Engineering yeast for producing human glycoproteins: where are we now?

Summary

Yeast has advanced as an alternative for mammalian cell culture for the production of recombinant therapeutic glycoproteins. Engineered yeast strains not only allow to mimic the human N-glycosylation pathway but also specific types of human O-glycosylation. This is of great value for therapeutic protein production and indispensable to determine the structure-function relationships of glycans on recombinant proteins. However, as the technology matures, some limitations have come up that may hamper biomedical applications and must be considered to exploit the full potential of the unprecedented glycan homogeneity obtained on relevant biopharmaceuticals. In this special report, we focus on the recent developments in N- and O-glycosylation engineering in yeasts of industrial importance, to produce recombinant therapeutics with customized glycans.

Introduction

Expression hosts for humanized therapeutic glycoproteins

The current market for recombinant bio-therapeutics comprises over 300 products, with sales exceeding $100 billion and an expected annual growth of 19% [1]. The majority of the products are monoclonal antibodies, hormones and growth factors of which more than two-thirds are glycoproteins. Protein glycosylation is important for proper folding and it can greatly influence pharmacokinetic and pharmacodynamic properties.

N-glycosylation is the covalent attachment of an oligosaccharide to selected Asparagine (N) residues within the sequence N-X-S/T of proteins passing through the secretory pathway of eukaryotes. In addition, O-glycosylation occurs on Ser/Thr residues and is likely dictated by structural determinants [2]. However, the details of this process are still unclear. Glycosylation patterns on proteins from cultured mammalian cells closely resemble those found on human endogenous proteins. Therefore, the majority of biopharmaceuticals are currently being produced by mammalian expression hosts such as Chinese Hamster Ovary cells (CHO) [3].

However, the glycosylation pattern of mammalian hosts is typically very heterogeneous and can be influenced by many factors during production, with potential for considerable batch-to-batch variation [4]. In general, stringent process control is key to minimize variability [5]. Moreover, mammalian expression hosts are costly, often yield low protein titers, require long fermentation times and are prone to viral infections. These drawbacks have spurred incentives to develop alternative hosts such as yeast. Yeasts are attractive because they can grow to high cell densities at very large scale in relatively inexpensive media, are capable of secreting large amounts of protein and perform most post-translational modifications.

The N- and O-glycosylation pathway of mammalian cells versus yeast and fungi

The initial stages of the N-glycosylation pathway are well conserved in almost all eukaryotes. In the Endoplasmic Reticulum (ER), a lipid linked oligosaccharide (LLO) precursor is assembled and the oligosaccharyltransferase complex (OST) transfers this Glc₃Man₉GlcNAc₂ oligosaccharide to the nascent glycoprotein. Upon partial deglucosylation, the N-glycans can bind to the ER-folding factor Calnexin, which can be critical for folding of the glycoprotein at an enhanced rate. Depending on the folding state of the glycoprotein, the N-glycan is further modified [7]. When exiting the ER, the correctly folded glycoproteins mostly carry a Man₈GlcNAc₂ N-glycan that will undergo further species- and even cell type-specific processing (Figure 1).

Figure 1. Endogenous N-glycosylation in the ER.

The first step in N-glycosylation is the formation of a dolichol-linked oligosaccharide composed out of 14 monosaccharides which are added stepwise. After two GlcNAc residues and five mannose residues are added on the cytosolic side of the ER membrane, the resulting dol-PP-GlcNAc₂Man₅ is flipped to the ER lumen. There it gets elongated with four more mannose and three glucose residues. The dol-PP-GlcNAc₂Man₉Glc₃ precursor is transferred en bloc to an asparagine residue in the N-X-S/T consensus sequence of the nascent protein by the oligosaccharyltransferase (OST). After transfer, the glycan is deglucosylated by the glucosidases GLS-I and GLS-II. These glycans then participate in protein folding quality control. When the protein is not folded properly, the N-glycan gets glucosylated again by glucosyltransferases and will again pass through the quality control. Upon correct folding, one alpha-1,2-linked mannose will be removed by Mannosidase-I (Man-I) and the glycoprotein will exit the ER. Figure adapted from [34].

