Proteins trafficking through the Golgi apparatus are subjected to processing by various glycosyltransferases. Most recombinant proteins expressed in plants acquire complex N-glycans, with GnGnXF or GnGn as the predominant forms, depending on whether the host is wild-type or glycoengineered ΔXF plants, respectively. However, the removal of terminal GlcNAc residues can lead to truncated N-glycans such as MMXF or MM, collectively referred to as paucimannosidic glycans (Figure 1, Supplementary Table S1).

The formation of these structures is primarily mediated by β-N-acetylhexosaminidases particularly HEXO1 and HEXO3 (Strasser et al., 2007a; Liebminger et al., 2011). Among them, HEXO3, localized at the plasma membrane, has been identified as the main enzyme responsible for trimming terminal GlcNAc and N-acetylgalactosamine (GalNAc) (Strasser et al., 2007b; Liebminger et al., 2011; Shin et al., 2017; Alvisi et al., 2021). In contrast, HEXO2 exhibits strict β-galactosaminidase activity (Alvisi et al., 2021).

Recombinant proteins expressed in N. benthamiana exhibit varying sensitivity to β-hexosaminidase activity. For instance, in full IgG1 antibodies and Fc fragments, the N-glycosylation site buried within the CH2 domain is largely protected from enzymatic trimming. Conversely, glycoproteins such as bovine follicle-stimulating hormone (Dirnberger et al., 2001), human α1-antitrypsin (A1AT) (Castilho et al., 2014), human secreted alkaline phosphatase (SEAP) (Becerra-Arteaga and Shuler, 2007), and the helminth glycoprotein Omega-1 (Wilbers et al., 2017) have shown a high proportion of paucimannosidic glycans.

These truncated glycans are often considered undesirable in therapeutic applications due to their potential to increase immunogenicity and adverse immune responses. Consequently, several strategies have been developed to suppress their formation. These include (i) the use of Arabidopsis thaliana hexo3 knockout mutants (Castilho et al., 2014), (ii) RNA interference to silence HEXO3 (Shin et al., 2017) and (iii) CRISPR-based genome editing of β-hexosaminidase genes (Blumberg et al., 2025).

An alternative strategy involves capping terminal GlcNAc and GalNAc residues with fucose using FUT9a. This strategy was used to make glycans resistant to hexosaminidase activity and enhanced the accumulation of fucosylated LDN motifs in helminth glycoproteins (Wilbers et al., 2017). Glycan processing by endogenous glycosidases should in most cases be avoided, but sometimes the action of glycosidases can be beneficial to fine-tune a specific glycan structure. Interestingly, paucimannosidic glycans can also be advantageous. Exposed mannose residues promote rapid receptor-mediated clearance from the bloodstream (Yang et al., 2015) and have shown enhanced efficacy in mannose receptor (MR)-mediated cellular uptake of biopharmaceuticals (Limkul et al., 2016; Sariyatun et al., 2021). Thus, for certain therapeutic proteins, especially those used in enzyme replacement therapy (ERT), paucimannosidic N-glycans are desirable. To enrich these glycans in plant-derived ERT proteins (e.g., human β-glucocerebrosidase, acid α-glucosidase), proteins have been targeted to the plant storage vacuole, where terminal residues are naturally removed to produce paucimannosidic glycans (Shaaltiel et al., 2007; Liebminger et al., 2011; Tekoah et al., 2013). Additionally, glycoengineered A. thaliana mutants lacking ALG3 (an α1,3-mannosyltransferase involved in early N-glycan precursor assembly) have been used to produce recombinant human acid α-glucosidase with predominantly MM-type glycans (Sariyatun et al., 2021).

A landmark in the field is taliglucerase alfa, a β-glucocerebrosidase produced in carrot cell suspension cultures by Protalix Biotherapeutics (Carmiel, Israel). ELELYSO® was the first plant-made pharmaceutical approved by the U.S. Food and Drug Administration (FDA) for human use.

