Sodium chloride impacts glycosylation and N‐ and O‐glycan site occupancy of an Fc‐fusion protein

Fc‐fusion proteins are highly complex molecules, difficult to manufacture at scale. In this work, undesired proteoforms were detected during the manufacture of a therapeutic fusion protein produced in CHO cells. These species were characterized using gel electrophoresis, size exclusion chromatography and liquid chromatography‐mass spectrometry leading to the identification of low molecular weight proteoforms presenting low N‐ and O‐glycan site occupancy, as well as a low sialylation content. Upstream process parameters were investigated, and fusion protein quality was shown to be linked to the sodium chloride content of the medium. A mitigation strategy was developed to avoid formation of unwanted glyco‐variants, resulting in an increased yield of highly glycosylated Fc‐fusion protein. The effect of sodium chloride was shown to be independent of the osmolality increase and was hypothesized to be linked to a modulation of Golgi acidity, which is required for the correct localization and function of glycosyltransferases. Altogether, this study highlights the importance of the salt balance in cell culture media used to produce highly sialylated and occupied glycoproteins, helping to maximize the yield and increase robustness of processes aiming at producing biopharmaceutical complex therapeutic molecules.

half-life.Additionally, the presence of a Fc part in fusion proteins facilitates downstream manufacturing operations, due to the possibility to use protein A based chromatography as an initial purification step.The antigen recognition part of fusion proteins is responsible for the specificity of the molecule and can be of various nature including receptors, cytokines, hormones, antibody fragments, or peptides.The region corresponding to the fusion of the proteins is essential for stability as well as activity of the molecule.Indeed, peptide linkers are often used to ensure the three-dimensional structure and favor the correct folding of proteins.However, in some cases these unnatural regions may lead to aggregation and may sometimes be prone to degradation, illustrating why it is essential to carefully design linkers ensuring protein functionality (Duivelshof et al., 2021).
Even though fusion proteins are commonly used to treat a variety of diseases, many challenges remain regarding their production, stability, and efficacy.These glycoproteins are typically manufactured in mammalian cells such as Chinese hamster ovary (CHO) cells because of their human like glycosylation machinery but are often considered difficult to express proteins due to their size and complexity.Limitations in the secretory pathway, particularly during protein processing in the endoplasmic reticulum (ER) and Golgi apparatus are considered the main bottlenecks for production at yields comparable to classical mAbs.This increased burden on the secretory machinery may also result in improper folded proteoforms, prone to aggregation and proteolytic degradation.As an example, the physicochemical and functional investigation of process fractions obtained during the development of an etanercept biosimilar revealed a correlation between reduced potency and incorrect disulfide bridging between specific cysteines in the receptor domain (Lamanna et al., 2017).
Development of appropriate cell culture media and feeds used in production processes of such difficult to express proteins is another crucial parameter to control the quality of the biotherapeutics and ensure maximum productivity.Similarly, process parameters such as pH, dissolved oxygen, cultivation time, and feeding regime can impact the critical quality attributes (cQAs) and thus need to be carefully considered.Comprehensive investigations were reported following the detection of clipped VEGFR1 Fc-fusion proteoforms.Multiple process parameters, cell culture media components, addition of protease inhibitor cocktail, and anticlumping agent were tested to identify the root cause explaining the proteolysis of fusion proteins.
Ultimately, lowering of the culture temperature from 37°C to 30°C was identified as the best lever considering quality, cost, and regulatory aspects (Chakrabarti et al., 2016).
Another extensive process development study was performed to obtain correctly folded erythropoietin (EPO) Fc-fusion protein (Ghorbani Aghdam et al., 2019).The final process included a twostep temperature shift, which led to an upregulation of genes encoding unfolded protein response specific transcriptional activators and ER-resident proteins associated with processing and folding in the secretory compartments (Torres et al., 2021).Finally, a study focusing on GLP1 Fc-fusion reported that media and feed formulations can impact specific posttranslational modifications such as serine phosphorylation or lysine hydroxylation (Hou et al., 2019).
