Impact of SARS-CoV-2 RBD Mutations on the Production of a Recombinant RBD Fusion Protein in Mammalian Cells

SARS-CoV-2 receptor-binding domain (RBD) is a major target for the development of diagnostics, vaccines and therapeutics directed against COVID-19. Important efforts have been dedicated to the rapid and efficient production of recombinant RBD proteins for clinical and diagnostic applications. One of the main challenges is the ongoing emergence of SARS-CoV-2 variants that carry mutations within the RBD, resulting in the constant need to design and optimise the production of new recombinant protein variants. We describe here the impact of naturally occurring RBD mutations on the secretion of a recombinant Fc-tagged RBD protein expressed in HEK 293 cells. We show that mutation E484K of the B.1.351 variant interferes with the proper disulphide bond formation and folding of the recombinant protein, resulting in its retention into the endoplasmic reticulum (ER) and reduced protein secretion. Accumulation of the recombinant B.1.351 RBD-Fc fusion protein in the ER correlated with the upregulation of endogenous ER chaperones, suggestive of the unfolded protein response (UPR). Overexpression of the chaperone and protein disulphide isomerase PDIA2 further impaired protein secretion by altering disulphide bond formation and increasing ER retention. This work contributes to a better understanding of the challenges faced in producing mutant RBD proteins and can assist in the design of optimisation protocols.


Introduction
Coronavirus disease , an infectious disease caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has been declared a global pandemic by the World Health Organization (WHO) in March 2020 [1]. About two years later, nearly 500 million confirmed cases and over 6 million deaths have been reported worldwide [1,2]. Tremendous efforts have been devoted to the development of diagnostic tools, vaccines and therapeutics to address this pandemic. Researchers have identified the receptor-binding domain (RBD) of the SARS-CoV-2 spike protein as a major target of neutralising antibodies, and thus as a protein domain of choice for the design and development of diagnostics, vaccine and therapeutic antibodies [3][4][5][6][7][8][9][10][11][12]. Notably, several studies reported the development and characterization of recombinant RBD proteins (amino acids 331-524 of the spike protein) fused to a C-terminal tag. These studies showed that recombinant RBD protein (amino acids 331-524) efficiently binds the angiotensin-converting enzyme 2 (ACE2) receptor, is efficiently recognised by neutralising antibodies from COVID-19 convalescent individuals, and elicits a strong neutralising antibody response when administered as an immunogen in BALB/c mice [13][14][15][16]. Of particular interest, Sun et al. expressed SARS-CoV-2 RBD (amino acids 331-524) either fused to a histidine (His) tag as a monomeric protein or fused to a human IgG1 Fc fragment to produce a Y-shaped dimeric protein. They showed that the dimeric RBD-Fc fusion protein bound the ACE2 receptor with an approximate 10-fold enhanced avidity compared to the monomeric RBD-His protein, suggesting an improved conformation of the dimeric RBD-Fc protein [16]. . The respective RBD dom were expressed in fusion with an N-terminal optimised signal peptide to allow protein secr and a C-terminal mouse IgG2a Fc fragment to facilitate purification on protein A affinity resin to produce Y-shaped protein dimers. RBDmfc proteins are produced as dimers linked by disulp bonds between their Fc fragments. One RBDmfc monomer is 437 amino acids long and has a dicted molecular weight of 49.2 kDa (upon cleavage of the signal peptide). (b) Concentratio RBDmfc proteins secreted in the culture medium of Expi293F cells, measured by Octet RED96 expressed in μg/mL culture. Values are mean (standard deviation) of two independent experim RBD mutants carrying the E484K mutation were expressed at lower levels in the mediu Expi293F cells. The level of secreted B.1.351 RBDmfc was 27% that of WT RBDmfc. By con mutation N501Y had no impact on the secreted protein production yield.

Plasmid Production and Purification
Plasmids were transformed into One Shot TOP10 Chemically Competent E. (ThermoFisher Scientific, Carlsbad, CA, USA; C404010) with the rapid chemical tran mation procedure proposed by the manufacturer. Transformed E. coli were spread o . The respective RBD domains were expressed in fusion with an N-terminal optimised signal peptide to allow protein secretion and a Cterminal mouse IgG2a Fc fragment to facilitate purification on protein A affinity resin and to produce Y-shaped protein dimers. RBDmfc proteins are produced as dimers linked by disulphide bonds between their Fc fragments. One RBDmfc monomer is 437 amino acids long and has a predicted molecular weight of 49.2 kDa (upon cleavage of the signal peptide). (b) Concentration of RBDmfc proteins secreted in the culture medium of Expi293F cells, measured by Octet RED96 and expressed in µg/mL culture. Values are mean (standard deviation) of two independent experiments. RBD mutants carrying the E484K mutation were expressed at lower levels in the medium of Expi293F cells. The level of secreted B.1.351 RBDmfc was 27% that of WT RBDmfc. By contrast, mutation N501Y had no impact on the secreted protein production yield.

