Modulating antibody effector functions by Fc glycoengineering

Abstract

Antibody based drugs, including IgG monoclonal antibodies, are an expanding class of therapeutics widely employed to treat cancer, autoimmune and infectious diseases. IgG antibodies have a conserved N-glycosylation site at Asn297 that bears complex type N-glycans which, along with other less conserved N- and O-glycosylation sites, fine-tune effector functions, complement activation, and half-life of antibodies. Fucosylation, galactosylation, sialylation, bisection and mannosylation all generate glycoforms that interact in a specific manner with different cellular antibody receptors and are linked to a distinct functional profile. Antibodies, including those employed in clinical settings, are generated with a mixture of glycoforms attached to them, which has an impact on their efficacy, stability and effector functions. It is therefore of great interest to produce antibodies containing only tailored glycoforms with specific effects associated with them. To this end, several antibody engineering strategies have been developed, including the usage of engineered mammalian cell lines, in vitro and in vivo glycoengineering.

1. Function and impact of IgG glycosylation in effector functions

Antibodies, also known as immunoglobulins (Ig), are important mediators of homeostasis and host defense. The biological activity of antibodies is driven by two distinct biological processes. First, antibodies can recognize certain antigens with great affinity and specificity. Second, they are able to interact with a variety of receptors expressed by innate immune system cells to respond to infecting agents (Hogarth and Pietersz, 2012). By doing so, antibodies aid in pathogen identification and internalization and increase or reduce cellular activation and the subsequent immunological effector functions (Gunn and Bai, 2021). In addition, antibodies have become one of the most potent and successful pharmacological classes for treating a number of human diseases, including various cancers, autoimmune and infectious diseases (Stanley et al., 2015). There are many different types of antibody-based treatments, such as monoclonal antibodies (mAbs), bispecific antibodies, antibody-drug conjugates (ADCs), and antibody fragments, being mAbs the most widely employed in clinical settings (Golay et al., 2022; Wang et al., 2018). There are currently about 100 mAbs approved by the FDA, and represent the class of therapeutics with the fastest rate of market expansion (Mullard, 2021).

1.1. Antibody classes and subclasses: similarities and differences

All monomeric antibodies are comprised of four peptide chains: two light chains and two heavy chains. The structure of antibodies is functionally classified into two regions: the fragment antigen-binding (Fab) region and the fragment crystallizable (Fc) region. The Fab region is implicated in antigen recognition. The Fc region interacts with Fc receptors and other effector proteins, notably complement components, to mediate effector functions. While the Fab region is composed of one variable and one constant domain (CH1 from each of the heavy and light chains, the Fc domain is a homodimer that compromises two constant domains (CH2 and CH3) from each heavy chain. The Fab and Fc domains are connected by a flexible hinge region.

In humans, antibodies are divided into five classes mainly based on constant region structure, oligomerization state and glycosylation profiles: IgG, IgA, IgM, IgE, and IgD (Figure 1). IgG is the most prevalent immunoglobulin (approximately 75%) and the most abundant glycoprotein in human serum (around 10 mg mL⁻¹) (Arnold et al., 2007; Schroeder and Cavacini, 2010). According to how common each subclass of IgG antibodies is in serum, they are numbered as IgG1, IgG2, IgG3 and IgG4 (Nimmerjahn and Ravetch, 2021, 2005). The sequence similarity among these four IgG subclasses is greater than 90%, yet each subclass has a unique functional profile with variable effector functions, complement activation and half-life (Vidarsson et al., 2014).

Figure 1. Schematic representation of human immunoglobulin subclasses.

Glycosylation sites (human numbering) are indicated. N-glycosylation sites occupied by CT and HM type glycans are shown with the structures of N-glycans G2S2 and Man6, respectively. O-glycosylation sites are shown with the structure of core 1 O-glycans. IgA2 is shown as a dimer, in complex with the joining chain (J-chain) and the secretory component (SC). IgA1 is shown in its monomeric form, but can also for dimers, whereas IgM forms pentamers. IgE has an additional N-glycosylation site at N383 which is unoccupied (Plomp et al., 2013). IgM has been reported to have a 30–40% occupancy at N563 (Chandler et al., 2019). N207 glycosylation site of IgA2 is only present in the IgA (n) and IgAm (2) allotypes, but not in the IgA2m (1) allotype (Plomp et al., 2018).

1.2. IgG effector functions

IgG is crucial for adaptive immunity, since it helps the body to rapidly produce antibodies in response to the presence of exogenous antigens. IgG promotes a variety of effector functions, such as the onset of antibody-dependent cell cytotoxicity (ADCC), complement-dependent cytotoxicity (CDC), antibody-dependent cellular phagocytosis (ADCP), and neutrophil activation, via Fc contact with various target molecules and receptors (Jefferis, 2009; Majewska et al., 2020). Antibodies are not only involved in defending the host against infection, but can also be the cause of unfavorable reactions including allergic reaction, autoimmune disease and inflammatory disease (Gunn and Bai, 2021; Vidarsson et al., 2014). Glycosylation, both N-linked and O-linked, can significantly modify the effector functions of each antibody type, playing a role in the maintenance of their structure, fine-tuning of effector functions, activation of co-receptors and co-factors, enhancement of their stability and alteration of antigenicity (Yamaguchi et al., 2006). One important aspect in which the Fc N-glycans play a role is receptor binding. Many IgG functions are mediated through Fc gamma receptors (FcγRs), which are classified into type I receptors (FcγRI, FcγRIIa, FcγRIIb, FcγRIIIa, FcγRIIIb) and type II receptors (DC-SIGN, CD23) and control downstream activating or inhibiting signaling cascades (Thulin and Wang, 2018). Diverse glycoforms attached to the Fc region of antibodies, as well as different IgG subclasses, interact differently with each receptor subtype, favoring the pro- or anti-inflammatory pathways controlled by a subset of FcγRs (Wang and Ravetch, 2019). All N-glycans share a common sugar core, composed by the sequence of sugars Manα1–6 (Manα1–3)Manβ1–4GlcNAcβ1–4GlcNAcβ1. The composition and number of antennae that decorate this core are, however, very diverse (Figure 2). In eukaryotes, N-glycans are classified in three types depending on the composition of their α(1,3) and α(1,6) antennae, namely high-mannose (HM) type, complex type (CT), and hybrid (Hy) type (Stanley et al., 2015). HM type glycans have between five to nine Man resides attached to the chitobiose core of the glycan, with both antennae fully composed of mannose (Man) sugars. CT glycans have N-acetylglucosamine (GlcNAc) residues attached both to α(1,3) and α(1,6) antennae, and might have an additional galactose (Gal) and N-5-acetlyneuraminic acid (Neu5Ac) residues attached to either or both antennae. They exist in bi-, tri- and tetraantennary forms, and can have a bisecting GlcNAc at their mannosyl core or a fucosylation on the first GlcNAc of the core. Hy type glycans have a dual nature, with an α(1,3) antenna that resembles that of CT glycans and an α(1,6) antenna equivalent to HM type glycans.

Figure 2. IgG N297 glycosylation and effector functions.

Cartoon representation of the overall structure of a human IgG1 antibody (PDB code 1HZH). The glycoforms that can be attached to N297 glycosylation site are shown. The dotted shapes indicate that there may or may not be a carbohydrate present, highlighting the considerable structural diversity of the N-glycans. The influence on antibody effector functions associated with the presence of each carbohydrate moiety is indicated (ADCC: antibody-dependent cellular cytotoxicity; ADCP: antibody-dependent cellular phagocytosis). The percentage of IgG containing either Fuc or bisecting GlcNAc is shown. The presence of either Fuc or bisecting GlcNAc usually involves the absence of the other carbohydrate.

1.3. IgG glycosylation: Asn297

An N-linked glycosylation site is conserved in the Cγ2 domain of the Fc region of each IgG subclass, located at Asn 297 (N297). The glycan occupying this location mostly has a CT biantennary structure. According to structural studies, Fc N-glycans are essential for preserving an open conformation of the Fc domain which facilitates interactions with FcγRs and C1q (Subedi and Barb, 2015). The elimination of this glycosylation site disrupts binding to C1q and FcγRs as well as antibody effector functions (Tao and Morrison, 1989). IgG Fc N-glycans are highly variable, with more than 37 different glycoforms being found on IgG in healthy individuals (Pucić et al., 2011; Stadlmann et al., 2010). This variety in composition includes core-fucosylation, galactosylation, sialylation, presence of bisecting GlcNAc and mannosylation (Reily et al., 2019). Despite the great level of conservation of the Fc glycans across IgG subclasses, the impact of IgG1 N-glycosylation on FcγR binding affinity has been the focus of most research (Van Coillie et al., 2022). IgG Fc glycan composition has been shown to have a major impact on effector functions, immunoreactivity, antigen affinity, pharmacokinetics, stability, and aggregation (H. Liu et al., 2017; Stadlmann et al., 2010). Disease, infection and different genetic or environmental factors have all been linked to modifications in IgG glycosylation, indicating that IgG glycosylation is altered in a dynamic way during different physiological processes (Cobb, 2020). The presence or absence of the core fucose (Fuc) and the terminal sialic acids, respectively, have the largest effect on IgG1 Fc-mediated activity by altering type I and type II FcγR binding affinity (Ferrara et al., 2011; Pincetic et al., 2014; Vattepu et al., 2022).

1.3.1. Fucosylation

In human serum, more than 90% of IgG N-glycans have a Fuc residue attached to the GlcNAc core structure (Kobata, 2008). The fucosyltransferase FUT8 operates as a catalyst for antibody fucosylation in the medial/trans-Golgi. Increased fucosylation levels have been associated with liver diseases such as liver fibrosis and cirrhosis (Mehta and Block, 2008), as well as with tumor progression in several cancers, such as hepatocellular carcinoma (Norton and Mehta, 2019), gastric cancer (Kodar et al., 2012) and many others (Bastian et al., 2021, p. 8). In contrast, decreased fucosylation has been observed in some autoimmune diseases such as autoimmune thyroid diseases, systemic lupus erythematosus and multiple sclerosis (Ząbczyńska et al., 2021). The crystal structures of fucosylated Fc and afucosylated Fc in complex with FcγRIIIa revealed that the lack of core Fuc residue favors carbohydrate-carbohydrate and carbohydrate-protein interactions between the afucosylated Fc N-glycans and FcγRIIIa glycans (Ferrara et al., 2011), increasing IgG affinity for the human FcγRIIIa and FcγRIIIb receptors found on macrophages, dendritic cells, and natural killer cells, which in turn enhances ADCC (Ferrara et al., 2006b; Iida et al., 2006; Liu et al., 2015). Therapeutic antibodies that are afucosylated, such as anti-HER2, exhibit 100-fold greater ADCC than fucosylated anti-HER2 antibodies (Shields et al., 2002). IgG1 Fc afucosylation is a significant proinflammatory alteration overall. In particular, well-characterized afucosylated IgG-FcγRIIIa interactions mediate a number of activating biological functions, including increased cytotoxicity in tumor models, myeloid cell maturation, and NK cell activation (Bournazos et al., 2016). For this reason, the use of afucosylated antibodies with improved ADCC activity has enormous potential in immunotherapy (Wang et al., 2019). The FDA has so far authorized seven antibodies with low or no Fc core fucosylation levels, five of which are for cancer therapy and two of which are for treatment of immune/inflammatory diseases (Golay et al., 2022). Several more antibodies are undergoing clinical trials or pre-clinical research (Golay et al., 2022). Obinutuzumab, a glycoengineered anti-CD20 antibody, received approval in 2013 as the first therapeutic antibody with reduced Fuc for the treatment of B-cell neoplasia (Golay et al., 2013). Extensive comparison with Rituximab in Phase III studies for chronic lymphocytic leukemia and B-cell non-Hodgkin lymphoma has either demonstrated increased activity for Obinutuzumab or, in a few instances, non-inferior activity to Rituximab (Golay et al., 2013; Sehn et al., 2020; Tobinai et al., 2017).

1.3.2. Galactosylation

IgG Fc N-glycans can be agalactosylated, monogalactosylated, or digalactosylated, with each glycoform having a different effector function. Galactosylation has been shown to increase binding to FcγRIIa, FcγRIIb, FcγRIIIa and FcγRIIIb two-fold (Subedi and Barb, 2016). Agalactosylated glycoforms are associated with a proinflammatory response, whereas mono- and di-galactosylated N-glycans are less related to inflammation (Maverakis et al., 2015). Consistent with this, many chronic inflammatory and autoimmune disorders like systemic lupus erythematosus, multiple sclerosis, rheumatoid arthritis, autoimmune vasculitis, active spondyloarthropathy, Crohn’s disease, inflammatory bowel disease and psoriatic arthritis are associated with a decrease in IgG galactosylation in human serum (Ercan et al., 2010), and increased galactosylation is linked to low inflammatory activity (Schwab and Nimmerjahn, 2013), probably because galactosylated IgG promotes complement activation by interacting with C1q (Raju, 2008). Age and inflammation both cause a decrease in IgG galactosylation (Alter et al., 2018), but the pace of change occurs more quickly with the onset of inflammatory conditions than it does with normal aging (Gudelj et al., 2018).

1.3.3. Sialylation

About 10–15% of serum IgG in healthy people are sialylated, with the bulk of them having monosialylated glycoforms (Gudelj et al., 2018). In fact, during inflammatory and autoimmune diseases, this percentage decreases (D. Li et al., 2021). Studies have shown that IgG Fc N-glycans with terminal α−2,6 sialic acid have an enhanced binding to type II FcγRs that trigger anti-inflammatory signaling (Wang, 2019), and are associated with an anti-inflammatory effect and a reduced affinity for type I FcγRs (Anthony et al., 2008a). Combined Fc sialylation and core fucosylation favor a closed conformation of the antibody with higher flexibility (Sondermann et al., 2013). Many FcγRs are thought to be less able to bind IgG Fc in this shape, although FcγRII is still able to do so (Vattepu et al., 2022).

