Recombinant protein biopharmaceuticals are highly effective and specific, and therefore essential in the area of healthcare. The advancement of biotechnology made their production feasible, their share in the market has grown steadily in the last decades and is predicted to keep growing (Facts and Figures 2021: The Pharmaceutical Industry and Global Health; Walsh, 2018). The production of high-quality therapeutic proteins is still a complex process. For this the biosynthesis machinery from cells is required, and the choice of the production platform is highly associated with the product´s requirements and quality (Tripathi and Shrivastava, 2019). Proteins are frequently post-translationally modified. Particularly, protein N-glycosylation, a very common post-translational modification (PTM) in most eukaryotes, is of great importance as most protein biopharmaceuticals need a correct glycosylation to achieve the desired therapeutic efficacy (Solá and Griebenow, 2010) and to prevent immunogenic effects by the pharmaceutical (Zhou and Qiu, 2019). Mammalian (esp. Chinese Hamster Ovary (CHO)) cell lines, have dominated the recombinant biologics industry since the 1990s, largely because their PTMs resemble human ones (Walsh, 2018; Tripathi and Shrivastava, 2019). However, high production costs of these systems and the increasing demand of newly designed protein therapeutics, driven by the growing knowledge of molecular mechanisms of diseases, reveal the need for alternative platforms for tailored production. The current COVID-19 pandemic highlights particularly the urgent need to expand the production capacities for vaccines, diagnostic reagents and therapeutical proteins, such as neutralizing antibodies. Plant-based production of biopharmaceuticals offers an interesting alternative. For this, plants combine several advantageous properties like their ability to produce, fold and post-translationally modify complex proteins, a high range of scalability combined with cost-effective cultivation and the lack of human pathogens which provides inherent safe products (Buyel, 2019). Currently, one plant-produced recombinant therapeutic is on the market (Elelyso^®, a β-glucocerebrosidase for the treatment of Morbus Gaucher, Grabowski et al., 2014) and many promising plant-made biopharmaceuticals are in clinical trials. Among them are the HIV-neutralizing human monoclonal antibody 2G12 produced in Nicotiana tabacum (Ma et al., 2015), the Nicotiana benthamiana-derived virus-like particles as candidate vaccines against influenza, dengue fever or COVID-19, respectively (Ward et al., 2020, 2021; Ponndorf et al., 2021) or α-galactosidase for enzyme replacement therapy in Morbus Fabry treatment produced in the moss Physcomitrella (Shen et al., 2016; Hennermann et al., 2019). Physcomitrella provides several beneficial features for biopharmaceutical production (reviewed in Decker and Reski, 2020): its high rate of somatic homologous recombination enables easy genome engineering (e.g., Strepp et al., 1998; Wiedemann et al., 2018), it is able to produce complex recombinant human proteins and multi-level synthetic complement regulators (Reski et al., 2015; Michelfelder et al., 2017; Top et al., 2019; Ruiz-Molina et al., 2021), can be cultivated under Good Manufacturing Practice (GMP) conditions in suspensions with volumes up to 500 L in photobioreactors (Reski et al., 2018). The promising biopharmaceutical candidates mentioned above demonstrate the potential of plant-based systems in this field. All the plant-derived biopharmaceuticals approved or in advanced clinical trials have in common, that their efficacy is not impaired by the lack of mammalian-typical N-glycosylation patterns, which differ from those produced in plants. The early processing of N-glycans in plants and mammals is conserved, while their maturation in the Golgi apparatus differs (Gomord et al., 2010). Plant and human N-glycans share the identical heptasaccharide GlcNAc₂Man₃GlcNAc₂ (GnGn, Figure 1) di-antennary complex-type core structure, while fucosylation of the Asn-linked N-acetylglucosamine (GlcNAc) is α1,3-linked in plants and α1,6-linked in humans. In humans though not in plants, the GnGn core is extended via β1,4-linked galactose, which is often terminally capped with α2,6-linked sialic acid. In plants, the GnGn core is substituted with a β1,2-linked xylose, a sugar not produced in humans, and it is terminally extended by β1,3-linked galactose and α1,4-linked fucose, both linked to the outer GlcNAc residues, forming the trisaccharidic Lewis A (Le^a) epitope. This epitope as well as the plant-specific β1,2-attached xylose and the α1,3-attached fucose have been associated with antibody formation in humans (Fitchette et al., 1999; Wilson et al., 2001). Antibodies recognizing a therapeutic protein can affect its efficacy by altering the pharmacokinetics and pharmacodynamics, and represent an additional safety risk (Tourdot and Hickling, 2019). Therefore, to avoid potential immunogenicity of plant-made therapeutical proteins, plant-specific N-glycan residues have already been tackled. Plant-specific N-glycan xylosylation and fucosylation were eliminated in several plant-based systems by knockout (KO) or downregulation of the genes encoding the respective xylosyltransferases (XT) and fucosyltransferases (FT) (Koprivova et al., 2004; Strasser et al., 2004, 2008; Cox et al., 2006; Sourrouille et al., 2008; Shin et al., 2011; Hanania et al., 2017; Mercx et al., 2017; Jansing et al., 2018). Additionally, Le^a epitope formation was abolished in Physcomitrella by knockout of the β1,3-galactosyltransferase 1 (GalT1) encoding gene (Parsons et al., 2012). The triple KO of xt, ft and galt1 in Physcomitrella resulted in an outstanding N-glycan homogeneity, with a strongly predominant GnGn glycosylation pattern (Parsons et al., 2012). This provides a suitable platform for the further glyco-optimization, comprising β1,4-galactosylation and sialylation.

