Proteomics Reveals Distinctive Host Cell Protein Expression Patterns in Fed‐Batch and Perfusion Cell Culture Processes

Abstract

Chinese hamster ovary (CHO) cells are widely used to produce recombinant proteins, including monoclonal antibodies (mAbs), through various process modes. While fed‐batch (FB) processes have been the standard, a shift toward high‐density perfusion processes is being driven by increased productivity, flexible facility footprints, and lower costs. Ensuring the clearance of process‐related impurities, such as host cell proteins (HCPs), is crucial in biologics manufacturing. Although purification processes remove most impurities, integrated strategies are being developed to enhance clearance of some high‐risk HCPs. Current understanding of HCP expression dynamics in cell culture is limited. This study utilized data‐independent acquisition (DIA) proteomics to compare the proteomic profiles of cell culture supernatants from 14 FB clones and three perfusion clones, all expressing the same mAb from the same host cell line. Results showed that perfusion processes enhance cell growth and productivity, exhibiting distinct proteomic profiles compared to FB processes. Perfusion processes also maintain a more comparable HCP abundance profile across clones, especially for 46 problematic HCPs monitored. Cluster analysis of FB proteomics revealed distinct abundance patterns and correlations with process parameters. Differential abundance analysis identified significant protein differences between the two processes. This is the first extensive study characterizing HCPs expressed by clones under different process modes. Further research could lead to strategies for preventing or managing problematic HCPs in biologics manufacturing.

Graphical Abstract and Lay Summary

This study utilizes data‐independent acquisition (DIA) proteomics to compare host cell protein (HCP) profiles in fed batch (FB) and perfusion processes using Chinese hamster ovary (CHO) cells producing the same monoclonal antibody. Perfusion processes exhibited enhanced productivity, and stable HCP profiles across clones. In contrast, the FB process showed significant variability in HCP profiles, with distinct clusters correlating with cellular viability and process parameters.

1. Introduction

Recombinant proteins including monoclonal antibodies (mAbs) stand at the forefront of therapeutic protein development within the biopharmaceutical industry, offering targeted treatments for a diverse array of diseases. Integral to their production are Chinese hamster ovary (CHO) cells, which have solidified their position as the preferred host organism [1]. For decades, fed‐batch (FB) cell culture processes utilizing CHO cells have served as the gold standard for manufacturing mAbs, boasting operational simplicity and high yields. This mode of operation has become proficient in routinely achieving high titers ranging from 3 to 6 g/L, going up to as high as 10–15 g/L in some cases [2]. Recently, the industry is witnessing a notable transition toward perfusion processes, signaling a paradigm shift in mAb production methodologies [3]. Perfusion processes offer the potential to enhance productivity, manufacturing flexibility, and sustainability. By integrating continuous downstream processing with end‐to‐end perfusion upstream cell culture, these processes provide opportunities to intensify volumetric productivity while minimizing investment requirements [4].

In FB processes, nutrients are added to the bioreactor intermittently, and the accumulated product over time in the culture is harvested altogether. Conversely, in perfusion processes, nutrients are continuously supplied to the bioreactor while product‐containing media is simultaneously harvested, allowing for a “steady‐state operation” over a longer period and potential reduced variability in product quality. In addition to boosting volumetric productivity, the controlled environment and brief product residence time of biologic molecules within the bioreactor help reduce the likelihood of unwanted chemical or enzymatic alterations [5]. This shortened residence time is especially beneficial for molecules prone to instability, such as enzymes or bispecific antibodies [6].

Apart from high productivity and product quality of the biologics, demonstration of clearance of host cell protein (HCP) levels is critical in the final drug substance. HCPs are proteins produced by the host cell required for its function that gets released into the cell culture fluid either through active secretion or cell lysis. Some HCPs, in the final drug substance, can pose a risk to patient safety by eliciting an immune response [7, 8]. Additionally, HCPs can compromise product stability by degrading excipients, which in turn can affect the therapeutic potency or stability of the product [1]. Many of the “high‐risk” HCPs exhibit reductase, glycosidase, or protease activity, which can lead to the breakdown of the product or formulation excipients and/or potentially promote aggregation in the drug formulation, thereby impacting its shelf life [9]. For instance, lipases such as lipoprotein lipase (LPL), Sialate O‐Acetylesterase (SIAE), and phospholipase A2 group VII (LP‐PLA2), have the ability to degrade excipients in the final product formulations, specifically polysorbate 80 (PS‐80) and polysorbate 20 (PS‐20) [10, 11].