1.3.4. Bisection

Only a modest percentage (10–15%) of antibodies in healthy individuals contain bisecting GlcNAc (Gudelj et al., 2018). The addition of bisecting GlcNAc to the internal Man residue is catalyzed in the Golgi by β−1,4-N-acetylglucosaminyltransferase III (GnT-III) (Kizuka and Taniguchi, 2018). ADCC activity and FcγRIIIa binding are both increased in recombinant IgG Fc with bisecting GlcNAc (Umaña et al., 1999). The bisecting GlcNAc, however, prevents the addition of a core Fuc residue because of steric hindrance, and the suppression of Fuc attachment seems to be responsible for the enhanced ADCC (Shinkawa et al., 2003). Decreased levels of bisection have been linked to tumor progression in gastric cancer, with higher levels of bisection being linked to better survival rate (Kodar et al., 2012). Conversely, an increment in bisecting N-glycans has been observed in type II diabetes (Lemmers et al., 2017), hypertension (Kifer et al., 2021), dyslipidaemia (D. Liu et al., 2018) and obesity (Greto et al., 2021). Bisecting GlcNAc glycoform levels have been shown to vary with age, with decreased levels of bisection being associated with familial longevity (Ruhaak et al., 2010).

1.3.5. Mannosylation

The Fc region can also bear N-linked HM type glycans, which confer a reduced half-life due to an increased clearance attributed to binding to Man and asialoglycoprotein receptors (Jones et al., 2007). This is important for immunotherapeutic mAbs, since HM type glycoforms lead to a potentially lower therapeutic efficacy. High-mannosylation levels are linked to oncogenic transformation in some cancers such as cholangiocarcinoma (Park et al., 2020). Pregnancy and ovarian cancer have also been associated with augmented levels of HM glycans on antibodies (Gerçel-Taylor et al., 2001). In malignant melanoma, an increase in Fc mannosylation of autoantibodies has been observed as the disease progresses, which has been suggested to enhance the survival and growth of malignant melanoma cells (Selim et al., 2011).

1.4. IgG glycosylation: other sites

It is also known that 15% to 25% of the IgG Fab domains include N-glycans (Holland et al., 2006; Stadlmann et al., 2010). Contrary to the Fc N-glycosylation, the function of Fab N-glycosylation is poorly understood. The Fab N-glycans might play a role in immunomodulation, since they affect the avidity and affinity of antibodies for antigens (Wallick et al., 1988; Xu et al., 2012), antibody half-life (Huang et al., 2006), antibody aggregation (Courtois et al., 2016) and immune complex formation (van de Bovenkamp et al., 2016). In contrast to Fc N-glycans, the biantennary CT N-glycans on Fab domains have been characterized as heavily sialylated (Mimura et al., 2007). It has been established that sialylation of the Fc region is responsible for the anti-inflammatory effects of intravenous immunoglobulins (Kaneko et al., 2006). However, more and more evidence points to the involvement of the Fab component (Trinath et al., 2013; Wiedeman et al., 2013). Fab N-glycans have also been shown to display a higher incidence of galactosylation and bisecting GlcNAc than Fc-bound N-glycans, albeit lower levels of fucosylation (Bondt et al., 2014). Moreover, it has been reported that the Fab section may bear HM type glycans (Anumula, 2012). These glycosylation site sometimes shows aberrantly elevated levels of HM glycans in some B cell cancers such as Burkitt’s lymphoma, follicular lymphoma and diffuse large B cell lymphoma (Sachen et al., 2012; Zhu et al., 2003) and in malignant melanoma (Selim et al., 2011).

Additionally, O-glycosylation is present in the hinge region of IgG3 (Plomp et al., 2015). IgG3, which makes about 8% of all the IgG in human serum (Morell et al., 1970), differs from other IgG subclasses due to its lengthened hinge region (Roux et al., 1997), stronger affinity for C1q (Redpath et al., 1998), and higher overall affinity for the Fc receptors (Hogarth and Pietersz, 2012). The elongated hinge region contains up to a triple repeat sequence, which accounts for the broader and more flexible angle between the two Fab arms as well as an enhanced flexibility between the Fab and the Fc portion (Dangl et al., 1988). It has been shown that the triple repeat sequence in the hinge region of each IgG3 heavy chain contains Thr residues that are partially O-glycosylated (Plomp et al., 2015). IgG3 from different sources has been found to include non-, mono-, and disialylated core 1 type O-glycans. Polyclonal IgG3 obtained from donor serum contains O-glycans on about 10% of the hinge repeat Thr residues. Contrary to N-glycosylation profiles, which demonstrate interindividual variability (Baković et al., 2013), the degree and type of O-glycosylation appears to be homogeneous across individuals (Plomp et al., 2015). Although the IgG3 hinge region O-glycans have not been fully characterized, earlier research raises a number of possibilities for their possible functions. First, proteolytic degradation may be prevented by the hinge glycosylation. Numerous bacterial or endogenous proteases have been shown to target the IgG hinge (Brezski and Jordan, 2010), and it is likely that the elongated shape of IgG3 makes it more vulnerable to proteolytic degradation compared to the other IgG subclasses (Baici et al., 1980; Virella and Parkhouse, 1971). The ability of the O-glycans to protect the hinge region from proteolytic cleavage has been demonstrated by the discovery that tryptic IgG3 O-glycopeptides are resistant to endoprotease AspN digestion but the corresponding nonglycosylated peptides are not. Another purpose for the hinge region O-glycans could be to aid in the maintenance of the extended conformation of the hinge region, which might increase the flexibility and orientation of the Fab domains and favor the divalent binding of the Fab to target antigens. Furthermore, IgG3 has been reported to be partially N-glycosylated at N392 in the Cγ3 domain (Stavenhagen et al., 2015). Afucosylated and bisecting glycoforms have a much higher occurrence in N392 compared to N297 (Stavenhagen et al., 2015). The biological role of this glycosylation site is, however, poorly understood.

2. Fc glycoengineering strategies

Therapeutic mAbs, specially IgG antibodies, are an expanding class of therapeutic drugs used to treat many clinical indications, including cancer, autoimmune diseases, inflammatory diseases and bacterial and viral infections. Glycan composition is a critical attribute of therapeutic mAbs and requires effective control strategies to both reduce variability to meet regulatory requirements (Ma et al., 2020) and to influence the antibody-mediated effector functions that are typically key contributors to therapeutic efficacy. IgG antibodies, including most made for clinical use, are frequently composed of several glycoforms, which have a substantial impact on their efficacies, stabilities, and effector functions. Therefore, there is an interest in creating methodologies to obtain uniformly N-glycosylated IgG antibodies which harbor tailored glycoforms with specific effector functions.

Human IgG N297 glycosylation, as observed in the serum of healthy individuals, consists of mainly biantennary core-α−1,6-fucosylated structures. β−1,4-galactosylation of the branches is incomplete (and is affected by systemic inflammation). Partially because of this, only about 5–10% of the glycans carry mainly α−2,6-sialic acid, and the structure with both branches galactosylated and sialylated is exceedingly rare (<0.5%). About 10–20% of the structures carry a bisecting GlcNAc. Higher-antennary structures are very rare (in most cases undetectable) in healthy human plasma on this conserved N-glycosylation site. Typically, we do not detect any appreciable level of HM N-glycans (Bondt et al., 2014; Z. Zhang et al., 2019).

Recent progress in glycobiology and glycoprotein engineering means that it is now feasible to produce homogeneously glycosylated antibodies with tailored effector functions (Li et al., 2017; Wang et al., 2019). The production of afucosylated IgG has been the primary focus of glycoengineering thus far and several strategies have been developed over the years (Dammen-Brower et al., 2022). The most common strategies include cell culture-based glycoengineering, which can be used to quickly produce desired glycoforms on large scale (Mastrangeli et al., 2019), and in vitro chemoenzymatic glycosylation remodeling.

2.1. Cell-based glycoengineering

It is a long-standing goal of glyco-biotechnology to enable the safe and cost-effective manufacturing of Fc-containing antibodies in a multitude of industrial recombinant protein expression systems.

2.1.1. Expression systems for antibody expression and glycosylation control

Antibody glycosylation is influenced by the cells from which they are manufactured. Cell culture systems for antibody production have been developed using cells from a wide variety of organisms, from mammals to plants to bacteria. Here we describe the key attributes, as well as advantages and disadvantages of many of the most common cell culture systems.

2.1.1.1. Mammalian cells

2.1.1.1.1. Current cell types used.

Non-human expression systems, mainly Chinese hamster ovary (CHO), remain predominant for the production of many commercial manufacturing processes for antibody-drug conjugates and mAbs, and to a lesser extent other Rodentia cell lines such as NS0, SP2/0 and YB2/0 cells (Blundell et al., 2020). CHO cells are robust hosts that grow well in suspension culture, can easily be adapted to media, and can produce and secrete several recombinant antibodies in bioreactors at high production rates (up to 25 g L⁻¹ of mAbs after 16 days of cell culture; Xu et al., 2017). Also, they are less susceptible to contamination by human pathogens. Despite dominating the manufacturing of antibodies, CHO cells still differ from human plasma cells (which produce the bulk of plasma IgG) in terms of N-glycan biosynthetic capability. CHO cells lack the expression of GnT-III, and hence do not produce bisecting GlcNAc. In addition, CHO cells produce only α−2,3-sialylated N-glycans, due to the lack of α−2,6-sialyltransferase in these cells (Dicker and Strasser, 2015), whereas human cells produce both α−2,3- and α−2,6-sialylated N-glycans. While most mammalian cells produce a mixture of Neu5Ac and N-glycolylneuraminic acid (Neu5Gc), human cells only produce the former as they lack a functional CMAH protein (Chou et al., 1998) which hydroxylates Neu5Ac to Neu5Gc. Moreover, anti-Neu5Gc antibodies have been detected in human sera (Samraj et al., 2015; Zhu and Hurst, 2002). Fortunately, the CHO cell lineages commonly in use in biopharmaceutical manufacturing are fortuitously low in endogenous Neu5Gc synthesis. Especially when using bovine serum-free media (as is standard in pharmaceutical production), Neu5Gc is typically in the low single digit % of the sialic acid pool. Furthermore, antibodies overproduced in CHO cells almost always have even much lower levels of branch β (1,4)-galactosylation than found in human serum IgGs (Baković et al., 2013; Ehret et al., 2019; Nguyen et al., 2021; Stadlmann et al., 2008). Hence, sialylated glycans constitute only a tiny percentage of the N297-linked N-glycan pool. A very small % Neu5Gc on this very small % of sialylated glycans clearly does not cause significant safety issues, as the extensive long-term use of CHO-made Fc-containing molecules attests to. If however in cellulo engineering attempts are made in CHO (and other non-human mammalian cell lines) to enhance the level of glycan sialylation, it is probably wise to combine this with CMAH gene knockout to avoid a higher level of Neu5Gc modification.

Two other mammalian cells used especially in the past for antibody productions were the murine NS0 and Sp2/0 cells, respectively myeloma and hybridoma cells. Originally, they were derived from immunoglobulin producing tumor cells (Barnes et al., 2000; Köhler and Milstein, 1975; Pornnoppadol et al., 2021). In addition, rat myeloma derived cells YB2/0, owing to their naturally lower expression of FUT8 were used to produce two mAbs, oledumab and ublituximab, with a low content of Fuc core which are currently in clinical trials (Sharman et al., 2017). Beyond having CMAH-based Neu5Gc production, in contrast to CHO cells, murine cells are also capable of incorporating the α−1,3-Gal epitope, which is highly immunogenic in humans, with many people producing high levels of pre-existing anti-αGal antibodies (Chung et al., 2008). This has been a major impediment to the use of these cells for therapeutic antibody production for applications requiring multiple administrations, and was one of the key factors in making CHO cells a more popular choice. Of course, with the advent of easy to use genome engineering technology, this concern can be alleviated (Saleh et al., 2020; Tearle et al., 1996). Meanwhile however, CHO cells have gained such prominence, so much resources have been invested in the industrialization and efficient genetic engineering of this system, cultivation media and process development is extremely optimized and scaled up, and the path for regulatory affairs compliance is now so well established that any other mammalian cell based production system faces a huge cliff to surmount to gain acceptance as a manufacturing platform. Indeed, CHO cells can now fairly routinely be generated with extraordinary specific productivity of over 100 pg/cell/24h (e.g. secreting about the same protein mass as total protein content of the cell itself), sustained over typical fed batch duration of 12–14 days at high levels of cell viability.

There is very little incentive to change from CHO cells as a Fcγ-containing molecule manufacturing platform to any other mammalian cell line. Alternative mammalian cell lines also suffer from the same fundamental challenges as CHO: expensive proprietary serum-free media, slow growth, contamination risk with animal/human viruses, highly complex manufacturing installations, lack of a cell wall that makes the cells less robust to high-intensity processes.

Almost all Fabs used in therapeutic IgGs are non-glycosylated. Indeed, during discovery campaigns of candidate therapeutic IgGs, the presence of N-glycans is a strong down-selection criterium for almost all antibody developers, and O-glycosylation sites with any level of site occupancy more than about 1% are as well. During a high-throughput discovery campaign, it is almost always possible to find Fabs against any given target that are suitable in terms of target engagement but which are not modified with glycans, and developers (wisely) avoid the added complexity of these post-translational modifications unless there is no other option. Of course, if an Fcγ fusion therapeutic is constructed with a fusion partner that is in itself modified with glycans, this can change the picture considerably. In such cases, human embryonic kidney cell (HEK293) expression systems (and those derived from HEK293, Expi293F and 293FreeStyle) have been used. They are the second-most developed mammalian expression system and the mainstay for the manufacturing of viral gene transfer vectors and virally vectored vaccines. As human-derived cells they don’t carry the risk of producing immunogenic glycans. HEK293 cells ensure a human glycan profile in the production of therapeutic mAbs. However, these cells can introduce a high degree of heterogeneity in glycan structures, as reported in IgE producing 30 different glycoforms, many of which may introduce unwanted effects (Ilieva et al., 2019). These cells are particularly apt at producing multi-antennary N-glycans, including poly-LacNAc modifications, which often results in highly complex N-glycan mixtures that can be very difficult to characterize (and manufacture consistently). In addition, human cell lines have other limitations, including the capacity to synthesize sialyl-Lewis^x, which binds to endothelial selectins to cause inflammation (Smith and Bertozzi, 2021) and may adversely affect the biodistribution and pharmacokinetics of antitumor mAbs. Moreover, HEK293 cells carry the risk of contamination by and transmission of human pathogens, mainly viruses, which may explain why CHO cells are still the preferred cell line used for therapeutic mAbs by the pharmaceutical industry.

