Construction of expression vectors for pEAQ-Cetuximab (pEAQ-Cx, two independent expression vectors, one for the heavy chain and one for the light chain), pTra-Cetuximab (pTra-Cx, one expression vector carrying two expression cassettes) and magnICON-Cetuximab (magnICON-Cx, two independent expression vectors, one for the heavy chain and one for the light chain) was described recently (Eidenberger et al., 2022). The expression vector for rituximab (Rx) was provided by Medicago (Li et al., 2016) and is referred to as pCAM-Rx (one expression vector carrying two expression cassettes) in this study. For expression in leaf epidermal cells of N. benthamiana, the pEAQ-Cx and magnICON-Cx plasmids, were introduced into Agrobacterium tumefaciens strain GV3101, while for pTra-Cx A. tumefaciens strain GV3101 pMP90RK was used. Mammalian cell produced Rx (Rituxan) was kindly provided by Friedrich Altmann and described previously (Stadlmann et al., 2008).

Cloning and expression of LmSTT3D has been described previously (Castilho et al., 2018). For cloning of LdOST, the coding sequence (GenBank: TPP46432.1) was amplified by PCR from synthetic DNA using the forward primer 5’-CTTCCGGCTCGTTTGTCTAGAATG-3’ and the reverse primer 5’-AAAAACCCTGGCGGGATCC-3’. The PCR fragment was digested with XbaI/BamHI and subcloned into XbaI/BamHI-digested binary expression vector pPT2 (CaMV35S promoter, Strasser et al., 2005) to obtain pPT2-LdOST or p48 (ubiquitin-10 promoter, Hüttner et al., 2014) to obtain p48-LdOST-TAG. In the p48-LdOST-TAG construct a stop codon was included at the end of the LdOST coding sequence to prevent expression of an RFP-fusion protein. To generate p48-LdOST for expression of LdOST-RFP, the LdOST coding sequence without a stop codon was amplified by PCR using the forward primer 5’-CTTCCGGCTCGTTTGTCTAGAATG-3’ and the reverse primer 5’- TATAGGATCCCACCTCACCAAGAGTCCT-3’ and cloned into p48. To generate the human OST4-GFP expression construct, the OST4 coding sequence was amplified by PCR from human cDNA using forward primer 5’-TATATCTAGAATGATCACGGACGTGC-3’ and reverse primer 5’-tataGGATCCTTCCTGCTTCTTGGG-3’, XbaI/BamHI-digested and cloned into p47 (ubiquitin-10 promoter, GFP-tag, Hüttner et al., 2014). For ectopic expression, the plasmids were introduced into A. tumefaciens strain UIA143 (Strasser et al., 2005).

N. benthamiana plants were grown at 23°C under long-day conditions (i.e., 16 h light/8 h dark). Besides wild-type plants a XT/FT-knockout line (ΔXF KO) was used (manuscript in preparation). Infiltration into leaves of 5-week-old N. benthamiana was done as previously described (Göritzer et al., 2017). Briefly, the respective Agrobacteria were grown in LB-medium overnight at 29°C. Bacteria were centrifuged, resuspended in infiltration buffer (10 mM MgSO₄, 10 mM MES and 0.1 mM acetosyringone) and the suspension was used for infiltration. In case of pEAQ-Cx and magnICON-Cx, where light chain and heavy chain are on separate plasmids, Agrobacteria suspensions with OD₆₀₀ of 0.15 were infiltrated for the heavy chain while the OD₆₀₀ of Agrobacteria suspensions for infiltration of the light chain was 0.1 unless otherwise stated. For pTra-Cx and pCAM-Rx Agrobacteria suspensions with OD₆₀₀ of 0.15 were used for infiltration. Binary constructs containing either LdOST or LmSTT3D, respectively, were infiltrated with Agrobacteria suspensions with OD₆₀₀ of 0.1.