Upstream parameters can significantly influence the composition and level of HCPs, which, in turn, can affect the downstream separation process [12, 13]. For example, higher cell concentrations and extended culture durations have been shown to increase the release of HCPs into the culture supernatant [13]. Additionally, differences in cell lines have been found to cause greater variations in HCP levels compared to cell age [14]. Notable distinctions in process parameters between perfusion and FB cultures, such as nutrient and byproduct concentrations, gas flow rates, and the use of a cell retention device, contribute to significant metabolic changes and variations in process performance between these two modes [15]. In a study conducted by Templeton et al., the specific productivity of HCPs, measured by ELISA, was reported to be 58% lower in FB cultures compared to perfusion. In a comparison between FB and semi‐steady state perfusion, the ELISA‐based HCP level was found comparable after downstream purification [4]. However, the behavior of problematic and high‐risk HCPs in perfusion and FB processes remains underexplored at the individual scale.

In this work, we used data‐independent acquisition (DIA) based quantitative proteomics analysis to evaluate the proteomic profile of cell culture fluid in both FB and perfusion process modes. The DIA‐based proteomics achieves high reproducibility, sensitivity, and quantification across a broad dynamic range compared to the DDA (data dependent acquisition)‐based approaches [16, 17]. We applied this method to analyze 14 clones in FB and three clones in perfusion runs, all expressing the same monoclonal antibody (mAb) derived from the same cell line to gain insights into 46 problematic HCPs and compared their behavior across the process modes to get useful insight. Overall, our study highlights the distinct HCP clustering in the FB, steady HCP levels in perfusion, and functional differences in HCP profiles between the two process modes.

2. Materials and Methods

2.1. Cell Line

Fourteen clones, each expressing the same monoclonal antibody (IgG mAb) were used for this study. All the clones were derived from a glutamine synthetase (GS) knockout cell line. Each of these clones was passaged in shake flasks in a humidified Multitron Cell incubator at 36.5°C, 5% CO₂, 70% humidity, and 140 rpm using a chemically defined CHO medium containing 12.5 µM methionine sulfoximine (MSX, Sigma–Aldrich).

2.2. Cell Culture Operations

For the FB runs, clones were initially propagated in shake flasks using CHO medium before being transferred to AMBR15 bioreactors (Sartorius Stedim, Gottingen, Germany), which feature two pitched blade impellers and an open pipe sparger. Cells in the exponential growth phase were inoculated at a density of 0.5 × 10⁶ cells/mL in 14.5 mL of Dynamis culture media (Gibco, NY, USA). Starting from Day 3, two proprietary feeds (Feed A and Feed B in a 9:1 ratio) were added daily at 3% v/v. Glucose levels were maintained at 6 g/L by supplementation whenever they dropped below 3 g/L. On Day 5 or upon reaching a cell density of 15 × 10⁶ cells/mL, the bioreactor temperature was reduced from 36.5 to 33°C. The cells were cultured at agitation speed of 900 rpm, pH set point of 6.9 ± 0.2, and dissolved oxygen (DO) set point of 30%. Six hundred and fifty microliters of samples were collected for cell counts and metabolite analysis with Flex2 analyzer (Nova Biomedical). Cultures were harvested on Day 14 or when cell viability decreased below 50% by centrifugation followed by filtration through the 0.22‐micron filter. At least three biological replicates were conducted for clones used in comparisons with perfusion runs (i.e., Clones C9, C13, and C14), while other clones were tested once. The top three clones selected for perfusion were chosen based on a combination of desired product quality attributes, productivity in both process modes, and optimal process performance characteristics. Key selection criteria included viability, cell growth characteristics, cell retention filter performance, and lactate production, evaluated in a prior experiment in ambr250 and ambr15.