The few still ongoing efforts to develop alternative mammalian cell expression systems beyond CHO and HEK293 are mainly driven from a ‘freedom to operate’ perspective, e.g. arising from a perception (and to some extent, present fact) that the industrialized versions of these cells are only accessible from a few dominant suppliers, at substantial access and licensing costs. However, this can largely or completely be overcome by careful consideration of cell lineage history and ownership or lack thereof of certain cell line collection deposits, and careful Current Good Manufacturing Practice (cGMP)-compliant re-engineering, and also one of our labs (N.C.) is working towards this goal, aiming at open-accessing these critical expression systems.

2.1.1.1.2. Mammalian cell expression system N-glycosylation engineering for customized antibody functionality.

Neuraminic acids, or sialic acids, are found at the termini of the branches of fully synthesized mammalian N-glycans. High-dose Intravenous immunoglobulin (IVIG) treatment is often used to treat autoimmune diseases that at least in part are driven by auto-antibodies. While this at least in part appears to work by diluting/competing with auto-antibodies for FcγR-binding, engineered increased N297-glycan Fc sialylation of the IVIG product has been linked to a potential for higher potency in certain models (Anthony et al., 2008a, 2008b; Schwab et al., 2014, 2012; Washburn et al., 2015), taking inspiration from earlier basic immunology studies, although the relevance of sialylation in this context is contested (Campbell et al., 2014; Guhr et al., 2011; Leontyev et al., 2012; Nagelkerke et al., 2014), and in other studies no role for enhanced sialylation in potency improvement was detected. It could be that it takes almost full α−2,6-sialylation of the N297 glycan (Washburn et al., 2015) to achieve robust results, or the details of the antibody source and models used are important for observed outcomes either way. Eventually, only human clinical trials would tell whether there is translational relevance to these findings. Substantial expense would also be involved in preparing the large dosages at scale of hypersialylated IVIG; a very significantly enhanced potency would be needed to achieve a cost-competitive position vs. the well-established standard IVIG therapy; possibly more than the 10x reduced dosage required that was reported even in the studies that reported the largest potency effect. Recently, the well-established increased potency of non-core-fucosylated Fc in binding to FcγRIIIa was studied as an avenue for enhanced IVIG potency (Mimura et al., 2022); that study made use of two enzymatic hydrolysis steps of the native IVIG antibody preparation, followed by an enzymatic transglycosylation of semisynthetic oxazoline-functionalized donor N-glycans. While this is a highly performant technology for small-scale preparation of experimental reagents, IVIG therapy uses very high dosage indeed, even when 10–20x enhanced potency could be achieved. It remains to be demonstrated that these chemoenzymatic remodeling technologies could be technically feasible and cost-effectively scaled to the levels required for reliable supply to a fairly large patient population.

A high branch substitution level with β−1,4-Gal of the N297 glycan is necessary for subsequent sialylation in case this is what needs to be achieved for more potent IVIG. Moreover, galactose content itself is also linked with increased CDC activity (Boyd et al., 1995; Hodoniczky et al., 2005; Raju, 2008), which can be beneficial for certain target product profiles of antibody therapy. Overexpressing the galactosyltransferase, β4GalT1, is the core implementation to improve galactose content on IgG. When the transferase is stably expressed, galactosylation has been increased to 68% – 87%, depending on the promoter (Chang et al., 2019; Nguyen et al., 2021; Schulz et al., 2018; Stach et al., 2019) and genomic integration site, with further improvements to 90% when the SLC35D1 transporter and biosynthetic genes for UDP-Gal (Stach et al., 2019), such as GALK1 are included. In transient productions of antibodies though, Gal is often poorly incorporated, with quite typically only 10% galactosylation. Co-transfection with a vector for β4GalT1 expression increased the percentage to 52%. However, the transferase DNA amount needs to be carefully controlled as increases lead to antibody yield losses. Nevertheless, galactose incorporation improved to 73% at optimal co-transfection ratios by further supplying the cells during expression with uridine, manganese and Gal, indicating limitations in precursor biosynthesis (Zhong et al., 2019). In transient HEK expressions of IgG, the same strategy can be implemented: co-transfection of transferase alone boosts galactosylation from 20% to 70%, and to 82%, with Gal supplementation of growth media (Dekkers et al., 2016). Oppositely, galactosylation reduces to 9% with analogs like 2FG (Dekkers et al., 2016). An alternative approach can be used for control in both directions: by first knocking out the endogenous galactosyltransferase and next knocking in its coding sequence under control of an inducible promoter, galactosylation levels can be controlled from undetectable to 87% (Chang et al., 2019). Next to glycoengineering the host cell for specific glycoforms, certain mutations in the Cγ2 domain of antibodies favor the accessibility of the glycosyltransferases, GalT (and SiaT) to the N297 glycosylation site, leading to more decorated glycoforms (Lund et al., 1996). Using multiple mutations of the Fc region in a double knockout mutant mammalian cell lines (KO in genes encoding α−2,3-sialyltransferase), Betenbaugh and coworkers produced hypergalactosylated IgGs (Chung et al., 2017a). Co-overexpression of α−2,6-sialyltransferase and β−1,4-galactosyltransferase in combination with an F243A mutation at the Fc domain, generated an IgG with more than 80% Fc sialylation in which the majority was in the α−2,6-sialylated form (Chung et al., 2017b, 2017a; Raymond et al., 2015), as presumably desired for IVIG treatment.

In terms of pharmacokinetics, for highly solvent-exposed N-glycans, sialic acid modification shields Gal, which typically results in increased half-life of the carrier glycoprotein, thanks to reduced interaction with the endocytosis-mediating asialoglycoprotein receptor (ASGPR) on hepatocytes in the liver. However, this is of only limited concern with regard to the conserved Fc-glycan on IgG, as this biantennary, incompletely processed CT glycan interacts with the Fc protein backbone and appears to be largely shielded from ASGPR recognition. Indeed, native hIgG has a circulatory half-life of 21 days with only minimal N297 N-glycan sialylation. ASGPR binding also reaches high affinity only upon multivalent binding to a Gal-modified glycan/glycoprotein ligand, which is less likely on the IgG Fc, as branch galactosylation is incomplete. IgG pharmacokinetics is governed dominantly by the FcRn-mediated recycling to the circulation upon uptake in endothelial cells. Only when the N297 glycan branch nature is altered in such a way as to substantially change the interaction with the protein backbone, PK effects can come about. Such is the case with HM N-glycans (Goetze et al., 2011; Kanda et al., 2007b), where indeed a somewhat faster beta-phase of blood clearance has been observed in mice, but oddly only at late time points; FcRn still apparently dominating the kinetics of clearance, perhaps until the antibody starts loosing conformational integrity in circulation over time, and the high-mannose glycan is being exposed for mannose receptor mediated clearance. In humans, similar effects of N297 glycan nature on PK can be expected but it remains rather understudied.

While the β1–2 linked GlcNAc residues of the N297 glycan are generally very efficiently installed in CHO cells, both IgG circulating in blood and those made by mammalian biotechnological expression systems typically do not carry structures with more than 2 antennae. As already mentioned above, bisecting GlcNAc modification is present on a fraction of native hIgG N297 glycan. GnT-III installs (Schachter, 1991) this β1–4 linkage onto the innermost Man, after a first GlcNAc has been installed on the α1–3 Man, and thereby blocks core-fucosylation (Umaña et al., 1999), mannosidase II catalyzed trimming of Man₅ (Schachter, 1986) and further branching (Brockhausen et al., 1988a, 1988b; Miyagawa et al., 2001) with additional GlcNAc residues. Overexpression of this GnTIII has been used for suppressing core-fucosylation (see below).

Even though GlcNAcylation to generate the A2 glycoform (Figure 3) is generally efficient in CHO cells, the biantennary nature of the N-glycan yields additional substrates for downstream enzymes, such as GalT and SiaT, to work on. Their lack of optimal efficiencies consequently yields higher heterogeneity, which can be circumvented either by removing the Gal and sialic acid incorporation or by knocking out GnT-II or mannosidase-II (Schulz et al., 2018), resulting in no further processing of the α1–6 arm and fewer substrates for downstream enzymes. However, the α1–6 mannose remains uncapped, or carries the two mannosidase-II substrate alpha-mannoses then, making the N-glycan a substrate for mannose receptors, which shortens half-life (Kanda et al., 2007b).

Figure 3. Schematic overview of (humanized) antibody N-glycosylation.

Panel A shows the frequency of antibody glycoforms encountered on plasma IgG, recombinant mAb produced in CHO cells or recombinant antibody secreted by the current “CHO-ized” Pichia strains. Alongside the graphs depicting relative abundances of glycan composition, the represented glycoforms are schematically depicted below. Panel B shows the strain engineering to convert yeast-type N-glycans in Pichia pastoris to the di-GlcNAcylated glycoform (A2) commonly found in CHO cells. The different monosaccharides are represented by colored circles, triangles, and squares: GlcNAc (blue square), galactose (yellow circle), mannose (green circle), fucose (red triangle) and neuraminic acid (sialic acid, purple diamond).

HM structures constitute only a very minor (often undetectable) fraction of the IgG N297 glycan pool in the plasma of healthy humans. Nevertheless, in biotechnologically overproduced antibodies (including in CHO), depending on the clone and the physiology imposed by the manufacturing process, HM glycans resulting from incomplete glycan processing along the early Golgi stages of the N-glycan biosynthesis can make up a substantial fraction. Keeping this as low as possible is a goal in antibody manufacturing and part of the batch release control analytics.

A last approach to customize Fcγ N297 N-glycans is by introducing ENGases in production cells. EndoT, for instance, placed in the cis-Golgi of MGAT1, also known as GnTI, knockout cells (HEK293GnTI(−)), will trim the N-glycan to a single GlcNAc (Meuris et al., 2014; Wang et al., 2020). This GlcNAc sometimes carries an α1–6 fucose, depending on whether EndoT acted before FUT8 or not. Indeed, when GnTI is knocked out, its acceptor Man₅GlcNAc₂ substrate is present at sufficiently high concentration for FUT8 to sometimes modify this substrate, whereas it strongly prefers glycans that are modified with GnTI and in wild type cells; HM glycans with a core Fuc are therefore typically not seen. The EndoT-generated single GlcNAc is then further extended with Gal and α−2,3-sialic acid (results from HEK293 cells; Meuris et al., 2014). If this extension is undesired, a supplementary inactivation of UDP-Gal/GalNAc synthesis in the cells provides a solution that removes all mucin-type O-glycosylation (Callewaert lab, forthcoming). We have called this engineering as resulting in ‘GlycoDelete’ cell types, although it in fact concerns more of a glycan remodeling technology rather than a true deletion of the glycans. The SiaLacNAc N-glycan that is formed provides for a cellularly installed chemo-orthogonal handle on the Fc for chemical conjugation purposes (see below), which perhaps forms its greatest utility in the case of the N297 N-glycan on antibodies.

In yeast cells, the GlcNAc stump generated by EndoT remains unmodified. While this technique effectively removes most of, if not all, of the heterogeneity of yeast-made glycoproteins, in the case of antibodies the removal of the CT glycan results in conformational alteration of the Fc that completely inactivates binding to FcγRs and complement, similarly to completely non-glycosylated antibodies (Mazor et al., 2007; Robinson et al., 2015; Simmons et al., 2002). In vitro treatment with glycosynthases can restore the N-glycan homogeneously if fully defined glycan donors are used in the transglycosylation reaction (Fairbanks, 2013). At small scale in the lab, this technique has shown promise. However, synthesis of the chemically defined oxazoline-functionalized glycan donor may prove challenging to upscale.

2.1.1.2. Non-mammalian cells

Because mAb production in mammalian cell lines is a relatively expensive process, alternative platforms using yeast, plant, protists, insects and bacteria have been developed and, in some cases, carefully optimized for the humanization of N-glycosylation pathway.

2.1.1.2.1. Yeast-based expression systems

While CHO-cell mediated IgG and Fcγ-fusion protein production is now a mature industrial technology that satisfies many demands, there are certainly areas of application where mammalian cell-based manufacturing technology for Fc-fusion therapeutics is not satisfactory, and limiting what can be dreamed of. In particular, non-mammalian production could be implemented in applications where the Cost of Goods needs to be an order of magnitude or less cheap than is achievable using even the best CHO technology (e.g. 10s of thousands of euros per gram rather than 100–200 euros per gram), or where manufacturing needs to be performed at an order of magnitude larger scale, or where time from cell bank to released drug lots needs to be faster (weeks instead of months). This is the area of biotechnology where micro-organism-based manufacturing prevails, both for pharmaceuticals (e.g. antibiotic production) and for food ingredients and industrial catalysts and biomaterials.

One key example application would be the very rapid re-manufacturing of rapidly scaling batches of an antibody prophylactic/therapeutic against a pandemic-threat infectious disease outbreak. As antibodies and especially Fc-containing molecules, are complex oligomeric molecules with multiple disulfide bonds, prokaryotic expression is still fraught with a lot of difficulties. Hence, the field has largely focused on fungal expression systems. Filamentous fungi often have enormous secretory capacity for their endogenous substrate-digesting enzymes, but as these include a multitude of proteases, success in producing other recombinant proteins has been difficult to achieve. While also here, multiplex genome engineering likely has the potential to substantially alleviate this problem, most of these improved filamentous fungus systems are tightly held by a select few companies, unfortunately not making it easy to independently verify the scope of utility or to access these systems for unrestrained exploration and further development. Furthermore, there is almost no regulatory safety track record of any of these systems especially for systemically administered (injected) protein therapeutics.