In this study, a complex Fc-fusion protein containing one N-glycosylation site on the Fc part and up to 2 and 13 N-and O-glycosylation sites on linker or the antigen recognition domain was studied.An early analytical investigation of the protein produced in CHO cells indicated the presence of a double band within a nonreducing SDS-PAGE.To understand the nature of the low molecular weight proteoforms (LMW), multiple orthogonal methods were applied including size exclusion chromatography (SEC), middleup liquid chromatography-mass spectrometry (LC-MS) and released glycan analysis.Results confirmed that the LMW band was linked to glycan microheterogeneity as well as differences in both N-and O-glycan site occupancy.A root cause analysis was started to identify upstream process parameters driving the formation of these protein variants.Surprisingly, the cell culture medium and more specifically the sodium chloride (NaCl) content of the medium was identified as main lever that can be easily varied to reduce formation of the unwanted glyco-variants and ultimately increase the yield of the highly glycosylated fusion protein.

| Reagents
If not stated otherwise, all cell culture media, feeds and other raw materials were purchased from Merck, Darmstadt, Germany.The tested cell culture media and feed formulations were chemically defined.Tested feeds contained all the required nutrients to support cell growth and productivity except glucose that was added separately on demand.CHO clones used in this study were generated using a suspension GS-/-CHOZN ® host cell line using the GS expression system, single cell cloning and further classical cell line development approaches.The transgene was inserted using random integration.

| Cell culture and spent media analysis
Suspension CHOZN ® clones, producing a recombinant Fc-fusion protein were expanded using a 2/2/3 days split scheme.Cells were passaged at 3 × 10 5 cells/mL for 2 days and 2 × 10 5 cells/mL for 3 days.On the first day of the experiment, cells were inoculated at 3 × 10 5 cells/mL in spin tubes with vented caps (TPP, Techno Plastic Products AG) and cultivated at 37°C, 5% CO 2 , 80% humidity and 230 rpm in a shake incubator (Kühner).Samples used to measure cellular performance, recombinant protein titer, metabolite concentrations and osmolality were taken daily.Supernatants of Day 7 in batch and additionally days 10, 12, and 14 in fed-batch were harvested and purified for cQA analysis.While various feeds and feed schemes, ranging from a total feed volume of 20%-30% V/V were tested throughout screening, 5% V/V of Feed 1 on days 3, 5, 7, 10, 12, 14 was applied for the final fedbatch experiment.In fed-batch, glucose was replenished to 6 g/L if the concentration was below 4 g/L and to 12 g/L before weekends.
To adjust osmolality in batch cultivation, required amounts of sterile filtered NaCl, Na 2 SO 4 or Ectoin were added as highly concentrated stock solution on respective days.For the confirmatory batch and fed-batch experiments, NaCl was added on Day 3 to minimize the negative impact of the osmolality increase on Day 0 on growth and productivity.Cell performance (VCD, viability) was assessed using Vi-CELL™ XR 2.04 cell counter (Beckman Coulter) with automated trypan blue dye exclusion protocol.
Glucose, titer and other metabolites were monitored using CEDEX BIO HT (Roche).Osmolality was assessed by freezing point using Gonotec ® Osmomat Auto (Gonotech).

| Recombinant protein purification
Expressed fusion proteins were purified from the cell culture supernatant using Protein A PhyTips ® (PhyNexus Inc., PTR91-40-01) with RAININ E4 XLS and Pure Speed Protocol (Mettler Toledo, J1413990U).Equilibration was performed using 140 mM NaCl in 10 mM phosphate buffer and 3 mM KCl at pH 7.4.After capture (10 μg protein/μL Protein A resin), proteins were washed in a first step with 140 mM NaCl in 10 mM phosphate buffer and 3 mM KCl at pH 7.4 and in a second step with 140 mM NaCl.Finally, proteins were eluted using 30 mM citric acid at pH 3.0 and immediately neutralized with 375 mM Tris base at pH 9.0.
For this, 1 μL enzyme solution was applied per 100 μg protein (native or previously desialylated and/or O-deglycosylated) and samples were incubated at 50°C for 10 min.