Analysis of subcellular localization via confocal microscopy was done as previously described (Schoberer et al., 2019). In short, the acquisition of live-cell confocal images was done on a Leica SP5 microscope using an oil immersion objective (Leica 63x/1.4 NA). For GFP excitation an argon laser at 488 nm and for RFP a diode laser at 561 nm was used. For detection a photomultiplier tube (PMT) and Hybrid detectors (HyD) at 500-530 nm and 600-630 nm respectively were employed.

For analysis of underglycosylation via densitometry, IgG1 constructs were infiltrated into leaves of N. benthamiana and harvested 3 days post infiltration (dpi). Plant material was mechanically disrupted using a ball mill and 4 µL of ice-cold extraction buffer (0.1 M TRIS, 0.5 M NaCl, 1 mM EDTA, 40 mM ascorbic acid, 2% (w/v), pH 6.8) added to 1 mg of plant material. After removal of cell debris via centrifugation Laemmli buffer was added to the extracts and the samples heated to 95°C for 5 min. Samples were then diluted 1:5 before applying them to SDS-PAGE (12% acrylamide). After electrophoretic separation, proteins were blotted to a nitrocellulose membrane and subsequently detected using an anti-IgG antibody conjugated to horseradish peroxidase (Promega). Proteins were visualized using ECL substrate on a Fusion instrument (Vilber). Densitometric analysis of bands corresponding to glycosylated and unglycosylated IgG1 heavy chain was done using the Evolution-Capt software (Vilber). Background subtraction was done using the rolling ball method of the software.

For analysis of N-glycans, IgG1 reporters were expressed as described above and enriched via Protein A sepharose as described previously (Castilho et al., 2018). Elution was done using 0.1 M glycine pH 2.8. Following reduction and carbamido-methylation the enriched proteins were subjected to a tryptic in-solution digest (Kolarich and Altmann, 2000) using Trypsin Platinum, Mass Spectrometry Grade (Promega). 10-20 µg protein are routinely digested with 0.5-1.0 µg trypsin for 16 h at 37°C. The resulting glycopeptides were analyzed using an Orbitrap Exploris 480 (Thermo Scientific) and the obtained data analyzed using Skyline Version 22.2 software (MacLean et al., 2010). ANOVA with Tukey post-hoc tests were carried out using GraphPad Prism 9.2 software.

To quantify the degree of underglycosylation of plant produced IgG1 via mass spectrometry purified proteins were either measured intact or the fraction of unglycosylated heavy chain analyzed from trypsin-digested glycopeptides. For the intact measurement, 1 µg of protein was applied directly to a Waters BioResolve Column (2.1 x 5 mm). An acetonitrile gradient was used ranging from 15% to 80% (v/v) acetonitrile in 0.1% (v/v) formic acid. Gradient time was 15 minutes, flow rate was set to 400 µL/min at 80°C. Detection was performed on a Q-TOF instrument (Agilent Series 6560 LC-IMS-QTOFMS) equipped with the JetStream ESI source in positive mode. Data analysis was done using MassHunter BioConfirm B.08.00. Spectrum deconvolution was performed using MaxEnt. For quantification on the peptide level, glycopeptides were deglycosylated using Protein-N-glycosidase A (PNGase A, Europa Bioproducts Ltd) which leads to deamidation of the glycosylated asparagine residue present in the N-glycosylation site. The resulting mass shift of + 0.984 can be monitored via mass spectrometry. For normalization, a synthetic peptide EEQYNSTYREEQYDSTYR (JPT Peptide Technologies) was used as described (Castilho et al., 2018). Analysis of mass spectrometry data was done using Skyline Version 22.2. ANOVA with Tukey post-hoc tests were carried out using GraphPad Prism 9.2 software.