As noted, sensitivity to β-hexosaminidase trimming is highly correlated with the accessibility of glycosylation sites. While glycosylation sites on proteins such as A1AT are fully accessible to β-hexosaminidase activity, others like the Asn297 buried in the CH2 domain of antibodies are structurally shielded from enzymatic access. Interestingly, the presence of core α1,3-fucose appears to enhance the exposure of certain glycans to β-hexosaminidases. Co-expression of FUT11 significantly increases the formation of paucimannosidic glycans in Fcabs (MMF, Supplementary Table S1), but not in full antibodies. This observation suggests that structural rearrangements in Fcabs may expose the glycosylation site, rendering it more susceptible to glycan trimming, particularly when core fucosylation alters glycan conformation and enhances accessibility to glycosidases. Of note, enhancing the synthesis of paucimannosidic N-glycans by core fucosylation is not observed when the α1,6-fucosyltransferase is used (Shin et al., 2017).

Man5 and hybrid N-glycans play crucial roles in both normal physiology and disease. Man5 structures are central to glycoprotein quality control, structural integrity, and therapeutic relevance, while hybrid N-glycans contribute to neuronal function, immune modulation, and protein stability. Alterations in either glycoform are linked to pathological processes, underscoring their biological and biomedical importance (Hall et al., 2016).

To promote the synthesis of high-mannose-type N-glycans (Man7-9), which resemble those found on many helminth glycoproteins and are desirable for the production of recombinant vaccine antigens in plants, strategies such as ER retention of the glycoprotein or treatment with kifunensine, have been successfully employed (Roychowdhury et al., 2018; Izadi et al., 2023a). Similarly, inhibition of GMII using swainsonine has been used to generate Man5 glycans (Choi et al., 2018), which can also be produced by expressing glycoproteins in GNTI-deficient plants, as previously discussed. However, precise engineering of plant systems to produce defined mannosidic or hybrid glycans remains challenging, as current glycoengineering strategies still require further optimization. One possible strategy to control N-glycan processing could be the use of chimeric proteins carrying specific CTS regions that interfere with glycan maturation at a defined step. Mislocalization or interaction of exogenous glycosyltransferases with endogenous enzymes via their stem regions can modulate N-glycan processing by limiting downstream enzyme access. For instance, GNTI, a key enzyme in the cis/medial-Golgi, initiates hybrid N-glycan formation by transferring a GlcNAc residue to Man5 (Figure 1). This terminal GlcNAc is essential for subsequent processing by medial- and trans-Golgi glycosyltransferases to form complex N-glycans. Interestingly, overexpression of the CTS region from GNTI and MNS1 fused to mRFP was shown to disrupt normal glycan processing, resulting in the accumulation of Man₅ structures on recombinant glycoproteins that would otherwise carry complex glycans (GnGnXF) (Schoberer et al., 2023). This pioneering work demonstrated that transient expression of specific CTS domains can redirect glycosylation from complex to mannosidic forms. A similar approach could be employed to selectively block further processing in the medial-Golgi, facilitating the production of hybrid or mono-antennary N-glycans by preventing recognition by downstream glycosyltransferases (Figure 4). Supporting this concept, targeting of B4GALT1 to the medial-Golgi (^XYLTB4GALT1 and ^FUT11B4GALT1) resulted in the formation of hybrid (Man₄A and Man₅A) and mono-antennary (MA) β1,4-galactosylated N-glycans (Supplementary Table S1) (Dicker et al., 2015). Thus, the overexpression of specific CTS regions fused to an inactive catalytic domain could promote the synthesis of unprocessed glycans (M5Gn, Man4Gn and MGn, Figure 4).

In plants, the only galactosylated native N-glycan structure is the Lewis A (Le^a) motif. These galactosylated glycans result from the activity of GALT1 that transfers galactose in β1,3-linkage to terminal GlcNAc residues, followed by the action of α1,4-fucosyltransferase (FUT13), which adds fucose in an α1,4-linkage to GlcNAc, thereby completing the synthesis of the Le^a epitope (Strasser et al., 2007a) (Figure 1, Supplementary Table S1). Le^a‐containing N‐glycans are present only on a limited number of, mostly unidentified, glycoproteins in plants, and their biological significance remains unclear (Wilson, 2001; Beihammer et al., 2021). Although rare on endogenous plant proteins, Le^a structures are frequently found on complex N-glycans of some plant-produced recombinant glycoproteins, such as human erythropoietin (EPO), SARS-CoV-2 spike protein, and hepatitis B virus (HBV) antigen, even when absent on the native versions of these proteins (Weise et al., 2007; Parsons et al., 2012; Jez et al., 2013; Balieu et al., 2022; Pantazica et al., 2023). Notably, overexpression of GALT1 significantly increases the abundance of Le^a-containing N-glycans on a monoclonal antibody, without the need to overexpress FUT13 (Nguyen et al., 2023).

