Genesee Scientific). The cells were shaken at 130 RPM at 37°C in 5% of CO2. One day post-transfection, 1× of anticlumping agent (Gibco) and 0.6 mM valproic acid sodium salt (Sigma-Aldrich) were added to each well. Some cells were harvested on day 2 post transfection for RNA extraction. Viability was performed on day 4 post-transfection and supernatant was harvested.

qPCR to Confirm Gene Overexpression

RNA extraction was done day 2 post transfection following the RNeasy Mini kit protocol (Qiagen) to confirm overexpression of the validation genes. RNA was converted to complementary DNA (cDNA) following the SuperScript II RT protocol (Thermofisher) with 1 μg of RNA and RNAseOUT Recombinant Ribonuclease Inhibitor (Life Technologies). Quantitative PCR (qPCR) was performed using iTaq™ Universal SYBR Green Supermix (Bio-Rad) using primers targeting the validating genes purchased from IDT DNA (USA) (Table 1) with Gnb1 as the housekeeping gene in 96-well PCR Plates (Bio-Rad). The samples were run as follows: 95°C for 2 min; 40×: 95°C for 10 sec, 60°C for 30 sec; 65°C for 5 sec, 95°C for 5 sec. Gene expression levels were analyzed using the ΔΔCt method, first normalized to the reference gene (ΔCt) and then to the empty vector control group (ΔΔCt) to determine relative fold-changes. Each experiment included no-template controls and was performed using technical duplicates.

Table 1:

Primers of validating genes for qPCR

Primers Sequence (5’- 3’)
Acsl4_Fw GGAATGACAGGCCAGTGTGA
Acsl4_RV CAGCACATGAGCCAAGGGTA
Agpat4_Fw CTCCGGGACTACCCAGAGAA
Agpat4_Rv TCTCGCAAGCACTTAACGGT
Ephx1_Fw ACAAGGAGAACATGGGCCAG
Ephx1_Rv TGTGCAGTAGCTCACAAGGG
LOC100774522_Fw CTGGAAGAGCATCCTGGTGG
LOC100774522_Rv TTGGACAGTTCCCGAGCATC
Nsdhl_Fw AGCCAATCCCTTTCTGGACG
Nsdhl_Rv GGACTGACCACCATCACCAG
Pgrmc1_Fw GGAAGAGATGCATCCAGGGG
Pgrmc1_Rv CCTTCAGCAGCTTACCCACA
Psme2_Fw TGGACATCCCCATCCCAGAT
Psme2_Rv CTACAGCAAACAGGGCCAGA
Open in a new tab

Fc-ELISA

Rituximab secretions (supernatant) were quantified by Fc-ELISA on day 4 post transfection of the validation genes. Ninety-six-well microplates (Maxisorp; Thermo Fisher) were coated with 2 μg/ml of Goat Anti-Human IgG (I1866; Sigma-Aldrich) in coating buffer (Elisa Reaction Buffer, Carbonate-Bicarbonate, 50 mM; BioWorld) and incubated at 4°C overnight. After the plates were washed 3× with 0.05% PBST, the plates were blocked with 1% BSA in PBS for 2 hours at 37°C. The plates were washed 3× with 0.05% PBST. The supernatant was tested in duplicate at 1:600 along with the standard IgG human serum (I4506; Sigma-Aldrich) in blocking buffer. The samples were incubated at 37°C for 2 hrs. After washing 3× with 0.05% PBST, the samples were incubated with anti-human IgG (Fc specific) HRP (A0170; Sigma-Aldrich) at 1:5,000 in PBS for 1 hour at 37°C. The plates were washed 3× with 0.05% PBST before adding TMB substrate (T8665; Sigma-Aldrich) and incubated for 4–5 minutes. The reaction was stopped with 1M sulfuric acid. The absorbance was read at 450 nm and 595 nm using the BioTek Synergy MX microplate reader (Agilent, Santa Clara, CA, USA). The concentrations were calculated by curve fitting in GraphPad Prism software with hyperbola.