Secreted and Intracellular Protein Fraction Preparation
Transfected cells were enumerated and harvested four days after transfection and cell suspensions were centrifuged for 15 min at 5000× g. Protein-containing medium supernatants (secreted RBDmfc protein fraction) were filtered on 0.22 µm PVDF membrane (Millipore/Merck, Darmstadt, Germany; S2GVU05RE) and stored at 4 • C. Cell pellets were stored at −20 • C until cell lysate preparation.

RBDmfc Purification and Quantification
Cell media and cell extract supernatants containing RBDmfc proteins were purified using a MabSelect protein A resin (GE Healthcare, Uppsala, Sweden; 17819902) on an AKTA FPLC Basic 10 (GE Healthcare, Uppsala, Sweden). Resin was equilibrated in PBS pH 7.2, supernatants or cellular extracts were loaded onto the resin and washed with 20 column volumes (CV) PBS pH 7.2. RBDmfc proteins bound to the resin were eluted with 10 CV of a proprietary acidic buffer. Eluted fractions were concentrated on Vivaspin For purified RBDmfc protein analysis by SDS-PAGE, protein samples (0.1 µg/mL in PBS) were mixed 1:4 in NuPAGE LDS Sample Buffer (4×), and 10 µL denatured and heated samples were loaded and separated on SDS-PAGE, as above.
Coomassie staining of proteins separated by SDS-PAGE was performed with the Gel-Code Blue Safe Protein Stain (ThermoFisher Scientific, 24596) following the manufacturer's standard protocol. Silver staining of proteins separated by SDS-PAGE was performed with the SilverQuest Silver Staining Kit (ThermoFisher Scientific, LC6070) using the manufacturer's Fast Staining protocol.

Western Blotting
Protein transfer for western blotting was done with the Trans-Blot Turbo Mini 0.2 µm Nitrocellulose Transfer Packs (Bio-Rad Laboratories, Hercules, CA, USA; 1704158) using a Trans-Blot Turbo Transfer System (Bio-Rad Laboratories) with its pre-programmed 1.5 MM GEL protocol.
Alkaline phosphatase activity detection was performed by incubating membranes with 7 mL 1-Step NBT/BCIP Substrate Solution (ThermoFisher Scientific, 34042) until desired signal intensity and blocked by three washes in 100 mL demineralised water.

Image Acquisition
Image acquisition was performed using the ChemiDoc XRS+ Gel Imaging System (Bio-Rad Laboratories, Hercules, CA, USA) and the Image Lab Software (Bio-Rad Laboratories). The Image Lab software was set to the colorimetric gel acquisition mode for Coomassie-stained gels, the silver stain mode for silver-stained gels, and the colorimetric blot acquisition mode for western blots. The signal intensity of western blot bands was quantified, and the relative signal was determined using the Image Lab Software (Bio-Rad Laboratories). Uncropped SDS-PAGE images are shown in Appendix A.

Protein Analysis by Enzyme-Linked Immunosorbent Assay (ELISA)
Purified secreted WT and B.1.351 RBDmfc proteins (100 µL at 5 µg/mL) were coated onto wells of a 96-well microplate (Greiner Bio-One, Frickenhausen, Germany; 655001) overnight at room temperature. Microplate wells were washed three times with 300 µL PBS, 0.05% Tween 20 and blocked for 3 h at room temperature with 200 µL PBS, 1% BSA (Proliant Biologicals, Ankeny, IA, USA; 68100). Wells were washed three times as before and incubated for 2 h at room temperature with 100 µL of the respective human sera (1:200 in PBS) or controls, as follows: the WT SARS-CoV-2 serum (same as that used in western blot) was added to duplicate wells coated with purified WT RBDmfc, the B.1.351 SARS-CoV-2 serum (same as that used in western blot) was added to duplicate wells coated with purified B.1.351 RBDmfc, a pre-pandemic human serum (EFS, Saint-Denis, France) used as a negative control was added to duplicate WT-and B.1.351 RBDmfc-coated wells, and PBS (blank) was added to four WT-and B.1.351 RBDmfc-coated wells. Microplate wells were washed three times as above and RBD-specific SARS-CoV-2 IgG were detected by incubation for 2 h at room temperature with 100 µL of a proprietary alkaline phosphatase mouse anti-human IgG (0.16 µg/mL in PBS, 0.05% Tween 20). Microplate wells were washed five times with 300 µL PBS, 0.05% Tween 20 and specific immunocomplexes were revealed by incubating for 20 min at 37 • C in 100 µL 4-Nitrophenyl phosphate disodium salt hexahydrate (PNPP; Sigma-Aldrich/Merck, Darmstadt, Germany; P5994) at 1 mg/mL in a proprietary diethanolamine (DEA)-HCl buffer. Coloration was stopped by adding 100 µL 1N NaOH (VWR, Radnor, PA, USA; 35256-1L). Absorbance was measured at 405-490 nm on a BioTek ELX808 microplate reader (Agilent, Santa Clara, CA, USA) and data were analysed using the BioTek Gen5 Microplate Reader and Imager Software 3.10 (Agilent).