In contrast, the two most-often used biotechnological yeast hosts, Saccharomyces cerevisiae and Komagataella phaffii (previously named and still best known as Pichia pastoris, hereafter in this review: Pichia), have been widely available for several decades, resulting in widespread familiarity in both academia and industry, with dozens of protein therapeutics in clinical use in all major markets. Compared with mammalian cells, yeast can be cultured at a higher cell density, which makes glycoprotein production at lower cost and higher efficiency via a scalable fermentation process, together with lower risks of human pathogen contamination. P. pastoris has a glycosylation pathway that produces glycoproteins with HM type glycans (Gemmill and Trimble, 1999), which reduce the in vivo half-life and compromise the therapeutic function of mAbs. Due to the strong difference in Golgi N-glycosylation pathway (but, crucially, not in ER-) between these Ascomycetous yeasts and mammalian cells, they have so far been almost exclusively used for the manufacturing of non-glycosylated protein therapeutics. However, the glycosylation pathway in yeast can be engineered to avoid fungal type glycosylation. It is clear that there are strong incentives to overcome this limitation of yeast protein production systems, as they are of similar regulatory/industrial maturity in microbial biopharmaceutical manufacturing as CHO/HEK293 systems are in the mammalian-cell based world. P. pastoris, which normally cannot produce GDP-Fucose, has been glycoengineered to eliminate fungal type glycans and to produce complex biantennary N-linked glycans. Given the relative simplicity of Fcγ N297 N-glycans in human plasma, and especially in CHO, converting the Pichia Golgi N-glycosylation pathway to the synthesis of biantennary GlcNAc-terminated N-glycans appears as a very attainable goal. Indeed, as CHO cells also do not produce bisecting GlcNAc by nature, there is no urgent need to build this into Pichia (Figure 3). Furthermore, CHO cell-made IgGs always have a very large proportion of non-galactosylated structures, showing that these are safe and efficacious to use. Pichia has been developed for glycoengineered humanized antibodies, including for the production of the anti-CD20 antibody rituximab (Li et al., 2006) and anti-HER2 trastuzumab (C. P. Liu et al., 2018), by knocking out four yeast-specific glycosylation genes and introducing 14 heterologous glycosylation genes (Hamilton et al., 2006). Even though β−1,4-galactosylation can quite easily be achieved in Pichia, one can also reasonably choose to forego this altogether and instead focus on achieving as high as possible homogeneity of the GlcNAc-terminated biantennary structure. Sialylation is indeed very minor anyway on the N297 glycan: why try to synthesize CMP-Neu5Ac, transport it to the Golgi apparatus and get it incorporated with a high degree of homogeneity, if CHO-made molecules with almost no such modification have been demonstrated to be efficacious and safe repeatedly in the past? Indeed, while this has been achieved in concept (Hamilton et al., 2006), we speculate that its industrial translation has been very complex, as the company that invested most resources in this (Merck, after their acquisition of Glycofi Inc.) appears to have terminated development, and instead reported on attempts to scale up production using less-advanced glycosylation-engineered strains (Potgieter et al., 2009; Ye et al., 2011).

Similarly, while human plasma Fcγ N297 N-glycans are almost homogenously modified with the core-α−1,6-fucose modification (Figure 3), many if not most applications where having CT N-glycans on the Fc is critical, actually benefit from not having this core-fucose. Indeed, having CT glycans on the Fc is necessary for obtaining the full immune-effector functionality imparted by the Fc, and at least one of these, ADCC through interaction with hFcγRIIIa/mFcγRIV, is potentiated by up to 100-fold when the core-fucose is not there on the N297 N-glycan (Kanda et al., 2007b; Shinkawa et al., 2003). Again, it is perfectly possible to engineer Pichia to produce GDP-Fucose, its transporter and the required fucosyltransferase (Patent WO2008112092A2), but there is a lot of value in the simpler system that just produces the GlcNAc₂Man₃GlcNAc₂-Asn structure as homogenously as possible. Notably, UDP-GlcNAc is then the only sugar nucleotide required in the building of the branches, and this is in plentiful supply in fungal cell’s intermediary metabolism, as it is the donor in the production of chitin, the defining highly abundant cell wall polymer of the fungal cell wall. In our experience, transport of UDP-GlcNAc to the Golgi is not limiting when an appropriate choice is made of targeting signals and their fusion points to the GlcNAc-transferases required to build both branches of the N-glycan (Jacobs et al., 2009), although others have found it necessary in their particular implementation of the technology (Choi et al., 2003). If no effector functions are required, which is the other main target product profile of Fcγ-containing molecules, we are of the opinion that engineering the Fc not to have N-glycosylation, while restabilizing it e.g. by the introduction of a compensatory disulfide bridge (Callewaert lab, forthcoming), is the more effective solution when manufacturing Fc-fusions in fungal systems.

Indeed, while N-glycosylation ‘humanization’ (a more accurate descriptor would be ‘CHOization’) of Pichia may appear simple in principle (Figure 3), it is actually quite complex to make sufficiently homogenous and robust in industrial practice. In contrast to CHO and human cells, yeast N-glycans only consist of HM-type, with a type of glycoform not seen in humans: hypermannosylation. The key enzyme involved in the hypermannosylation, OCH1p, installs mannose α1–6 linked on the inner α−1,3-linked mannose of the Man₈GlcNAc₂ ER-pathway end product that is the last common structure with the mammalian cell pathway. This first yeast-specific α1–6 linked mannose serves as a starting point for extension of a poly-α1–6 linked mannan backbone, of which the residues carry oligo-α1–2-mannose branches that can be capped with mannosylphosphoryl substitution, or with β-mannose. Most of what we know about the detailed structure of the Ascomycete yeast mannan structure derives from studies on S. cerevisiae and we lack detailed knowledge on relative quantities of substitutions on the mannan even for Pichia, where indeed it would not be a surprise if we had not yet discovered all types of substitution. Mutating the OCH1 gene (Choi et al., 2003) to a large extent, though not completely, abolishes the initiation of the mannan chain. Yet, fully knocking out the gene results in a suboptimal phenotype with increased flocculation and temperature sensitivity. Serendipitous observations on an unintended mutation in our early-generation GlycoSwitch OCH1 knock-in gene disruption constructs have revealed that, at least in Pichia, this can be largely avoided by mutating the coding sequence such that an N-terminally truncated form of the protein is produced at a low level (Patent US9617550B2). This results in a lack of mannan formation while retaining wild type growth. How this can be explained is a matter of further research. When a strong fungal or mammalian Golgi α−1,2-mannosidase is targeted to the ER or ER-Golgi boundary in such OCH1 mutation-engineered Pichia, this competes with remaining non-Och1p mannosyl (phosphoryl)transferases: the mannosidase trims down the branches of the Man_8/9GlcNAc₂ structure that is generated on fully folded proteins exiting the ER, resulting in Man₅GlcNAc₂. This deprives the endogenous mannosyltransferases of substrate. However, even this is not robustly sufficient across different N-glycosylation sites to completely abolish off-target modifications. In particular, we have found that α−1,3-glucosylation can occur in such Man₅GlcNAc₂-strain, almost surely on the inner α−1,3-mannose that is exposed upon α−1,2-mannosidase trimming. This residue is then further modified by β−1,2/3-mannosylation and capped by an α−1,2-glucose residue. The nature of glycosyltransferases performing these off-target modifications is to be determined, but the initiating α−1,3-glucosyltransferase is likely to be the UDP-Glc:glycoprotein glucosyltransferase (UGGT) enzyme involved in glycoprotein folding in the ER. The Glycofi group has reported a structure in which the β-mannosylation occurs directly onto the Man₅GlcNAc₂. Regardless of the exact structures being formed on each specific N-glycosylation site, these observations serve as a caveat that it is often not so difficult in synthetic biology to introduce a heterologous pathway and achieve some level of the target product. However, it can be exceedingly difficult to avoid any off-target product formation as there may be many unknown endogenous enzymes that start recognizing the novel on-target pathway intermediates that are completely novel to the organism and for the avoidance of which the organism’s enzymes hence have never been evolutionarily selected. In the case of Pichia, knocking out the main β-mannosyltransferase gene BMT2 avoids most of this modification. We speculate that this then gives sufficient time for the ER glucosidase 2 to remove the initiating α−1,3-glucose, resulting in a robustly cleaner Man₅GlcNAc₂ glycan profile (Hopkins et al., 2011). Achieving such strain that produces >90% pure Man₅GlcNAc₂ robustly on many different therapeutic glycoproteins and across variations of bioreactor cultivation conditions and carbon sources used, is actually the most challenging step in ‘CHOization’ of yeast N-glycosylation. To subsequently achieve the key GlcNAc₂Man₃GlcNAc₂-Asn Fc N297-linked CT glycan structure, GlcNAc-transferase I, Mannosidase-II and GlcNAc-transferase II are overexpressed. To achieve targeting to the yeast Golgi apparatus, the targeting signals of these higher-eukaryotic (mammalian/insect cell-derived) enzymes are replaced by those of yeast Golgi mannosyltransferases. The best results are achieved by careful consideration of the topology of the fusion point between the targeting signal and the catalytic domain, as the so-called stalk region that spaces the membrane from the catalytic domain is often involved in interactions with other similar-topology Golgi proteins. Careful consideration of where the stalk ends and the catalytic domain fold begins can avoid the large effort in brute-force screening of a large number of less-informed constructs, which is the other approach to solving this protein engineering problem. When all of this is carefully implemented, we now typically achieve high levels of pathway efficiency >95% conversion at each step. However, as the starting substrate (Man₅GlcNAc₂) is only about 90% pure, and we have 4 conversion steps to reach GlcNAc₂Man₃GlcNAc₂, the expected yield of the final product is 0.90*0.95⁴=0.69, or about 73%. This indeed closely corresponds to what we observe. The intermediates between Man₅GlcNAc₂ and GlcNAc₂Man₃GlcNAc₂ are intermediates of the human pathway and are of little to no concern in a therapeutic Fc-fusion. Although their terminal mannose may result in somewhat faster clearance from the bloodstream (Kanda et al., 2007b). These so-called Hy type glycans are sufficient to impart the conformation of the Fc that is important for FcγR interactions. Of more concern are the large diversity of low-abundance remaining high/hyper-mannosylstructures. Indeed, larger glycans beyond Man₉GlcNAc₂ have so many possible linkage isomers and further modifications (such as mannosylphosphoryl-substitutions) that they become very difficult to detect and quantify in Och1-mutant strains, when using the common high-resolution analytical profiling methods such as HPLC, CE-LIF or mass spectrometry. Each structure is present in only a tiny quantity that often falls below the detection limit, but all structures added up together can make up for a substantial fraction of the entire N-glycan pool. When the first synthetic biology attempts at this pathway were undertaken in the early 2000s in several laboratories, including in our own laboratory, this analytical limitation was underappreciated and misguided researchers: glycan patterns look amazingly clean, while careful analysis making use of HM-selective enzymes such as EndoH, with SDS-PAGE analysis of the mass shift induced by removal of the N-glycans, demonstrates that these ‘humanized’ N-glycan proteins still contain substantial fractions (20–30%) of non-CT structures. It is mainly this problem, next to the poor growth phenotype of full-knockout och1 strains in some of the labs involved, that has precluded the successful industrialization of glycan-humanized Pichia using this first generation of technology. Recently, we have developed an effective work-around to this stalemate by making use of a Pichia-made tool enzyme: the EndoT enzyme. This enzyme, naturally secreted by the industrial biotech fungus Trichoderma reesei (anamorph Hypocrea jecorina), removes all non-human CT N-glycans and can be added in minute quantities to Pichia fermentation harvests prior to downstream processing. A simple hold step allows for complete digestion, upon which protein purification starts without any further added complexity or cost. In this way, glycoproteins are manufactured with approx. 75% GlcNAc₂Man₃GlcNAc₂ N-glycans and ca. 25% single-GlcNAc N-glycans (Callewaert lab, forthcoming). The functional characteristics of this fully characterizable glycosylation profile on Fc biophysics and effector function profile is a subject of the present study.

With the approval of the first Pichia-made hIgG (non-N-glycosylated) in the FDA-regulated market in 2021 for a long-term application in a broad patient population (Eptinezumab ant-CGRP for migraine), the regulatory road is wide open for antibody manufacturing in this organism. Rather than producing complex heterotetrameric hIgG, we are finding that the much simpler single-gene encoded VHH-Fc antibody format (with highly stable single-domain camelid VHHs instead of Fabs) is a much more robust match to the Pichia secretory system capacity. Hence, by both deconstructing the antibody format and the required N297 N-glycosylation to its essence (next to the option of abolishing it for effector-dead antibodies), a route to a more important role for yeast-based manufacturing of glycosylated Fc-fusion antibodies is now opening up, potentially fulfilling the need for a faster, cheaper and more highly scalable manufacturing system than achievable in CHO cells, with further expansion of the application potential of antibody therapy.

Alternatively, to humanization by in cellulo engineering of yeast, EndoT processing can also be achieved through carefully tuned expression of this enzyme in the production yeast itself. This results in homogenous glycosylation with a single GlcNAc residue. As detailed elsewhere in this review, such single GlcNAc N-glycans can serve as acceptor substrates for transglycosylation with CT glycan oxazoline donors. Whether this can be scaled to the levels required for biopharmaceutical manufacturing in a cost-effective way remains to be demonstrated. Likely, it could be mainly attractive when the transglycosylated product carries extra therapeutic functionality, e.g. glycan-linked drug payloads. Even there, such manufacturing method will have to compete with alternatives in which the glycan is installed in cellulo and then functionalized enzymatically for drug coupling.

While we have focused here on yeasts and in particular on Pichia as leading non-mammalian eukaryotic expression system in the biopharmaceutical industry, also less frequently used hosts have been engineered along similar lines over the past 20 years. Insect cell glycosylation has been partially humanized, and so has plant (cell) glycosylation (Geisler et al., 2015; Loos and Steinkellner, 2012). While the issues of rewiring the glycosylation in these systems are conceptually similar (mutating out undesired modifications, dealing with negative effects on growth of certain of these mutations, remaining heterogeneity of the glycan profile after humanization), the specifics are of course peculiar to each system. Insect cells share many of the production economics limitations with mammalian cells and plant-based systems suffer from regulatory constraints (depending on the exact system and jurisdiction) and a current lack of available cGMP-compliant production facilities. Nevertheless, as always with core technology development, key applications often turn up unexpectedly over time and the availability of a matching manufacturing technology then makes for a winning combination.