For perfusion runs, cells were passaged in CD CHO medium and then cultured in an in‐house perfusion media in 3 L bioreactors with a working volume of 1.8 L. Cells were inoculated at an initial density of 0.5 × 10⁶ cells/mL and cultured to target viable cell densities (VCD) of 50, 100, and 150 × 10⁶ cells/mL. An alternating tangential flow (ATF) filtration device equipped with a 0.2 µm polyethersulfone (PES) hollow fiber membrane (Repligen, Waltham, MA) was used for the perfusion bioreactors. The perfusion rate was gradually increased to a maximum of approximately two vessel volumes per day (VVD) at the start of the harvest phase. Cells were continuously bled to maintain the target VCD for up to 28 days. For Days 0–5, agitation was controlled at 260 rpm, and 400 rpm for the rest of the days. DO was maintained at 40 %, and the temperature was kept at 36.5°C. Glucose levels were maintained at 3.5 g/L by supplementation whenever they dropped below 3 g/L. The bioreactor pH was regulated between 6.9 and 7.3 using sodium bicarbonate and sparged carbon dioxide as needed.

2.3. Cell Culture Sample Analysis and Analytical Methods

Daily cell culture samples were collected for immediate analysis. Viable cell density (VCD) and cell viability were measured using the trypan blue exclusion method with a Cedex HiRes cell counter (Roche Diagnostics GmbH, Mannheim, Germany) for perfusion runs and Flex2 analyzer (Nova Biomedical) for the FB runs. Concentrations of glucose, lactate, glutamine, and glutamate were determined with a Flex2 analyzer (Nova Biomedical). Antibody titer (g/L) was analyzed using affinity (HPLC) based Protein A column as described previously [18]. Cation‐exchange chromatography was used to separate mAb charge variants based on their charge differences. The mAb samples were loaded onto a YMC BioPro IEX SF column (YMC America) in a low‐salt buffer (18 mM Na₂HPO₄, pH 6.0, 10% ACN) at a pH below the isoelectric point and eluted using a salt gradient (0–270 mM NaCl). Detection was carried out at 280 nm, and the percentage of the main peak, acidic variants, and basic variants was quantified by peak area normalization.

2.4. Proteomics Analysis

Proteomics sample preparation, data acquisition, and data processing were performed as previously described [16]. Briefly, the cell culture sample (20 µL) was incubated with 4 M urea, 25 mM TCEP‐HCl, and 50 mM acrylamide in 50 mM tris‐HCl pH 8.0 (total volume: 50 µL) at 60°C for 30 min. An aliquot (20 µL) was incubated with Sera‐Mag SpeedBead (10 µL, 200 µg) and acetonitrile (70 µL) at 20°C for 10 min with agitation at 500 rpm. The beads were washed three times with 80% EtOH (200 µL) and then incubated with 50 mM tris‐HCl pH 8.0 (20 µL) containing 0.4 µg trypsin and 0.1 µg LysC at 37°C for 18 h. The peptide concentration was determined with the BCA assay (Thermo Scientific) and adjusted to 0.02 g/L with 0.1% formic acid before manually loading 20 µL (400 ng) on Evotip Pure (Evosep) according to the manufacturer's instructions (https://www.evosep.com/wp‐content/uploads/2024/03/PR‐001C‐Sample‐loading‐protocolWEB.pdf). Peptides were separated on the Evosep One LC system with a Pepsep C18 column (8 cm × 100 µm, 3 µm) using the 100 samples per day method (mobile phase A: 0.1% formic acid in water; mobile phase B: 0.1% formic acid in acetonitrile; gradient time: 11.5 min; cycle time: 14.4 min; flow rate: 1.5 µL/min) and analysed on the Bruker timsTOF Pro 2 system in the diaPASEF mode. The diaPASEF method set both TIMS accumulation and ramp times at 100 ms and covered the ion mobility range 0.7–1.4 1/K ₀ and the mass range 300–1200 m/z with 16 varying‐sized windows, resulting in the cycle time estimate of 0.95 s. Spectral library was generated from repeated measurements of a sample pool, which was a mixture of aliquots from all samples, via the ion mobility–coupled gas phase fractionation (IM–GPF) approach. Peptides were separated on the Evosep One LC system with a Pepsep C18 column (8 cm × 100 µm, 3 µm) using the 60 samples per day method (mobile phase A: 0.1% formic acid in water; mobile phase B: 0.1% formic acid in acetonitrile; gradient time: 21.0 min; cycle time: 24.0 min; flow rate: 1.0 µL/min) and analyzed on the Bruker timsTOF Pro 2 system in the diaPASEF mode. The IM–GPF method set both TIMS accumulation and ramp times at 100 ms and covered the ion mobility range 0.7–1.4 1/K ₀ and the mass range 300–1200 m/z with 50 effective 1‐m/z windows (width: 2 m/z; overlap: 1 m/z) over 18 injections. No retention time calibration standard was used. Note that the Evosep separation methods are standardized and fully reproducible on the platform. Acquired spectra were processed with DIA‐NN v1.8.1 [19] with an in‐house sequence database. Mass accuracies were set to 10 ppm for spectral library generation and to 15 ppm for all other processing. MBR (match‐between runs) was disabled for spectral library generation and enabled for all other analyses. The global precursor and global protein FDR thresholds were set to 1% for all analyses.