Statistical Analyses

Statistical analysis of BAR proteomic data was performed using the DEP package (version 1.24.0) in RStudio (version 4.3.3) and Microsoft Excel. RNA sequencing data was statistically analyzed in RStudio. Gene Ontology (GO) was performed using the clusterProfiler package (version 4.10.1) in RStudio with genome annotation provided by the org.Hs.eg.db package (version 3.18.0). Data visualization included volcano plots, RNA-Seq PCA plots, and enrichment pathway plots, which were generated using the ggplot2 package (version 3.5.0). BAR proteomic PCA plots were created with the DEP package. Bar graphs for qPCR and validated gene expression data were generated using GraphPad Prism using paired t-test (version 10.4.0). All figures were designed and assembled using BioRender.com.

Supplementary Material

Supplementary Table 1
Supplementary Table 3
Supplementary Table 2
Supplementary Figure 1

Supplementary Figure S1: Rituximab correlation of mRNA heavy chain (HC) and light chain (LC) expression with productivity (pg/cell/day), clone secretion confirmation by western blot, and rituximab E9 overexpression viability day 4 post-transfection (A) qPCR was performed to quantify the mRNA expression of the HC and LC of rituximab. Productivity (pg/cell/day) was determined with SPR of each rituximab-producing clone. Spearman rank correlation analysis between HC mRNA expression and productivity revealed a low positive correlation that was not statistically significant (r = 0.4185, p = 0.2277). In contrast, a significant correlation was observed between LC mRNA expression and productivity, with an outlier identified among the cell lines (r = 0.7378, p = 0.0186). (B) Rituximab secretion from clone supernatants was confirmed using western blot analysis. The membrane was stained with IRDye 800CW Goat-anti-Human IgG Secondary Antibody and imaged with a LI-COR infrared (IR) scanner, confirming rituximab secretion and demonstrating the expected light chain (~25 kDa) and heavy chain (~55 kDa) bands. Rituximab clone B2 was not included. (C) Cell count and viability were assessed on day 4 post-transfection with the validation genes. Paired t-test analysis showed no significant difference (p > 0.05) in viable cell count and cell viability between overexpression of the validated genes and the control (empty vector backbone).

Acknowledgements

This work was supported by generous funding from NIH (R35 GM119850), NSF (CBET-2030039), and the Novo Nordisk Foundation (NNF20SA0066621). Light microscopy was performed at UCSD School of Medicine Microscopy Core, which is supported by Grant P30 (NS047101). The authors would also like to thank Henning Gram Hansen for his support in generating the rituximab clones.

Footnotes

Declaration of Interests

NEL is a scientific advisor for CHO Plus, and co-founder of NeuImmune, Inc. and Augment Biologics, Inc.

Data Availability

Study findings are supported by data available upon request from the corresponding author. Mass spectrometry proteomics data, raw and processed, has been uploaded to theMassIVE (member of the proteome Xchange consortium) repository at https://doi.org/10.25345/C5ZZ0V, reference number PXD063116. Raw and processed transcriptomic data are deposited in GEO (GSE295486). The code for the FcBAR proteomic and transcriptomic processing and analysis is at [https://github.com/lewiscelllabs/FcBAR_Rituximab_Proteomic_and_Transcriptomic_Analysis].