A 20-mer peptide (P20) of the extracellular loop of human CD20 was used as an antigen (Blasco et al., 2007). 1 μg/mL of P20 (diluted in PBS buffer, pH 7.4) was coated (50 µL/well) to 96 well microplates (MicroWell™ MaxiSorp™ Merck) for 16 h at 4°C, then saturated by incubation 100 µL/well with 3% fat free milk powder, dissolved in PBS-T (PBS with 0.05% Tween 20) for 1.5 h at 22°C. Solutions of rituximab diluted in blocking solution and applied to the coated plates in two-fold serial dilutions starting from 1000 µg/mL were then added (50 µL/well) and incubated for 2 h at 22°C to obtain calibration curves. Peroxidase-conjugated goat anti-human gamma chain antibody (Merck) was added at a dilution of 1:5000 and the plates were incubated for a further hour at 22°C. Every step was followed with three-time PBS-T washing. 50 µL/well of the substrate 3,3’,5,5’-tetramethylbenzidine (Merck) was added and plates were incubated for 5 to 10 min. The reaction was stopped with 2 M H₂SO₄ and absorbance (λ = 450 nm) with reference to 620 nm was measured with an ELISA reader (Tecan Spark® spectrophotometer). The binding curve was generated using GraphPad Prism 9.2 software.

FcγRIIIa (CD16a; F158 allotype) expressing TZM-bl cells were used to assess the binding affinity of Rx variants. Flow cytometry was carried out exactly as described (Forthal et al., 2010). The binding curves were generated by plotting the mean fluorescence intensity of positive cells indicating receptor binding as a function of antibody concentration. Unspecific binding to wild-type cells was subtracted from binding to FcγRIIIa-expressing cells. Each antibody concentration was run in duplicate. Binding experiments were repeated three times.

In vitro binding experiments of Rx variants to the extracellular domain (amino acids 17-208) of FcγRIIIa/CD16a was performed using a Biacore T200 (Cytiva). First the sensor chip surface was captured with anti-His antibody with the His Capture Kit (Cytiva) to a CM5 chip as described in the manufacturers’ protocol. The capturing of anti-His antibody reached 32000 RU. Secondly, immobilization of the His-tagged human FcγRIIIa (F158 allotype) or FcγRIIIa (V158 allotype) (AcroBiosystems) on the chip surface was performed for 60s with a concentration of 1 μg/mL and a flow rate of 10 μL/min in HEPES-EP running buffer. 40 RU units were achieved. The immobilization step was previously optimized in order to avoid avidity effects using lower concentration of FcγRIIIa and shorter times. Flow cell 2 remained unmodified and served as a reference cell for the subtraction of systematic instrument noise and drift. Rx binding curves were generated in multi-cycle kinetic experiments at seven different concentrations ranging from 7.8 to 500 nM with 60 seconds association and 60 seconds dissociation time at a flow rate of 10 μL/min. After each run, surface regeneration was accomplished using 10 mM glycine, pH 1.7, for 120 seconds at a flow rate of 30 μL/min. Binding affinities (K_D) were calculated with Biacore T2 Evaluation software using a 1:1 binding in kinetics (Rx, Rx + LdOST, Rx + LmSTT3D) or steady state (Rituxan). All experiments were repeated as three independent kinetic runs.

Protein stability measurements were carried out by differential scanning fluorimetry (DSF) using the CFX96 Real-Time PCR Detection System (Bio-Rad) with a final dilution of 1:500 of the SYPRO Orange dye (Molecular Probes). Fluorescence of a 25 μL sample (final concentration 0.4 mg/mL) in PBS was recorded from 10 - 95°C (0.5°C increments, 10 seconds hold per step) using the FRET channel. The thermograms, both the normalized relative fluorescence units (RFU) and the normalized derivative of relative fluorescence units (d(RFU)/dT) with respect to temperature (T) were recorded and compared. The peaks of the d(RFU)/dT-T thermogram are regarded as the melting temperatures (Tm) of the corresponding protein. Triplicate measurements were performed for each protein. Consistent with previous literature, the first peak of the mAb was attributed to the unfolded Fc fragment, while the second transition was interpreted as the melting of the Fab fragment.

