BGAL1 depletion boosts the level of β‐galactosylation of N‐ and O‐glycans in N. benthamiana

Summary Glyco‐design of proteins is a powerful tool in fundamental studies of structure–function relationship and in obtaining profiles optimized for efficacy of therapeutic glycoproteins. Plants, particularly Nicotiana benthamiana, are attractive hosts to produce recombinant glycoproteins, and recent advances in glyco‐engineering facilitate customized N‐glycosylation of plant‐derived glycoproteins. However, with exception of monoclonal antibodies, homogenous human‐like β1,4‐galactosylation is very hard to achieve in recombinant glycoproteins. Despite significant efforts to optimize the expression of β1,4‐galactosyltransferase, many plant‐derived glycoproteins still exhibit incomplete processed N‐glycans with heterogeneous terminal galactosylation. The most obvious suspects to be involved in trimming terminal galactose residues are β‐galactosidases (BGALs) from the glycosyl hydrolase family GH35. To elucidate the so far uncharacterized mechanisms leading to the trimming of terminal galactose residues from glycans of secreted proteins, we studied a N. benthamiana BGAL known to be active in the apoplast (NbBGAL1). Here, we determined the NbBGAL1 subcellular localization, substrate specificity and in planta biological activity. We show that NbBGAL1 can remove β1,4‐ and β1,3‐galactose residues on both N‐ and O‐glycans. Transient BGAL1 down‐regulation by RNA interference (RNAi) and BGAL1 depletion by genome editing drastically reduce β‐galactosidase activity in N. benthamiana and increase the amounts of fully galactosylated complex N‐glycans on several plant‐produced glycoproteins. Altogether, our data demonstrate that NbBGAL1 acts on galactosylated complex N‐glycans of plant‐produced glycoproteins.


Introduction
Glycosylation is an important protein modification in all eukaryotes, and the impact of different glycan modifications on the function of glycoproteins has been extensively reviewed (Dalziel et al., 2014). Nicotiana benthamiana is one of the most widely used host plants to produce therapeutically relevant glycoproteins (Montero-Morales and Steinkellner, 2018). However, the production of recombinant proteins in plants can lead to the synthesis of aberrant glycosylation that can impair protein biological activity.
In planta galactosylation has been accomplished by transient or stable expression of a late-Golgi targeted GalT using a chimeric protein consisting of cytoplasmic tail, transmembrane domain and stem (CTS) region of rat a2,6-sialyltransferase (ST) ( ST GalT, Strasser et al., 2009). Previous results revealed that approximately 40% of N-glycans on endogenous total soluble proteins (TSPs) are galactosylated, but only 17% of N-glycans are galactosylated on secreted proteins present in the apoplastic fluid (AF) (Schneider et al., 2015).
Plant BGALs have numerous biological functions. So far, studies of plant BGALs have been focusing on their involvement in physiological processes such as fruit ripening, pollen development and seed germination and in cell wall modifications (Dwevedi and Kayastha, 2010). A recent report demonstrated that secreted NbBGAL1 contributes to immunity against pathogenic bacteria by releasing immunogenic peptides from glycosylated flagellin containing a terminal mVio residue (Buscaill et al., 2019). Importantly, NbBGAL1 null mutants generated by CRISPR/ Cas9 (bgal1) have drastically reduced b-galactosidase activity in the AF (Buscaill et al., 2019). Despite the numerous studies on BGAL functions, a direct evidence for the involvement of specific BGAL(s) in trimming N-glycans is still missing.
In this investigation, we cloned the cDNA sequence of NbBGAL1 in different expression vectors and determined its (i) subcellular localization, (ii) enzymatic activity/specificity towards galactosylated glycans and (iii) in planta biological activity. Finally, we assessed the impact of suppressing NbBGAL1 activity by transient RNA interference or using bgal1 null mutants on the generation of recombinant glycoproteins with di-galactosylated N-glycans.

Results
We postulated that all recombinant proteins are efficiently galactosylated when expressed in ΔXTFT GAL plants but that apoplast-resident glycosyl hydrolases, such as b-galactosidases, subsequently trim terminal b1,4-galactosyl residues. Two secreted active BGALs (NbBGAL1 and NbBGAL2) were previously identified in the extracellular space of N. benthamiana using ABPP (Chandrasekar et al., 2014). NbBGAL1 was detected with an apparent molecular weight (MW) of 45-kDa, lower than its theoretical MW for the mature protein (89.7-KDa), probably indicating that this is a processed protein. NbBGAL1 accumulates in the apoplast and is a functional b-galactosidase that cleaves galactose from fluorescein di-beta-D-galactopyranoside (FDG) and selectively catalyses hydrolyses of 4-nitrophenyl-conjugates of b-galactose but no other monosaccharide conjugates (Buscaill et al., 2019). Furthermore, null mutants of N. benthamiana lacking NbBGAL1 generated by genome editing (bgal1) have substantially reduced apoplastic b-galactosidase activity as compared to wild-type (WT) plants (Buscaill et al., 2019). According to these findings, NbBGAL1 is a good candidate to start investigating the involvement of secreted b-galactosidases in the processing of complextype galactosylated N-glycans decorating recombinant proteins produced in N. benthamiana.