References

  1. Baldo BA (2016). Monoclonal Antibodies Approved for Cancer Therapy. Safety of Biologics Therapy, 57–140. 10.1007/978-3-319-30472-4_3 [DOI] [Google Scholar]
  2. Bar DZ, Atkatsh K, Tavarez U, Erdos MR, Gruenbaum Y, & Collins FS (2018). Biotinylation by antibody recognition-a method for proximity labeling. Nature Methods, 15(2), 127–133. 10.1038/nmeth.4533 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Belviso BD, Mangiatordi GF, Alberga D, Mangini V, Carrozzini B, & Caliandro R (2022). Structural Characterization of the Full-Length Anti-CD20 Antibody Rituximab. Frontiers in Molecular Biosciences, 9, 823174. 10.3389/fmolb.2022.823174 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bosch JA, Chen C-L, & Perrimon N (2021). Proximity-dependent labeling methods for proteomic profiling in living cells: An update. Wiley Interdisciplinary Reviews. Developmental Biology, 10(1), e392. 10.1002/wdev.392 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Brown AJ, Gibson S, Hatton D, & James DC (2018). Transcriptome-Based Identification of the Optimal Reference CHO Genes for Normalisation of qPCR Data. Biotechnology Journal, 13(1), 1700259. 10.1002/biot.201700259 [DOI] [PubMed] [Google Scholar]
  6. Caldas H, & Herman GE (2003). NSDHL, an enzyme involved in cholesterol biosynthesis, traffics through the Golgi and accumulates on ER membranes and on the surface of lipid droplets. Human Molecular Genetics, 12(22), 2981–2991. 10.1093/hmg/ddg321 [DOI] [PubMed] [Google Scholar]
  7. Chen Y-J, Knupp J, Arunagiri A, Haataja L, Arvan P, & Tsai B (2021). PGRMC1 acts as a size-selective cargo receptor to drive ER-phagic clearance of mutant prohormones. Nature Communications, 12(1), 5991. 10.1038/s41467-021-26225-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Dinnis DM, & James DC (2005). Engineering mammalian cell factories for improved recombinant monoclonal antibody production: Lessons from nature? Biotechnology and Bioengineering, 91(2), 180–189. 10.1002/bit.20499 [DOI] [PubMed] [Google Scholar]
  9. Edin ML, Hamedani BG, Gruzdev A, Graves JP, Lih FB, Arbes SJ, Singh R, Orjuela Leon AC, Bradbury JA, DeGraff LM, Hoopes SL, Arand M, & Zeldin DC (2018). Epoxide hydrolase 1 (EPHX1) hydrolyzes epoxyeicosanoids and impairs cardiac recovery after ischemia. The Journal of Biological Chemistry, 293(9), 3281–3292. 10.1074/jbc.RA117.000298 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Gautheron J, & Jéru I (2021). The Multifaceted Role of Epoxide Hydrolases in Human Health and Disease. International Journal of Molecular Sciences, 22(1), Article 1. 10.3390/ijms22010013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Gautheron J, Morisseau C, Chung WK, Zammouri J, Auclair M, Baujat G, Capel E, Moulin C, Wang Y, Yang J, Hammock BD, Cerame B, Phan F, Fève B, Vigouroux C, Andreelli F, & Jeru I (2021). EPHX1 mutations cause a lipoatrophic diabetes syndrome due to impaired epoxide hydrolysis and increased cellular senescence. eLife, 10, e68445. 10.7554/eLife.68445 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. González-Hernández Y, & Perré P (2024). Building blocks needed for mechanistic modeling of bioprocesses: A critical review based on protein production by CHO cells. Metabolic Engineering Communications, 18, e00232. 10.1016/j.mec.2024.e00232 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Grav LM, Lee JS, Gerling S, Kallehauge TB, Hansen AH, Kol S, Lee GM, Pedersen LE, & Kildegaard HF (2015). One-step generation of triple knockout CHO cell lines using CRISPR/Cas9 and fluorescent enrichment. Biotechnology Journal, 10(9), 1446–1456. 10.1002/biot.201500027 [DOI] [PubMed] [Google Scholar]
  14. Huber W, von Heydebreck A, Sültmann H, Poustka A, & Vingron M (2002). Variance stabilization applied to microarray data calibration and to the quantification of differential expression. [DOI] [PubMed]