Upon entering the Golgi of higher eukaryotes, the Man₈GlcNAc₂ structure is first trimmed to a Man₅GlcNAc₂ intermediate. After the addition of a first N-acetylglucosamine (GlcNAc) residue and the removal of two more mannose-residues, the newly exposed N-glycan termini can be further modified with one or more GlcNAc residues that can be sequentially substituted with galactose (Gal) and sialic acid (Sia). This results in so-called 'complex type' N-glycans, which can be highly branched and further modified with fucose and other cell-type specific modifications (Figure 2, panel A). As glycoproteins progress through the secretory apparatus, different glycosyltransferases can act on the glycoprotein. Because the expression of these glycosyltransferases can strongly be influenced by culture conditions and that they can also compete for the same substrate, a myriad of glycoforms is created. This can explain the heterogeneity observed in mammalian expression hosts [8]. For a glycoprotein’s evolved role in its natural context, glycan heterogeneity is evidently not a ‘problem’. However, this is quite different from a biotechnology/molecular medicine perspective, in which we attempt to modify glycoproteins with the specific glycan structures that impart specific therapeutic functionality and easy manufacturing and post-manufacturing characterization. For these purposes, highly homogenous glycosylation is desirable.

Figure 2. Glycan processing in the Golgi apparatus.

A. Mammalian cells. In the mammalian Golgi, the Man₈GlcNAc₂ structure is shortened to Man₅GlcNAc₂ through the action of Mannosidase-I (Man-I). This is followed by the addition of a GlcNAc residue by N-acetylglucosaminyltransferase-I (GnT-I). Subsequently, one α-1,3 and one α-1,6-mannose residue is removed by Mannosidase-II (Man-II). The resulting GlcNAcMan₃GlcNAc₂ structure is further modified by several glycosyltransferases, resulting in a complex-type N-glycan containing galactose and sialic acid residues. B. In S. cerevisiae, Man₈GlcNAc₂ N-glycans imported from the ER are modified by the α-1,6-mannosyltransferase Och1p, which adds an α-1,6-mannose to the α-1,3-mannose of the trimannosyl core. This mannose initiates the formation of hyperglycosyl-type glycans by elongating the outer chain with α-1,6-mannose residues. This action is catalysed by two protein complexes, mannan polymerase-I (M-Pol-I) and mannan polymerase-II (M-Pol-II). The α-1,6-mannose backbone is further modified by the addition of α-1,2-mannoses, followed by mannosylphosphate residues and capping α-1,3-mannoses. Although core-type N-glycans do not often have outer chain elongation, they can be modified with α-1,2-mannose, mannosylphosphate and be capped with α-1,3-mannose. Figure adapted from [68].

In contrast, most fungi and especially the yeasts in common biotechnological use will further elongate the oligo-mannose Man₈GlcNAc₂ N-glycan in the Golgi with additional mannose and mannosylphosphate residues, leading to hypermannosyl-structures. This elongation is initiated by the Och1p (Outer Chain Elongation1) α-1,6-mannosyltransferase (Figure 2, panel B). Notably, following extension, S. cerevisiae can further incorporate α-1,3-mannose, whereas P. pastoris incorporates β-1,2-mannose residues, which both could be immunogenic [9,10]. Furthermore, the fission yeast Schizosaccharomyces pombe (S. pombe) which generates Man₉GlcNAc₂ instead of Man₈GlcNac₂ and also incorporates galactose in addition to mannose [11].

The most prevalent O-glycosylation type on mammalian proteins is initiated with a N-acetylgalactosamine (GalNAc) residue in the Golgi and is further modified with a range of different modifications and capped with several structures, sometimes yielding a bewildering complexity. In general, yeasts and fungi will exclusively perform O-mannosylation. After transfer of a single mannose to serine or threonine residues in the ER, the glycan will be extended with additional α-mannose residues in the Golgi. Further modifications (e.g. phosphomannosylation,β-mannosylation or galactosylation) resemble those seen on N-glycans and are again species-specific [12].

Wild-type fungal N- and O-glycans are ligands for mannose receptors such as the Macrophage Mannose receptor (MMR) or the ‘Dendritic Cell-Specific Intercellular adhesion molecule-3 Grabbing Non-integrin (DC-SIGN) found on antigen presenting cells. Extensive N- or O-mannosylation on proteins can enhance antigen presentation and subsequent T-cell stimulation. This has been exploited for yeast-derived recombinant vaccines [13,14].