Le^a determinants are sometimes referred to as human-type glycosylation, since both Le^a and sialyl-Le^a epitopes are significantly elevated in the sera of patients with colon cancer (Terracciano et al., 2005). However, they are rarely present on glycoproteins in healthy adult humans. To eliminate Le^a formation in plants, the synthesis pathway has been successfully disrupted in Physcomitrella patens by knocking out the α1,4-fucosyltransferase and β1,3-galactosyltransferase genes (Parsons et al., 2012).

In contrast to plant β1,3-galactosylated N-glycans, mammalian and helminth glycoproteins carry N-glycans extended with antennary β1,4-galactose residues. This widespread glycan modification can significantly influence the efficacy of therapeutic glycoproteins. For instance, Fc galactosylation has been shown to affect the conformation of IgG1 and its binding to Fc receptors (Houde et al., 2010; Raju and Jordan, 2012). Moreover, β1,4-galactosylated N-glycans serve as essential acceptor substrates for terminal sialylation, a modification known to enhance glycoprotein stability, prolong serum half-life, and improve therapeutic performance (Chia et al., 2023). However, the efficient synthesis of bi-antennary β1,4-galactosylated N-glycans (AA) varies depending on the target protein (Kriechbaum et al., 2020). In fact, except for monoclonal antibodies, achieving homogeneous human-like β1,4-galactosylation in recombinant glycoproteins remains challenging. Efforts to fine tuning the expression of B4GALT1 included CTS swapping to target the enzyme to the late Golgi (Table 3) and optimizing expression levels through selection of weaker promoters, as discussed above. These approaches have enabled improved synthesis of sialylated N-glycans as well as Lewis X (Le^X) and blood group structures on plant-produced glycoproteins.

The synthesis of Le^X ([AF], Supplementary Table S1) structures found on native helminth glycoproteins was achieved through co-expression of chimeric ^ST6B4GALT1 and ^ST6FUT9a (Figure 2A) (Wilbers et al., 2017). Additionally, two fucosyltransferases from the parasite Schistosoma mansoni, FucTD and FucTE, have also been shown to synthesize Le^X (van Noort et al., 2020). The predominance of monoantennary Le^X-carrying N-glycans suggests that overexpression of GNTII may be necessary to enhance glycan branching and support bi-antennary Le^X formation, as discussed above (Wilbers et al., 2017).

Apart from the common β1,4GalGlcNAc (LacNAc or LN) motif, many invertebrates, such as helminths, express glycans containing GalNAc, an amino sugar derivative of galactose, in the form of LDN motifs (Supplementary Table S1), which also occur in vertebrates and are present on several mammalian hormones.

Engineering GalNAc-carrying glycans in plants was first explored to produce mammalian mucin-type O-glycans (Daskalova et al., 2010; Yang et al., 2012; Castilho et al., 2012) by introducing the enzymatic machinery to transfer GalNAc residues to O-glycosylation sites of recombinant proteins. These studies revealed that, among the three proteins involved in the process (UDP-GlcNAc C4 epimerase, Golgi UDP-GalNAc transporter, and β1,4-N-acetylgalactosaminyltransferase), only GalNAcT was strictly necessary for GalNAc-glycan synthesis (Castilho et al., 2012; Dicker et al., 2016). This was later confirmed by the successful synthesis of LDN on the N-glycans of S. mansoni kappa-5 in N. benthamiana using only a C. elegans GalNAcT (CeGalNAcT) (Wilbers et al., 2017) (Figure 2A). This shows that, even though GalNAc is not native to plant N-glycans, UDP-GalNAc is present within the correct Golgi compartment and can be incorporated into the antennae of N-glycans of recombinant glycoproteins. Plant-derived LDN motifs can be further modified by adding an α1,3-fucose to the distal GlcNAc (via FucT9a, SmFucTD, or SmFucTE) to yield LDN-F and an additional α1,3-fucose to the terminal GalNAc (via SmFucTF) to produce F-LDN-F, resulting in mono- and bi-fucosylated LDN motifs (Figure 2A, Supplementary Table S1) (Wilbers et al., 2017; van Noort et al., 2020; van der Kaaij et al., 2022; Zwanenburg et al., 2023). Fucosylation of LDN motifs has been shown to enhance the accumulation of GalNAc-containing N-glycans, likely by increasing resistance to HEXO-mediated removal of unsubstituted GlcNAc and/or GalNAc residues (Strasser et al., 2007a).