NbBGAL1 overexpression and subcellular localization
To confirm the accumulation of NbBGAL1 in the apoplast and investigate other possible subcellular localizations, NbBGAL1 was C-terminally tagged with the monomeric red fluorescent protein (mRFP). cDNA of NbBGAL1 was cloned either including its endogenous signal peptide (SPb-BGAL1) or as a chimeric sequence where the signal peptide was substituted by the signal peptide of the barley alpha-amylase (SPa-BGAL1) ( Figure S1). SPb-BGAL1-mRFP and SPa-BGAL1-mRFP fusion proteins consist of the GH35 domain, a Gal-lectin domain and the fluorescent tag at the C-terminus, with an expected molecular weight of 115.1-kDa ( Figure 1a). SPb-BGAL1-mRFP and SPa-BGAL1-mRFP were transiently expressed in N. benthamiana leaves by agroinfiltration in the absence of silencing inhibitor p19. At two days postinfiltration (dpi), the fusion proteins are stable and detected as a~130-kDa mRFP-tagged protein in both AF and TSP (Figure 1b). In AF, several other bands are also detected and most probably represent degradation products. At 6 dpi, the fusion proteins are no longer detected ( Figure 1b). The higher MWs of the two mRFP fusion proteins when compared to the calculated MW are most probably due to protein glycosylation (see section on characterization of NbBGAL1). To determine NbBGAL1 subcellular localization, leaf epidermal cells expressing SPb-BGAL1-mRFP and SPa-BGAL1-mRFP were analysed by live-cell confocal microscopy at 2 dpi. The fluorescence signal detected for both fusion proteins is typical of secreted proteins (Shin et al., 2017) (Figure 1c), confirming that NbBGAL1 is an apoplastic beta-galactosidase.
Full-length NbBGAL1 (25-846 amino acids) and the NbBGAL1-GH35 domain (25-360 amino acids) were expressed in N. benthamiana using tobacco mosaic virus (TMV)-based magnICONâassembled vectors carrying either the bor a-signal peptide (SPb-BGAL1, SPa-BGAL1 and SPa-BGAL1-GH35; Figure S1). Coomassie staining of secreted proteins isolated from AF and subsequent peptide mapping demonstrates that SPa-BGAL1 accumulates as three protein bands: a 95-kDa corresponding to the full-length protein sequence; a 48-kDa band assigned to the GH35 domain lacking its C-terminus and a 38-kDa protein band representing a mixture of N-and C-protein truncations (Figure 2a). In contrast, SPb-BGAL1 expressed is detected as a single 48-kDa protein and all the identified peptides were originated from the GH35 catalytic domain and not from the C-terminal half of this protein ( Figure 2a). To evaluate whether this discrepancy is due to differences in the expression level between SPa-BGAL1 and SPb-BGAL1, we co-expressed the p19 RNA silencing suppressor protein, known to increase levels of transient expression (Voinnet et al., 2003). Indeed, p19 significantly boosted SPb-BGAL1 protein accumulation, which is now detected as a 95-kDa fulllength protein in addition to the truncated 48-kDa protein ( Figure 2b). Notably, despite several attempts to express the GH35 catalytic domain of NbBGAL1 (SPa-BGAL1-GH35) no protein was detected in AF or in TSP ( Figure S2).