Hydrophobicity Determination by HIC-MALS
Hydrophobic interaction chromatography hyphenated to multi-angle light scattering detection (HIC-MALS) measurements were performed using an UltiMate 3000 HPLC and UHPLC System (ThermoFisher Scientific) equipped with a quaternary RS pump, a WPS-3000FC thermostated sampler and fraction collector, and a DAD-3000RS Diode Array Detector set at 280 nm. Multi-angle light scattering data were acquired from a Dawn 8+ MALS detector and interpreted with the Astra 7 software (version 7.3.2.19; Wyatt Technology, Santa Barbara, CA, USA). RBDmfc sample molar mass calculations were performed using a UV extinction coefficient ε (0.1%, 280 nm) of 1.357 for WT RBDmfc protein and 1.386 for B.1.351 RBDmfc protein, as described above (Section 2.1.6).
Analytical HIC measurements of the RBDmfc samples were performed using an AdvanceBio HIC column (4.6 mm inner diameter × 100 mm length, 3.5 µm particle size; Agilent). The autosampler and fraction collector was set at 5 • C, 10 µg protein sample was injected, and the mobile-phase flow rate was set at 0.5 mL/min. A 20-min linear gradient was applied from 100% 2 M ammonium sulphate containing 0.05 M phosphate buffer pH 7.0 to 100% 0.05 M phosphate buffer pH 7.0, followed by a 5 min washing step with 100% 0.05 M phosphate buffer pH 7.0 to assure that all bound impurities were removed. The column was then re-equilibrated at starting mobile phase conditions using a 5 min gradient and 10 min step. Fraction collection started at 23 min and ended at 32 min, using a collection period of 30 s. Collected fractions were stored in a Nunc 2.0 mL U-bottom Polypropylene DeepWell 96-well microplate (ThermoFisher Scientific, 278752) within the thermostated collector until the end of HIC separation, then immediately loaded on SDS-PAGE for silver staining analysis, as described above.

Size-Exclusion Chromatography
Size-exclusion chromatography (SEC) is a relative method allowing estimation of the apparent MW of proteins in solution, based on the retention time. Analytical SEC was performed on an H-Class Acquity Ultra Performance Liquid Chromatography system (UPLC, Waters, Milford, CT, USA). Briefly, the SEC column (Waters, BEH200) was preequilibrated in PBS 3× pH 6.0 for 30 min at 0.4 mL/min, as described by the manufacturer. Purified WT and B.1.351 RBDmfc protein samples were filtered (Ultrafree MC-GV Durapore PVDF 0.22 µm), and a sample volume of 10 µL was injected into the column at 30 • C at a flow rate of 0.4 mL/min. UV absorbance of the eluate was monitored at a wavelength of 280 nm.

Thermal Stability Determination by Differential Scanning Calorimetry
Differential scanning calorimetry (DSC) was performed using a CSC-6300 NanoDSC III calorimeter, and data were acquired and interpreted using the DSCRun 4.4.9 and NanoAnalyze 3.7.5 software, respectively (TA Instruments, New Castle, DE, USA). Purified, dialysed RBDmfc protein samples were adjusted to 0.605 mg/mL with PBS. The instrument sample cell was filled with the protein sample (exactly 299 µL), and the reference cell was filled with PBS (the same PBS batch as used for sample dialysis and concentration adjustment). A PBS blank was also performed for further baseline correction.
Data were collected using a heating rate of 1 • C/min from 20 • C to 100 • C under 3 atmospheric pressure and after a pre-equilibration step of 600 s at 20 • C. The sample's raw thermogram (µJ/s vs. temperature) was baseline subtracted versus the PBS thermogram, then converted to molar heat capacity (defined as the amount of energy in the form of heat needed to raise the temperature of one mole of purified RBDmfc protein by one Kelvin) using their respective molecular weight to obtain kJ/mol.K vs. temperature thermograms. The integration of transition peaks was then achieved using a sigmoidal baseline method before calculating thermodynamic parameters.