The ABH(O) blood group antigens are specific glycan determinants on glycoproteins and glycolipids that also carry GalNAc residues, which are crucial for their antigenicity and the determination of blood types (Stowell and Stowell, 2019). In plants, synthesis of blood group A (type 1 and type 2) structures was recently achieved on the RBD of the SARS-CoV-2 spike protein expressed in N. benthamiana (Konig-Beihammer et al., 2021). These modifications were generated by co-expressing either the native A. thaliana GALT1 (type 1) or a chimeric ^ST6B4GALT1 (type 2) with a human α1,2-fucosyltransferase targeted to late Golgi (^ST6FUT1 or ^ST6FUT2) to produce H antigens (α1,2FucGalβ-R), followed by extension with α1,3-GalNAc residues via N-acetylgalactosaminyltransferase (ABO A). Interestingly, in this case co-expression of a Yersinia enterocolitica UDP-GlcNAc 4-epimerase (YeGNE), which converts UDP-GlcNAc to UDP-GalNAc, and a Caenorhabditis elegans UDP-GlcNAc/UDP-GalNAc transporter (Castilho et al., 2012) further improved the biosynthesis of blood group A type 2 trisaccharide structures (Beihammer et al., 2021) (Figure 2A, Supplementary Table S1). The efficiency of ABH antigen synthesis appears to depend on glycosite accessibility. Mono-antennary structures suggests partial removal of terminal galactose and GalNAc by endogenous apoplastic galactosidases (Kriechbaum et al., 2020), whereas terminal fucosylation protects these residues from enzymatic trimming, as also observed for fucosylated LDN motifs (Wilbers et al., 2017). Despite significant efforts to optimize terminal galactosylation, many plant-derived glycoproteins still display incompletely processed N-glycans with heterogeneous β1,4-galactosylation due to trimming of terminal galactose residues in the apoplast (Castilho et al., 2014; Wilbers et al., 2017; Bohlender et al., 2020; Kriechbaum et al., 2020). Consequently, studies assessing the impact of glycosylation on protein function or activity (other than monoclonal antibodies) rarely include β1,4-galactosylated variants (Jez et al., 2013; Schneider et al., 2014; Loos et al., 2014; Montero-Morales et al., 2019).

β-galactosidases (BGALs) exhibit specificity for β1,3-, β1,6-, or β1,4-galactosidic linkages (Ahn et al., 2007). In N. benthamiana, BGAL1 (NbBGAL1) is an active enzyme targeting both N- and O-glycans, capable of removing β1,4-linked galactose from N-glycans as well as terminal β1,3-galactose from Lewis-A epitopes and T-antigen O-glycans (Kriechbaum et al., 2020).

Recombinant proteins expressed in N. benthamiana exhibit varying sensitivity to β-galactosidases activity. While glycans on the IgG Fab fragment are fully exposed to the surrounding solvent, those in the Fc fragment are largely shielded by the opposing CH2 domain, making galactosylation at this site inherently more efficient. Interestingly, similar to HEXO3 sensitivity, the susceptibility of Fc glycans to β-galactosidase activity appears to be influenced by plant-specific core α1,3-fucosylation (Kriechbaum et al., 2020). This plant-specific modification seems to relax N-glycan or protein structural constraints, rendering terminal sugar residues more accessible to glycosidases, a phenomenon once again not reproduced by core α1,6-fucosylation (Shin et al., 2017; Kriechbaum et al., 2020).

Transient downregulation via RNA interference (RNAi) or complete knockout of NbBGAL1 significantly reduces β-galactosidase activity, leading to higher levels of fully bi-galactosylated complex N-glycans on multiple plant-produced glycoproteins (Kriechbaum et al., 2020). More recently, NbBGAL3B was identified as a second major contributor to undesired β-galactosidase activity, displaying similar specificity for β1,4-linked galactose on N-glycans (van der Kaaij et al., 2025). BGALs hydrolyze terminal β-D-galactosyl residues from diverse substrates, including polysaccharides and glycoproteins, playing key roles in cell wall structure and function (Sampedro et al., 2012; Hoang et al., 2023). Consequently, knockout of β-galactosidase genes can markedly affect plant development due to their essential biological functions. In transient glycoengineering, approaches to achieve near-complete galactosylation on glycoproteins prone to galactosidase trimming have focused on exploiting the protective effect of capping sugars. This was demonstrated by producing recombinant proteins carrying terminal fucose or sialic acid residues, followed by selective removal of the capping sugars with specific fucosidases or sialidases to expose terminal β1,4-galactose (Castilho et al., 2014; van der Kaaij et al., 2025).