| SDS-PAGE
Nonreducing samples were prepared according to the NuPAGE ® Bis-Tris protocol (Life Technologies; MAN0007894).Five hundred nanograms of each preparation was loaded to the 4%-12% Bis-Tris BOLT gel.Five microliter of Magic Mark XP (Thermo Fisher Scientific; LC5602) served as MW standard.Samples were separated using 1x MES buffer (Life Technologies; J62138.K2) for 50 min at 200 V constant.After separation, samples were fixed for 30 min in 50% ethanol and 7% acetic acid, stained for 1 h with SYPRO ® Ruby Protein Gel Stain (Thermo Fisher Scientific; S12000), washed for 30 min with 10% ethanol and 7% acetic acid and visualized at 610 nm using Fusion FX (Vilber Lourmat).
Calibration was carried out using the internal lockmass at 1221.9906.
Charge state deconvolution was performed with Data Analysis software (v5.3;Bruker) using the maximum entropy algorithm.

| Statistics
Statistical and graphic analyses were performed with Prism 9.1.2software (GraphPad).

| RESULTS
CHO cells engineered to express an Fc-fusion protein were cultivated in a classical fed-batch process over 17 days resulting in a titer of roughly 1.4 g/L (Figure 1a).Characterization of the affinity purified protein using SDS-PAGE under nonreducing conditions revealed two distinct bands (Figure 1b).One band, corresponding to the intact protein, was detected at an apparent molecular weight (MW) of ~135 kDa whereas a second, predominant band, was observed at an apparently lower MW of ~120 kDa.Size heterogeneity was also visible in native SEC with the highest abundance of LMW species on Day 7 (67%), relative to 14% high molecular weight proteoforms (HMW), and only 19% main peak (Figure 1c,d).Screening of nine different feeds across two different basal media in fed-batch mode (Figure 1e), as well as investigating two different clones in three different basal media in batch mode (Figure 1f) led to the identification of basal medium 1 as major root cause for observed LMW species.Since feeds demonstrated only a minor effect on LMW content, batch mode was selected as a simplified model to further investigate and more importantly mitigate size heterogeneity for this protein.
To better understand the observed protein size heterogeneity, the standard (1) was treated with multiple glycosidases to create LMW species in vitro (Figure 2a).The size of resulting bands was compared to the ~120 kDa band obtained for the Day 7 sample, produced using medium 1 in batch (Figure 2b).All enzymatic treatments resulted in one homogeneous band per preparation.A band at ~125 kDa was obtained after desialylation of N-and Oglycans using a mixture of sialidases hydrolyzing α2-3, α2-6, and α2-8 linked sialic acids (SialEXO, [2]), whereas a band at ~120 kDa was detected after removal of N-glycans using the amidase PNGase F (3).
The successive desialylation and removal of N-glycans (4), or desialylation and hydrolysis of core 1 O-glycans using an endoglycosidase (OglyZOR, [5]) resulted in bands of ~115 and ~120 kDa respectively.Finally, the consecutive application of all three enzymes (6) resulted in proteins lacking N-and O-glycans and migrating at ~100 kDa.Generated samples were analyzed using native SEC and the overlaid traces (Figure 2c) suggest that the Day 7 sample heterogeneity may be due to both differences in N-and O-glycan site occupancy and sialylation.
To get a deeper understanding of the glycosylation heterogeneity, a middle-up LC-MS approach was applied to the Day 7 sample produced in medium 1 and compared to the standard.
Results indicated major differences in O-glycan site occupancy between both samples (Figure 2d).Whereas 11%, 34%, 43%, and 10% of the proteoforms in the standard exhibited 8, 9, 10, and 11 occupied O-glycosylation sites per monomer, only 7%, 8%, 6%, and 2% of proteoforms presented such high site occupancy in the Day 7 sample.In contrast, 37% and 24% of the proteoforms produced in medium 1 had zero or only 1 occupied O-glycosylation site per monomer.A similar lack of site occupancy was observed for Nglycans (Fc region), as 35% of the proteoforms showed unoccupied N-glycosylation sites in the Day 7 sample, whereas the standard was fully occupied (Figure 2e).No difference in N-glycan site occupancy was observed for the two sites present on the antigen recognition domain (data not shown).Finally, N-glycan microheterogeneity was studied in both samples using a released glycan method.Results (Figure 2f As initial experiments (Figure 1e) suggested that the occurrence of LMW proteoforms was dependent on the cell culture medium, batch data obtained from multiple formulations were compiled and correlations between the relative abundance of the LMW species and key cellular performance parameters (peak viable cell density [VCD], viability, titer, productivity) or key media parameters (glucose, lactate, NH 4 + , phosphate concentrations, and osmolality) were studied (Supporting Information: Figure 1).