Impact of RBD Mutations on the Production Yield of Recombinant RBDmfc Fusion Proteins in 293F Cells
The RBD domain (amino acids 331-524) of SARS-CoV-2 isolate Wuhan-Hu-1 (hereafter referred to as wild type or WT) and variants B.  (Figure 1b). These observations indicate that the single RBD mutation N501Y had no impact on protein yield while mutation E484K had a strong impact, further enhanced by the presence of the K417N mutation.
To identify the cause of this difference in production level, we focused on B.1.351 RBDmfc, which showed the lowest secretion level, and further analysed its expression characteristics and biochemical properties, in comparison to WT RBDmfc. Because differences in disulphide bond arrangements might impact protein function, the functionality of the purified secreted WT and B.1.351 RBDmfc recombinant proteins was verified. RBDmfc proteins were developed and produced for use in immunoassays Accordingly, their activity was evaluated as to their ability to be recognised by SARS- Because differences in disulphide bond arrangements might impact protein function, the functionality of the purified secreted WT and B.1.351 RBDmfc recombinant proteins was verified. RBDmfc proteins were developed and produced for use in immunoassays. Accordingly, their activity was evaluated as to their ability to be recognised by SARS-CoV-2-specific IgG under both denaturing (western blot) and non-denaturing (enzyme-linked immunosorbent assay [ELISA]) conditions ( Figure S1). Western blot analyses demonstrated that sera from patients infected with SARS-CoV-2 (Wuhan isolate [WT] or B.1.351 variant) recognised the three protein species (of WT and B.1.351 RBDmfc, respectively), with a stronger detection of the upper MW band ( Figure S1a). Similarly, ELISA performed on coated RBDmfc proteins (WT or B.1.351) using COVID-19 patient sera, demonstrated strong antigen-antibody (IgG) recognition signals (about 100-fold over negative serum background) for both RBDmfc proteins ( Figure S1b). These experiments thus demonstrate that both WT and B.1.351 RBDmfc recombinant proteins are reactive toward SARS-CoV-2specific IgG from patient sera.