Only few biopharmaceuticals with fungal N-glycosylation are commercially approved, such as recombinant Granulocyte Macrophage Colony Stimulating Factor (Leukine^®, Sanofi) which is produced in S. cerevisiae and recombinant Interferon-α2b (Shanferon^®, Shantha Biotech/Sanofi) that is manufactured in P. pastoris. Although the latter molecules create precedents to launch yeast-derived bio-therapeutics, glycans on therapeutic glycoproteins produced in yeast would generally need to be converted to a structure that is function-neutral or -optimized before they can be administered to patients [15].

Engineering Human-like N-glycans for biotherapeutics

N-glycosylation humanization (i.e. the incorporation of the mammalian-type N-glycosylation pathway) has been performed mostly in S. cerevisiae and P. pastoris but is also pursued in Hansenula polymorpha, Yarrowia lipolytica and Ogataea minuta [16]. All of the approaches to reconstruct the mammalian N-glycosylation pathway focus on generating the appropriate substrate for heterologously expressed mammalian glycosyltransferases by generating the mammalian Man₅GlcNAc₂ intermediate (Figure 3, panel A) or by tailoring a Man₃GlcNAc₂ precursor by intervening in the lipid linked oligosaccharides (LLO) biosynthesis (Figure 3, panel B).

Figure 3. Glyco-engineering of N-glycans in yeast.

A. Preventing hypermannosylation. To eliminate elongation of the outer chain, an α-1,2-mannosidase, which trims the α-1,2-mannose residues, is targeted to the ER or early-Golgi. Disruption of the OCH1, MNN4 and MNN1 genes prevents further elongation of the N-glycan to hypermannose structures. Further modification to human like N-glycans can be obtained by introducing several glycosyltranferases. B. Production of phosphorylated N-glycans. To increase the degree of phosphorylation on a Man₈GlcNac₂ glycan, Mnn4p or its homologous can be overexpressed in a Δoch1-strain. To obtain efficient endocytosis of the recombinant protein via the CI-M6PR, “capping” mannoses and terminal α-mannoses must be removed. This is achieved by treating the N-glycans with glycosidases produced by Cellulosimicrobium cellulans (CcGH92_5, CcGH92_4). C. Interference in LLO-biosynthesis. After deletion of the gene encoding the mannosyltransferase Alg3p, an ER-type Man₅GlcNAc₂ is transferred by the oligosaccharyltransferase complex (OST) to the nascent protein. Knock-out of the OCH1 gene is necessary since the ER-type Man₅GlcNAc₂ can also be a substrate for OCH1p. To generate substrate for GnT-I, a heterologously expressed α-1,2-mannosidase can trim the ER-type Man₅GlcNAc₂ to Man₃GlcNAc₂ before transfer to the Golgi. This N-glycan can then be further modified by the addition of GlcNAc and galactose residues. Another approach is to generate Man₃GlcNAc₂ by the deletion of ALG11 in addition to ALG3. To increase glycosylation efficiency in this Δalg3Δalg11 mutant, flipping of the truncated LLO across the ER membrane can be improved by overexpressing Rft1p or Flc2p* in H. polymorpha and S. cerevisiae respectively. A protozoan oligosaccharyltransferase (POT) which has a relaxed substrate specificity can further improve the transfer of the truncated LLO to a folding protein.

Preventing hypermannosylation

A key step in hyperglycosylation of yeast N-glycans is the addition of the first α-1,6-mannose by Och1p. In theory, disruption of the OCH1 gene would lead to a more homogenous N-glycan profile. But in practice, additional mannosyltransferases can act on the core N-glycan. Therefore, a triple mutant (Δoch1Δmnn1Δmnn4) devoid in phosphomannosyltransfer and α-1,3-mannosyltransferase activity was necessary to obtain more or less homogenous Man₈GlcNAc₂ N-glycans in S. cerevisiae, but resulted in impaired growth [17–19]. This was later compensated for by random mutagenesis and screening for improved growth. The selected strains had an altered cell wall- and plasma membrane composition and also expressed more protein compared to the parental Δoch1Δmnn1Δmnn4 strain [20]. Disruption of the OCH1 locus in P. pastoris yields Man₈GlcNAc₂ N-glycans next to other larger structures. To achieve a more homogenous Man₈GlcNAc₂ glycosylation, additional knock-outs in genes important for phosphomannosylation (Δpno1 and Δmnn4B) and β-mannosylation (Δbmt2) were necessary. Fortunately, this did not affect growth or strain productivity [21]. In Yarrowia lipolytica, it was found that Och1p and Mnn9p act sequentially during the initiation of hyperglycosylation. Nevertheless, the Δoch1 mutant could be used to obtain sufficiently homogenous N-glycans upon further engineering [22]. A similar approach in S. pombe is problematic and requires knock-out of 10 galactosyl-transferase genes in addition to OCH1. Knocking-out the Golgi UDP-galactose transporter (gms1⁺) is sufficient to eliminate galactosylation but of course precludes the introduction of human-type galactosylation [23,24].