Among the tested combinations, a weak correlation (R ² = 0.6) was detected between the relative abundance of LMW proteoforms and osmolality of the cell culture medium (Supporting Information:  | 3169 titers were comparable for all conditions, regardless of the timepoint of NaCl application, except for Day 0 addition with repeatedly decreased titers (Figure 3d).
Affinity purified samples showed decreasing size heterogeneity as a function of the timepoint of NaCl addition.The best Fc-fusion quality was obtained for conditions where the hyperosmotic stimulus was applied as early as possible, for example, on Day 3 (Figure 3e) for which a simultaneous decrease in LMW proteoforms and an increase in N-glycan maturity was observed (Figure 3f).
Thus, NaCl addition had a time dependent impact on protein quality when applied at equimolar conditions on different days.
Acceptable titers and less heterogeneity in terms of glycan related LMW species were obtained by applying the hyperosmotic shock (+100 mOsm/kg) on Day 3 instead of Day 0. Addition on later days (4, 5, 6, 7) did not further impact titers compared to the addition on Day 3 but resulted in a lower protein quality with increasing LMW proteoforms generated over time.Conclusively, applying a hyperosmotic shock of +100 mOsm/kg through addition of 50 mM NaCl on Day 3 was selected as a simple way to drastically improve protein quality with no negative impact on fusion protein titer.
To apply these findings and validate the observed NaCl dose dependency while keeping titers constant, a NaCl dose response was applied on Day 3 in a new batch experiment (Figure 4).
The addition of 10-50 mM NaCl (Figure 4a) confirmed no negative impact on cellular performance (VCD and titer, Figure 4b) and resulted in a dose dependent decrease of LMW proteoforms, from 66% in the control to 24% in the 50 mM NaCl condition (Figure 4c).O-glycan site occupancy drastically increased with increasing concentrations of NaCl (Figure 4d), for example the percentage of proteoforms with 8, 9, and 10 occupied Oglycosylation sites increased about 13%, 24%, and 20% respectively in the 50 mM NaCl condition.Along with a much higher occupancy in total, no unoccupied O-glycosylated proteoforms were detected in the highest NaCl condition.N-glycosylation site occupancy of the monomeric Fc part also increased about 34% and was thus restored to the maximum possible degree (Figure 4e).
Similarly, terminal N-glycan species (Figure 4f) were impacted in a dose dependent manner and 50 mM NaCl addition on Day 3 resulted in elevated terminal sialylated species (+22%) on cost of terminal mannosylation (−16%) and terminal galactosylation (−12%), when compared to the control.
Ectoin was preferred over other osmolality controls such as mannitol (Romanova et al., 2022) due to its high solubility in water (~4 mol/L [Zaccai et al., 2016]), the absence of counter ions (Ectoin is a zwitterionic compound [Rieckmann et al., 2019]) as well as its property to be biocompatible and taken up by cells (Graf et al., 2008;Pastor et al., 2010;Rieckmann et al., 2019), similarly to the salts being tested.Osmotically active substances (NaCl, Na 2 SO 4, and Ectoin) were added again on Day 3 to maintain high titers (Figure 5).), whilst 100 mM Ectoin resulted in slightly higher osmolality than 50 mM NaCl (Figure 5a).Regardless of these osmolality differences, cell performance (VCD and titer) was comparable amongst all tested conditions (Figure 5b).Albeit all tested compounds were able to modulate some protein quality attributes, NaCl was the most efficient lever with a decrease of LMW proteoforms by 39%.In contrast, Na 2 SO 4 addition resulted in a decrease of LMW species by only 23%.Interestingly, although osmolality was highest using Ectoin, this osmolyte had no noticeable impact (−3%) on LMW proteoforms (Figure 5c), indicating that the increase in osmolality was not responsible for the decrease in size heterogeneity.O-and N-glycan site occupancy (Figure 5d,e) analysis confirmed that the highest site occupancy state was obtained after addition of 50 mM sodium in form of NaCl, followed by Na 2 SO 4 and 100 mM Ectoin.Released N-glycan profiles also revealed the same ranking with the highest impact for NaCl, resulting in an increase in mature (+19% sialylated) on cost of immature (−15% mannosylated and −10% galactosylated) N-glycan species (Figure 5f).