De novo β1,4-galactosylation of plant N-glycans provides a substrate for further elongation (e.g., blood group antigens and sialylation), but these newly introduced β1,4-linked galactoses can also serve as anchors for other sugars, leading to the formation of aberrant N-glycans. Several studies have shown that recombinant proteins co-expressed with B4GALT1 carry β1,4-galactosylated N-glycans decorated with additional pentoses and hexoses (Bohlender et al., 2020; Kittur et al., 2020). In N. tabacum and Physcomitrella patens, multiple pentosylations on β1,4-galactosylated N-glycans were identified as α-linked arabinoses (Bohlender et al., 2022). This modification likely arises because the β1,4-linked galactose creates a non-native acceptor site for endogenous arabinosyltransferases, which normally target arabinose to cell wall polysaccharides but can act promiscuously on engineered N-glycans. Arabinoses are absent in humans and therefore potentially immunogenic (Leonard et al., 2005). In addition, they may interfere with the full N-glycan humanization of plant-derived glycoproteins. Comprehensive characterization of this aberrant N-glycosylation and identification of the responsible glycosyltransferases remain essential future steps to enable their elimination and ensure the safety of biopharmaceuticals produced in plant-based systems.

An essential step for generating complex human-type glycans is the formation of multi-antennary structures, common in human but absent in plant N-glycans. As stated above, transgenic Arabidopsis and N. benthamiana plants expressing the human GNTIV or/and GNTV enable the synthesis of proteins with tri- and tetra-antennary N-glycans (Gn[GnGn], [GnGn]Gn and [GnGn][GnGn]) (Figure 2A, Supplementary Table S1), when the enzymes are targeted to the medial-Golgi via CTS swapping (^FUT11GNTIV, ^FUT11GNTV and ^XYLTGNTIV, ^XYLTGNTV, Table 3) (Nagels et al., 2011; Nagels et al., 2012). Low abundance of multi-antennary structures on endogenous proteins may reflect HEXO activity toward β1,2-, β1,4-, and β1,6-linked GlcNAc residues (Nagels et al., 2012). Transient expression of GNTIV and V confirmed that medial-Golgi targeting resulted in more homogeneous multi-antennary glycans on recombinant proteins such as Butyrylcholinesterase (BChE), A1AT, and EPO, though efficiency depended on protein and glycosylation site accessibility (Castilho et al., 2011b; Schneider et al., 2014; Castilho et al., 2014). For example, a monoclonal antibody with an additional glycosylation site in the Fab region (such as Cetuximab), and an EPO protein fused to the Fc domain of an IgG1 (EPO-Fc), showed differing levels of accessibility to ^FUT11GNTIV, ^XYLTGNTIV and ^FUT11GNTV. While glycans on the Fab region and on EPO were successfully branched with GlcNAc residues, the N-glycosylation site on the Fc fragment remained unmodified (Castilho et al., 2011b, Castilho et al., 2015).

Addition of a bisecting GlcNAc to the β-mannose of the tri-mannosyl core via a β1,4-linkage, catalyzed by GNTIII, represents another critical branching step (Figure 2A). Unlike other branches, bisecting GlcNAc is not elongated and often inhibits further processing by GnTIV/V and other glyco-processing enzymes (Schachter, 1986). This was demonstrated in early efforts to produce bisected N-glycans by expressing full-length human GNTIII in Nicotiana tabacum and BY2 cells, which primarily resulted in the formation of hybrid bisected glycans (Cox et al., 2006; Rouwendal et al., 2007; Karg et al., 2010). Similarly, the expression of chimeric GNTIII targeted to the medial Golgi (^GMIIGNTIII, ^XYLTGNTIII and ^FUT11GNTIII) in wild-type plants led to the production of incompletely processed N-glycans lacking core fucose and xylose (Man5Gnbi, Supplementary Table S1) (Castilho et al., 2011a; Castilho et al., 2011b). In contrast, fully processed bisected N-glycans were achieved when GNTIII was targeted to the trans-Golgi (^STGNTIII, Table 3) (Castilho et al., 2011a; Castilho et al., 2011b; Castilho et al., 2015). Finally, co-expression of the sequentially targeted ^XYLTGNTV, ^FUT11GNTIV, and ^STGNTIII enabled the synthesis of branched and bisected glycans on recombinant proteins with accessible glycosylation sites, such as human transferrin and EPO (Castilho et al., 2011b). As observed with other glycoengineering strategies, accessibility of Fc glycans to branching and bisecting enzymes significantly improved with α1,3-core fucosylation (FUT11), whereas α1,6-core fucosylation (FUT8) had a weaker effect (Castilho et al., 2011a; Castilho et al., 2015).