2.5. Statistical Analysis

Differential expression analysis was conducted for each clone using the limma R package [20], fitting protein group‐wise linear models to the log2‐transformed LFQ (Label‐Free Quantification) values with the lmFit function, and comparing the manufacturing modes at each time point with the contrasts.fit function. Moderated t‐statistics were computed using the eBayes function with the robust option. p values were extracted with the topTable function, combined across time points via the Fisher's method, and adjusted for multiple testing according to the Benjamini–Hochberg procedure. Fold‐changes at each time point were averaged and used in the GSEA (Gene Set Enrichment Analysis) [21]. Unsupervised analyses including hierarchical clustering analysis and principal component analysis (PCA) used log2‐transformed and standardized LFQ values, that is, z‐scores as input. Pearson's correlation coefficient (r) was used to calculate distance matrices (1–r) for hierarchical clustering analysis following convention. Spearman coefficient was used when assessing the correlation between process parameter and HCP abundance as it is robust against outliers.

3. Results and Discussion

3.1. Perfusion Process Boosts Cell Growth and Volumetric Productivity

Different clones expressing the same recombinant protein are known to show significant differences in growth rate, viability, protein productivity, and product quality attributes. This can be attributed to differences in gene integration sites and copy numbers of the gene of interest that get integrated into the genome. Due to these differences between clones, we expanded the number of clones we evaluated in this study accounting for a range of viabilities and protein productivities. However, it is not currently well‐known how expression profiles of different HCPs compare across clones. To capture the biological variations and improve broader applicability of our analysis, we studied 14 clones run on FB mode and three clones run in perfusion mode. All the clones were derived from the same host cell line and expressed the same mAb. Figure 1 shows the process performance indicators such as viability, integrated viable cell density (IVCD) or VCD, titer, and specific productivity (Qp) of the clones run in both FB (Figure 1A) and perfusion mode (Figure 1B) at the harvest time point(s) as appropriate. In the FB runs conducted in ambr15 bioreactors, the Day 14 viabilities for clones ranged from ∼48% to ∼96% and mAb titers ranged from 2.2 to 7.2 g/L. Three out of these 14 clones were also evaluated in perfusion process (highlighted in red in Figure 1). The perfusion cultures were maintained at a target VCD starting from ∼Day 10. The Day 25 titer ranged from 0.97 g/L to a max. of 2.4 g/L and viability was ∼90% for two clones and 63% for another (C14). Perfusion cultures exhibited, on average, a 4‐fold increase in peak VCD and greater cell viability, leading to an 8‐fold increase in IVCD relative to the FB process. This led to 4‐ to‐11‐fold increase in daily HCCF (harvested cell culture fluid) productivity and 6 (C13), 10 (C14), and 16 (C9) fold increase in total productivity. This was consistent with existing literature suggesting generality of the observation [22, 23]. For the same clones operated in both the process modes, there was a moderate increase of Qp in perfusion on average, but other clones run only in FB mode exhibited comparable Qp to the perfusion suggesting a clone‐specific behavior or difference in media rather than process mode specific behavior.

FIGURE 1.

Cell culture process profiles of viability (%), integrated viable cell density (IVCD), titer (g/L), and specific productivity (Qp) for clones run in fed‐batch and perfusion modes. Red symbols indicate clones run in both fed‐batch and perfusion modes. All the clones in fed‐batch run were harvested on day 14 except for clone C10 which was harvested on day 13.