The impact of terminal N-glycan residues on efficacy and functional role of protein therapeutics has been extensively reviewed (Jefferis, 2009; Li and d’Anjou, 2009; Tan et al., 2018). Terminal N-glycan sialylation increases the protein surface charge and hides the underlying sugars galactose, GlcNAc and mannose. Renal filtration and elimination rates are retarded for highly charged proteins (Solá and Griebenow, 2010). Additionally, liver asialoglycoprotein receptors recognizing terminal galactose, as well as mannose receptors, mainly on immune cells, recognizing terminal mannose or GlcNAc, are responsible for a rapid clearance of non-sialylated glycoproteins from serum (Datta-Mannan, 2019).

To reach N-glycan sialylation, which has already been stably attained in N. benthamiana and Physcomitrella (Kallolimath et al., 2016; Bohlender et al., 2020), the galactosylated N-glycan acceptor should be provided as a first step. In planta N-glycan β1,4-galactosylation has been achieved via expression of heterologous coding sequences (CDSs) of different versions of β1,4-galactosyltransferases (β1,4-GalT), including the sequences of various animal species along with the human one and chimeric varieties thereof (Palacpac et al., 1999; Bakker et al., 2001, 2006; Misaki et al., 2003; Huether et al., 2005; Fujiyama et al., 2007; Hesselink et al., 2014; Kittur et al., 2020; Kriechbaum et al., 2020). In these various approaches in different plant species, it has become evident that galactosylation efficiency and quality is influenced by diverse factors. Among them, localization of the enzyme within the Golgi apparatus plays an important role. When localized too early in the Golgi sub-compartments, the β1,4-GalT activity interferes with the activities of the α-mannosidase II (GMII) or the N-acetylglucosaminyltransferase II (GnTII), impeding further N-glycan maturation and leading to incompletely processed mono-antennary galactosylated N-glycans (Strasser et al., 2009; Schneider et al., 2015; Kallolimath et al., 2018). The localization of a protein anchored in the endomembrane system is dependent on the N-terminal cytoplasmic, transmembrane and stem (CTS) domain (Czlapinski and Bertozzi, 2006; Schoberer and Strasser, 2011; Welch and Munro, 2019). Accordingly, the CTS of the human β1,4-GalT, which is apparently localized in the early to medial plant Golgi apparatus was replaced by CTS sequences with an assumed late trans-Golgi localization. Chimeric variants of the β1,4-GalT with different CTS domains, like the CTS of the human sialyltransferase (Strasser et al., 2009), the CTS of the Arabidopsis β1,3-galactosyltransferase 1 (Kriechbaum et al., 2020) or the CTS of the Physcomitrella α1,4-fucosyltransferase (FTGT) (Bohlender et al., 2020) have been described and led to higher shares of di-antennary galactosylated N-glycans. Furthermore, the target glycoprotein itself influences its galactosylation efficiency (Kriechbaum et al., 2020), probably based on conformation-related accessibility.

In this study we analyzed the galactosylation efficiency of the chimeric β1,4-galactosyltransferase FTGT, which consists of the CTS domain of the moss α1,4-fucosyltransferase fused to the catalytic domain of the human β1,4-GalT (Bohlender et al., 2020). Differing from the protein employed in our previous study, here we use recombinant human erythropoietin (rhEPO). Human EPO is a highly glycosylated protein hormone which inhibits apoptosis of erythroid progenitor cells and stimulates their differentiation, increasing the number of circulating mature red blood cells (Jelkmann, 2013). Recombinant hEPO is widely used for the treatment of severe chronic anemia especially associated with chronic kidney disease and chemotherapy (Jelkmann, 2013). Additionally, non-sialylated rhEPO (asialo-rhEPO) is of pharmacological interest due to its tissue-protective activity devoid of erythropoietic activity (Peng et al., 2020).