Sialylation, the attachment of sialic acid residues to glycoproteins, is a critical modification for therapeutic proteins, enhancing their stability, half-life, and biological activity (Zhu et al., 2024). It plays a vital role in various cellular processes and interactions, making it a key focus in the development and optimization of biopharmaceuticals. Ensuring proper sialylation can significantly improve the therapeutic efficacy and pharmacokinetics of glycoprotein drugs (Jeong et al., 2008; Bork et al., 2009). The broader applications of therapeutic N-glycan engineering are driven by the goal of optimizing α2,3- and α2,6-linked sialylation of glycoproteins (Zhong et al., 2022). Among these efforts, α2,6-sialylation has garnered particular attention in glycoengineering, especially in the context of antibodies. Research indicates that the anti-inflammatory activity of antibodies is largely attributed to Fc-sialylation, highlighting the significance of this modification in enhancing therapeutic efficacy (Anthony et al., 2008).

De novo sialic acid biosynthesis occurs in the cytosol through a four-step pathway. The key enzyme UDP-GlcNAc 2-epimerase/ManNAc kinase (GNE) initiates the process by converting UDP-GlcNAc into ManNAc-6P, which is subsequently condensed with phosphoenolpyruvate (PEP) by Neu5Ac-9-phosphate synthase (NANS) to form Neu5Ac-9P. This intermediate is then dephosphorylated by Neu5Ac-9-phosphate phosphatase (NANP) to yield free N-acetylneuraminic acid (Neu5Ac). Neu5Ac is activated in the nucleus by CMP-Neu5Ac synthetase (CMAS) to form CMP-Neu5Ac, which is then transported into the Golgi lumen via the CMP–sialic acid transporter (CSAT). In the final step, α2,3- or α2,6-sialyltransferases (ST3GAL4 and ST6GAL1) transfer Neu5Ac from CMP-Neu5Ac to β1,4-galactosylated acceptor substrates in the trans-Golgi, thereby completing the synthesis of complex-type mammalian N-glycans (Zhu et al., 2024).

Since plants naturally lack this pathway and its corresponding enzymes, engineering sialylation requires extensive metabolic and subcellular modifications to enable the formation of human-type sialylated glycans. Fine-tuning of the plant sialylation pathway commonly includes the use of a truncated CMAS variant lacking the 40 N-terminal amino acids, which destabilizes the enzyme, and a feedback-insensitive GNE mutant (GNE^R263L) to circumvent CMP-Neu5Ac–mediated negative feedback regulation (Castilho et al., 2010; Kallolimath et al., 2016; Bohlender et al., 2020). Interestingly, stable or transient co-expression of GNE, NANS, and CMAS in plants results in the accumulation of Neu5Ac rather than Neu5Ac-9-P, indicating that an endogenous plant phosphatase with NANP-like activity catalyzes the dephosphorylation step (Castilho et al., 2008; Castilho et al., 2010). However, co-expression of NANP can further increase Neu5Ac production (Bohlender et al., 2020).

In the past years several approaches have been performed to establish sialylation in plants. In N. benthamiana functional N-glycan sialylation could be achieved by MUTE of the sialylation pathway-genes (Izadi et al., 2023a). This resulted in an efficient sialylation (up to 90%) of several recombinant proteins produced in plants (Castilho et al., 2010; Castilho et al., 2014; Jez et al., 2013; Schneider et al., 2014; Kallolimath et al., 2016; Goritzer et al., 2019; Montero-Morales et al., 2019; Kallolimath et al., 2020; Izadi et al., 2025).