Characterization of NbBGAL1 b-galactosidase activity
Many enzymes involved in protein glycosylation are themselves glycoproteins. Using the NetOGlyc 4.0 Server (www.cbs.dtu.dk/se rvices/NetNGlyc), we predicted four potential glycosylation sites on NbBGAL1, located at Asn 26 , Asn 255 , Asn 580 and Asn 723 ( Figure S3a). We therefore evaluated the glycosylation profile of NbBGAL1 expressed in DXTFT and DXTFT GAL plants. Trypsin digestion of full-length SPa-BGAL1 isolated from AF (95-kDa, Figure 2a) allowed the identification of three out of the four glycopeptides (GP1, 3 and 4; Table S1). Glycoprofiling of SPa-BGAL1 by LC-ESI-MS showed that all GPs are decorated with complex glycans with one or two terminal GlcNAc residues. Interestingly, SPa-BGAL1 expressed in DXTFT GAL plants lacks terminal b1,4-galactose residues in all GPs ( Figure S3b).
Next, NbBGAL1 was evaluated for b-galactosidase activity using a fluorogenic substrate for b-galactosidases. We confirmed that SPa-BGAL1 has b-galactosidase activity since it can cleave galactose from fluorescein di-b-D-galactopyranoside (FDG) when overexpressed in apoplast of N. benthamiana (Figure 2d). Approximately 0.6 mU of endogenous NbBGAL1 (WT) is naturally active in 1 mL of AF, and this is markedly increased to up to 300 mU/mL upon overexpression of SPa-BGAL1 (Figure 2d).
Studies of the enzyme kinetics determined the apparent K M and V max values of AF-derived SPa-BGAL1 for FDG as 29.8 mM and 2.7 µM/min/mL of AF, respectively.
b-Galactosidase activity assayed by FDG hydrolysis at various pH shows that SPa-BGAL1 optimal hydrolysis is obtained at pH5.0. The FDG hydrolytic activity was halved at near-neutral pH   (2) co-expressing p19 with SPb-BGAL1. Protein bands identified as BGAL1 by peptide mapping are marked (*). Protein size marker is shown in kilo Dalton (kDa). (d) b-Galactosidase activity was measured in AF isolated from leaves of N. benthamiana wild-type (WT) and from N. benthamiana transiently expressing SPa-BGAL1 by FDG assay. Error bars represent SEM of n = 3 biological replicates. (e) FDG assay was used to measure the optima pH for SPa-BGAL1 activity. Maximum activity at pH 5.0 was set to 100% to calculate the relative activity at other pHs. (f) The effect of various metal ions on the activity of AF-derived SPa-BGAL1 was studied by FDG in the presence of metal ions. Maximum activity obtained with no supplement (MES) was set to 100% to calculate relative activity in the presence of divalent metal ions (Fe 2+ Ca 2+ , Cu 2+ ) and EDTA. A b-Galactosidase purified from Aspergillus oryzae (276 U/mL) was used as a standard to quantify SPa-BGAL1 activity in mU/mL of AF. (g) The effects of different pHs, temperatures and metals on the b-galactosidase activity of SPa-BGAL1 were assayed with a galactose-binding lectin (RCA). Sialylated human A1AT (NaNa, negative control) was digested with neuraminidase to expose galactose residues (AA, positive control). The levels of galactosylation on A1AT N-glycans were compared before and after incubating the asialo-A1AT (AA) with SPa-BGAL1 in different conditions (see also Figure S6) and markedly reduced at extreme acidic pH (7.6% at pH 2.0), and only 1.5 % of full enzyme activity was observed at pH 9.0 ( Figure 2e).
The effects of various metal ions and EDTA on enzyme activity were studied by incubating AF from leaves overexpressing SPa-BGAL1 with FDG in the presence of 5.0 mM metal ions. EDTA has no significant effect on the hydrolytic activity; Mg 2+ and Ca 2+ slightly reduce SPa-BGAL1 activity (30%), while Fe 2+ and Cu 2+ significantly inhibited SPa-BGAL1 activity (90% inhibition for Fe 2+ and no activity detected in the presence of Cu 2+ ) ( Figure 2f). We confirmed the optimal pH and the impact of metal ions on SPa-BGAL1 activity by the ability of the enzyme to remove b1,4galactose residues present on plasma-derived alpha-1 anti-trypsin (A1AT) after removing terminal sialic acids by sialidase treatment (Figure 2g). Incubation of galactosylated A1AT (AA: Gal 2 Glc-NAc 2 Man 3 GlcNAc 2 ) with AF-derived SPa-BGAL1 at different temperatures shows that the SPa-BGAL1 optimum activity is at 37 ᵒ C. SPa-BGAL1 enzymatic activity is reduced at lower temperature (25 ᵒ C); lower at relatively increased temperatures (up to 50 ᵒ C); and inhibited at higher temperatures (from 60 ᵒ C and above) ( Figure 2g).
The in vitro b-galactosidase activity assay described above does not discriminate if both forms of SPa-BGAL1 (full-length and GH35 domain) protein are active. The active state of glycosidases isolated from apoplast has been previously monitored, indicating that the truncated 48-kDa NbBGAL1 is an active b-galactosidase (Buscaill et al., 2019;Chandrasekar et al., 2014). It is thought that that trimming of the C-terminal half of BGAL generates a mature and active protein consisting only of the GH35 catalytic domain.
Here, we aimed to investigate whether the full-length version of the protein, carrying the C-terminal domain, is active. However, our attempts to use a C-terminal Strep-tag to purify the full-length SPa-BGAL1 protein were unsuccessful. To overcome this shortcoming, we used glycosidase activity profiling with a fluorescent glycosidase probe (Chandrasekar et al., 2016) to specifically identify active b-galactosidases in AFs expressing p19 or co-expressing p19 and SPb-BGAL1. Both AF samples showed a 45-kDa fluorescent protein signal corresponding to the GH35 domain of NbBGAL1, but we detected an additional signal at 95 kDa upon overexpression of SPb-BGAL1 ( Figure 2c). These data demonstrate that full-length NbBGAL1 has b-galactosidase activity.