Impact of PDIA2 Expression on the Production and Secretion of WT and B.1.351 RBDmfc Fusion Proteins in 293F Cells
The concentration of intracellular and secreted WT and B.1.351 RBDmfc proteins was then evaluated upon expression in Expi293F cells. The intracellular concentration of WT RB-Dmfc was low while its concentration in the culture medium was high (Figure 3, 100% WT RBDmfc condition), demonstrating that WT RBDmfc is efficiently secreted. By contrast, the level of secreted B.1.351 RBDmfc was low, as described above (Figure 1b), while its intracellular concentration was higher compared to that of WT RBDmfc-expressing cells (Figure 3, 100% B.1.351 RBDmfc condition). This suggests that a pool of B.1.351 RBDmfc proteins is retained inside the cells and that the lower yield of protein secretion is likely due to impaired trafficking rather than impaired translation. These observations also confirm that WT and B.1.351 RBDmfc secreted proteins are processed differently via disulphide bond formation and isomerisation during maturation in the ER. Disulphide bond formation and isomerisation being influenced by protein folding (and vice-versa), these results in turn suggest that WT and B.1.351 RBDmfc proteins might differ in their folding, resulting in impaired processing of B.1.351 RBDmfc along the secretory pathway, possibly explaining its reduced yield compared to WT RBDmfc upon expression in Expi293F cells.
Previous studies have shown that overexpression of chaperones, including protein disulphide isomerase PDIA2, can reduce the amount of aggregated misfolded proteins and promote protein refolding and secretion [46][47][48]. To address the possible implication of protein misfolding and disulphide bond rearrangements in the impaired secretion of B.1.351 RBDmfc, we tested the effect of overexpression of the protein disulphide isomerase and chaperone PDIA2 on B.1.351 RBDmfc protein expression and secretion in Expi293F cells.
Overexpression of PDIA2 led to a dose-dependent increase of the intracellular concentration of both WT and B.1.351 RBDmfc proteins (Figure 3). This increase in intracellular RBDmfc was accompanied by a strong decrease in the concentration of secreted WT RB-Dmfc, while the level of secreted B.1.351 RBDmfc remained low (Figure 3). This observation indicates that overexpression of PDIA2 interfered with the processing and secretion of WT RBDmfc, possibly by altering disulphide bond formation and isomerisation, while it did not improve secretion of B.1.351 RBDmfc. In addition, the strong accumulation of intracellular B.1.351 RBDmfc upon overexpression of PDIA2 suggests a possible protection and stabilisation of the RBD protein by PDIA2, but no promotion of folding and secretion, as anticipated. These results also confirm that the low yield of secreted B.1.351 RBDmfc was not due to impaired translation but occurred post-translationally.
On the whole, these results demonstrate that WT and mutant RBDmfc proteins are differentially processed along the secretory pathway and suggest possible differences in protein folding and disulphide bond arrangements during maturation in the ER.
vation indicates that overexpression of PDIA2 interfered with the processing and secretion of WT RBDmfc, possibly by altering disulphide bond formation and isomerisation, while it did not improve secretion of B.1.351 RBDmfc. In addition, the strong accumulation of intracellular B.1.351 RBDmfc upon overexpression of PDIA2 suggests a possible protection and stabilisation of the RBD protein by PDIA2, but no promotion of folding and secretion, as anticipated. These results also confirm that the low yield of secreted B.1.351 RBDmfc was not due to impaired translation but occurred post-translationally. In co-transfection experiments, the respective proportion of plasmid expressing RBDmfc, PDIA2 or empty plasmid (to complete to 1 μg plasmid DNA) is indicated below each bar. Accordingly, 100%, 95%, 70% or 5% plasmid DNA indicates that 1, 0.95, 0.7 or 0.05 μg plasmid DNA (per ml cell suspension) was used for transfection, respectively. Intracellular and secreted protein fractions were prepared four days after transfection. Protein concentration was measured by Octet RED96 and expressed in μg/mL culture. Values are mean (standard deviation) of duplicate Octet RED96 measurements of one representative experiment (out of two independent experiments).
On the whole, these results demonstrate that WT and mutant RBDmfc proteins are differentially processed along the secretory pathway and suggest possible differences in protein folding and disulphide bond arrangements during maturation in the ER.
Western   An anti-Fc blot on the intracellular protein fractions (Figure 4b), revealed the presence of smeary signals around 50 kDa and 100 kDa, likely corresponding to proteins lacking or with improper disulphide bonds. The apparent MW of 50 kDa of the intracellular monomer species also suggests that the protein is not or partially glycosylated (as opposed to the monomer of 59 kDa shown in Figure 2). These signals were weaker for WT RBDmfc (Figure 4a Aggregation of misfolded proteins in the ER is responsible for the ER-stress response UPR [23,42,49]. One of the outcomes of UPR is the induction of chaperones, meant to promote protein folding. Western blot analysis revealed an increase in the protein level of endogenous PDIA2 (Figure 4b Figure 3 were analysed by Western blot using antibodies directed against the recombinant RBDmfc proteins (anti-RBD, anti-Fc), against chaperones (anti-PDIA2, anti-HSPA5, anti-HSP90B1), and against βactin as loading control. M, protein ladder (apparent MW in kDa indicated on the left); *, endogenous PDIA2 protein (anti-PDIA2 blot); arrow, RBDmfc proteins (anti-RBD and anti-Fc blots in panel (b)) likely lacking disulphide bonds and mature glycosylations, and thus running in SDS-PAGE with an apparent MW of 50 kDa (compare to 59 kDa reduced, mature proteins shown in Figure 2, lanes 3 and 4). Quantification of signal intensity of chaperones is shown in Figure S2. Data are from one representative experiment (out of three independent experiments). Uncropped blot images are shown in Appendix Figures A3 and A4.

Glycosylation Profile of Intracellular and Secreted WT and B.1.351 RBDmfc Proteins
To determine whether intracellular B.1.351 RBDmfc proteins indeed accumulate into the ER and do not transit to the Golgi apparatus, we investigated the sensitivity of intracellular and secreted RBDmfc proteins to the deglycosylases PNGase and EndoH. PNGase deglycosylates all N-glycosylations, thus glycosylated proteins are expected to be sensitive to PNGase as they proceed through both the ER and Golgi apparatus. On the other hand, glycosylated proteins are expected to lose their sensitivity to EndoH during maturation in the Golgi apparatus. Purified secreted and intracellular WT and B.1.351 RBDmfc proteins were treated with PNGase or EndoH and analysed by SDS-PAGE and Coomassie staining ( Figure 5). RBDmfc proteins, whether WT or mutant, secreted or intracellular, were sensitive to PNGase, resulting in a downward shift (Figure 5a, lanes 2, 4, 6, 8 vs. 1, 3, 5, 7, respectively), indicating that all proteins were glycosylated. As expected, secreted  Figure 2, lanes 3 and 4). Quantification of signal intensity of chaperones is shown in Figure S2. Data are from one representative experiment (out of three independent experiments). Uncropped blot images are shown in Appendix A Figures A3 and A4. An anti-Fc blot on the intracellular protein fractions (Figure 4b), revealed the presence of smeary signals around 50 kDa and 100 kDa, likely corresponding to proteins lacking or with improper disulphide bonds. The apparent MW of 50 kDa of the intracellular monomer species also suggests that the protein is not or partially glycosylated (as opposed to the monomer of 59 kDa shown in Figure 2). These signals were weaker for WT RBDmfc (Figure 4a Altogether, these results demonstrate that WT and B.1.351 RBDmfc present distinct disulphide bond arrangements, likely resulting in misfolding and enhanced intracellular retention of a portion of B.1.351 RBDmfc proteins. The observation that the protein disulphide isomerase PDIA2 exacerbated the accumulation of proteins with improper disulphide bonds further supports the idea that aberrant disulphide bond arrangements are responsible for the retention of B.1.351 RBDmfc in the ER (in the absence of PDIA2 overexpression).
Aggregation of misfolded proteins in the ER is responsible for the ER-stress response UPR [23,42,49]. One of the outcomes of UPR is the induction of chaperones, meant to promote protein folding. Western blot analysis revealed an increase in the protein level of endogenous PDIA2 (Figure 4b, *), HSPA5, and to a lesser extent HSP90B1, in cells expressing B.1.351 RBDmfc compared to WT RBDmfc (Figure 4b, lanes 4-6 vs. lane 1). HSPA5 (and to a lesser extent HSP90B1) also appeared upregulated in WT RBDmfc cells overexpressing PDIA2 (Figure 4b, lanes 2 and 3 vs. lane 1). These results suggest that the UPR was induced in cells expressing B.1.351 RBDmfc, likely as a response to the ER accumulation of misfolded mutant proteins.