  15. Ishikawa C, Barreyro L, Sampson AM, Hueneman KM, Choi K, Philbrook SY, Choi I, Bolanos LC, Wunderlich M, Volk AG, Watowich SS, Greis KD, & Starczynowski DT (2025). Ubiquitin-conjugating enzyme UBE2N modulates proteostasis in immunoproteasome-positive acute myeloid leukemia. The Journal of Clinical Investigation, 135(10). 10.1172/JCI184665 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kaneyoshi K, Kuroda K, Uchiyama K, Onitsuka M, Yamano-Adachi N, Koga Y, & Omasa T (2019). Secretion analysis of intracellular “difficult-to-express” immunoglobulin G (IgG) in Chinese hamster ovary (CHO) cells. Cytotechnology, 71(1), 305–316. 10.1007/s10616-018-0286-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Karagiota A, Chachami G, & Paraskeva E (2022). Lipid Metabolism in Cancer: The Role of Acylglycerolphosphate Acyltransferases (AGPATs). Cancers, 14(1), 228. 10.3390/cancers14010228 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Karp NA, Huber W, Sadowski PG, Charles PD, Hester SV, & Lilley KS (2010). Addressing accuracy and precision issues in iTRAQ quantitation. Molecular & Cellular Proteomics: MCP, 9(9), 1885–1897. 10.1074/mcp.M900628-MCP200 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kim DI, Jensen SC, Noble KA, Kc B, Roux KH, Motamedchaboki K, & Roux KJ (2016). An improved smaller biotin ligase for BioID proximity labeling. Molecular Biology of the Cell, 27(8), 1188–1196. 10.1091/mbc.E15-12-0844 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kvien TK, Patel K, & Strand V (2022). The cost savings of biosimilars can help increase patient access and lift the financial burden of health care systems. Seminars in Arthritis and Rheumatism, 52, 151939. 10.1016/j.semarthrit.2021.11.009 [DOI] [PubMed] [Google Scholar]
  21. Malm M, Kuo C-C, Barzadd MM, Mebrahtu A, Wistbacka N, Razavi R, Volk A-L, Lundqvist M, Kotol D, Tegel H, Hober S, Edfors F, Gräslund T, Chotteau V, Field R, Varley PG, Roth RG, Lewis NE, Hatton D, & Rockberg J (2022). Harnessing secretory pathway differences between HEK293 and CHO to rescue production of difficult to express proteins. Metabolic Engineering, 72, 171–187. 10.1016/j.ymben.2022.03.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Masson HO, Samoudi M, Robinson CM, Kuo C-C, Weiss L, Shams Ud Doha K, Campos A, Tejwani V, Dahodwala H, Menard P, Voldborg BG, Robasky B, Sharfstein ST, & Lewis NE (2024a). Inferring secretory and metabolic pathway activity from omic data with secCellFie. Metabolic Engineering, 81, 273–285. 10.1016/j.ymben.2023.12.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Masson H, Tat J, Giusto PD, Antonakoudis A, Shamie I, Baghdassarian H, Samoudi M, Robinson CM, Kuo C-C, Koga N, Singh S, Gezalyan A, Li Z, Movsessian A, Richelle A, & Lewis NE (2024b). A reconstruction of the mammalian secretory pathway identifies mechanisms regulating antibody production (p. 2024.11.14.623668). bioRxiv. 10.1101/2024.11.14.623668 [DOI]
  24. Mellahi K, Cambay F, Brochu D, Gilbert M, Perrier M, Ansorge S, Durocher Y, & Henry O (2019). Process development for an inducible rituximab-expressing Chinese hamster ovary cell line. Biotechnology Progress, 35(1), e2742. 10.1002/btpr.2742 [DOI] [PubMed] [Google Scholar]
  25. Minami Y, Kawasaki H, Minami M, Tanahashi N, Tanaka K, & Yahara I (2000). A Critical Role for the Proteasome Activator PA28 in the Hsp90-dependent Protein Refolding*. Journal of Biological Chemistry, 275(12), 9055–9061. 10.1074/jbc.275.12.9055 [DOI] [PubMed] [Google Scholar]
  26. Ohashi M, Mizushima N, Kabeya Y, & Yoshimori T (2003). Localization of mammalian NAD(P)H steroid dehydrogenase-like protein on lipid droplets. The Journal of Biological Chemistry, 278(38), 36819–36829. 10.1074/jbc.M301408200 [DOI] [PubMed] [Google Scholar]
  27. Pagliuso A, Valente C, Giordano LL, Filograna A, Li G, Circolo D, Turacchio G, Marzullo VM, Mandrich L, Zhukovsky MA, Formiggini F, Polishchuk RS, Corda D, & Luini A (2016). Golgi membrane fission requires the CtBP1-S/BARS-induced activation of lysophosphatidic acid acyltransferase δ. Nature Communications, 7, 12148. 10.1038/ncomms12148 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Peng R-W, Abellan E, & Fussenegger M (2011). Differential effect of exocytic SNAREs on the production of recombinant proteins in mammalian cells. Biotechnology and Bioengineering, 108(3), 611–620. 10.1002/bit.22986 [DOI] [PubMed] [Google Scholar]
  29. Pereira RC, Abreu PH, Rodrigues PP, & Figueiredo MAT (2024). Imputation of data Missing Not at Random: Artificial generation and benchmark analysis. Expert Systems with Applications, 249, 123654. 10.1016/j.eswa.2024.123654 [DOI] [Google Scholar]