The next step is conversion of Man₈GlcNAc₂ to Man₅GlcNAc₂, a key intermediate for the production of hybrid or complex N-glycans. In mammalian cells, several Mannosidase-I (Man-I) enzymes trim the remaining α-1,2-linked mannose residues of Man₈GlcNAc₂. Remarkably, overexpression of an ER-retained α-1,2 mannosidase from Trichoderma reesei (T. reesei) to mimic the Man-I activity in P. pastoris was largely sufficient to convert the heterogeneous N-glycosylation of the Δoch1-strain to a useful Man₅GlcNAc₂-profile, without the need for additional gene knock-outs [25]. This is likely because the mannosidase rapidly removes the α-1,2-linked mannose residues before endogenous glycosyltransferases can act. In addition, an ER- to cis Golgi-targeted Mannosidase-1B from C. elegans fused to the ScMns1p targeting signal was also successful to obtain Man₅GlcNAc₂ in P. pastoris [26]. Similar approaches, using ER-targeted fungal α-1,2-mannosidases were also reported for S. cerevisiae, H. polymorpha and Y. lipolytica [19,22,27].

In mammalian cells, the Man₅GlcNAc₂ structure is modified by N-acetylglucosaminyltransferase-I (GnT-I) to GlcNAcMan₅GlcNAc₂. To generate such hybrid N-glycans in P. pastoris, heterologous expression of a ScKre2p-leader fused to human GnT-I (hGnT-I), targeting the cis-to-medial Golgi compartments, was sufficient for Vervecken et al. to obtain near quantitative conversion to GlcNAcMan₅GlcNAc₂ [25]. However, in other studies using a cis-Golgi targeted ScMnn9p-hGnT-I fusion, an additional Golgi UDP-GlcNAc transporter from Kluyveromyces lactis (K. lactis) was necessary to ensure full conversion [26,28]. In order to generate bi-antennary complex-type sugars, first the α-1,3- and α-1,6-mannose residues of the hybrid N-glycan must be removed by Mannosidase-II (Man-II). Next, a β-1,2 linked GlcNAc must be transferred by N-Acetylglucosaminyltransferase-II (GnT-II). To replicate this in P. pastoris, a combinatorial library was set up to select for properly targeted enzymes, to achieve efficient complex-type N-glycosylation. The highest GlcNAc₂Man₃GlcNAc₂ homogeneity was obtained when Man-II and GnT-II (from Drosophila melanogaster and Rattus norvegicus respectively) were fused to ScMnn2p for targeting the medial-Golgi [28]. Interestingly, introduction of Man-II alone seems to hamper growth and increases N-glycan heterogeneity. However, this could be largely resolved by the introduction of GnT-II. Presumably, the GlcNAcMan₃GlcNAc₂ product of Man-II is a substrate of unknown glycosyltransferases, thereby interfering with yeast physiology.

Human complex-type N-glycans are usually galactosylated by β-1,4-galactosyl transferase (GalT). However, for efficient galactosylation in P. pastoris, it was necessary to fuse human GalT-I to a UDP-Galactose 4-epimerase from Schizosaccharomyces pombe (S. pombe) to generate a sufficiently large UDP-galactose pool in the Golgi for GalT to function. By generating a triple fusion with the ScMnn2p leader-sequence, bi-antennary galactosylated N-glycans could be produced [29]. This type of N-glycan is of particular interest for monoclonal antibodies (mAbs) to improve antibody dependent cellular cytotoxicity (ADCC) [30]. To this end, Pichia glyco-engineered anti-HER2 (human epidermal growth factor receptor-2) was compared with Trastuzumab (Herceptin^®), now used for treatment of breast cancer. It was shown that the glyco-engineered product from P. pastoris had similar in vitro and in vivo properties as the CHO-produced counterpart [15,31][101]. Recently, certain components of this engineering, known as GlycoSwitch^® technology have become commercially available [32].