Glycosylation Profile of Intracellular and Secreted WT and B.1.351 RBDmfc Proteins
To determine whether intracellular B.1.351 RBDmfc proteins indeed accumulate into the ER and do not transit to the Golgi apparatus, we investigated the sensitivity of intracellular and secreted RBDmfc proteins to the deglycosylases PNGase and EndoH. PNGase deglycosylates all N-glycosylations, thus glycosylated proteins are expected to be sensitive to PNGase as they proceed through both the ER and Golgi apparatus. On the other hand, glycosylated proteins are expected to lose their sensitivity to EndoH during maturation in the Golgi apparatus. Purified secreted and intracellular WT and B.1.351 RBDmfc proteins were treated with PNGase or EndoH and analysed by SDS-PAGE and Coomassie staining ( Figure 5). RBDmfc proteins, whether WT or mutant, secreted or intracellular, were sensitive to PNGase, resulting in a downward shift (Figure 5a, lanes 2, 4, 6 (Figure 5b, lanes 7 and 8, *), probably due to its lower abundance compared to WT RBDmfc (see Figure 2). These results confirmed that the pool of RBDmfc proteins detected intracellularly is retained in the ER, and is likely prevented from proceeding to secretion because of improper disulphide bonds and misfolding.
Altogether, these data demonstrate that, compared to WT RBDmfc, only a small fraction of B.1.351 RBDmfc proteins, presenting the correct disulphide bond arrangements and proper folding, is secreted from transfected Expi293F cells, while a pool of misfolded proteins is retained in the ER, resulting in the UPR ER-stress response. intracellular WT RBDmfc proteins in transit through the Golgi along the secretory pathway. Such a band was not detected in B.1.351 RBDmfc-expressing cells (Figure 5b, lanes 7  and 8, *), probably due to its lower abundance compared to WT RBDmfc (see Figure 2). These results confirmed that the pool of RBDmfc proteins detected intracellularly is retained in the ER, and is likely prevented from proceeding to secretion because of improper disulphide bonds and misfolding. Altogether, these data demonstrate that, compared to WT RBDmfc, only a small fraction of B.1.351 RBDmfc proteins, presenting the correct disulphide bond arrangements and proper folding, is secreted from transfected Expi293F cells, while a pool of misfolded proteins is retained in the ER, resulting in the UPR ER-stress response.