Conclusively, results demonstrate that the salt concentration (sodium and/or counter ions), rather than the osmolality increase, is involved in the mitigation of the Fc-fusion heterogeneity.Full dose response data for Na 2 SO 4 and Ectoin addition are presented in Supporting Information: Figures 3 and 4.
Finally, as the ultimate manufacturing process required application of fed-batch operation to maximize yield, the effect of Day 3 NaCl addition (50 mM) on Fc-fusion protein product quality was validated in a 14-day fed-batch process (Figure 6).
A slightly reduced growth was observed in the NaCl condition, but the final Fc-fusion titer was comparable (Figure 6a), indicating a higher cell specific production rate for the test condition.Monitoring of the protein size by nonreducing SDS-PAGE confirmed the presence of one major band with an apparent MW of ~135 kDa for all days in the NaCl condition (Figure 6b).The increase in fully glycosylated proteoforms was confirmed by SEC since LMW proteoforms were reduced by 42% on Day 7.Although LMW species decreased in a time dependent manner, the difference observed on Day 14 was still 20% (Figure 6c).In this final process, HMW species were detected for both conditions.HMW bands were sharper for the 50 mM NaCl condition compared to the respective control harvested on the same day, which is aligned with the higher site occupancy resulting in decreased size heterogeneity in this condition.Aggregates were, most likely, due to intermolecular disulfide mispairing as suggested by their complete disappearance in presence of DTT as reducing agent (Supporting Information: Figure 5).Focusing on Oglycosylation site occupancy in fed-batch mode (Figure 6d), NaCl was also able to drastically increase, for example the percentage of proteoforms carrying 8, 9, and 10 O-glycans about 11%, 29%, and 24% respectively on Day 7 and about 3%, 11%, and 8% respectively on Day 14. Site occupancy of the monomeric Fc part was also restored to the maximum possible degree for days 10, 12, and 14 (Figure 6e).Released N-glycan profiles showed a change towards more mature glycans, with an 18% increase in sialylated glycoforms on Day 14 at simultaneously decreased galactosylation (Figure 6f).
Altogether, results obtained in fed-batch mode confirm the success of the developed mitigation strategy aiming at decreasing size heterogeneity and highlight a significantly increased yield of the highly glycosylated Fc-fusion protein.

| DISCUSSION
Fc-fusion proteins are large and complex molecules that are difficult to produce at scale in the required quality and with an acceptable yield.During process development aiming at producing a 135 kDa Fcfusion protein, two bands were observed using nonreducing SDS- An investigation was started to understand the nature of the proteoforms and the root cause for their formation, enabling the establishment of a mitigation strategy.
In the literature, Fc-fusion size heterogeneity is known to be caused by proteinogenic fragments, resulting from the activity of host cell-related proteases such as cysteine endopeptidase (Mitsuo et al., 1990) or matrix metalloproteases (Sandberg et al., 2006).
Several mitigation strategies have been developed including changes in bioprocess conditions (culture temperature and time, partial pressure of CO 2 [Chakrabarti et al., 2016 composition (addition of ferric citrate, or protease inhibitor cocktail [Chakrabarti et al., 2016;Hou et al., 2019]).Other recent studies on complex N-and C-terminal Fc-fusion proteins (Datola et al., 2023) showed that LMW proteoforms may be caused by reshuffled disulfide bridges and can lead to protein misfolding responsible for aggregation and precipitation (Zhang et al., 2011), further affecting the biological activity and safety of the therapeutic molecule.