2.1.1.2.2. Conamax platform

The Conamax platform is based on a marine protist in the genus Aurantiochytrium (Conagen, Inc). This organism is robustly fermentable, reaching high volumetric productivity (>1 g L⁻¹ d⁻¹ mAb) production with high purity. This host produces and secretes glycoproteins generally decorated with small, HM glycans (fewer than nine Man molecules branching from the GlcNAc₂ stem), and recently, this host has been manipulated to produce a less-immunogenic paucimannose proteoglycan (GlcNAc₂Man₃). Conamax can be engineered to secrete functional mAbs that have been successfully used to make ADCs via cysteine-linked conjugation. Conamax and CHO-derived anti-HER2 antibodies exhibited similar efficacy and function (Tawfiq et al., 2020).

2.1.1.2.3. Plant-based expression systems

As a main advantage, compared with mammalian cells, glycoproteins produced in plant cells show a high degree of homogeneity (mainly containing two predominant N-glycans) (Chiang et al., 2016), which helps to establish the detailed contribution of glycans to the physicochemical and biological properties of antibodies. A disadvantage of plant cells is that glycoproteins produced by such cells usually contain potentially immunogenic plant-specific core α−1,3-Fuc and β−1,2-Xyl, as well as Lewis A-type structures (Strasser, 2016), all of which are absent in humans. Using knock-down and knock-out approaches, plant lines have been generated with mutations in xylosyltransferase and fucosyltransferase genes in numerous species, including Arabidopsis thaliana, the aquatic plant Lemna minor, Nicotiana benthamiana and the moss Physcomitrella patens, which have been used to produce functional humanized mAbs, such as rituximab CD20 (Kang et al., 2021), anti-HIV mAb 2G12 (Strasser et al., 2008), the anti-Ebola ZMAPP h-13F6 antibody cocktail (lack of core Fuc) (Castilho et al., 2011), and Trastuzumab (Castilho et al., 2011), as well as other recombinant human proteins such as lysosomal acid alpha-glucosidase (Hintze et al., 2021).

2.1.1.2.4. Insect cells expression systems

Insect cells produce glycoproteins with high levels of paucimannosidic N-glycans (GlcNAc₂Man₃) with no terminal Gal or Neu5Ac (Altmann et al., 1999; Hollister et al., 2002; Loos and Steinkellner, 2012). The advantage and disadvantage of the baculovirus-insect cell expression system are similar to those of plant cells. It is also a high-yielding expression system and easy to use (Irons et al., 2018), but can incorporate non-human complex N-glycan structures, including the core α−1,3-fucosylation to the innermost GlcNAc residue of target glycoprotein, which is absent in vertebrates (Walski et al., 2017). Some reports have demonstrated the efficient generation of different mAb glycoforms in insect cells with comparable effector functions as reported in CHO-produced counterparts (human HIV anti-gp41 antibody 3D6 using SweetBac) (Palmberger et al., 2012). Recently, silkworm (Bombyx mori) pupae were shown to be viable antibody factories for successful production of recombinant human IgGs with amino-acid-selective isotope labeling (Yagi et al., 2020).

2.1.1.2.5. Bacterial expression system

Bacteria have recently attracted great interest in their potential use for protein glycoengineering as a fast, simple, and low-cost expression system (Harding and Feldman, 2019). However, glycans in bacteria are significantly different from those in humans (Kightlinger et al., 2018). For glycoengineering purposes, the specific inverting glycosyltransferases (GT) NGT system, with high similarity to eukaryotic acceptor sites, from a proteobacteria Actinobacillus pleuropneumoniae (ApNGT), has been functionally transferred into E. coli and exploited for designer glycoprotein synthesis (Naegeli et al., 2014). A human Fc-IgG1 antibody was shown to be homogenously glycosylated at the naturally occurring site, although such glycans may still need some downstream chemoenzymatic modifications to produce human-like antibodies (C. P. Liu et al., 2018).

2.1.2. Cell culture conditions

Beyond the particular cell type, the cell culture environment can be manipulated during the fermentation process to generate desirable profiles for therapeutic antibodies. IgG N-glycosylation is influenced by several contributing stochastic factors that may act intracellularly or through the external environment in which cell culture is performed at the level of endoplasmic reticulum stress (ERS), as well as changes in temperature, pH, osmolality, and trace metals, which at extreme levels has been shown to affect glycosylation (Goetze et al., 2011; Majewska et al., 2020). Numerous computational models of protein glycosylation have been generated and optimized to provide guidance on the design of optimal strategies to obtain a target glycosylation profile with desired properties (Pawlowski et al., 2018; L. Zhang et al., 2021). In the manufacturing process of mAbs, using newly developed intensified fed-batch processes, it is possible to reach a high titer of mAbs after several days of cell culture (Yang et al., 2016). However, an extended culture duration can significantly decrease the completeness of glycan processing. In general, towards the end of the bioreactor run, the glycan processing becomes less efficient, due to the limitation on nutrients and reduction of cell viability, which is commonly observed as increased levels of HM type, agalactosylated as well as afucosylated glycans (Sumit et al., 2019). Addition of trace elements, such as Zn²⁺, Cu²⁺, and Mn²⁺ in chemically defined media, can increase cell culture performance during the production of mAbs and modulate glycan maturation/processing (Graham et al., 2019).

2.1.2.1. Factors affecting glycan maturation which modulate HM glycan content

The levels of HM glycans in most recombinant mAb biopharmaceuticals are between 1% to 20%, whereas in endogenous human IgG1 are scarce (<0.1%) (Goetze et al., 2011). HM N-glycans in mAbs are problematic due to their low efficacy and high immunogenicity, as well as a potential risk for off-target toxicity due to the uptake by C-type lectin receptors (Goetze et al., 2011). Some factors, such as increasing osmolality, through the addition of salt to the media, or extending cell culture duration have been observed to increase the Man₅ levels and also decrease cell growth, viability, and titer. It was reported that supplementation of Mn²⁺ or the osmoprotectant betaine can reduce this increase of HM induced by osmolality (Kang et al., 2015). Besides, due to the delicate pH gradients (cis/trans Golgi network pH 6–6.7 and cytosol and ER pH ~7.2) between the different cellular compartments, both high and low pH changes can modulate the glycosylation pathway efficiency and thus result in variations of the HM content. In addition, the selected culture media can directly impact the HM levels resulting in different HM levels in the range of 5–16% (Powers et al., 2020). Other components such as choline, spermine, ornithine, and arginase I have also been reported to impact the HM content in mAbs (Kang et al., 2015). By replacing glucose (Glc) entirely by Man, a significant increase of Man₉-Man₆ was reported, likely due to the inhibition of αmannosidases (Slade et al., 2016). In addition, a limitation of glutamine leads to a decrease in the UDP-GlcNAc formation, thereby impairing intracellular glycan maturation. Supplementation to the basal medium (up to 1 μM) of Mn²⁺ accelerated the GT (GnTI) turnover rate and, thus, reduced HM levels, whereas the replacement of Mn²⁺ with Mg²⁺ has been reported to have the opposite effect, decreasing the GnTI catalytic activity by 60% (Graham et al., 2019). Copper and cobalt ions are strong inhibitors of α−1,2-mannosidases Ca²⁺-dependent enzymes (including mannosidases IA, IB, and II) (Graham et al., 2019). Thus, copper-deficient conditions induce an accumulation of the substrates Man₉-Man₆ and a decrease in Man₅ content (Mastrangeli et al., 2020).

2.1.2.2. Modulating Gal and Neu5Ac content

A recent computational model, combining cell metabolism, antibody production, nucleotide sugar synthesis, and glycosylation, yielded a process strategy that increased antibody galactosylation to 93% for all glycans with no reported negative process outcomes (Kotidis et al., 2019). Supplementation of culture media with Gal increased the expression of GnTI and galactosyltransferases and the level of the nucleotide sugar precursor UDP-Gal. This upregulation accelerates the glycan maturing by ensuring that the Man₅ glycan is converted into hybrid forms, which feeds downstream into the galactosylation reaction of N-glycan synthesis (St. Amand et al., 2014). In other studies, it has been observed that raffinose downregulates galactosyltransferase (Brühlmann et al., 2017) and Zn²⁺ (at ≥100 μM) decreases mAb galactosylation in CHO cells (Prabhu et al., 2018). It was also reported that the addition of uridine and MnCl₂, could increase terminal Gal (Chiang et al., 2016), whereas the addition of UDP-GlcNAc and using serum-free culture media were associated with increases in sialylation of IgG1 antibodies (Patel et al., 1992). An approximate 7% increased sialylation level was detected through the addition of micromolar concentrations of dexamethasone in combination with a mixture of Mn²⁺, uridine and Gal (Ehret et al., 2019).

2.1.2.3. Modulating Fuc content

Interestingly, the supplementation of a high concentration (10 mM) of D-Ara in the cell culture resulted in almost complete arabinosylation (>98%) at the core of the produced antibodies (Hossler et al., 2017). D-Ara, which is different from L-Fuc in lacking the 5-methyl group, can be incorporated efficiently into the biosynthetic pathway to compete with core fucosylation during antibody expression. These fully arabinosylated antibodies showed a similar increase in ADCC activity as that of afucosylated antibodies.

2.1.3. Use of small molecule inhibitors of glycan biosynthesis

This approach is cost effective, simple and applicable to different cell lines and genetic backgrounds in cell culture systems. Small molecule inhibitors successfully allow manipulation and control of the levels of specific glycans and monosaccharides associated with mAbs, usually by inhibiting one or more glycoside hydrolase (GH) or GTs involved in N-glycan biosynthesis. On the other hand, the effective use of small molecules in cell culture also could have some issues, including reduced antibody yield, lack of potency, poor cell uptake, limited chemical stability, target specificity/promiscuity, and on- and off-target toxicities that may preclude the application of certain inhibitors in cells (Choi et al., 2018).

One of the small molecule inhibitors most used to modulate glycan biosynthesis pathway for the glycoengineering of mAbs is kifunensine. This alkaloid, originally extracted from the bacterium Kitasatosporia kifunense, acts as a potent inhibitor of α-mannosidase I located in the cis-Golgi (Elbein et al., 1990). Kifunensine and other similar GH inhibitors, swainsonine (inhibitor of α-mannosidase II in the medial Golgi) and castanospermine (blocks the removal of the first D-Glc residue (s) during the initial trimming step in the ER), have also been used to modulate the glycan profile of diverse antibodies (e.g, IgG1 antibody DP-12) (Krahn et al., 2017). In one study, kifunensine was shown to increase the amount of HM glycoproteins containing Man_7–9-GlcNAc₂ oligosaccharides, while castanospermine resulted in HM with attached Glc glycans, and swainsonine allowed for fucosylation of hybrid structures with and without sialylation (Krahn et al., 2017). Kifunensine has also been used successfully for the generation of afucosylated Rituximab in the N. benthamiana transient expression platform (Kommineni et al., 2019) and afucosylated IgG-like bispecific and multivalent anti-HIV-1 molecule, 4Dm2m-F in CHO cells (Chen et al., 2009). However, the application of kifunensine for the generation of afucosylated mAbs may be limited by the concomitant increase in HM glycans (Goetze et al., 2011). Several promising inhibitors to modulate galactosylation of mAbs are based on analogues of UDP-Gal donor of galactosyltransferases (GalTs), located in the medial- or trans-Golgi, such as 1 and 5-FT UDP-Gal, able to block a conformational change in the active site of GalTs that is essential for catalysis (Pesnot et al., 2010). Because mammalian sialyltransferases (STs) use CMP-sialic acid or CMP-Neu5Ac as their donor substrate, the fluorinated donor analog CMP-3Fax-Neu5Ac was reported as a potent ST inhibitor but is not a cell-permeable drug. To circumvent this issue, a peracetylated metabolic precursor SiaFAc was developed (analog to CMP-sialic acid), which is able to passively cross the membrane and efficiently reduce global sialylation in human HL-60 cells (Rillahan et al., 2012). Inhibition of fucosylation by depleting intracellular pools of GDP-L-Fuc using a fluorinated L-Fuc analogue resulted in nearly completely defucosylated IgG1-based mAbs, not only in cell culture, but also in vivo in mice (Okeley et al., 2013a). An operacetylated derivative of 2-fluoro-L-Fuc was developed as a prodrug to optimize cellular uptake. Treatment of CHO cells in Fuc-deficient media with either 2-fluoro-L-Fuc or 2-fluoro-L-Fuc per-O-acetate produced nonfucosylated antibodies, which was also used to produce an anti-CD40 antibody (Okeley et al., 2013a). Fucosylation was reduced in a dose-dependent manner in two different mAbs, by addition of a 6,6,6-trifluorofucose (fucostatin-I) operacetylated, to the CHO and hybridoma cell lines cultures (McKenzie et al., 2018). By co-crystallization assays of GDP-Man 4,6-dehydratase (GMD) and fucostatin (Allen et al., 2016), Allen and coworkers revealed that the interaction occurs at an allosteric binding site. Fucostatin I was found to be incorporated in place of Fuc at low levels (<1%) in the glycans of recombinantly expressed antibodies. For this reason, a Fuc-1-phosphonate analog, fucostatin II, was developed that inhibits fucosylation with no incorporation into antibody glycans. However, its inhibitory potency towards protein fucosylation is weaker than that of fucostatin-I (EC50 ~30 μM vs ~4 μM). A range of other fluorinated Fuc analogues has also been used as metabolic inhibitors of cellular fucosylation, including 5-thio-L-Fuc and 6-alkynyl-L-Fuc (Allen et al., 2016). These metabolic inhibitors of cellular fucosylation may also be applicable for mAb glycoengineering (S. Li et al., 2021). In addition, also 1,2,4-triazole-3-thiol motif might represent a common pharmacophore for fucosyltransferase inhibition (X. Zhang et al., 2019).