Hydrophobicity and Thermostability of WT and B.1.351 RBDmfc Fusion Proteins
To further characterise WT and B.1.351 RBDmfc proteins, purified secreted proteins were fractionated according to their hydrophobicity status using hydrophobic interaction chromatography hyphenated to multi-angle light scattering detection (HIC-MALS). WT and B.1.351 RBDmfc protein preparations exhibited different hydrophobicity profiles (Figure 6a), as expected from their differences in amino acid composition. As opposed to WT RBDmfc, which eluted as one main peak (at 27-29 min), B.1.351 RBDmfc revealed two protein populations (eluting at 26-28 min and 28-30 min, respectively), thus presenting differences in hydrophobicity (Figure 6a).
were fractionated according to their hydrophobicity status using hydrophobic interaction chromatography hyphenated to multi-angle light scattering detection (HIC-MALS). WT and B.1.351 RBDmfc protein preparations exhibited different hydrophobicity profiles (Figure 6a), as expected from their differences in amino acid composition. As opposed to WT RBDmfc, which eluted as one main peak (at 27-29 min), B.1.351 RBDmfc revealed two protein populations (eluting at 26-28 min and 28-30 min, respectively), thus presenting differences in hydrophobicity (Figure 6a).  MALS analysis confirmed similar absolute molar mass for the different protein populations (Figure 6a). This homogeneity in molar mass was confirmed by intact mass spectrometry analysis using MALDI-TOF (Figure 7) and by size-exclusion chromatography ( Figure 8). Moreover, the 56 kDa mass observed for monomers under reducing conditions by intact mass spectrometry (Figure 7, "+DTT") also confirmed the presence of post-translational modifications (such as glycosylation) that occurred during expression in HEK293 cells, resulting in increased molecular weight compared to the predicted 49 kDa of RBDmfc monomers (see Figure 1). MALS analysis confirmed similar absolute molar mass for the different protein populations (Figure 6a). This homogeneity in molar mass was confirmed by intact mass spectrometry analysis using MALDI-TOF (Figure 7) and by size-exclusion chromatography ( Figure 8). Moreover, the 56 kDa mass observed for monomers under reducing conditions by intact mass spectrometry (Figure 7, "+DTT") also confirmed the presence of post-translational modifications (such as glycosylation) that occurred during expression in HEK293 cells, resulting in increased molecular weight compared to the predicted 49 kDa of RBDmfc monomers (see Figure 1).   Finally, thermograms of the purified WT and B.1.351 RBDmfc secreted proteins were established by differential scanning calorimetry (DSC). After baseline subtraction and integration of the signals, each tested variant showed two independent denaturation events at a half-denaturation temperature (Tm) of approximatively 46-49 • C and 77 • C, respectively (Figure 9b). Independent DSC experiments using mouse IgG2a and its Fab and Fc fragments showed that the thermal denaturation event occurring at Tm 77 • C corresponded to the Fc domain (Figure 9a, green curve), and consequently that the first denaturation event at Tm 46-49 • C was related to the RBD subdomain (Figure 9b). Tm and reaction enthalpy (ΔH) were measured from the thermograms for the two denaturation events and compared between WT and B.  Tm and reaction enthalpy (∆H) were measured from the thermograms for the two denaturation events and compared between WT and B.