3.2. Cluster Analysis of Proteomics Profiles in FB Processes Exhibits Distinct HCP Grouping

HCP expression profiles for different clones in FB and perfusion processes were analyzed using LC‐MS/MS proteomics. For all samples, the amount of protein injected into the LC‐MS instrument was kept constant. Approximately, 1800 proteins in the HCCF were identified with high confidence (global precursor and protein FDRs < 0.01). We focused on 46 high‐risk and frequently observed HCPs due to their challenges in downstream separation and the potential risk due to their presence in the final drug substance [13, 24]. The abundance distribution of these HCPs is depicted in the heatmap in Figure 2A. Most HCPs have comparatively similar abundance across the clones. For example, HCPs such as CLU, HSPA5, PDIA3, and LPL were relatively highly abundant, whereas PLA2G15, TGFB1, PLA2G7, and PRDX3 were low abundant HCPs in the secreted fluid. However, some HCPs, such as CTSA, LIPA, and CTSZ, exhibited variable abundance that was clone‐specific. For instance, LIPA abundance varied from 7.2 to 11.2 on the log2 scale (∼16‐fold difference on linear scale).

FIGURE 2.

Proteomics profile of HCCF fluid in fed‐batch mode. (A) Heatmap of high‐risk and frequently seen HCPs for 14 clones with LFQ values (log2 scale) from Day 14. Clones C9, C13, and C14 were run at least three times and each profile displayed individually. (B) Heatmap of the entire HCCF proteomics profile, where each protein is scaled across the clones and hierarchical clustering was performed on the pairwise correlation scores (Pearson). standardized protein abundance LFQ values were labeled as z‐scores and dense colors of each process parameter represents higher values. (C) Density plots of rho values (spearman correlation scores) between various parameters and HCPs by cluster. (D) Correlation plot between process parameters and selected HCP abundances, with significant correlations (p value < 0.05) marked by *. (E) Cellular localization of HCPs in each cluster by enrichment p values. HCCF, harvested cell culture fluid; HCPs, host cell proteins.

To better capture the behavior of the HCPs as a group, we performed cluster analysis of the proteomic dataset. We scaled protein abundance across clones, performed pairwise Pearson correlation analysis, and used hierarchical clustering to identify protein clusters with similar abundance patterns. More specifically, HCPs in a cluster exhibit similar abundance patterns across each clone. Interestingly, 63% of the problematic HCPs (29 out of 46) belonged to cluster 4 (Figure 2B). This does not mean that cluster 4 genes are always upregulated or downregulated. Instead, if an HCP in cluster 4 has higher abundance, it is likely that other HCPs in the same cluster will also have a higher abundance, and vice versa. Majority of the high‐risk HCPs reported in Jones et al. were in fact in the same cluster 4 [24]. We then analyzed how protein abundances in each cluster correlate with the process parameters using Spearman correlation and plotted a density distribution of the correlation coefficients (Figure 2C). When considering HCPs in a particular cluster, they follow similar patterns with process parameters as seen in Figure 2D. Specifically, HCPs in clusters 1 and 2 were correlated with cell diameter and specific productivity (Qp). We observed that cluster 4 was distinctly correlated with viability and peak VCD and negatively correlated with the end‐day lactate level (Figures 1S and 2C). In contrast, cluster 3 was positively correlated with the end‐day lactate level, glutamine level and negatively correlated with glutamate level, likely suggesting that cluster 3 is metabolically different than cluster 4.

Cellular localization of HCPs belonging to different clusters indicated that cluster 4 HCPs were primarily located in the plasma membrane or were extracellular in nature, as indicated by the p value score. All other clusters were either cytoplasmic or nuclear, with cluster 3 being significantly nuclear (Figure 2E). To discern the functional distinctiveness of cluster 4, we performed Gene Ontology (GO) term analysis to identify the distinct biological properties of the cluster (Figure 2S). Several proteins in this cluster, including LGMN, CTSA, CTSD, CTSB, CTSZ, and HTRA1, are involved in proteolytic processes, characterized by the peptidase activity GO term. Proteins such as MMP19, MMP9, MMP14, LAMB1, LAMC1, and CSPG4 play crucial roles in the structure and remodeling of the extracellular matrix, which may contribute to the activation of cell surface receptors (GO terms: Cell Surface Receptor Signaling Pathway, Regulation of Response to Stimulus) [25]. Notably, the physical subnetwork interaction analysis reveals that matrix metalloproteinases form a tightly interconnected physical interaction network within Cluster 4 (Figure 3S). Many proteins in this cluster, including ANXA2, CLU, NUCB2, ACTB, and PLTP, are involved in vesicle‐mediated transport. The increased lipid metabolism, represented by HCPs such as LIPA, LPL, PLA2G7, and PLA2G15, suggests that clones upregulating this cluster of proteins likely rely on an extracellular lipid source, which is internalized by vesicle‐mediated transport, as evidenced by PLTP's role in lipid transport. Unlike the Cluster 4 proteins which were cation binding, Cluster 3 proteins were mostly anion binding in nature (Figure 2S). Many of them have nucleic acid and RNA binding behavior which would contribute to the GO terms “Regulation of gene expression” and “Regulation of RNA biosynthetic process”. HCPs in this cluster were involved in managing cellular response to the stress, predominantly by protein folding activity. When combined with the low viability, this suggest that in cases where cluster 3 HCPs are more abundant, the cells are likely experiencing greater stress and undergoing cell lysis.