2.1.4. Genetic modifications of the N-glycans biosynthetic pathway

To improve their therapeutic efficacy, antibody glycoforms can be altered by modulating the host N-glycosylation pathway. Substrate availability has been modified by multiple and diverse genetic techniques generating knockout or knockdown via gene silencing using small interference RNA (siRNA) technology and gene editing with CRISPR/Cas-9, ZFNs and TALEN approaches (Chan et al., 2016; Choi et al., 2018). In addition, overexpressing specific nucleotide sugar transporters or genetically inactivating or increasing one or more GT activities is directly responsible for the transfer of single monosaccharides to glycan structures or even altering the localization pattern of specific GTs (Li et al., 2017).

2.1.4.1. Production of antibodies with increased galactose or sialic acid content

Co-overexpression of α−2,3-sialyltransferase and β−1,4-galactosyltransferase was shown to elevate IgG sialylation and galactosylation (Costa et al., 2014). Co-transfection of cytidine monophosphate-sialic acid synthase (CMP-SA synthase), cytidine monophosphate-sialic acid transporter, and α−2,3-SiaT significantly increased the intracellular CMP–SA level and improved the SA content of the recombinant protein in CHO cell lines (Li et al., 2017). Moreover, in several studies, it was shown that certain single or multiple mutations in the CH2 domain of the Fc region, favored the accessibility of GalT and ST to the glycosylation site, leading to more decorated glycoforms (Wang et al., 2019). Using multiple mutations of the Fc region in a double knockout mutant mammalian cell line in genes encoding α−2,3-sialyltransferase (Chung et al., 2017b, 2017a), Betenbaugh and coworkers produced hypergalactosylated IgGs. Co-overexpression of α−2,6-sialyltransferase and β−1,4-galactosyltransferase in combination with an F243A mutation at the Fc domain, generated an IgG with more than 80% Fc sialylation in which the majority was in the α−2,6-sialylated form (Chung et al., 2017b, 2017a; Raymond et al., 2015).

2.1.4.2. For production of afucosylated antibodies

The area of Fc-glycan engineering that has had the largest impact on the field of therapeutic antibody production no doubt is that of manipulating the core-fucosylation level of the N297-linked N-glycan, which is why we have reserved the different engineering approaches in which this has been achieved for this separate section. Currently, around 40 different glycoengineered antibodies, with their Fc fucose partially or completely removed, have been investigated in animal models, and many of them have been studied in clinical trials, whereas a few of them had been approved for use in clinical practice (Pereira et al., 2018).

IgG N297-linked glycans found natively in plasma or recombinantly produced in non-engineered mammalian cells are substituted almost homogenously with a core-α1–6 fucose, e.g. attached to the protein-proximal GlcNAc. However, the presence of this modification dramatically decreases the affinity between the antibody and CD16a (hFcγRIIIa, mouse ortholog mFcγRIV), the key FcγR involved in antibody-dependent cell-mediated cytotoxicity (ADCC). While the basic mechanism was unknown at the time when biotechnologists first started exploring altered FcγR interactions upon engineering of N297 glycan engineering, it was later established that the fucose clashes with the N162 N-glycan of CD16a. The strongly decreased affinity in the interaction results in the suboptimal activity of therapeutic monoclonals that depend at least in part upon FcγRIIIa binding for their potency, in particular against tumors. Such potential for higher ADCC is also evident by the superior efficacy (Cartron et al., 2002; Musolino et al., 2008; Treon et al., 2005) of antibodies in patients with the higher affinity (Koene et al., 1997) CD16a 158V variant.

The first glycan engineering method (GlycoMab^™; (Umaña et al., 1999)) that achieved high-ADCC antibodies was discovered at ETH-Zürich and then developed at its spinoff GlycART (now Roche-GlycART, Schlieren, CH). The discovery involved overexpression of the enzyme that generates bisecting GlcNAc, GnT-III. Initially, it was thought that this modification itself resulted in the high ADCC profile of antibodies made in these cells. However, after the reporting of similar impact on ADCC of core-fucosylation removal, it was soon realized that overexpression of GnT-III kinetically siphons off N-glycan substrates for the core fucosyltransferase FUT8 (Umaña et al., 1999). However, pushing the GnT-III expression level too much resulted in growth inhibition of the CHO cells and loss of antibody productivity. At intermediate expression levels of GnT-III, compromising with a 30% loss of antibody yield, only half of the product carries no fucose. The level of fucosylation could be further reduced in subsequent efforts to 10% by redirecting the GnT-III enzyme in the Golgi by using the localization domain of mannosidase II (Ferrara et al., 2006a). In the hypothesized explanation, the preferential location of the chimera allows for easier shuttling of glycoproteins from GnT-I. Yet, despite lowered fucosylation, installing high levels of bisecting GlcNAc does not only interfere with fucosylation, as altered galactosylation and an increase of Hy type glycans (Umaña et al., 1999) have been reported, which may have the potential to affect pharmacokinetics and CDC activity to some extent.

The conceptually most straightforward implementation of removing core-fucosylation is implemented in the POTELLIGENT^™ technology (BioWa, a member of the Kyowa Kirin Group; (Yamane-Ohnuki et al., 2004)), in which the responsible fucosyltransferase gene (FUT8) is knocked out in CHO cells. In a remarkable achievement at the time in the early 2000s, this was achieved many years prior to the advent of the first robust genome engineering technologies for mammalian cell lines, through classical genetic engineering (sequential homologous recombination) and screening.

Both technologies have resulted in important antibody therapies approved for clinical use, with the GlycART technology incorporated in the Roche-Genentech platform, and the BioWa technology accessible through a licensing business strategy. As both technologies are now more than 20 years old, the end of the lifetime of key patents has been reached, but in the past two decades there has of course been a strong incentive for inventors to explore independent ways of achieving non-core-fucosylated glycosylation of antibodies. It is in this light that many of the follow-on work needs to be seen, beyond scientific creativity.

As FUT8 is only responsible for core α1–6 fucosylation, other fucosylation products such as the Lewis antigens are preserved, as opposed to changes in the fucose metabolism such as GFUS/FX and GMDS modifications (Figure 4). Knocking out or mutating these genes results in non-core-fucosylated antibodies by blocking the de novo synthesis pathway of GDP-L-fucose, the activated sugar donor for FUT8 (Imai-Nishiya et al., 2007). GDP-mannose 4,6 dehydratase (GMDS) catalyzes the conversion of GDP-D-mannose into GDP-4-keto-6-deoxy-D-mannose, which is subsequently converted into GDP-L-fucose by GDP-L-fucose synthase (GFUS/FX; (Louie et al., 2017; Taupin, 2011)), the TSTA3 gene product. Yet another alternative approach is to overexpress the bacterial GDP-6-deoxy-D-lyxo-4-hexulose reductase (RMD) (Mabashi-Asazuma et al., 2014; Palmberger et al., 2014; von Horsten et al., 2010) which will metabolize GMDS’ product into GDP-D-rhamnose, in this way diverting the pathway away from GDP-L-fucose. These methods have produced the FDA-approved anti-CD20 mAb obinutuzumab (Gazyva^®) for treatment of non-Hodgkin’s lymphoma and chronic lymphocytic leukemia (Goede et al., 2015). With a different approach, Sandig and coworkers (von Horsten et al., 2010) produced almost completely afucosylated antibodies by heterologous expression of a prokaryotic enzyme GDP-6-deoxy-D-lyxo-4-hexulose reductase, which converts GDP-4-keto-6-deoxy-D-to GDP, providing feedback inhibition to GMD activity and generating low levels of the GDPdonor. Interestingly, using localization motifs of other enzymes to change the localization pattern of GnT-III resulted in a 70% decrease in core-fucosylation levels of antibodies (mogamulizumab) which has been approved for the treatment of adult T cell leukemia/ lymphoma (Ferrara et al., 2006a). Yet, such overexpression system that rewires the metabolic destination of GDP-fucose rather than simply killing its synthesis, may not generate non-fucosylated antibodies as robustly as knock-outs. Moreover, the strategies circumventing the de novo pathway only work successfully when Fuc is absent from the growth medium, as it can be taken up and phosphorylated for subsequent activation with GDP, in the so called salvage pathway. In an attempt to block both pathways, the GDP-L-fucose transporter (Kanda et al., 2007a; Mabashi-Asazuma et al., 2014) in the Golgi was targeted. While the knockdown proved successful, the fucosylation levels did not decrease, indicating redundancy for fucose transport into the Golgi apparatus. While irrelevant for antibody manufacturing at larger scale, the salvage pathway offers the possibility in the FX and GMDS modified CHO cells to express antibodies with restored fucosylation as experimental control by externally supplying fucose. With another genetic approach, Weiss and coworkers tuned IgG fucosylation and galactosylation levels. They knocked out endogenous α−1,6-fucosyltransferase FUT8 and β−1,4-galactosyltransferase (β4GALT1) genes in CHO cells, and then exerted precise control of exogenous FUT8 and β4GALT1 gene expression using orthogonal small molecule inducible promoters, generating antibodies with desired fucosylation (0–92%) and galactosylation (2–81%) levels (Chang et al., 2019).

Figure 4. Metabolic pathways leading to core-fucosylation.

GDP-L-fucose, the sugar donor for core-fucosylation, can be obtained through two pathways. In the de novo pathway, activated D-mannose is converted into GDP-L-fucose, while in the salvage pathway L-fucose is energized directly. Discussed interventions to reduce core-fucosylation in mammalian cells are indicated with a black outline. Human enzymes are colored in red, in contrast to the insect enzyme RMD. Abbreviations: GLUT1: glucose transporter 1, HK: hexokinase, GPI: glucose-6-phosphate isomerase, MPI: Mannose-6-phosphate isomerase, PMM: phosphomannomutase (1/2), GMPPA: mannose-1-phosphate guanyltransferase alpha, GMDS: GDP-mannose 4,6 dehydratase, RMD: GDP-6-deoxy-D-mannose reductase, GFUS: GDP-L-fucose synthase, FUCT1: GDP-fucose transporter 1, FUT8: alpha-(1,6)-fucosyltransferase, FCSK: L-fucose kinase and FPGT: fucose-1-phosphate guanylyltransferase.

A number of ultimately less practical, effective or scalable approaches have also been described and mentioned here for completeness. An absolute requirement for core-fucosylation is the presence of GlcNAc on the α1–3 branch (Boruah et al., 2020; Longmore and Schachter, 1982). In the maturation pathway of N-glycans, GnT-I installs GlcNAc on Man5GlcNAc2 following α−1,2-mannose trimming of higher mannose glycoforms. By removing GnT-I (Byrne et al., 2018), no core-Fuc should be transferred onto the N-glycan, but in reality some often is, as the higher concentration of Man₅GlcNAc₂ can be a (very) slow-converted substrate for FUT8 (Boruah et al., 2020). As the subsequent enzymes (ManII, GnT-II) do not accept Man₅ as substrate, the N-glycan remains as Man₅GlcNAc₂ (with minimal core fucosylation) and carries terminal mannoses, which ultimately lower the antibody’s half-life.

Culturing CHO cells with fucosylation inhibitors such as 2F-peracetyl-fucose can be attempted as well (Okeley et al., 2013a). While introducing this compound does not reduce cell viability or antibody titers, core-fucosylation is only reduced to 20%, making this approach less attractive than the genetic interventions. Alternatively, small molecule control (Chang et al., 2019) has also been used to make core-fucosylation inducible. This involved first knocking out endogenous FUT8 and subsequently re-introducing FUT8 under a doxycycline or abscisic acid inducible promoter, allowing for fucosylation control during expression. Yet, employing small molecules complicates manufacturing by requiring their inactivation or removal from the final product.

2.2. Chemoenzymatic glycoengineering of IgG Fcs

The methods described above enable the production of antibodies with enriched glycoforms. However, the complexity of the glycan biosynthetic pathways limits the nature of glycoforms that can be obtained through these strategies. The usage of engineered mammalian cell lines requires the development of special cell lines for the production of each specific glycoform of interest. This problem can be avoided through in vitro glycoengineering, in which the addition or deletion of specific sugar moieties is achieved by using specific GTs and GHs. In vitro glycoengineering methods might be chemoenzymatic or enzymatic.

Enzymatic in vitro glycoengineering is based on the stepwise addition or removal of sugars. For example, by adding UDP-Gal and β−1,4-galactosyltransferase-1 (B4GalT1) to a mixture of antibodies, the G2F glycoform with terminal galactosylation can be obtained. Further addition of CMP-NANA and α−2,6-sialyltransferase (ST6Gal1) allows the production of the terminally sialylated G2S2F glycoform (Washburn et al., 2015). Antibodies obtained using this strategy have been shown to effectively attenuate autoimmune disease in rheumatoid arthritis and systemic lupus erythematosus mice models (Bartsch et al., 2018). Conversely, asialylated and agalactosylated glycoforms can be obtained by treating antibodies with neuraminidase and β−1,4 galactosidase enzymes (Tayi and Butler, 2018).

IgG antibody chemoenzymatic glycoengineering is a process that takes place in two steps. Recombinant expression of IgGs in mammalian culture results in a core fucosylated biantennary CT glycan with varying number of Gal residues (G0, G1, G2) at the conserved N297 site on Fcs (Jefferis, 2005). In order to eliminate heterogeneity, the first step of a chemoenzymatic remodeling reaction requires cleavage of the existing glycans using endo-β-N-acetylglucosaminidases (ENGases). ENGases cleave between the first and second GlcNAc residues, leaving only the core GlcNAc and Fuc if present. They can be derived from bacteria, fungi, or Caenorhabditis elegans and exhibit varying efficiency and glycan specificity (Fairbanks, 2017). Although several ENGases have been identified and studied, the unique properties of EndoS and EndoS2 make them ideal options for efficient cleavage of IgG Fc glycans.