For other, non-IgG therapeutics, introduction of the human sialylation pathway may be useful in the case where sialylation is needed to improve serum half-life of recombinant therapeutics. This requires the introduction of CMP-Neu5Ac biosynthesis, transport of sialic acid to the Golgi and transfer to the acceptor N-glycan. Researchers at Glycofi (USA) succeeded in incorporating the five steps of this pathway in a single vector. When transformed to P. pastoris, this resulted in the synthesis of a sizeable portion of bi-antennary sialylated N-glycans on recombinant human Erythropoietin (rhuEPO) (Figure 3. Panel A and Table 1). Reportedly, the growth characteristics and the production levels of such glyco-engineered strain did not differ from wild-type P. pastoris, at least at small scale [21,33] [102].

Table 1. Yeast N-glycosylation engineering.

This table provides details of selected strains that have been described in the manuscript and some of the most important features of the engineering process (e.g. knock-outs, heterologous glycosyltransferases, targeting signals, promoters and selection markers). All strains appear in the order in which they are first mentioned.

¹ Strains YGLY1703, YGLY 3159 are also mentioned in [33] but these are intermediates of the final strain as shown in the table. For their precise genotype we refer to WO/046855 (2011) [102].

* The URA-marker was used and recovered (marker-rescue) for further engineering.

Abbreviations used: (n.a.) not applicable, (n.d.) not disclosed, (S.c.) Saccharomyces cerevisiae, (CSS) CMP-sialic acid synthase, (GNE) UDP-GlcNAc 2-epimerase/N-acetylmannosamine kinase, (SPS) N465 acetylneuraminate-9-Phosphate synthase, (CST) CMP-sialic acid transporter, (α-2,6-siaT) α-2,6-sialyltransferase catalytic domain. (Sia) Sialic acid, (Gal) Galactose, (GlcNAc) N-Acetylglucosamine, (Man) Mannose, (ManPi) Mannose-phosphate.

Currently, a variant of rhuEPO that contains two additional N-glycosylation sites is produced in CHO-cells (Darbepoetin, Aranesp^®). This engineered form is modified mainly with tri- and tetra-antennary sialylated N-glycans. Pre-clinical research demonstrated that glyco-engineered rhuEPO produced in Pichia had superior in vitro activity but was cleared faster in vivo than Aranesp^®. The lower in vivo stability could be counteracted by PEGylation, leading to nearly identical in vivo and in vitro characteristics as Aranesp^® [33].

However, it should be realized that genetic stability of such a heavily engineered strain can be rather difficult to maintain. Moreover, recent work has illustrated that minor non-targeted N-glycan species can often still be present (Laukens et al., Manuscript in preparation). These result from the many generated pathway intermediates non-native to yeast. An exhaustive approach to eliminate these glycoforms, by identifying and inactivating the responsible glycosyltransferases, is impractical. Therefore, simple approaches to solve this problem would be of great interest.

Intervening in the LLO-biosynthesis

In an alternative approach to N-glycan engineering, one can interfere in the ER-associated lipid linked oligosaccharide (LLO) biosynthesis pathway. A tailored (ER-type) Man₅GlcNAc₂ LLO is transferred to nascent glycoproteins where it can act as a direct substrate for heterologously expressed glycosyltransferases. Consequently, there is no need to introduce mammalian mannosidase-II. In practice, deletion of the Dol-P-Man:Man₅GlcNAc₂-PP-Dol α-1,3-mannosyltransferase (ALG3) that catalyses the addition of an α-1,3-linked mannose residue to Man₅GlcNAc₂-PP-Dol intermediate abrogates the LLO-biosynthesis on the luminal side of the ER causing the transfer of this intermediate to be transferred to the nascent glycoprotein (Figure 3, Panel C). However, the Δalg3 (ER-type) Man₅GlcNAc₂ is not efficiently glucosylated in the ER and hence not efficiently transferred to proteins by OST. This was tackled in Y. lipolytica by improving LLO-glucosylation [34].