3.3. Steady‐State HCP Abundance in Perfusion Mode Contrasts Dynamic Changes in FB Processes

Unlike the FB process, where the proteomic profile of HCCF fluid changes dynamically over time in terms of HCP number, and abundance [26, 27], most HCPs, including problematic ones, maintain steady abundance throughout the culture in perfusion mode (Figure 3A). On average, there was a slight increase of abundance for the HCPs PLTP, CTSB, CALR, HSP90B1, CCL2, PRDX4 whereas slight decrease of abundance for ANXA5, NID1, MMP9, RACK1, LPL as the days progress. Although there were changes in the process output such as viability, Qp, mAb titer, the three clones run in the perfusion mode maintained similar abundance of HCPs (Figure 3B). Similar to the observation in FB, HCPs such as CLU, PPIA, GSTP1, HSPA5 have higher abundance in perfusion mode whereas HCPs such as PRDX3, PLA2G15, MMP9, TGFB1 had low abundance across the clones in perfusion. However, unlike the FB, the HCPs did not cluster densely as a group when clustered based on the similarity of abundance pattern (data not shown). This is likely because the HCP abundance profile of the perfusion run is at a steady state for all the clones analyzed resulting in a lack of correlation. This can be attributed to the continuous harvest of cell culture fluid and the simultaneous addition of feeds, which dilute the accumulation of HCPs at any given time. Consequently, the dynamic equilibrium established in perfusion systems prevents significant temporal or clonal variability in HCP profiles.

FIGURE 3.

Proteomics profile of HCCF fluid in perfusion mode. (A) Time course abundance trends of high‐risk and frequently seen HCPs stratified by clone and HCP, fitted using a generalized additive model. (B) Heatmap of proteomics profiles of clones by day, displaying LFQ values (log2 scale) with hierarchical clustering. HCPs are labeled near their clusters. HCCF, harvested cell culture fluid; HCPs, host cell proteins.

3.4. Distinct Proteomic Profiles Between FB and Perfusion Processes and Their Functional Implications

To evaluate the impact of process mode on HCP expression patterns, we assess the data from three clones run in both FB and perfusion processes. Multivariate data analysis was then performed to analyze these patterns. At a high level, the proteomics profile of the clones run in the FB process was distinct from that of the perfusion run on a PCA plot, with differentiation observed in PC2 (17%) (Figure 4S). This distinction occurred despite similar amounts of protein being loaded and recovered from the LC‐MS/MS run, suggesting a change in the abundance pattern of the proteins. The proteins contributing most to the PC2 difference are listed in Figure 4S and have roles in stress regulation, protein quality control, RNA metabolism, and apoptosis. For instance, HSPA4, PTMA, and CHID1 (downregulated) are involved in protein folding and stress response, while BAX and NIBAN1 (mitigates apoptotic signals promoting survival pathways), which were upregulated in perfusion, are involved in apoptosis. SRSF1, which functions in mRNA splicing and RNA metabolism, was downregulated in perfusion HCCF. Proteins responsible for RNA processing and splicing were enriched in the FB process HCCF. This suggests a potential increase of cellular stress in FB, likely due to nutrient starvation or byproduct accumulation, resulting in altered gene expression and upregulation of RNA processing proteins. This stress can lead to cell lysis, causing protein leakage into the culture fluid.