Low levels of human-like b1,4-galactosylation in planta
Full monoclonal antibodies co-expressed with ST GalT or expressed in ST GalT transgenic plants are up to 83% galactosylated from which 60% are di-galactosylated (AA) (Schneider et al., 2015). These results contrast strongly with the low galactosylation levels of other glycoproteins in N. benthamiana.
To illustrate this, we expressed five different recombinant glycoproteins in ΔXTFT plants stably transformed with a modified version of the human b1,4-galactosyltransferase. These recombinant plants are similar to those described previously ( ST GalT-DXF) (Schneider et al., 2015), but in this new host plant, the catalytic domain of the GalT was targeted to the late Golgi using the cytoplasmic tail, transmembrane domain and stem (CTS) region of the Arabidopsis b1,3-galactosyltransferase (GALT1). ΔXTFT plants stably transformed with GALT1 GalT (ΔXTFT GAL ) behave very similar to ST GalT-DXF with no major differences in growth or developmental phenotypes (Schneider et al., 2015).
We next used viral-based binary vectors (magnICONâ) to transiently express five different recombinant proteins that carry a different number of glycosylation sites: i) human transferrin (TF); ii) erythropoietin fused to an Fc fragment (EpoFc); iii) human alpha-1 anti-trypsin (A1AT); iv) monoclonal antibody cetuximab (Cx-IgG); and v) the Fc fragment with engineered HER2/neubinding sites (Fcab-HER2 or in short Fcab) (Figure 3a and Table S1, Castilho et al., 2011bCastilho et al., , 2013Castilho et al., , 2014Castilho et al., , 2015Jez et al., 2012). The recombinant proteins were transiently expressed in ΔXTFT GAL plants and either purified from TSP (Fcab-Her2, EpoFc, Cx-IgG) or collected in AF (TF and A1AT) (Figure 3b). Liquid chromatography-electrospray ionization mass spectrometry (LC-ESI-MS) revealed that only 8% of N-glycans are di-galactosylated for the Cx-IgG glycosite located on the variable fraction (Fab) of the heavy chain (GP1) ( Figure S4 and Figure 3c). By contrast, the GP2 site of Cx-IgG shows up to 50% of di-antennary galactosylated (AA) structures. Galactosylation of the Fc glycosite on Fcab-Her2 is also not efficient as only 36% of the N-glycans are fully digalactosylated. Similarly, galactosylation of the three glycosites of the Epo fragment on EpoFc is very inefficient and only 12% of the sites are di-galactosylated. Although TF can be galactosylated with mono-antennary hybrid structures, di-galactosylation was very poor in TF and not detected in A1AT. All together, these data indicate that the efficiency of galactosylation is poor and varies from protein to protein, and even between different glycosites of a particular protein.

Core fucosylation suppresses galactosylation
Most mammalian glycoproteins carry complex N-glycans decorated with a core a1,6-fucose residue, while in plant proteins, glycans can be decorated with a a1,3-fucose residue. We previously demonstrated that the a1,3-fucose residue promotes sialylation in DXTFT plants that express the human sialylation pathway . Galactosylation is the intermediate step necessary for protein sialylation. To determine whether fucosylation also affects galactosylation, Cx-IgG was transiently expressed in ΔXTFT GAL plants in the absence of core fucose or coexpressed with a1,3-FucT and a1,6-FucT fucose transferases. Glycoprofiling showed that core fucosylation negatively influences Fc-galactosylation: The level of bi-antennary galactosylated glycans is reduced upon co-expression with a1,6-FucT (27%, AAF 6 ) or a1,3-FucT (12% AAF 3 ) when compared to 50% digalactosylated Fc-glycans in the absence of fucosyltransferases ( Figure 3d and Figure S5). These data indicate that the core fucosylation negatively affects the maintenance of galactosylated glycans.
NbBGAL1 removes terminal b1,4-galactose residues on recombinant N-glycoproteins As shown above, the galactosylation levels of plant-derived IgG-Fab, EpoFc, Fcab-Her2, TF and A1AT differ considerably from the high levels observed on IgG-Fc. To address the question whether galactosylated N-glycans are substrates for NbBGAL1 and evaluate its ability to remove terminal b1,4-galactose residues in vitro, we incubated galactosylated proteins with AF from WT N. benthamiana and from N. benthamiana overexpressing SPa-BGAL1 and compared their galactosylation levels using a galactosebinding lectin Ricinus communis agglutinin (RCA). The results for IgG-Fc showed lower levels of galactosylation upon incubation with recombinant SPa-BGAL1, but the relative levels of galactosylation do not differ significantly after incubation with endogenous BGAL activity present in the AF (Figure 4a). To assess the activity of NbBGAL1 also on Fcab-Her2, TF and A1AT, we used sialylated versions of the proteins and removed sialic acid by  Figure S6). When compared to IgG-Fc, these proteins are more susceptible to the activity of NbBGAL1 (Figure 4a). Protein glycosylation profiles reveal a significant reduction of di-galactosylated glycans for all glycopeptides due to the activity of native (AF) and overexpressed NbBGAL1 (AF + SPa-BGAL1) (Figure 4b, Figure S6 and Table S2), demonstrating that galactosylated Nglycans are substrates for NbBGAL1.
Next, to evaluate whether recombinant NbBGAL1 is active in vivo, we co-expressed reporter glycoproteins with or without SPa-BGAL1 in DXTFT GAL plants and compared the b1,4-galactosylation levels by LC-ESI-MS. To avoid expression of competitive virus, reporter glycoproteins were co-expressed with SPa-BGAL1 cloned into non-viral-based binary vectors ( Figure S1). Overall, a drastic reduction of mono-and di-galactosylated glycans was observed upon co-expression of SPa-BGAL1 or SPb-BGAL1 ( Figure 5 and Figure S7), demonstrating that NbBGAL1 is active in vivo. Moreover, analysis of secreted endogenous proteins present in AF from ΔXTFT GAL plants showed that galactosylation levels of 18% are further reduced to 6% upon expression of SPa-BGAL1 ( Figure S8). All together, these data demonstrate that NbBGAL1 can remove terminal b1,4-galactose residues on recombinant N-glycoproteins.
NbBGAL1 removes terminal b1,3-galactose residues from Lewis a epitopes on N-glycans and from mucin-type Oglycans In contrast to the occurrence of b1,4-galactosyl residues in animal complex-type N-glycans, plant-derived glycoproteins, especially secreted-type glycoproteins, may carry a b1,3-galactosyl residue Here, we have expressed EpoFc in WT N. benthamiana and detected significant amounts of structures compatible to Le-a by immunoreaction to JIM84.
The b1,3-galactosyltransferase gene responsible for the biosynthesis of the Lewis a in Arabidopsis has been identified (Strasser et al., 2007a), but the b-galactosidase responsible for the degradation of N-glycans harbouring the Le-a epitope remains to be identified.
To investigate whether NbBGAL1 can remove the b1,3galactose residues present in the EpoFc Le-a, we incubated the purified EpoFc protein with AF isolated from WT N. benthamiana and monitored the amounts of Le-a using the anti-Le-a antibody JIM84. A decrease in the Le-a signal was immediately detected after two hours of incubation, which became more evident after 4 hours (Figure 6a), indicating that NbBGAL1 can remove the b1,3-galactose residues present in the EpoFc Le-a. Next, to evaluate the activity of NbBGAL1 in planta, EpoFc was transiently co-expressed in N. benthamiana with and without SPa-BGAL1. The 55-kDa band corresponding to the intact EpoFc is detected using JIM84 only when SPa-BGAL1 is not overexpressed (Figure 6b), indicating that the synthesis of Le-a in EpoFc is prevented or reduced by SPa-BGAL1 activity. These results were confirmed by LC-ESI-MS where glycan profiles showed that Le-a epitopes are not detected on EpoFc upon co-expression with SPa-BGAL1 ( Figure S9). All together, these data demonstrate that NbBGAL1 removes terminal b1,3-galactose residues from Lewis a epitopes on N-glycans.
O-glycosylation is a common post-translational modification of serine and threonine residues of secreted and membrane-bound proteins (Strasser, 2012;Strasser, 2013). Many glycoproteins carry glycans initiated by GalNAc attached to the hydroxyl of Ser or Thr residues, also called O-glycans.
Although plants do not contain endogenous glycosyltransferases that perform mammalian-type Ser/Thr glycosylation, the synthesis of plant-derived EpoFc carrying mucin-type O-glycan terminated by galactose (core 1 or T-antigen) was previously described .
To investigate whether NbBGAL1 is able to act on EpoFc Oglycans to remove the terminal b1,3-galactose residue present on T-antigen glycans, we transiently co-expressed EpoFc with a mammalian GalNAc-transferase (GalNAc-T2) and a drosophila core 1 b1,3-galactosyltransferase (C1GALT1) (Castilho et al., Detailed mass spectrometry for Fcab-Her2 and A1AT is shown in Figure S6. For interpretation of glycoforms present in assigned peaks, see Figure S13. . After protein A purification and confirmation of the synthesis of T-antigen (GalNAc + Gal) by mass spectrometry, EpoFc was incubated at 37ᵒC with AF from N. benthamiana plants expressing SPa-BGAL1. Glycan analysis revealed that NbBGAL1 is able to reduce the T-antigen present in EpoFc from 60% to 40% in 1 h and to 14% in 4 h ( Figure S10), demonstrating that NbBGAL1 can remove terminal b1,3-galactose residues from mucin-type O-glycans.