FTGT expression led to a galactosylation efficiency of about 66% on rhEPO N-glycans, and 65% of the galactosylated fraction consisted of mature di-antennary galactosylated structures. However, up to five additional pentoses were found to be attached to about 92% of all β1,4-galactosylated N-glycans. Pentosylation on β1,4-galactosylated N-glycans was recently reported in N. tabacum and Physcomitrella (Bohlender et al., 2020; Kittur et al., 2020), indicating that this modification might affect different plant-based production systems; but so far no reports are available elucidating its identity. Here, we identified the unknown pentoses as α-linked arabinofuranoses. The arabinose identity was verified by immunoblot-based detection on rhEPO with an anti-α1,5-arabinan antibody and specific digestion of the pentoses from rhEPO with α-L-arabinofuranosidase, confirmed via immunoblot and mass spectrometry analysis.

Arabinoses are not present in humans, and therefore potentially immunogenic (Anderson et al., 1984; Steffan et al., 1995; Leonard et al., 2005). Moreover, they might interfere with the efficient establishment of in planta sialylation. In this regard, the characterization of the undesired pentosylation as α-L-arabinosylation is an indispensable step towards the identification of the responsible glycosyltransferase and thus to provide plant-based glyco-engineered biopharmaceuticals with tailored N-glycosylation patterns.

Glycosylation, a frequent and complex posttranslational modification of proteins, is a critical quality feature for glycoprotein-based therapeutics, as it influences their conformation, solubility, activity, pharmacokinetics and antigenicity (Arnold et al., 2007; Solá and Griebenow, 2010). The composition of the respective N-glycans is dictated by intrinsic characteristics of the protein itself, such as conformation, as well as by the glycan-processing enzymes of the production platform (Clausen et al., 2015; Suga et al., 2018). N-glycosylation of most biopharmaceutical production hosts, even the predominantly used mammalian cell systems, differ to their human counterparts to different extents (Wang et al., 2015). For instance, N-glycolylneuraminic acid (Neu5Gc), a sialic acid not existing in humans and consequently associated with antibody formation (Tangvoranuntakul et al., 2003; Padler-Karavani et al., 2011), can be found on N-glycans of glycoproteins produced in some non-human mammalian cell lines (Varki, 2001; Ghaderi et al., 2012). Although plant N-glycosylation differs from the human pattern, its humanization, which includes the removal of plant-specific sugar residues, the introduction of a β1,4-galactosylation capacity and the final establishment of terminal N-glycan sialylation, has been performed to varying degrees in different plant systems (reviewed in Montero-Morales and Steinkellner, 2018). These studies have demonstrated a great flexibility of plants towards glyco-engineering. Especially the moss Physcomitrella offers the additional advantages of a high rate of homologous recombination in mitotic cells, a characteristic feature used for efficient precise genome editing, and a haploid gametophytic tissue, which enables immediate implementation of glyco-modifications (Parsons et al., 2012; Decker et al., 2014; Wiedemann et al., 2018).

The β1,4-linked galactoses on N-glycans provide the anchor for sialic acid, but terminal galactose also plays an important role in non-sialylated glycoproteins. For example, asialo-EPO was proposed to be neuroprotective (Erbayraktar et al., 2003; Peng et al., 2020) and on the Fc domains of monoclonal antibodies terminal N-glycan galactosylation increases complement-dependent (Hodoniczky et al., 2005) as well as antibody-dependent cytotoxicity (Thomann et al., 2016).

In terms of de novo β1,4-galactosylation in plants, its efficiency and degree of maturation (mono- or di-antennary) depends on the expression level of the respective galactosyltransferase (Kallolimath et al., 2018), its Golgi localization, determined by the N-terminal CTS domain (Strasser et al., 2009; Hesselink et al., 2014; Kriechbaum et al., 2020), and the reporter glycoprotein itself (Schneider et al., 2015; Kriechbaum et al., 2020).

In this study, we established β1,4-galactosylation on rhEPO produced in moss devoid of plant-specific sugar residues. To target the β1,4-GalT activity to the late Golgi compartments, the catalytic domain of this enzyme was fused to the CTS domain of the moss-endogenous α1,4-fucosyltransferase, whose activity is the last known in plant N-glycan maturation (Fitchette et al., 1999; Parsons et al., 2012).