For proteins with exposed glycosylation sites, terminal sialic acid protects against trimming by BGAL1 and HEXO3. Accordingly, full galactosylation on these sites can be achieved by sialidase treatment (Kriechbaum et al., 2020). In contrast, the impact of core α1,3-fucosylation on sialylation of glycosylation sites with difference exposure has been demonstrated with cetuximab, which carries N-glycans in both the Fab and CH2 domains. While glycan processing in the Fab domain is unaffected by fucosylation, the presence of plant-specific α1,3-core fucose markedly increases the levels of bi-antennary sialylated Fc glycans (Castilho et al., 2015; Kallolimath et al., 2016; Kallolimath et al., 2020; Eidenberger et al., 2022), an effect not reproduced by the over-expression of mammalian α1,6-core fucosylation (FUT8) where mainly monosialylated N-glycans are detected (ANaF, Supplementary Table S1) (Castilho et al., 2015).

Among the genes required to establish sialylation in plants, the correct subcellular localization of B4GALT1 and of ST3GAL4 or ST6GAL1 represents a major bottleneck. In moss, co-expression of B4GALT1 and ST6GAL1 targeted to the late Golgi by CTS swapping with FUT13 (FT4) led to the synthesis of mono-antennary sialylated glycans (MNa, Supplementary Table S1) (Bohlender et al., 2020).

Specific glycosylation patterns, including the degree of branching and sialylation, can modulate the biological activity of glycoproteins and enhance their pharmacokinetic properties (Teare et al., 2019). Overexpression of GNTIV and GNTV, increases the glycan antennary structure and potentially the number of terminal GlcNAc residues available for sialylation. For example, the introduction of tri-antennary sialylated glycans on EPO significantly enhanced its hematopoietic activity in vivo (Yin et al., 2015). In plants, the combined expression of GNTIV and GNTV with genes required for protein α2,6-sialylation enabled the production of tri- and tetra-antennary sialylated N-glycans on several recombinant proteins (Figure 2A, Table 3) (Castilho et al., 2012; Castilho et al., 2013; Castilho et al., 2014; Schneider et al., 2014).

Glycans are typically mono-sialylated, but sialic acids can also form homo-polymers of α2,8- or α2,9-linked residues known as polysialic acid (PSA) (Colley et al., 2014). Among the few naturally polysialylated proteins identified, the neural cell adhesion molecule (NCAM) is the best studied, carrying α2,8-linked NeuAc PSA on two of its N-glycans (Close et al., 2003). The addition of PSA to proteins, termed polysialylation (polySia), is a complex post-translational modification that can improve the pharmacokinetic behavior of therapeutic proteins while reducing their immunogenicity. Unlike synthetic polymers such as PEG, PSA is biodegradable and non-immunogenic, making it a safer alternative for therapeutic protein modification (Zhang et al., 2023; Chia et al., 2023).

Recent advances in biosynthetic engineering have enabled the polysialylation of recombinant proteins in both bacteria (Keys et al., 2017) and plants (Kallolimath et al., 2016). In bacterial systems, a biosynthetic pathway for site-specific polysialylation was established in Escherichia coli, in which a cytoplasmic polypeptide-glycosyltransferase installs a defined primer onto the target protein, followed by elongation with bacterial glycosyltransferases to generate long-chain polySia. This strategy has been successfully applied to modify green fluorescent proteins and therapeutic DARPins (designed ankyrin repeat proteins) (Keys et al., 2017).

In N. benthamiana, transient co-expression of two human polysialyltransferases (polySTs: ST8SiaII and ST8SiaIV) together with genes required for α2,6-sialylation demonstrated that the synthesis of biologically active α2,8-linked PSA is feasible (Figure 2A) (Kallolimath et al., 2016). Bacterial polysialyltransferases are generally more promiscuous than their mammalian counterparts (Willis et al., 2008; Lindhout et al., 2011), and engineered bacterial variants capable of generating oligo- and polysialic acids of defined length represent promising candidates to further enhance in planta synthesis of site-specific polySia on diverse recombinant glycoproteins.

Although substantial progress has been made in plants and other heterologous systems, several biochemical and engineering challenges remain. These include achieving precise control over polymer length, ensuring efficient transfer of defined polySia to target proteins, and optimizing downstream purification processes to handle highly hydrophilic polysialylated products. Continued development in biosynthetic pathway engineering, enzyme design, and analytical methods will be critical to fully realize the potential of polysialylation in plant-based expression platforms.