Discussion
We investigated the impact of naturally occurring SARS-CoV-2 RBD mutations on the production and secretion of a recombinant Fc-tagged RBD protein (RBDmfc) transiently expressed in HEK 293 cells. We showed that mutation E484K and the association E484K+K417N, but not N501Y, greatly impaired the secretion of RBDmfc. The comparison of WT and B.1.351 RBDmfc proteins (the latter carrying mutations K417N, E484K and N501Y) demonstrated that impaired secretion of the B.1.351 RBDmfc was not due to a defect in protein translation but to its retention in the ER. WT and B.1.351 RBDmfc proteins showed differences in disulphide bond arrangements, hydrophobicity and thermostability, suggesting that both proteins differ in their conformation and folding. We, therefore, propose that improper disulphide bond formation and misfolding are responsible for the diminished secretion of B.1.351 RBDmfc, compared to WT RBDmfc, upon expression in HEK 293 cells.
We attempted to improve B.1.351 RBDmfc protein folding and secretion by co-expressing the chaperone and protein disulphide isomerases PDIA2, as successfully described for other mutated proteins [46,47]. Unexpectedly, PDIA2 overexpression resulted in the increased retention of both WT and B.1.351 RBDmfc proteins with altered disulphide bonds in the ER. Such "anti-chaperone" behaviour of PDIA2 has been previously reported [42,58,59]. The mechanisms controlling the chaperone vs. anti-chaperone activities of PDIA2 are not fully understood, but some evidence suggests that PDIA2 s anti-chaperone activity might preferentially target disulphide-rich proteins under certain cellular conditions [58,59]. Interestingly, our results resemble those reported by Davis et al. [58], demonstrating that overexpression of PDIA2 in mammalian cells was associated with the ER retention of the recombinant tumour necrosis factor receptor (TNFR)-Fc fusion protein (containing 14 disulphide bridges), but not of the globular protein IL-15 (containing two disulphide bonds) [58]. These authors showed that retention of TNFR-Fc in the ER by PDIA2 required the disulphide isomerase activity of PDIA2 [58]. These observations suggest that upon overexpression of PDIA2 in HEK 293 cells, RBDmfc proteins are retained in the ER and their disulphide bonds are rearranged by PDIA2. In the process, WT and B.1.351 RBDmfc proteins' disulphide bonds are differentially altered, likely due to differences in WT and B.1.351 protein folding.
In addition, the observation that the intracellular level of B.1.351 RBDmfc was further increased upon overexpression of PDIA2 suggests that aggregated misfolded B.1.351 RB-Dmfc might be normally eliminated, possibly via the ER-associated protein degradation (ERAD) pathway, in the absence of exogenous PDIA2, in order to preserve ER integrity and homeostasis [23,44,49]. This further support the idea that the overall lower yield of B.1.351 RBDmfc vs. WT RBDmfc was not due to a defect in protein translation but rather in trafficking along the secretory pathway.
The different disulphide bond arrangements detected between WT and B.1.351 RB-Dmfc are likely due to differences in protein conformation and folding caused by point mutations within B.1.351 RBD, especially E484K, and to a lesser extent K417N. This proposition is in agreement with the observations that mutation E484K affects the structure and conformation of the spike protein [54][55][56] and that disulphide bond formation plays an essential role in the structure and function of the RBD [53]. These differences in protein folding might be responsible for the improper disulphide bond formation, which is known to be influenced by the proximity and accessibility of cysteine residues [38,41]. These differences in protein structure were also evidenced by the hydrophobicity profile of the WT and B.1.351 proteins and might explain the reduced thermostability of the RBD domain of the B.1.351 RBDmfc fusion protein identified by differential scanning calorimetry, in line with the important role of disulphide bond formation in protein structure stabilisation [42,53]. The reduced thermal stability of the B.1.351 RBD domain compared to that of WT RBD detected in our study is in agreement with a previous report demonstrating the destabilising effect of E484K and K417N mutations, but not of the N501Y mutation, in thermal denaturation experiments [60].
Finally, the critical role of residue E484K was recently confirmed in our laboratory by preliminary experiments performed using an RBDmfc variant based on the recently emerged SARS-CoV-2 omicron (B.1.1.529 or BA.1) variant [19]. Among the 15 RBD mutations of the B.  Figure S3). The E484K mutation results in the substitution of a negatively charged amino acid (glutamic acid E) with a positively charged amino acid (lysine K), while the E484A mutation reverses the positively charged K to a neutral and hydrophobic amino acid (alanine A). This modification is likely to alter electrostatic interactions and the conformation of the RBD [55], as further suggested by the 3D structure comparison surrounding amino acid 484 ( Figure S4). Although we cannot exclude an additional impact of the 12 other RBD mutations in the production yield of B.1.1.529 RBDmfc, these preliminary data confirm the importance of amino acid 484 for the proper folding and secretion of RBD recombinant proteins.
It is worth noting that in the context of its C-terminal fusion with the Fc domain, our recombinant RBD protein allows principally only three of the four disulphide bonds identified in the native RBD (namely C336-C361, C379-C432 and C480-C488) [51][52][53]. Indeed, in our fusion protein, the RBD domain ends at amino acid 524, thus preventing the formation of the native disulphide bond C391-C525. Other recombinant Fc-tagged RBD proteins (amino acids 331-524), thus lacking C525, were previously described as retaining structural and biological properties of the native RBD domain, such as efficient binding to the human ACE2 receptor [14][15][16]. Here, we additionally demonstrated the antigenic property of RB-Dmfc, by showing its ability to be recognised by SARS-CoV-2-specific IgG from COVID-19 convalescent patients, in agreement with previous results based on the His-tagged version of recombinant RBD (amino acids 331-524) [13]. The missing C391-C525 disulphide bond did not impede the efficient production and secretion of WT RBDmfc. However, in the context of the B.1.351 mutant and its altered conformation, we cannot exclude that the lack of this amino acid might contribute to the reduced efficiency of protein secretion, by further affecting disulphide bond formation and isomerization during protein maturation in the ER. Additional experiments would be needed to address this question. Finally, the data presented here should serve as a basis for the design of optimisation protocols aiming to improve the production of B.1.351 RBDmfc and other difficult-to-express variant proteins, for instance by manipulating the UPR pathway, as described for unrelated recombinant proteins [23,24,33,34].

Conclusions
We showed that RBD mutation E484K has a major impact on the folding, disulphide bond formation and secretion of a recombinant RBD protein expressed in human 293F cells. Optimisation protocols, possibly manipulating the UPR, might permit us to overcome this challenge. Institutional Review Board Statement: The study was conducted in accordance with the Declaration of Helsinki, as revised in 2013. Human sera from COVID-19 patients were leftover samples of routine laboratory testing for SARS-CoV-2 infection. Accordingly, ethical review and approval were waived due to the use of leftover samples for research purposes, after written informed consent was obtained, in accordance with French regulations. The human negative serum was collected from a healthy donor by the Etablissement Français du Sang (EFS, France). EFS received the authorisation (AC-2017-2958) from the French ministry (Ministère de l'Enseignement Supérieur, de la Recherche et de l'Innovation) to collect and transfer samples to partners.

Informed Consent Statement:
Informed consent for the use of leftover blood samples for research purposes was obtained from all subjects involved in this research project.

Data Availability Statement:
The data presented in this study are available within the article and supplementary materials.