A genome edited β1,2-xylosyltransferase and core α1,3-fucosyltransferase N. benthamiana knockout line devoid of xylose and fucose carrying N-glycans (ΔXF KO), was used as expression host (manuscript in preparation). Two therapeutic IgG1 antibodies, namely rituximab (Rx) carrying only the conserved N-glycosylation site Asn297 and cetuximab (Cx) which has an additional N-glycosylation site in the Fab domain of the HC were characterized (Castilho et al., 2015; Li et al., 2016; Eidenberger et al., 2022). First, antibodies were transiently expressed under different conditions, including the use of various vector backbones (i.e. magnICON, pEAQ, pCAM, pTra), either as single genes or in tandem repeats, different bacterial concentrations for agroinfiltration and harvesting at diverse time points. Recombinant antibody expression was monitored by immunoblotting of total soluble protein extracts, with focus on the two HC associated bands, representing the glycosylated (~ 53 kDa band) and the underglycosylated HC (~ 50 kDa, Supplemental Figure S1). Densitometric analyses revealed significant amounts of underglycosylated HC in all settings, varying from 15-30% for Rx and 30-55% for Cx. These results are in line with previous observations (Castilho et al., 2018; Eidenberger et al., 2022) and demonstrate that changing expression conditions only had a minor impact on the N-glycosylation efficiency.

Recently we characterized LmSTT3D, a single-subunit OST from Leishmania major that increases the N-glycosylation efficiencies of plant-produced recombinant glycoproteins (Castilho et al., 2018). GFP tagged LmSTT3D transiently expressed in N. benthamiana was located in the ER and displayed punctate structures that resemble Golgi bodies as reported previously (Castilho et al., 2018). A more detailed analysis by confocal microscopy revealed that LmSTT3D-GFP exhibits unusual aggregate-like subcellular structures (Supplemental Figure S2A) and can alter the N-glycan profile of co-expressed recombinant antibodies towards mannosidic structures (Supplemental Figure S2B). While this negative impact on N-glycan processing was only occasionally observed, these findings indicate a possible interference of LmSTT3D with the plant endogenous glycosylation pathway that should be avoided for the robust production of homogenous recombinant glycoproteins.

To overcome this potential limitation of LmSTT3D, sequence databases of protists were searched for alternative single-subunit OSTs. From the obtained hits we selected the single-subunit OST from Leishmania donovani. While this protein termed LdOST displayed less than 80% amino acid sequence identity to LmSTT3D, crucial amino acid residues in the substrate binding region appear all conserved in LdOST and the predicted membrane protein topology is similar to LmSTT3D. This makes LdOST a potential single-subunit OST candidate to overcome IgG1 underglycosylation in plants (Supplemental Figures S3, S4).

We generated an LdOST-RFP fusion protein (Figure 1A), transiently expressed it in N. benthamiana and analyzed the subcellular localisation in leaf epidermal cells using confocal microscopy. In contrast to LmSTT3D-GFP (Supplemental Figure S2A), LdOST-RFP displayed exclusively ER labelling and no aberrant cellular structures that are indicative of aggregate formation or mislocalization. Co-localization with a GFP-tagged ER-marker (OST4-GFP) (Figure 1C) confirmed the subcellular localization (Figure 1B). When an untagged LdOST variant was co-expressed with pCAM-Rx and the ER-marker RFP-PDI5 (Shin et al., 2021), a typical ER labelling was observed for the ER marker suggesting that LdOST has no obvious adverse effect on the ER morphology (Figure 1D). By contrast, Rx plus LmSTT3D-GFP and RFP-PDI5 expression resulted in the formation of an aberrant ER with accumulation of the ER marker in punctuate structures (Figure 1E).