  30. Pfeiffer CT, Paulo JA, Gygi SP, & Rockman HA (2022). Chapter 11—Proximity labeling for investigating protein-protein interactions. In Shukla AK (Ed.), Methods in Cell Biology (Vol. 169, pp. 237–266). Academic Press. 10.1016/bs.mcb.2021.12.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Pristovšek N, Hansen HG, Sergeeva D, Borth N, Lee GM, Andersen MR, & Kildegaard HF (2018). Using Titer and Titer Normalized to Confluence Are Complementary Strategies for Obtaining Chinese Hamster Ovary Cell Lines with High Volumetric Productivity of Etanercept. Biotechnology Journal, 13(3), 1700216. 10.1002/biot.201700216 [DOI] [PubMed] [Google Scholar]
  32. Pristovšek N, Nallapareddy S, Grav LM, Hefzi H, Lewis NE, Rugbjerg P, Hansen HG, Lee GM, Andersen MR, & Kildegaard HF (2019). Systematic Evaluation of Site-Specific Recombinant Gene Expression for Programmable Mammalian Cell Engineering. ACS Synthetic Biology, 8(4), 758–774. 10.1021/acssynbio.8b00453 [DOI] [PubMed] [Google Scholar]
  33. Pybus LP, Dean G, West NR, Smith A, Daramola O, Field R, Wilkinson SJ, & James DC (2014). Model-directed engineering of “difficult-to-express” monoclonal antibody production by Chinese hamster ovary cells. Biotechnology and Bioengineering, 111(2), 372–385. 10.1002/bit.25116 [DOI] [PubMed] [Google Scholar]
  34. Radif Y, Ndiaye H, Kalantzi V, Jacobs R, Hall A, Minogue S, & Waugh MG (2018). The endogenous subcellular localisations of the long chain fatty acid-activating enzymes ACSL3 and ACSL4 in sarcoma and breast cancer cells. Molecular and Cellular Biochemistry, 448(1), 275–286. 10.1007/s11010-018-3332-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Rechsteiner M, & Hill CP (2005). Mobilizing the proteolytic machine: Cell biological roles of proteasome activators and inhibitors. Trends in Cell Biology, 15(1), 27–33. 10.1016/j.tcb.2004.11.003 [DOI] [PubMed] [Google Scholar]
  36. Rega C, Pardo M, Martin L-A, & Choudhary J (2025). Protocol to identify endogenous proximal proteins using biotinylation by antibody recognition. STAR Protocols, 6(1), 103547. 10.1016/j.xpro.2024.103547 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Schindelin J, Rueden CT, Hiner MC, & Eliceiri KW (2015). The ImageJ ecosystem: An open platform for biomedical image analysis. Molecular Reproduction and Development, 82(7–8), 518–529. 10.1002/mrd.22489 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Sears RM, May DG, & Roux KJ (2019). BioID as a Tool for Protein-Proximity Labeling in Living Cells. Methods in Molecular Biology (Clifton, N.J.), 2012, 299–313. 10.1007/978-1-4939-9546-2_15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Selewski DT, Shah GV, Mody RJ, Rajdev PA, & Mukherji SK (2010). Rituximab (Rituxan). AJNR: American Journal of Neuroradiology, 31(7), 1178–1180. 10.3174/ajnr.A2142 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Templeton N, Dean J, Reddy P, & Young JD (2013). Peak antibody production is associated with increased oxidative metabolism in an industrially relevant fed-batch CHO cell culture. Biotechnology and Bioengineering, 110(7), 2013–2024. 10.1002/bit.24858 [DOI] [PubMed] [Google Scholar]
  41. Tyanova S, Temu T, & Cox J (2016). The MaxQuant computational platform for mass spectrometry-based shotgun proteomics. Nature Protocols, 11(12), 2301–2319. 10.1038/nprot.2016.136 [DOI] [PubMed] [Google Scholar]
  42. Václavíková R, Hughes DJ, & Souček P (2015). Microsomal Epoxide Hydrolase 1 (EPHX1): Gene, Structure, Function, and Role in Human Disease. Gene, 571(1), 1–8. 10.1016/j.gene.2015.07.071 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Walsh G, & Walsh E (2022). Biopharmaceutical benchmarks 2022. Nature Biotechnology, 40(12), Article 12. 10.1038/s41587-022-01582-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Zhang X, Smits AH, van Tilburg GB, Ovaa H, Huber W, & Vermeulen M (2018). Proteome-wide identification of ubiquitin interactions using UbIA-MS. Nature Protocols, 13(3), 530–550. 10.1038/nprot.2017.147 [DOI] [PubMed] [Google Scholar]
  45. Zhukovsky MA, Filograna A, Luini A, Corda D, & Valente C (2019). The Structure and Function of Acylglycerophosphate Acyltransferase 4/ Lysophosphatidic Acid Acyltransferase Delta (AGPAT4/LPAATδ). Frontiers in Cell and Developmental Biology, 7, 147. 10.3389/fcell.2019.00147 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Table 1
NIHMS2099777-supplement-Supplementary_Table_1.xlsx (22.8KB, xlsx)
Supplementary Table 3
NIHMS2099777-supplement-Supplementary_Table_3.xlsx (101.9KB, xlsx)
Supplementary Table 2
NIHMS2099777-supplement-Supplementary_Table_2.xlsx (76.5KB, xlsx)
Supplementary Figure 1