EndoS is an ENGase secreted by Streptococcus pyogenes that is highly specific for N297-linked glycans on IgG Fcs (Collin and Olsén, 2001; Goodfellow et al., 2012). EndoS adopts a “V” shaped conformation and is composed of six different protein domains including, from N- to C-terminus (Trastoy et al., 2014, Trastoy et al., 2018; Du et al., 2020; Trastoy et al., 2022): (i) N-terminal three-helix bundle domain (N-3HB), (ii) a GH18 GH domain, (iii) a leucine-rich repeat domain (LRR), (iv) a hybrid Ig domain (hIg), (v) a β-sandwich domain and (vi) C-terminal three-helix bundle domain (C-3HB) (Figure 5A). Similarly, EndoS2 is a related protein identified from serotype M49 of S. pyogenes that has also been shown to be IgG Fc specific (Sjögren et al., 2013). EndoS2 is also a monomeric “V-shaped” protein, but lacks the N- and C-terminal 3HB that are present in EndoS. An important difference between the two enzymes lies in their glycan selectivity. While EndoS can only cleave CT glycans, EndoS2 shows relaxed substrate specificity and can cleave CT, Hy, and HM type glycans (Klontz et al., 2019; Trastoy et al., 2018; Trastoy et al., 2021). The reason for this difference in glycan selectivity is explained by the crystal structure of EndoS and EndoS2 with their different glycan substrates (Figure 5B). The X-ray crystal structure of the catalytically inactive mutant EndoS_D233A/E235L in complex with the biantennary CT N-glycan product Gal₂GlcNAc₂Man₃GlcNAc (G2) (PDB: 6EN3) reveals that each antenna of the CT glycan is positioned in one of two well-defined grooves present in the GH domain of EndoS. Groove 1, which is defined by loops 2, 3 and 4, holds the α(1,6) antenna, whereas groove 2, which comprises loops 1, 2 and 7, interacts with the α(1,3) antenna. The involvement of loop 2 in both grooves is due to the fact that loop 2 bisects the active site with the aromatic reside W153, which forms hydrogen bonds with Man (−3) and CH-π interactions with GlcNAc (−7). Similarly, the X-ray crystal structure of EndoS2 in complex with G2 glycan shows that the active site has two grooves determined by loops that are equivalent to those of EndoS, with both the α(1,3) and α(1,6) antennae of CT Ν-glycans fitting into the grooves (Klontz et al., 2019b) (PDB: 6MDS). Loops 3 and 4 of EndoS2, which interact with the α(1,6) antenna, are shorter than in EndoS, leaving more space in groove 2. The X-ray crystal structure of EndoS2 in complex with Man₉, a HM type glycan (PDB: 6MDV), shows that a bulky α(1,6) antenna can be accommodated in groove 1. In contrast to EndoS, EndoS2 has the histidine residue H109 in loop 2, instead of the W153 present in EndoS. Therefore, EndoS2 can process bisecting CT N-glycans, whereas the bulkier W153 on EndoS hinders the binding of these types of substrates. Alanine scanning mutagenesis experiments show that the interactions with the glycan core and with the α(1,3) antenna are responsible for the recognition of CT N-glycans by EndoS and the recognition of CT, HM type and Hy type N-glycans by EndoS2 (Klontz et al., 2019b; Trastoy et al., 2018a). In the case of EndoS2, the structural conformation adopted by loops 3 and 4 creates additional space that allows EndoS2 to bind glycans with different compositions in the α(1,6) antenna, such as HM type N-glycans. Additionally, the dual nature of Hy-type N-glycans makes EndoS2 able to cleave this type of glycans from IgG antibodies too.

Figure 5. Structural basis of N-glycan recognition by EndoS and EndoS2.

(A) Overall structure of EndoS (left) and EndoS2 (right), indicating the different domains of each enzyme. (B) Surface representation of the GH domains of EndoS_D233A/E235L in complex with the CT product product, Gal₂GlcNAc₂Man₃GlcNAc (left; PDB: 6EN3) and EndoS2 in complex with the CT product Gal₂GlcNAc₂Man₃GlcNAc (center; PDB: 6MDS) and the HM type product Man₇GlcNAc (right; PDB: 6MDV). The residues from each loop that establish interactions with the glycan product are annotated. For each structure, a close-up view ribbon representation of the active site is shown with the main residues interacting with the glycan labelled.

Once the antibody glycans have been removed with EndoS/S2, if an afucosylated glycoform is desired, an α-L-fucosidase can be used to cleave the core α(1,6)-Fuc on IgG Fc glycans. There have been several fucosidases reported, with AlfC from Lactobacillus casei being the most efficient (Klontz et al., 2020; Prabhu et al., 2021). Importantly, AlfC can only cleave the core Fuc if the glycan has been hydrolyzed first - such as by EndoS or EndoS2. It is interesting to note that there has yet to be an α-fucosidase identified that can cleave a core Fuc on an intact IgG Fc glycan. Although the human enzyme FucA1 has been reported to defucosylate an intact CT glycan, it is highly inefficient, requiring both large amounts of enzyme and prolonged incubation time (Prabhu et al., 2021). Thus, the high specificity and efficiency of EndoS and EndoS2 can be harnessed to hydrolyze the heterogeneous glycoforms of N297-linked IgG Fc glycans down to only the core GlcNAc, while an α-fucosidase like AlfC further removes the core Fuc, if desired.

After hydrolyzing the N297-linked glycans, the desired glycoform can then be added to the Fc using glycosynthases. Glycosynthases are variants of ENGases in which at least one mutation, usually in the active site, has been engineered to promote efficient catalysis of the transglycosylation reaction. Despite the fact that several glycosynthase mutants have been designed, EndoS and EndoS2 glycosynthase mutants with their high IgG Fc specificity are ideal candidates for transglycosylation. EndoS_D233A/Q and EndoS2_D184M variants have been shown to successfully add sialylated CT (SCT) glycans to the core GlcNAc on IgG Fcs (Giddens et al., 2018; Huang et al., 2012; Li et al., 2016). The SCT glycan can be isolated and purified from sialylglycopeptides that occur in commercially available egg yolk powder (L. Liu et al., 2017). The derived SCT glycan then needs to be activated with an efficient donor substrate, most commonly oxazoline (ox) (Lin et al., 2015). EndoS/S2 glycosynthase mutants can then add the ox-S2G2 onto IgG Fcs (Figure 6). However, there are several challenges associated with such transglycosylation reactions: (i) the oxazoline needed to derivatize the glycan is highly unstable, and (ii) glycosynthase mutants such as EndoS2_D184M retain appreciable residual hydrolytic activity (Li et al., 2016). To circumvent these issues, ox-S2G2 can be used in significant excess compared to the IgG substrate and the time of the reaction can be limited. Additionally, there have been attempts at bypassing the use of oxazoline with an EndoM_N175Q glycosynthase mutant which can directly form an activated intermediate from sialylglycopeptides (SGPs), allowing other glycosynthase mutants to add the glycan onto the substrate (Ivanova and Falcioni, 2022). This reaction scheme, however, still needs to be studied further, and a large excess of SGP remains necessary. In contrast to biosynthetic pathway manipulation and enzymatic glycoengineering methods, the chemoenzymatic approach is not limited to the manipulation of biantennary glycoforms and can also handle the modification of glycans with higher degree of branching. For instance, the usage of the bacterial ENGase EndoF3 permits the removal of bi- and tri-antennary glycoforms from core fucosylated antibodies, and its GT mutant EndoF3_D165A enables the transfer of tailored bi- and tri-antennary glycans (Giddens et al., 2016).

Figure 6. Strategies of Chemoenzymatic Glycoengineering of IgGs.

IgGs containing heterogeneous Complex-type N-glycans on Fc region can be recombinantly produced in cell culture in the presence or absence of the fucose inhibitor 2-deoxy-2-fluoro-L-fucose. EndoS2 can then be used to remove the glycan on Fc leaving only the core GlcNAc and fucose, if present. Upon affinity chromatography purification, homogeneous and chemically activated glycans can be added onto the deglycosylated IgG using the glycosynthase EndoS2 D184M. The glycoengineered antibodies can be further modified with several enzymes and glycosyltransferases as indicated.

SCT glycans can be further trimmed with the use of sialidases – also known as neuraminidases - and galactosidases. Clostridium perfringens secretes several sialidases, of which NanJ was reported to preferentially cleave the α−2,6-sialic acid linkages found in SCT glycans (Hsu et al., 2022). However, NanH from C. perfringens has also been used to successfully cleave the terminal sialic acid in IgGs (Li and McClane, 2014). Other commonly used sialidases include MvNA from Micromonospora viridifaciens and sialidase A recombinantly produced from an Arthrobacter ureafaciens gene (Cheng et al., 2014; Pagan et al., 2018). Similarly, galactosidases such BgaA from S. pneumoniae cleave the terminal β−1,4-linked Gal after sialidase treatment (Hsu et al., 2022). The unique glycosylation site on IgG Fcs allows for “one-pot” chemoenzymatic trimming of SCT glycans to produce G2 or G0 glycoforms. It is also a convenient and efficient way to produce multiple glycoforms from the same starting glycan. This method can only produce glycans with identical sugars on the α(1,3) and α(1,6) antennae. An alternative method to produce glycans with different sugar compositions on each antenna requires the use of total chemical synthesis and protecting groups (Lin et al., 2015). Other non-CT glycans can also be chemically synthesized and enzymatically added onto Fcs.

In vitro glycoengineering approaches could also be considered as alternative strategies to eliminate the HM content partially or completely from recombinant mAbs. HM-specific ENGases, such as EndoH, EndoF1, EndoBT-3987, can remove all Man residues, whereas the use of certain mannosidases, including the specific human α−1,2-specific Man₉-mannosidase (Moran et al., 1998) (convert Man₉ in Man₅); the broad spectrum mannosidase from Jack Bean, which is able to hydrolyze all α-linkages (α−1,2-, α−1,3- and α−1,6-) leaving the final core β-linked Man (Gnanesh Kumar et al., 2014) or the α−1,6 and α−1,2, α−1,3 mannosidases isolated from Xanthomonas (Wong-Madden and Landry, 1995) can also control HM content on Fc-IgG.

The final step of chemoenzymatic glycoengineering involves confirming the glycoforms present on the remodeled Fc. This has traditionally been achieved by releasing the glycans with enzymes like PNGaseF, labeling them, and using fluorescence or mass spectrometry (MS) for detection (de Haan et al., 2020). However, these methods suffer from lengthy sample preparation and loss of spatial information about the glycosylation site. Although the glycosylation site on IgGs is well defined, spatial information is useful to understand the “pairing” of glycans on both chains of an Fc. As such, glycan analysis of intact proteins is an attractive alternative. Lectin binding assays enable intact protein analysis; however, they are only useful for partial glycan detection, as the lectin usually binds a specific sugar moiety, such as an α−2,6-sialylic acid (Ito et al., 2018). More recently, technological advancements have enabled the increased use of MS-based methods for intact mAb glycan analysis. In such top-down analyses, intact mAbs are separated prior to injection into an MS ionization system. The resulting ions are then detected based on m/z ratios. The main challenge with the use of separation techniques for intact protein analysis is the increased adsorption. Most commonly, reverse-phase liquid chromatography (RPLC) employs high temperatures and a low flow rate of the mobile phase to enable intact mAb separation (Fekete et al., 2016). However, the development of a wide-pore stationary phase allowed for the recent characterization of whole proteins using hydrophobic interaction liquid chromatography (HILIC) that separates proteins based on specific and non-specific hydrophilic interactions (Periat et al., 2016). Similarly, capillary electrophoresis (CE) separation of monoclonal antibodies was made possible by using a neutrally coated capillary with a porous tip (Haselberg et al., 2018). After separation, the antibodies can be ionized either via electron spray ionization (ESI) which uses high voltage to aerosolize liquid, or matrix-assisted laser desorption/ionization (MALDI), which uses a laser (Toby et al., 2016; van der Burgt et al., 2019). Finally, the ion source can be coupled to a quadruple time of flight (QTOF), Fourier-transform ion cyclotron resonance (FTICR), or Orbitap analyzer that determines m/z ratio based on the ions’ time of flight, cyclotron resonance in a fixed magnetic field, or axial motion around a central electrode, respectively (García-Alija et al., 2022; Toby et al., 2016; van der Burgt et al., 2019).

Another approach to antibody glycoengineering is in vivo glycoengineering, e.g., the modification of the glycoforms in vivo circulating IgG antibodies. EndoS has been administered to mouse models of idiopathic thrombocytopenic purpura (ITP) and showed a therapeutic effect (Collin et al., 2008). Pretreatment of antibodies from arthritis murine models with EndoS has been shown to abolish the development of the disease by suppressing IgG-Fc receptor binding and altering the formation of immune complexes (Nandakumar et al., 2007). In vivo sialylation has been employed by making a fusion of GTs β4GalT1 and ST6Gal1 to IgG Fc (Pagan et al., 2018). This approach modifies the glycans in pathogenic antibodies, turning them into sialylated glycoforms with an anti-inflammatory effect and abolishing their pathogenicity. The sialylation of pathogenic antibodies occurs specifically at inflammation sites using CMP-sialic acid (CMP-SA) as a source of donor sugar (Pagan et al., 2018), but not in other antibodies in circulation.