In line with this, cellular component analysis showed that compared to FB, the perfusion culture had greater abundance of HCPs which are extracellular in nature throughout the different time points (Figure 5S). Moreover, the extracellular and plasma membrane‐associated protein abundance increased while the nucleus‐associated HCPs decreased as the culture progressed suggesting active secretion by more viable perfusion process. It is to be noted that in both perfusion and FB process, the HCPs which were extracellular in nature were associated with higher viability, supporting previous reports that high viability might lead to active secretion of these HCPs [13, 28].

We performed differential abundance analysis to identify proteins with significant differences. Using linear modeling, we compared each day (9, 11, 14, 18, 21, and 25) of each clone in the perfusion‐run with Day 14 of the FB run of the same clone. The p values for each protein from each comparison were combined using Fisher's Chi‐squared test. Figure 4A highlights the HCPs with significant differential abundance between the process modes. We found that most cathepsin proteins and MMP14 were relatively upregulated in the perfusion process, implying predominant proteolysis and remodeling of extracellular matrix components. Cathepsin D is known to cause antibody fragmentation and particle formation through its proteolytic activity [29, 30]. Although the perfusion process is thought to offer a more favorable intracellular redox environment compared to FB culture [31], which may reduce the risk of fragmentation, caution is still warranted if antibody degradation is a concern for a particular process. Additionally, the shorter residence time in perfusion may limit mAb exposure to cathepsins, potentially reducing the risk of degradation. Additionally, PLA2G15, LIPA, and PLBD2 were also significantly upregulated C13 and C14 clones in the perfusion process and similarly trended in C9. Consequently, the GO terms “membrane lipid catabolic process,” “positive regulation of proteolysis involved in protein catabolic process,” and “peptidase activity” were upregulated in perfusion process (Figure 4B). In contrast, PDIA3, PDIA4, and HSPA5, that are involved in proper protein folding, were less abundant in the cell culture fluid of the perfusion run compared to the FB process. Similarly, PRDX6, which provides antioxidant defense and is secreted during cellular stress, was detected less in the perfusion run HCCF fluid. Overall, this suggests that the perfusion process may be better at handling cellular stress. This is in line with Sinharoy et al. which showed that the favorable intracellular redox state dictated by the reduced glutathione is comparatively higher in perfusion than FB promoting a better mitochondrial health, thus leading to reduced intracellular stress [31]. Indeed, when we measured the total abundance of five previously established cell death markers (GAPDH, LGALS1, CFL1, TAGLN2, and MDH) [32], we found that the FB process trended higher in these markers compared to the perfusion process (Figure 4C). The biological GO term analysis also indicated that the perfusion run experienced decreased transcription, splicing, and mRNA processing GO terms in the HCCF, as evidenced by the lower ratio of related genes SRSF1, SRSF7, HNRNPL, and DHX9 (Figure 6S), suggesting less cellular stress.

FIGURE 4.

Comparative proteomics between fed‐batch and perfusion processes. (A) Volcano plot showing differential protein abundance of perfusion versus fed‐batch processes, highlighting selected HCPs with log2 fold change > 0.5 and adjusted p value < 0.01 in red. The horizontal dashed line indicates an adjusted p value < 0.05. (B) GSEA analysis of fold change abundance values of perfusion over fed‐batch for positive enrichment scores, displaying top GO terms for BP, MF, and CC on the left, with GSEA enrichment plots for selected GO terms on the right. (C) Average death marker percentage calculated as the fraction of death markers relative to total LFQ value for each sample. Each dot represents a sample within specific time points or conditions (perfusion and fed‐batch), with horizontal lines showing mean values. (D) The bar graph illustrates the average log fold change of PSDE (Polysorbate degradation enzyme) HCPs in perfusion compared to fed‐batch, with error bars representing the standard error of the mean for each protein. BP, biological processes; CC, cellular components; GO, gene ontology; HCPs, host cell proteins; MF, molecular functions.

Scientific literature indicates that certain HCPs preferentially bind to the mAb [33, 34, 35]. Charge variants of a mAb can influence mAb–HCP binding, especially since most HCPs are anionic (lower pI) [36] and the mAb (pI 8.8) is cationic at neutral pH. During a typical ProA capture and wash step in neutral pH buffer, these low pI HCPs could preferentially bind the acidic variant of the mAb. We compared the charge distribution of the mAb between the two process modes (Figure 7S) and found that the perfusion mode mAb had ∼2‐fold reduction in acidic variants compared to FB, which can impact HCP levels in the ProA pool.