In our earlier study with sialylating moss lines (Bohlender et al., 2020), FTGT-mediated overall galactosylation was up to 89%. From that fraction up to 34% comprised di-antennary galactosylated structures. In contrast, in the current study, 66% of all glycans carried galactoses, and 65% of these were di-antennary galactosylated, indicating a medial- to trans-Golgi localization of the FTGT enzyme. These values are very promising, considering that previous studies reported lower galactosylation efficiencies with up to 20 and 12% di-antennary galactosylated rhEPO produced in N. tabacum or N. benthamiana plants, respectively (Kittur et al., 2013; Kriechbaum et al., 2020). However, the galactosylation efficiency on rhEPO produced in N. benthamiana was increased by knocking out the β-galactosidase NbBGAL1, an enzyme responsible for galactose cleavage (Kriechbaum et al., 2020). A similar strategy might be applied to moss.

Accompanying the established human-like galactosylation, we detected the attachment of pentose residues on β1,4-galactosylated N-glycans. Up to three pentoses were attached to mono-antennary and up to five pentose residues to di-antennary galactosylated N-glycans, which indicates the building of short pentose chains. These were not present in the corresponding parental line with an intact β1,3-galactosyltransferase (Parsons et al., 2012), indicating that naturally occurring β1,3-galactosylated N-glycans do not display a substrate for this modification.

In planta N-glycan pentosylation on a recombinant protein upon the establishment of β1,4-galactosylation has also been observed in N. tabacum (Kittur et al., 2020), suggesting that this phenomenon is not restricted to Physcomitrella but rather affects plant-based production in general. Pentosylation was also observed in sialylating moss lines (Bohlender et al., 2020). However, in these plants either pentoses or sialic acid could be detected on galactosylated N-glycans, indicating that the pentosylation may interfere with the full N-glycan humanization of plant-derived glycoproteins. This observation confers importance to the elucidation of the respective pentose residues.

Based on immunodetection with LM6-M, a monoclonal antibody recognizing short chains of α1,5-linked arabinan (Cornuault et al., 2017), we could identify the pentoses on moss-produced rhEPO as arabinoses. Specific digestion of these pentoses with α-L-arabinofuranosidase, an enzyme specifically cleaving α1,2-, α1,3- and α1,5-linked arabinofuranoses from arabinan molecules, was verified via immunodetection and supported by MS analysis of rhEPO glycopetides. These findings confirm the identity of the pentoses as (short chains of) α-linked arabinofuranoses. Additionally, we found the arabinoses to be occasionally methylated. Some residual pentoses after α-L-arabinofuranosidase digest may be attributed to inefficient hydrolysis of α-1,5-linked arabino-oligosaccharides by the enzymes used. The fact that these residual pentoses (structures with up to two pentoses) present in the α-L-arabinofuranosidase-treated sample were not detected by the LM6-M antibody, can be due to the very low concentration of these residual sugars, and the characteristic of LM6-M antibodies that recognize arabinose₂ chains with much less avidity than longer arabinose chains, while no information about its avidity towards single arabinose residues is available (Cornuault et al., 2017).

Recently, the presence of arabinose and methylated arabinose on N-glycans of the microalga Chlorella sorokiniana has been described (Mócsai et al., 2020) and in very rare cases methylation of N-glycans in Physcomitrella wild-type strains was detected (Stenitzer et al., 2022). However, we never observed this sugar and any methylation on N-glycans of glyco-engineered Physcomitrella strains before the establishment of human-like β1,4-galactosylation. Evidently, an arabinosyltransferase from a different biosynthetic pathway recognizes the β1,4-galactosylated N-glycan as substrate. Plants display a wide diversity of cell-wall glycans and O-glycosylated hydroxyproline-rich glycoproteins (Seifert et al., 2021). This diversity originates from the combination of different monosaccharides and various linkages, generated by a huge variety of glycosyltransferases from which a considerable amount has not been thoroughly characterized yet (Showalter and Basu, 2016; Amos and Mohnen, 2019). Some enzymes responsible for the attachment of arabinoses to β1,4-linked galactoses on O-glycosylated arabinogalactan proteins as well as in cell-wall associated structures like rhamnogalaturan I have been described, but many still remain unknown (Léonard et al., 2010; Laursen et al., 2018; Ropartz and Ralet, 2020; Petersen et al., 2021). The identification of the enzyme or enzymes responsible for the arabinosylation of galactosylated N-glycans is therefore not a straightforward task.

For the application of plant-based biopharmaceuticals, this newly appearing N-glycan attachment bears the risk of immunogenicity in patients, as arabinose is a sugar not produced in humans (Anderson et al., 1984; Steffan et al., 1995; Leonard et al., 2005). To avoid arabinose attachment, the responsible arabinosyltransferases need to be identified and abolished by gene targeting to create stable lines devoid of N-glycan arabinosylation. To this aim, our study provides the first important step by elucidating the unknown pentose residues, which helps to ensure the production of safe biopharmaceuticals in plant-based systems.