In contrast to previous studies, the Fc-fusion size heterogeneity observed in this study was linked to glycan microheterogeneity as well as differences in both N-and O-glycosylation site occupancy.O-glycans as well as the effect on N-glycan microheterogeneity.The effect of NaCl was demonstrated to be independent from the increase in osmolality using Ectoin as an osmolality control.
In the literature, while the impact of N-glycan composition and branching on bioavailability, stability and in vivo potency of IgG molecules (Raju, 2008;Wada et al., 2019) or more complex therapeutic proteins (Arnold et al., 2007;Li and d'Anjou 2009) is widely established, very few studies focus on glycosylation site occupancy.N-glycan site occupancy was described as an important factor impacting the stability of Fc-fusion proteins (Alsenaidy et al., 2014) but less is known about the impact of O-glycan site occupancy on the physicochemical and biological properties of proteins.O-glycosylation had been related to the efficient secretion and increased resistance to proteolytic cleavage of recombinant human chorionic gonadotropin protein (Deng et al., 2019) and one study focusing on Etanercept reported that certain O-glycans enhance TNF-alpha binding affinity and consequently potency (Biel et al., 2022).Similarly, only a few levers have been described to impact O-and N-glycan site occupancy.In cell culture experiments, manganese and iron salts have been linked to both N-and Oglycosylation heterogeneity in recombinant proteins produced in mammalian cells (Clincke et al., 2011;Kaufman et al., 1994).
Intracellularly, N-acetylgalactosaminyltransferase 1 (Galnt1), and uridine diphosphate-glycose pyrophosphorylase 2 (Ugp2) have been identified as the major limiting enzymes for O-glycan sugar chain synthesis (Deng et al., 2019).Galnt1 is a member of the glycosyltransferase family adding the first N-acetylgalactosamine (GalNAc) onto Ser/Thr residues to form mucin like structures whereas Ugp2 is a key enzyme involved in the synthesis of activated sugar precursors from glucose (Deng et al., 2019).
In our study, the low O-glycan site occupancy was mitigated by the supplementation of medium 1 with NaCl which led us to review the literature linking NaCl and nucleotide sugar content or glycosyltransferase activity.
As nucleotide sugars are essential to ensure sugar transfer onto the nascent protein chain, a reduced synthesis of UDP-GalNAc might lead to the observed macroheterogeneity.Activated sugars are formed from glucose and low glucose levels may contribute to the glycan precursor limitation as previously described in CHO cells producing classical IgGs (Villacrés et al., 2015).In our study performed in fed-batch, glucose was supplemented on demand and was thus not a limiting factor.Interestingly, glucose may be imported in mammalian cells through sodium/glucose cotransporters (SGLT, coded by SLC5A family members) and thus a lower sodium content in medium 1 compared to medium 2 and 3 may slow down uptake, leading to suboptimal concentrations of nucleotide sugars.This could be verified by following the intracellular amounts of activated sugars in cells cultivated in both media.Nevertheless, a former study excluded limiting levels of nucleotide sugars as the root cause for the lower galactosylation and sialylation levels observed on TNFR-IgG (Gawlitzek et al., 2000).
The potential link between NaCl and glycosyltransferase activity is more complex.Analysis of the relationship between chloride ions and glycosylation heterogeneity revealed the existence of a Golgi resident multitransmembrane protein named Golgi pH Regulator (GPHR, encoded by GPR89) (Maeda et al., 2008).This voltage dependent anion-channel is ubiquitously expressed and is necessary to ensure Golgi acidification via altered chloride permeability and counter-ion conductance (Maeda et al., 2008).C27 mutant cells, defective in this channel, exhibit delayed protein transport, but most importantly impaired glycosylation.Multiple proteins studied in C27 cells such as CD59, CD44, or the lysosome-associated membrane glycoproteins (Lamp-2) were characterized by a smaller size which was related to defects in both N-and O-glycosylation (Maeda et al., 2008).
If applied directly to our study, these results imply that the lower level of chloride in medium 1 compared to medium 2 and 3 might impair the Golgi pH leading to the lower N-and O-glycosylation site occupancy.