Suppression of NbBGAL1 activity increases the abundance of b1,4-galactosylated N-glycans
To investigate whether the formation of b1,4-galactosylated Nglycans can be improved by silencing NbBGAL1, a hairpin construct for NbBGAL1 (RNAiBGAL1; Figure S1) was generated. Transient RNAiBGAL1 expression drastically reduces the accumulation of transiently expressed BGAL1 (Figure 7a), demonstrating that the RNAiBGAL1 construct is effective.
Co-expression of RNAiBGAL1 with the five mammalian reporter proteins increased the level of mono-and di-galactosylation in these proteins, especially for glycosites that are normally poorly galactosylated (Figure 7b and Figure S11). For instance, we detect an increase from 8% to 33% in the Cx-Fab glycosite.
To investigate whether galactosylation of recombinant proteins is also increased in the bgal1 null mutant plants (bgal1-1), we compared the galactosylation level of reporter glycoproteins transiently co-expressed in WT and in bgal1-1. Since both these host plant do not express the b1,4-galactosyltransferase, we coexpressed the recombinant proteins with ST GalT, driven by a weak promoter to promote enhanced galactosylation (Act:: ST GalT; Kallolimath et al., 2018). When expressed in bgal1-1 mutant plants, the levels of di-galactosylation are significantly improved for all reporter glycoproteins: EpoFc (from 18% to 37%); TF (from 0% to 12%); Cx-IgG (from 37% to 56% in Fc and 15% to 25% in Fab); and Fcab-HER2 (from 6% to 28%) (Figure 7c and Figure S12). Importantly, these levels of galactosylation also depend on the transient expression of ST GalT. In agroinfiltration experiments, we cannot assure that all cells are simultaneously expressing ST GalT and the reporter protein. In fact, we observed a significant discrepancy on the di-galactosylation levels in independent experiments. Nonetheless, our combined results using RNAiBGAL1 and bgal1-1 null mutant plants clearly demonstrate that N. benthamiana NbBGAL1 activity is a critical factor for the trimming of terminal b1,4-galactose residues on Nglycans of recombinant glycoproteins in plants.