4. Conclusions

HCPs are critical quality attributes that can impact the immunogenicity, activity, and stability of the final drug substance. This study aimed to examine HCP abundance at an individual level in relation to process mode across clonal diversity, providing insights into whether different HCPs are modulated collectively or exhibit systematic differences. We monitored the HCP abundances in the HCCF even though most of them can be removed in the downstream processing. The prominent reason is that it gives an overall impression of the HCP dynamics as a function of the behavior of the cell. For example, it has been postulated that the cell undergoing unfolded protein response may contribute to the aggregate formation which would be responsible for HCP persistence in the downstream polishing steps [37, 38]. Proteomics profile of the HCCF would capture the molecular behavior of the cell culture fluid to predict the mechanistic understanding of HCP persistence. It might also be able to provide a glimpse of variability of HCP level in the downstream processing steps for select HCPs which are mAb bound or extracellular in nature.

Our comprehensive proteomics analysis reveals significant differences in the HCP profiles between FB and perfusion processes in CHO cell cultures. The FB process exhibits a dynamic proteomic profile with pronounced changes in HCP abundance between clones. The clustering of the HCPs provides a glimpse into the biological behavior such as cellular stress and RNA processing activities along with process attribute differences which can be harnessed to modulate process conditions to minimize HCPs of concern. For instance, if one HCP in cluster 4 is abundant, others are likely to be as well. This predictability allows us to adjust process conditions to minimize concerning HCPs by targeting the entire cluster's abundance. Our finding indicates significant variability among clones in the FB cell culture process. Given these differences, clone selection must be emphasized as a critical step for effective high‐risk HCP control in FB culture. In contrast to the FB, the perfusion process maintains a more stable HCP profile, suggesting enhanced control over cellular stress responses and reduced RNA processing demands that usually happen to cope with the stress. These differences in protein abundance were evident, with stress‐related and protein‐folding enzymes being relatively less prevalent in the perfusion culture fluid. Additionally, there was an enrichment of GO terms such as “vacuole”, “lytic vacuole”, and “lysosome”, indicating enhanced proteolytic activity and improved handling of cellular stress [39]. The reduced charge variants of the mAb in perfusion highlights its potential for improved downstream processing efficiency with respect to reduction of HCPs coeluting with mAb. However, caution is necessary when running the process in perfusion mode if polysorbate degradation is a concern. The abundance of PSDEs (Polysorbate degradation enzymes) was amplified in the perfusion run, likely due to the secretory nature of these enzymes, which are more prevalent in the viable environment of perfusion culture [28]. Are the observed HCP differences solely due to process mode variations? Known factors such as cell viability and stress, which influence HCP distribution [28], were addressed by comparing clones with a wide viability range and standard process inputs. Although vessel size and geometry may impact HCP levels through effects on cell growth and mixing, the literature indicates minimal influence on HCP concentrations across scales [40]. Thus, the culture vessel's impact on our findings is likely minimal. Additionally, studies affirm similar intracellular proteomics between the 10 mL Ambr and 300 L stainless steel tanks [41]. In addition, the Ambr 15 and 3 L bioreactor is validated as reliable scale‐down models for FB and perfusion processes, respectively [42, 43]. Although ideally, a single medium type would be ideal to rule out media impact on HCP difference, our data and previous findings suggest that media differences minimally affect overall HCP trends for a given mAb, with only certain HCP species showing media sensitivity [28].

Our findings indicate that HCP abundance profiles remain consistent during the perfusion process across different clones and days, suggesting potential implications for the ease of clone selection in the context of HCP control. This study builds a new understanding of HCP dynamics in cell culture supernatant in different process modes.

Author Contributions

Ansuman Sahoo: Data Analysis, Visualization, Writing—original draft, Writing—review & editing, Taku Tsukiadate: Experiment, Data generation, Data Analysis, Visualization, Writing—review & editing, Ray Lin: Experiment, Data generation, Erin Kotzbauer: Experiment, Data generation, Jason Houser: Experiment, Data generation, Misaal Patel: Study design, Xuanwen Li: Writing—review & editing, Supervision, Sri R. Madabhushi: Conceptualization, Writing—review & editing, Supervision.

Conflicts of Interest

The authors report no conflicts of interest.

Supporting information

Supporting information

Data Availability Statement

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