2.2.1. Glycoengineering of other classes of antibodies

Although the field of therapeutic antibodies is dominated by IgG-based drugs, other classes of antibodies are also being developed for their use in clinical environments. For instance, IgA-based therapies have been suggested as potential treatments for bacterial and viral infections, as well as for inflammation and tumors (de Sousa-Pereira and Woof, 2019; Sterlin and Gorochov, 2021). IgA can be divided into two subtypes: IgA1 and IgA2. Subtype IgA2 can be further divided into three allotypes: IgA2m (1), IgA2m (2) and IgA2 (n). In contrast to IgG1, IgA1 contains two N-glycosylation sites, whereas IgA2m (2) contains four N-glycosylation sites and IgA2m (1) and IgA2m (n) have a fifth glycosylation site (van Tetering et al., 2020). Moreover, IgA1 is O-glycosylated in its hinge region, and secretory IgA on mucosal surfaces and external secretions is composed of a dimer of IgA monomers joined by a J chain which displays additional glycosylation sites (Ding et al., 2022). N-glycosylation patterns of IgA are therefore very heterogeneous, which results in an inconvenient production process that makes the use of IgA antibodies for tumor therapy challenging. In addition, many of these N-glycans contain terminal Gal or Man residues, which are susceptible to clearance by the asialoglycoprotein receptor and the Man receptors, respectively, accounting for the short half-life of IgA (Lee et al., 2002; Rifai et al., 2000). Thus, supplementing production cell lines with the enzymes needed to add terminal sialic acids has been explored as a way to extend IgA serum exposure time (Rouwendal et al., 2016). An alternative approach to IgA glycoengineering has been to simplify glycosylation and consequently enhance drug homogeneity by removing some of the N-glycosylation motifs, which did not negatively impact antigen binding or antibody effector functions (van Tetering et al., 2020). Finally, production of IgA in plants has been explored as a different method to arrange different glycosylation patterns. Monomeric IgA has been produced in Nicotiana benthamiana plants, obtaining antibodies with a more homogeneous glycosylation profile and composed mainly by CT biantennary glycoforms (Göritzer et al., 2017; Göritzer et al., 2020). IgM has also been expressed in different plant lines of N. benthamiana modified to produce human-like glycosylation patterns (Vasilev et al., 2016). Furthermore, production of recombinant IgE is also being developed for cancer immunotherapy (Sutton et al., 2019).

3. Synthesis of Fc glycoconjugates

Monoclonal antibodies and their derivatives such as bispecifics, cytokine-fusions and antibody-drug conjugates (ADCs) are well established therapeutic protein formats in the clinic. To generate ADCs, potent cytotoxic molecules are conjugated to the antibody scaffold, achieving highly targeted delivery to cells expressing the antibody’s antigen. Initially, random conjugation to lysines was implemented, but with suboptimal results both in terms of potency and pharmaceutically reproducible and characterizable manufacturability. Hence, more selective conjugation methods were and are sought. Currently, the majority of approved ADCs carry their payload linked to the interchain disulfides. While this cysteine specificity generates improved conjugates, the interchain disulfide bridges are no longer formed, which is disadvantageous for the stability of the antibody molecule. There is a demand for as simple as possible (and hence scalable) conjugation technologies with high conjugation site specificity. In this review, we will focus on the current state of the art in conjugation techniques in which the N297 glycan is used as a handle for coupling. An overview can be found in Figure 7. Aside from glycan-specific conjugation, amino acid targeted site specific alternatives have also been documented in the literature, and this has developed into a specialist field in and by itself. These methods all come with their advantages and disadvantages and various validation states with regard to clinical manufacturing and utility (most remain in early stage and have so far not been demonstrated in any product entering clinical development). They either target cysteines, naturally present or incorporated at engineered positions, or unnatural amino acids functionalized for chemo-orthogonal coupling chemistry, or peptide tags that serve as substrates for enzymes that generate a reactive tag or that directly produce the conjugate of interest. Discussing all of these methods in further detail exceeds the scope of this review. We refer to several reviews which extensively cover this topic (Beck et al., 2010; Chudasama et al., 2016; McCombs and Owen, 2015).

Figure 7. Schematic overview of antibody conjugation techniques employing N-glycans.

For each panel, an exemplary start structure is indicated in a gray box. In panel A, approaches with oxidation are depicted, both enzymatic with GaOx (galactose oxidase) and chemical with sodium periodate. Panel B shows techniques employing ENGases (endo-glycosidases) that trim off a broad range of substrates into a GlcNAc stump which is then either elongated with monosaccharides with functional groups by glycosyltransferases, or transglycosylated with oligosaccharides by (mutated) ENGases. Next to the two - step methods depicted on the right (ENGase + EndoS(2) or ENGase + GalT Y289L), EndoS2 can also be used in an one – step strategy in which the transglycosylated products (the trisaccharides) are no substrate for the hydrolase. Panel C illustrates the use of glycosyltransferases to introduce chemo-orthogonal groups on antibodies. Blue squares, green circles, yellow circles, purple diamonds and red triangles represent N-acetylglucosamine (GlcNAc), mannose, galactose, N-acetylneuraminic acid (sialic acid) and fucose residues respectively.

As most clinical monoclonals carry only two N-glycans, one on N297 in each of their Fc domains, conjugation onto the saccharides results in high site specificity. Nevertheless, these N-glycans cannot be used directly in orthogonal chemistry, as the functional groups they carry (mostly hydroxyl functionalities) are not uniquely found in carbohydrates. Yet, conjugations to these glycans can be achieved upon selectively oxidizing them into aldehydes, either chemically with sodium periodate or enzymatically. Periodate based oxidations usually target the extracyclic glycerol side chain of sialic acid, as this requires very low concentrations of the oxidizing agent and therefore minimizes side reactions on amino acid side chains such as methionine (O’Shannessy and Quarles, 1987; Wang et al., 2011; Zhou et al., 2014), which would negatively impact FcRn recycling. For other monosaccharides, such as Fuc (Hinman et al., 1993; Zhou et al., 2014), higher concentrations of periodate are required, making this reaction less favorable. Enzymatic oxidation with galactose oxidase targets galactosylated glycan termini selectively and results in C6-oxidation (Angelastro et al., 2022; Chua et al., 1984; Morell et al., 1966; Solomon et al., 1990; Stan et al., 1999). It is being intensively explored as a cost-effective way of functionalizing the N297 N-glycan or engineered versions of it (see below). The resulting aldehydes can subsequently be used in highly selective conjugations with aminooxy- and hydrazide-based linkers, among others, in conditions that are compatible with antibody integrity. Such simple click chemistry can be greatly catalyzed using aromatic amine catalysts (with aniline as the simplest, but with many much improved derivatives having been developed over the last decade). In contrast to observations on more limited stability of oxime conjugates to other types of protein-carried aldehydes (Agarwal et al., 2013), such oxime conjugations to glycans can be very stable in plasma (e.g. we have not observed any appreciably disintegration over more than 3 weeks of incubation in human or murine plasma of galactose-C6-oxime conjugates). One obstacle for obtaining homogenous conjugation products when targeting the N297 N-glycans, is their inherent heterogeneity. Indeed, galactose oxidase needs terminal Gal residues, uncapped with sialic acids, while for periodate based oxidation, homogenous sialylation is optimal. To achieve such specific glycoforms, IgG can be modified in vitro or in cellulo, as discussed above, including with GlycoDelete technology, which generates inherently more homogenous single-branched N-glycan ‘stumps’.

Unnatural saccharides, which already include handles for conjugation, can also be installed on N-glycans. A first approach is to incorporate them metabolically into antibodies during production. A prerequisite for this concept is that the native enzyme allows modifications to the saccharide donors they use as substrate for glycosylation. FUT8 for instance (Okeley et al., 2013b) allows the C6 of Fuc to be extended with a thiol. Providing 6-thiofucose as additive during production results in core-fucosylation onto which maleimide-based drugs can be attached. A drug to antibody ratio (DAR) of 9.3 and 1.3 can be achieved, depending on whether the antibody’s interchain disulfide bonds are reduced or not prior to the conjugation. To install fully orthogonally reactive tags, aldehyde- and alkyne-containing Fuc analogs can also be supplemented, with 80% and 3% incorporation efficiency, respectively. Aside from metabolically incorporating modified α−1–6-Fuc in core-fucosylation, in a promising very recently reported methodology, functionalized fucose derivatives can also be installed in an α−1,3-linkage onto galactose residues by employing an engineered fucosyltransferase from H. pylori in vitro (Yang et al., 2022). The authors first used the native biantennary partially galactosylated N-glycans as substrate. Yet, this reaction is inefficient and needs 16 h to reach completion. The reaction rate improves after trimming down the N-glycan with EndoS, thereby freeing up space in the CH2 cavity for subsequent enzymatic incorporation of β−1,4-Gal. This LacNAc structure is then an efficient substrate for the engineered fucosyltransferase that takes synthetic GDP-fucose analogs in which the Fuc is functionalized. On this N-glycan LacNAc structure, in reactions that can be as fast as five minutes, homogenous derivatized-fucose transfer occurs with fewer equivalents and with a wider variety of labels. Nevertheless, by removing the biantennary nature of the N-glycan, DAR of the final conjugate is reduced. As a solution, the number of toxins per conjugation site can be increased by using branched linkers (X. Zhang et al., 2021).

Another unnatural saccharide, sialic acid modified with an azido functionality, can be incorporated in vitro and in cellulo with wild type sialyltransferase ST6Gal-I (Luchansky and Bertozzi, 2004). For cellular azido-sialylation, CHO cells can be supplemented with N-azido-acetylmannosamine (ManNAz), a precursor for neuraminic acid which has a modification at the C2 position. Even though the compound is peracetylated to increase membrane permeability, the medium needs to be carefully monitored to reduce scavenging of neuraminic acid and its precursors from glycoproteins. In contrast, in vitro manipulation of antibodies (Li et al., 2014) with sialyltransferase and CMP-activated neuraminic acid analogues can be done in a more controlled fashion. Next to C2 functionalization, the C9 position can also accommodate azides onto which toxins can be conjugated. Li et al. found that ST6Gal I shows more favorable properties than ST3Gal IV, yielding antibodies with 4.3 sialic acids installed. However, full prior in vitro galactosylation with UDP-Gal and GalT was required and the in vitro protocol is only completed after several days.

N297 glycan branch galactosylation can also be engineered for coupling purposes. The β4Gal-T1 mutant Y289L transfers GalNAc onto GlcNAc and Glc (Ramakrishnan and Qasba, 2002), as efficiently as it transfers Gal. Moreover, the mutant is capable of handling C2 functionalized Gal analogs, such as C2-keto-galactose and azide-modified GalNAz (Boeggeman et al., 2007; Qasba et al., 2008). When applied on biantennary N-glycans, the reaction can be driven to near completion (e.g. to bigalactosylated product) but only when using high concentrations of UDP-Gal and enzyme, as the enzyme much prefers the branch (Ramakrishnan et al., 2011; Ramasamy et al., 2005).

A drawback of targeting the N-glycan building blocks as handles for conjugation is linked to the intrinsic heterogeneity of glycosylation. To generate conjugates efficiently, the antibody substrate needs to be treated to convert the majority into suitable glycoforms. For instance, to productively label monoclonals with GalT_Y289L, the main N-glycan is supposed to be G0 (F), requiring galactosidase treatment to convert the G1 (F) and G2 (F) glycoforms to G0 (F). Additionally, antibodies are also known to carry HM and Hy type glycans, which will not be substrates for in vitro galactosylation. Hence, a solution has been sought to remove this heterogeneity and to produce a simple conjugatable N297 glycan stump. ENGases that can remove the entire N-glycan but the first GlcNAc have been instrumental for this. EndoA for example hydrolyses HM and Hy type glycans, while EndoS allows biantennary complex type glycosylated IgG as substrate. EndoS2 is the most lenient as it removes the N297 glycan on IgG regardless of glycan structure (Klontz et al., 2019; Trastoy et al., 2018). After hydrolysis of the N-glycan, which generates a single GlcNAc, two approaches can be implemented. Firstly, the GlcNAc stump can be galactosylated with GalT_Y289L to incorporate C2 modified Gal residues (Sadiki et al., 2020; van Geel et al., 2015). This approach has been successfully applied to several antibodies by the startup Synaffix (Oss, The Netherlands), with three phase I clinical trials ongoing (Wijdeven et al., 2022), indicating that issues with scalability, more specifically the enzyme production and manufacturing of UDP-GalNAz, can be overcome. GlycoDelete technology, in which (Sia)LacNAc glycan stumps are produced already during expression of the glycoprotein, together with highly cost-effective periodate or galactose oxidase functionalization, not requiring any expensive (functionalized) sugar nucleotide, also offers substantial potential for achieving similarly structured antibody-drug conjugates (Breedam et al., 2021).

In a second concept starting from endoglycosidase-generated single-GlcNAc N297 N-glycans, an endoglycosidase extends the GlcNAc stump with a semisynthetic N-glycan carrying an orthogonal functionality. To drive this reaction, the glycan-donor is activated as an oxazoline. Approaches implementing this second concept can be classified as either one-pot reactions or sequential reactions, depending on whether the transglycosylated product can be hydrolyzed by the enzyme used. In a sequential approach, the antibody is first trimmed by a wild type ENGase to generate a single GlcNAc stump. After purifying the IgG, a mutant enzyme with transglycosylation activity, but without hydrolyzing activity, will install the glycoform with orthogonal reactivity. An example of this two-step approach employs EndoS_D233Q (Guo et al., 2017; Parsons et al., 2016; Tang et al., 2016; Wang et al., 2019), a mutant with reduced hydrolysis activity. However, this approach is not the most efficient, requiring larger amounts of oxazolines which increases the likelihood for glycation reactions onto the protein backbone. While this issue can be resolved by having smaller amounts of donor material being added several times, it is not as optimal as the approach using EndoS2. Its hydrolysis-incapable mutant EndoS2_D184M can be used to transfer modified glycoforms onto proteins. Due to its higher specific activity, the transglycosylation is completed faster and with less oxazoline (Ou et al., 2021). However, that high activity is also present in wild type EndoS2, requiring that all traces of wild type enzyme, used to generate the GlcNAc stump, are removed.

This is no issue when Endo S2 is used in one-pot reactions with oxazolines carrying mannosylated GlcNAc or galactosylated GlcNAc (Shi et al., 2022; X. Zhang et al., 2021), generating trisaccharides which are not hydrolysable. For both substrates, the orthogonal functionalities are incorporated in the terminal saccharide, respectively mannose and galactose. For Gal, the authors only tested azido incorporation at the C6 position, while up to six azides can be installed onto the mannose residue, using the C3, C4 and C6 positions and branched linkers. The latter therefore allows conjugates of DAR 12 with a 95% productivity yield.

• IgG monoclonal antibodies are an expanding class of therapeutics.

• Fc N-glycosylation at Asn297 fine-tune effector functions of IgG antibodies.

• Fc N-glycosylation at Asn297 impact the efficacy/stability of IgG antibodies.

• Fc glycoengineering strategies were developed to modulate IgG effector functions.