Discussion
Nicotiana benthamiana is currently the favourite and most widely used plant host when it comes to the production of recombinant proteins with engineered glycosylation. Tailored glycosylation profiles on N. benthamiana -derived therapeutic proteins have been reported including the extensive engineering of in planta multi-antennary sialylation Castilho et al., 2014;Schneider et al., 2013) and polysialylation (Kallolimath et al., 2016). However, except for monoclonal antibodies, none of these studies reported successful human-like b1,4-galactosylation.
Based on recent findings on the profiling of active glycosidases in the apoplast of N. benthamiana (Buscaill et al., 2019;Chandrasekar et al., 2014), we set up to evaluate and characterize NbBGAL1 as one of the potential enzymes involved in processing galactosylated glycans.
We used viral-based vectors to express NbBGAL1 (SPb-BGAL1 or SPa-BGAL1). Studies have shown that signal peptides can profoundly impact protein secretion (Kober et al., 2013). Thus, the efficiency of protein secretion can be improved by choosing an optimized signal peptide (Haryadi et al., 2015;Klatt and Konthur, 2012). This was observed for the overexpression of BGAL1 in N. benthamiana, where high level of expression of fulllength BGAL1 was only detected for the chimeric protein (SPa-BGAL1). However, the expression level of SPb-BGAL1 was efficiently improved by co-expression with silencing inhibitor p19.
In this study, we show that NbBGAL1 is a secreted and active beta-galactosidase acting on both N-and O-glycans. NbBGAL1 removes terminal b1,4-galactose residues on recombinant Nglycoproteins in vitro and in planta. In addition, NbBGAL1 removes terminal b1,3-galactose residues from Lewis a epitopes on N-glycans and from mucin-type O-glycans. Figure 6 NbBGAL1 removes terminal b1,3-galactose residues from Lewis a motifs. (a) EpoFc was expressed in Nicotiana benthamiana WT plants, and the synthesis of Lewis a motifs was monitored by glycan analysis. Purified EpoFc carrying N-glycans with Lewis a motifs was incubated with AF isolated from N. benthamiana WT plants for 2 or 4 h. The relative amounts of Lewis a motifs were monitored by Western blotting with anti-Lewis a antibodies (JIM84) (b) EpoFc was co-expressed in N. benthamiana without (À) and with (+) SPa-BGAL1. Ponceau staining show similar amount of proteins loaded. Bands corresponding to full-length fusion protein (EpoFc) and degradation product (free Fc) are marked. Cartoons representing N-glycans carrying Lewis a motifs are shown on the right. Mass spectrometry showing the relative levels of Lewis a epitopes for Epo GP3 is shown in Figure S9. For interpretation of glycoforms present in assigned peaks, see Figure S13. [Colour figure can be viewed at wileyonlinelibrary.com] Our attempts to determine whether the catalytic domain of BGAL1 (GH35) is sufficient for trimming terminal galactose residues were unsuccessful since expression of the NbBGAL1 GH35 domain lacking the C-terminus could not be achieved. Studies of the processing of a human lysosomal BGAL expressed in COS-1 cells suggested that the proteolytic processing of the full-length protein to remove the C-terminus is essential for activity (van der Spoel et al., 2000). Also here, expression of the human BGAL N-terminal catalytic domain without the C-terminal domain could not be achieved in COS-1 cells (van der Spoel et al., 2000). In contrast, transfection of COS-1 cells with DNA fragments representing the N-and C-terminal domains increased the overall b-galactosidase activity, implying that the catalytic activity of BGAL requires an interaction of the two domains.
Characterization of NbBGAL1 showed that the enzyme is a secreted glycoprotein with four potential glycosylation sites decorated with complex-type N-glycans. Interestingly, no galactosylation was detected on NbBGAL1 when expressed in ΔXTFT GAL plants, which could be due to the enzyme activity on its own glycans. The optimal in vitro NbBGAL1 activity was established at acidic pH (5.0), which corresponds to the leaf apoplastic pH (Grignon and Sentenac, 1991).
A common obstacle that hampers detailed research on the impact of glycosylation to protein functional activities is the incomplete knowledge of how glycan structures are generated and the unavailability of expression platforms that allow the synthesis of targeted glycoforms.
Studies on the impact of galactosylated N-or O-glycans on protein function are few since terminal galactose residues are either capped by sialic acid or trimmed off by galactosidases.
Examples are studies comparing the performance of asialo-and sialylated EpoFc , BChE (Schneider et al., 2013), IgM (Loos et al., 2014) and A1AT  that do not include the galactosylated (intermediate) glycoform. Another investigation comparing the synthesis core-fucosylated asialo-and sialylated Cx-IgG again excluded the fucosylated-  . The efficiency of glycoengineering depends on the protein and on the glycosylation site. A dramatic example was our efforts to b1,4galactosylate A1AT . Recombinant A1AT expressed in DXTFT and in DXTFT GAL is decorated mainly with paucimannosidic N-glycans contrasting with the highly efficient sialylation in DXTFT SIA . This indicates that capping b1,4-galactose residues with sialic acid prevents their exposure to galactosidases and consequently the accessibility of hexosaminidases to GlcNAc residues. Another example is EpoFc: the poor galactosylation observed for three glycosites located on the Epo fragment contrasts with efficient galactosylation on the Fc site. By contrast, EpoFc expressed in DXTFT SIA shows fully sialylated Epo glycosites, while Fc fragment is decorated mainly with mono-sialylated glycans . Even for monoclonal antibodies, glycoengineering greatly depends on the availability of the glycosite for glyco-modulating enzymes . While glycans connected to the IgG-Fab fragment of Cx-IgG are exposed to the surrounding solvent, the structures located in the Fc fragment are largely shielded by the opposing CH2 fragment, and therefore, galactosylation at this site is much more efficient. Interestingly, the exposure of Fc to glyco-modelling enzymes seems to depend on core fucosylation. Here, we showed that Fc N-glycans can also be converted into BGALs substrates upon plant-specific core a1,3-fucosylation. We have previously shown that N-linked glycans on IgG opposite chains (CH2) maintain the conformation of the Fc domain and impose constraints on the action of processing enzymes, thereby hampering synthesis of more complex glycans . However, plant-specific core fucosylation seems to alleviate these structural constraints making terminal sugar residues available not just to glycosyltransferases (e.g. GnT-III, IV V and ST)  but also to glycosidases (e.g. HEXO3; Shin et al., 2017).
In conclusion, active BGALs are a severe limitation for tailored glycosylation because these enzymes generate truncated glycans and compromise the synthesis of galactosylated N-and Oglycans. We showed that co-expression of the RNAiBGAL1 construct resulted in increased levels of galactosylated N-glycans for all tested glycoproteins and glycosites. Moreover, when expressed in bgal1 mutant plants with a WT background, galactosylation of recombinant proteins is greatly increased. The impact of terminal galactosylation on the function of recombinant proteins has only been determined for IgG1 (Thomann et al., 2016). The results presented here are highly encouraging for a future establishment of yet another DXTFT-based expression platform, depleted of b-galactosidase activity, thus enabling b1,4galactosylation on a diverse group of glycoproteins. This study opens a new area for plant-based glycoengineering.