Impaired glycosylation upon elevated Golgi pH was reported by multiple groups as a result of the mislocalization of Golgi glycosyltransferases (Axelsson et al., 2001;Rivinoja et al., 2009) and an impaired enzyme heterodimerization process (Hassinen et al., 2011).
Most impacted enzymes were the α(2,3)sialyltransferase, which was mislocated to the endosomal compartment and GalnT and Nacetylglucosaminyltransferase (MGAT), partially relocalized from the Golgi stack to endosomal structures or even the cell membrane.
GalnT is the main enzyme responsible for the transfer of GalNAc to the respective Ser/Thr residue to give rise to O-glycans, thus its relocalization is expected to lead to reduced O-glycosylation site occupancy as observed in this work.
Sodium can also be linked to Golgi acidity due to its involvement in the function of ion pumps such as the Na + /K + ATPase and the NHE 7/8 Na + /H + antiporter (Kellokumpu, 2019).The sodium transport mediated by these pumps is promoting the activity of the V-ATPase, an ATP proton pump, known as the main acidifier of the Golgi and the secretory pathway (Jefferies et al., 2008;Sun-Wada et al., 2004).
Furthermore, the sodium content was linked to the protein expression of the chloride transporter GPHR and thus to Golgi acidity in a proteomic study performed on CHO cells.In their work, an increase in sodium ions following the rise of pCO 2 commonly observed in bioreactors was shown to increase the GPHR protein expression (Nguyen Dang et al., 2019), further supporting the synergetic effect of both sodium and chloride addition in our study.
The observed microheterogeneity of N-glycans might also be linked to the NaCl content of the medium.Taking mannosylation as an example, a higher mannosylation baseline on Day 7 of the final fed-batch process was obtained in medium 1 (17.5%)compared to medium 1 supplemented with NaCl (5.8%).Over time, the percentage of mannosylation decreased in medium 1 whereas it increased with medium 1 supplemented with NaCl.When focusing on the dose response studies with NaCl or Na 2 SO 4 supplementation in batch, the supplementation led to an increase in osmolality and a decrease in the D7 mannosylation baseline.In the literature, mannosylation of classical monoclonal antibodies is known to increase with the duration of the fed-batch process (Huang et al., 2015;Pacis et al., 2011) and following an increase in osmolality (e.g., through addition of NaCl [Pacis et al., 2011] or selected sugars [Hossler et al., 2014]).This was also described by studies performed in our group (Ehret et al., 2019;Zimmermann et al., 2019).Thus, the results obtained in this study highlight the direct link between the amount of sodium and chloride ions in medium 1 and N-glycan mannosylation.
As one of the main enzymes affected by changes in Golgi pH is MGAT, it is tempting to suggest that the relocalization or lower activity of this enzyme leads to the increased mannosylation content observed in medium 1.This increase in mannosylation might be directly linked to the decrease in terminal GlcNac proteoforms, as MGAT commonly transfers GlcNac on mannosylated structure.
Similarly, the hypothesized impact of the Golgi pH on sialyltransferase relocalization may explain the low sialylation observed on the studied Fc-fusion protein and the corresponding increase in galactosylation (in medium 1).Alternatively, a decrease in sialylation has been reported previously as a result of an increase in osmolality (Lee et al., 2017) or through the release of sialidase in the cell culture medium (Gramer et al., 1995).
Concluding, the salt balance and in particular the amount of sodium and chloride ions in the cell culture medium used to grow CHO cells appears essential to ensure proper N-and O-glycan site occupancy and control microheterogeneity of highly complex Fcfusion proteins.This study was funded by Merck Life Science KGaA.

ACKNOWLEDGMENTS
) showed drastically reduced sialylation in the Day 7 sample (21% of terminal sialylated glycoforms vs. 49% in the standard).This shift towards less-mature glycoforms in the Day 7 sample coincided with the detection of more terminal galactosylated and mannosylated species and less terminal N-acetylglucosamine (GlcNAc) glycoforms.Altogether, the characterization of the Day 7 sample indicates that the LMW proteoforms detected by SDS-PAGE correspond to partially N-and O-glycosylated proteins.While O-glycosylation site occupancy is the most affected posttranslational modification, N-glycan site occupancy as well as sialylation are also significantly reduced.