Next, we transiently co-expressed untagged LdOST with pCAM-Rx and Cx, respectively, in N. benthamiana ΔXF KO. In crude leaf extracts and on affinity-purified antibodies, the faster migrating HC band at approximately 50 kDa, representing the underglycosylated HC, disappeared in the presence of LdOST suggesting the expression of an active OST subunit (Figures 1F, G). Three different MS-based methods were applied to exactly quantify glycosylated versus non-glycosylated Fc of purified antibodies. First, Rx and Cx antibodies were subjected to tryptic digestion and the respective Asn297 containing polypeptides analyzed by LC-ESI-MS (Figures 2A, 3A). Second, N-glycans were released from the glycopeptides by digestion with PNGase A and the mixture of deglycosylated (EEQYQSTYR) and unglycosylated (EEQYNSTYR) peptides was examined by LC-ESI-MS. A peptide harbouring both the unglycosylated and deglycosylated (N to Q conversion) amino acid sequence motif was used as an internal standard for quantification and processed in the same way as the samples (Figures 2B, 3B) (Castilho et al., 2018). Finally, intact fully assembled IgG1 was analyzed by MS (Figures 2C, D, 3C, D).

Collectively, in the absence of any OST subunit, both antibodies displayed substantial amounts of unglycosylated HC-Fc with clear differences between Rx and Cx (26% and 46% non-glycosylated, respectively) (Figures 2B, 3B). LmSTT3D and LdOST were both able to significantly reduce the underglycosylation at site Asn297 of both antibodies (down to 3 and 10%, respectively). Intact MS-measurement of the assembled antibodies showed considerable amounts of non- and hemi-glycosylated Rx (together 28%) (Figure 2C) and Cx (together 43%) (Figure 3C). These values were significantly reduced by OST subunit co-expression with up to 97% of the assembled antibodies being fully glycosylated. Interestingly, under all tested conditions, the N-glycan occupancy at the Cx Fab N-glycosylation site was more than 99% (Supplemental Figure S5).

Expression of a single-subunit OST might directly or indirectly affect the N-glycan composition on the antibody. The direct effect could be caused by the transfer of an incompletely assembled oligosaccharide precursor (Man₅GlcNAc₂ to Glc₂Man₉GlcNAc₂ intermediates) (Kelleher and Gilmore, 2006). The indirect effect could be linked to changes in the secretory pathway which may alter the organisation of glycosyltransferases in the Golgi apparatus and thus N-glycan processing. Therefore, the N-glycan composition was determined by LC-ESI-MS. While, the N-glycan profile did not change by overexpression of LdOST, LmSTT3D co-expression resulted in a small but significant increase in oligomannosidic N-glycans (Figure 4; Supplemental Figures S6, S7). This increase was also observed for the Cx Fab N-glycans (Supplemental Figures S5, S8) and is consistent with the observed tendency of LmSTT3D to interfere with ER-morphology.

Having established that LdOST is functional, we further examined the impact of LdOST expression on Rx function. An antigen binding ELISA showed equal binding for the three tested Rx variants - Rx alone, Rx co-expressed with LdOST and Rx co-expressed with LmSTT3D, respectively (Supplemental Figure S9). In accordance, in a cell-based assay, Rx co-expressed with LdOST or with LmSTT3D displayed comparable binding to cells expressing the low affinity FcγRIIIa (F158 allotype) (Supplemental Figure S10). To analyze the binding kinetics in more detail and determine the binding affinities, we carried out SPR with the ectodomains of FcγRIIIa (F158) and FcγRIIIa (V158) (Figure 5). Rx co-expressed with LdOST and LmSTT3D, respectively, displayed reduced K_D values with both receptors indicating that improved N-glycosylation site occupancy increases the binding affinity to the receptor (Supplemental Table S1). The K_D of the mammalian cell-derived Rx (rituxan) was at least 10-fold higher being consistent with the negative impact of the core fucose on FcγRIIIa binding (Shields et al., 2002; Ferrara et al., 2011). Finally, we compared the thermal stability of the antibodies. While all plant-produced Rx variants displayed melting points that were comparable to rituxan, the Rx co-expressed with LdOST variant exhibited a thermal unfolding transition of the CH2 domain at a slightly higher temperature (+ 1°C) than Rx alone (Table 1; Supplemental Figure S11).