Plant material
In this investigation, we used Nicotiana benthamiana wild-type plants (WT) and mutant plants lacking plant-specific core b1,2xylose and a1,3-fucose residues (DXTFT; Strasser et al., 2008). For in planta b1,4-galactosylation, the cytoplasmic tail, transmembrane domain and stem (CTS) region of the rat a2,6-sialyltransferase present in the ST GalT construct (Strasser et al., 2009) was replaced by the CTS region (amino acids 1-60) of the Arabidopsis a1,3-galactosyltransferase (GALT1). GalT1 F1/ GalT1 R1 primer pair (Table S3) was used to amplify the CTS region, and the resulting GALT GalT fragment was recloned in binary vector carrying an expression cassette for glyphosate tolerance (Dr. Koen Wetering, Bayer CropScience). Finally, DXTFT plants were stably transformed with GALT GalT construct to generate the DXTFT GAL expression platform.

Cloning of reporter glycoproteins used in this investigation
We expressed five different reporter recombinant glycoproteins in N. benthamiana. (TMV)-MagnICONâ-assembled viral vector system was used to express TF (Castilho et al., 2011b), EpoFc  and A1AT . Two noncompetitive virus vectors, TMV-and potato virus X (PVX)-based, were used to express the heavy and light chains of Cx-IgG . In addition, we recloned an Fc fragment with engineered HER2/neu-binding sites (Fcab-HER2; Jez et al., 2012) into the PVX-based magnICONâ-assembled vector. For that, we amplified the cDNA sequence, with flanking BsaI restriction sites, out of the previous vector using the primer pair Fcab F1/ Fcab R1 (Table S3), digested with BsaI and inserted it into the BsaI cloning site of the magnICONâ-assembled vector pICHa31150 (Marillonnet et al., 2005).
The full length of NbBGAL1 was commercially synthesized and previously cloned (Buscaill et al., 2019). Here, the NbBGAL1 cDNA (73-2541 nucleotides) was amplified with flanking BsaI restriction sites using the primer pairs BGAL F1/BGAL R1 (cDNA1) and BGAL1 F2/ BGAL1 R1 (cDNA2) that includes a C-terminal sequence for a StrepII-tag (WSHPQFEK) ( Table S3). After BsaI digestion, the cDNA1 was inserted into the BsaI sites of pICHb26211 and cDNA2 in the BsaI cloning site of pICHa26211, containing the barley a-amylase signal peptide (Schneider et al., The resulting viral-based vectors were termed SPb-BGAL1 and SPa-BGAL1, respectively ( Figure S1). To avoid virus competition and allow for co-expression of NbBGAL1 with reporter proteins already cloned in pICH26211a, we cloned NbBGAL1 into pPT2 vector (Strasser et al., 2007a). cDNA sequence from SPa-BGAL1 was amplified out of the viral-based vectors using the primer pair aSP F1/BGAL1 R3 (Table S3), digested with XbaI/ BamHI and ligated into pPT2 digested the same way ( Figure S1). Additionally, the cDNA corresponding to the GH35 domain of the NbBGAL1 (73-1080 nucleotides) was amplified with flanking BsaI restriction sites using BGAL1 F2/ BGAL1 R2 primers (Table S3). After BsaI digestion, the fragment was inserted into the BsaI cloning site of pICHa26211 (SPa-BGAL1-GH35) ( Figure S1).