F
Figure2).To confirm the involvement of medium osmolality in the formation of LMW species and glycan heterogeneity, increasing concentrations of NaCl were applied to medium 1 on Day 0.
Addition of 50 mM sodium in form of Na 2 SO 4 resulted in lower osmolality than the addition of equimolar concentrations of sodium in form of NaCl, explicable by the osmotic pressure introduced by counter ions (1 Cl − vs. 0.5 SO 4 2− Impact of sodium chloride addition on Day 3 on cell culture performance and Fc-fusion quality attributes.Osmolalities (a) and cellular performance (b) of a batch process with NaCl dose response applied on Day 3 as well as relative abundancies of SEC species (c).O-glycosylation (d) and N-glycosylation (e) site occupancy of protein A-purified samples using middle-up LC-MS.Analysis of released N-glycans (f) using Rapifluor™ labeling and LC-MS separation and quantification.Error bars represent standard deviations (n = 3 for a, b, c, f, and n = 2 for d and e).SEC, size exclusion chromatography.PAGE indicating the formation of low molecular weight proteoforms.
; Ghorbani Aghdam et al., 2019; Menthe et al., 2022]) or change in cell culture media F I G U R E 5 Impact of sodium chloride, disodium sulfate and Ectoin addition on Day 3 on cell culture performance and Fc-fusion quality attributes.Osmolalities (a) and cellular performance (b) of a batch process with NaCl, Na 2 SO 4, and Ectoin applied on Day 3 as well as relative abundancies of SEC LMW species (c).O-glycosylation (d) and N-glycosylation (e) site occupancy of protein A-purified samples using middle-up LC-MS.Analysis of released N-glycans (f) using Rapifluor™ labeling and LC-MS separation and quantification.Error bars represent standard deviations (n = 3 for a, b, c, f, and n = 2 for d and e).LMW, low molecular weight; SEC, size exclusion chromatography.

F
I G U R E 6 Impact of sodium chloride addition on cell culture performance and Fc-fusion quality attributes in final fed-batch manufacturing process.(a) VCD (rectangles) and Fc-fusion titers (triangles) in a fed-batch experiment with medium 1 (empty) and with addition of 50 mM NaCl on Day 3 (solid).Feed 1 was used in this experiment.(b) Nonreducing SDS-PAGE of purified fed-batch samples (MW reference: Magic Mark XP). (c) relative abundancies of LMW proteoforms.O-glycosylation (d) and N-glycosylation (e) site occupancy of protein A-purified samples using middle-up LC-MS.Analysis of released N-glycans (f) using Rapifluor™ labeling and LC-MS separation and quantification.Error bars represent standard deviations (n = 3 for a, c, and f).LMW, low molecular weight.BOHL ET AL. | 3173 Regarding O-glycosylation, a drastic reduction of Ser/Thr site occupancy was detected with more than 60% of the proteoforms presenting either zero or one occupied O-glycosylated sites, in contrast to the standard, for which 77% of the proteoforms contained eight or nine occupied O-glycosylated sites.Regarding N-glycosylation, up to 35% of the Fc monomer presented nonglycosylated Asn residues in contrast to the standard which was fully occupied.The microheterogeneity on N-glycans was also peculiar with a decrease in terminal GlcNAc and sialylation and an increase in terminal galactosylation and mannosylation compared to the standard.As the LMW species were preferentially formed in medium 1, root cause analysis focused on this formulation.Surprisingly, NaCl was demonstrated to be pivotal regarding Fc-fusion protein quality.Addition of NaCl on Day 3 in both batch and fed-batch processes mitigated the heterogeneity caused by the partial lack of N-and

Susanne
Bohl designed and performed the cell culture experiments, characterization experiments using SDS-PAGE and SEC, performed data analysis and wrote the manuscript.Maxime Le Mignon performed the protein characterization experiments using mass spectrometry.Thomas Kilian performed the released N-glycans experiments.Aline Zimmer supervised the study and reviewed the data and the manuscript.All authors have approved the final article.