Supplementary Figure S1: Rituximab correlation of mRNA heavy chain (HC) and light chain (LC) expression with productivity (pg/cell/day), clone secretion confirmation by western blot, and rituximab E9 overexpression viability day 4 post-transfection (A) qPCR was performed to quantify the mRNA expression of the HC and LC of rituximab. Productivity (pg/cell/day) was determined with SPR of each rituximab-producing clone. Spearman rank correlation analysis between HC mRNA expression and productivity revealed a low positive correlation that was not statistically significant (r = 0.4185, p = 0.2277). In contrast, a significant correlation was observed between LC mRNA expression and productivity, with an outlier identified among the cell lines (r = 0.7378, p = 0.0186). (B) Rituximab secretion from clone supernatants was confirmed using western blot analysis. The membrane was stained with IRDye 800CW Goat-anti-Human IgG Secondary Antibody and imaged with a LI-COR infrared (IR) scanner, confirming rituximab secretion and demonstrating the expected light chain (~25 kDa) and heavy chain (~55 kDa) bands. Rituximab clone B2 was not included. (C) Cell count and viability were assessed on day 4 post-transfection with the validation genes. Paired t-test analysis showed no significant difference (p > 0.05) in viable cell count and cell viability between overexpression of the validated genes and the control (empty vector backbone).

NIHMS2099777-supplement-Supplementary_Figure_1.pdf (1.9MB, pdf)

Data Availability Statement

Study findings are supported by data available upon request from the corresponding author. Mass spectrometry proteomics data, raw and processed, has been uploaded to theMassIVE (member of the proteome Xchange consortium) repository at https://doi.org/10.25345/C5ZZ0V, reference number PXD063116. Raw and processed transcriptomic data are deposited in GEO (GSE295486). The code for the FcBAR proteomic and transcriptomic processing and analysis is at [https://github.com/lewiscelllabs/FcBAR_Rituximab_Proteomic_and_Transcriptomic_Analysis].

RESOURCES