RNAiBGAL1 cloning
A RNA interference (RNAi) binary vector to knockdown the expression of NbBGAL1 was generated in three cloning steps. First, the intron 2 of A. thaliana XYLT (Shin et al., 2017) was amplified with primer pair intron F1/R1 introducing flanking XbaI-XhoI and KpnI-BamHI restriction sites. After XbaI/BamHI digestion, the intron was ligated into pPT2 vector digested the same way. This resulted in a RNAi 'preliminary' binary vector (pRNAi). Next, a 290-bp fragment corresponding to the coding sequence for amino acids 27-111 of NbBGAL1 was amplified using BGAL1 F3/ BGAL1 R4 primer pair creating XbaI-BamHI and XhoI-KpnI flanking restriction sites (Table S3). The PCR product digested BamHI/KpnI was cloned in KpnI/BamHI site of pRNAi generating an intron-antisense construct. Finally for the 'sense' fragment, PCR product was digested with XbaI/XhoI and inserted into XbaI/XhoI sites of the intron-antisense construct to generate a sense-intron-antisense hairpin vector (RNAiB-GAL1; Figure S1).

Transient expression, apoplastic fluid (AF) collection and protein purification
Recombinant proteins were transiently expressed in N. benthamiana leaves by agroinfiltration as described previously (Loos and Castilho, 2015). Agrobacteria containing magnICONâ vectors were infiltrated at an optical density at 600 nm (OD 600 ) of 0.1, while binary constructs were mixed at an OD 600 of 0.05.
Isolation of secreted protein from the apoplastic fluid (AF) and total soluble protein (TSP) extraction were as before (Schneider et al., 2015). Proteins were purified in small scale out of TSP by immunoaffinity with protein A as previously reported (Dicker et al., 2015).

SDS-PAGE and immunoblotting
Proteins fractionated by 12% SDS-PAGE under reducing conditions were either stained with Coomassie Brilliant Blue R-250 or transferred to Hybond enhanced chemiluminescence nitrocellulose membranes (GE Healthcare) to be analysed by immunoblotting using specific antibodies/lectins.
Clarity TM Western enhanced chemiluminescence reagents (BIO-RAD Laboratories, Inc., Hercules, CA, USA) were used as substrates. Finally, membranes blots were stained with Ponceau S (Sigma Aldrich, St. Louis, MO, USA) to visualize transferred proteins.

Confocal imaging of fluorescent protein fusions
Leaves of 4-to 5-week-old N. benthamiana plants were infiltrated with agrobacterium suspensions carrying binary plant expression vectors for mRFP-tagged proteins together with a plasma membrane marker EGFP-LTI6b (PM-GFP; Strasser et al., 2007b).
High-resolution images were acquired 2 days post-infiltration (dpi) on an upright Leica SP5 II confocal microscope using the Leica LAS AF software system. mRFP was excited with 561-nm laser and detected at 600-630 nm. Post-acquisition image processing was performed in Adobe Photoshop CS5.
In vitro b-galactosidase activity assay and NbBGAL1 kinetics Fluorescein di-b-D-galactopyranoside (FDG) was first dissolved in ethanol + DMSO (1:1) and stored as 2 mM stock solution in DMSO + ethanol+water (1:1:8). Eighty microlitres of AF isolated from N. benthamiana leaves was incubated for one hour at 25°C with 0.2 µM FDG and 50 mM MES buffer with pH 5 in the total reaction volume of 100 µL. The fluorescence of fluorescein, a product of FDG hydrolysis by b-galactosidase, was measured after one hour using a 96-well plate reader (1420 multilabel counter Victor 2 TM Wallac Oy) with an excitation wavelength of 485 nm and emission wavelength of 535 nm. A b-galactosidase purified from Aspergillus oryzae with a known concentration (276 U/mL) was used as a reference to quantify b-galactosidase activity in 1 mL of AF.
The effect of pH on the activity of NbBGAL1 was determined by FDG assay, where AF-derived SPa-BGAL1 was incubated with 0.2 µM FDG at 25°C for one hour in 50 mM MES buffer with pH ranging from 2.0 to 9.0. Similarly, the effect of metal ions on the activity of NbBGAL1 was examined by FDG assay by incubating the AF-derived SPa-BGAL1 in 50 mM MES buffer supplemented with 5 mM of FeCl 2 , CaCl 2 , CuSO 4 and EDTA.
NbBGAL1 in vitro activity was also assayed by incubating reporter proteins carrying terminal galactosylated glycans (IgG, A1AT, TF and Fcab) overnight at 37°C with AF from N. benthamiana and with AF from plants overexpressing SPa-BGAL1.
The effect of temperature on the activity of NbBGAL1 was examined by incubating galactosylated A1AT with AF-derived SPa-BGAL1 at different temperatures (25, 37, 50, 60 and 95°C). Fully galactosylated reporter proteins were generated by digestion of sialic acid residues with neuraminidase (New England Biolabs) according to the manufacturer's instructions. Commercially available human TF and A1AT are natural sialylated, and sialylated Fcab-

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.