Both in S. cerevisiae and Y. lipolytica, Och1p shows only low activity towards the Δalg3 Man₅GlcNAc₂ [22,35]. However, in P. pastoris and H. polymorpha, several high molecular weight structures were observed, most likely due to a more relaxed substrate specificity of OCH1p, thereby necessitating OCH1 knock-out for further engineering [35,36]. Although N-glycans on glycoproteins of Δalg3 mutants are predominantly Man₅GlcNAc₂, some distinct glucosylated structures were also found on secreted glycoproteins in Δalg3 mutants of P. pastoris, S. cerevisiae and Y. lipolytica, but less in H. polymorpha [35–37]. The remaining glucosylation, as seen in Y. lipolytica, could be efficiently removed by the heterologous expression of an ER-targeted glucosidase-II from Aspergillus niger [34].

To generate the Man₃GlcNAc₂ substrate out of the ER-type Man₅GlcNAc₂, an ER-targeted α-1,2 mannosidase can remove the remaining α-1,2-mannose residues on the Man₅GlcNAc₂ N-glycans. Further engineering is then possible because of the somewhat relaxed substrate specificity of GnT-I, which is also capable of using Man₃GlcNAc₂ as substrate in vitro [38,39]. Such GlcNAcMan₃GlcNAc₂ N-glycans were obtained in H. polymorpha by using an ER-retained α-1,2-mannosidase from Aspergillus saitoi in combination with a ScMnn9p-leader fused to hGnT-I for correct targeting. In P. pastoris, an ER- to cis-Golgi-targeted ScSec12p-mouse Mannosidase-I fusion construct in combination with a ScMnn9p-hGnT-I proved also functional to generate hybrid N-glycans. However, only for P. pastoris have there been reports about further engineering leading to the production of highly homogenous bi-antennary Gal₂GlcNAc₂Man₃GlcNAc₂ N-glycans [29,35,40]. Although the strain productivity for this latter strain was not affected, there was a slight growth defect [36]. Recently, homogenous Man₃GlcNAc₂ N-glycans were also obtained in Y. lipolytica by expressing an ER-retained α-1,2-mannosidase from T. reesei, but no further engineering was reported so far [34].

Further tuning of LLO-biosynthesis by deletion of ALG11 in addition to ALG3 prevents the assembly of the Man₅GlcNAc₂ precursor at the cytoplasmic side of the ER during LLO-biosynthesis (Figure 3, panel C). Consequently, such Δalg3Δalg11 mutants mainly produce Man₃GlcNAc₂ N-glycans on secreted glycoproteins [41,42]. However, this deep interference with LLO-biosynthesis often results in N-glycosylation site under-occupancy due to reduced flipping of the modified LLO across the ER membrane. This could be solved by overexpressing Flc2p or Rft1p, which are both involved in translocation of the LLO-across the ER-membrane, in S. cerevisiae and H. polymorpha respectively [41,42]. Further engineering of such strains in H. polymorpha has recently led to reporter proteins modified with up to 50% bi-antennary Gal₂GlcNac₂Man₃GlcNAc₂ [43].

Just recently, protozoan oligosaccharyltransferases (POTs) have been used to improve truncated LLO-transfer to glycoproteins in P. pastoris and S. cerevisiae [44,45] (Figure 3, panel C). Moreover, overexpression of a non-specific flippase, in combination with a POT could boost the N-glycosylation efficiency up to 100% in a Δalg3Δalg11 strain of S. cerevisiae. Unfortunately, further attempts to engineer this strain gave quite heterogeneous N-glycan profiles [46]. An overview of some selected strains is given in Table 1.

Phosphorylated N-glycans for Enzyme Replacement Therapy

Lysosomal storage diseases are hereditary deficiencies of lysosomal catabolism, causing the toxic accumulation macro-molecules into the lysosomes. Enzyme Replacement Therapy (ERT) is used to deliver the recombinant enzymes to the lysosomes by targeting the cation-independent mannose-6-phosphate receptor (CI-M6PR) that is present on the plasma membrane of many human cells. Efficient binding and internalisation requires high levels of mannose-6-phosphate (Pi-6-Man) to be present on the N-glycans of the therapeutic enzymes. Currently, these enzymes are produced in mammalian cells, which often results in only limited modification with phosphorylated glycans. Yeasts naturally substitute their N-glycans with Man-Pi-6-Man, which is a good starting point for further engineering [47]. To this end, an Δoch1 strain can be used to restrain yeast-specific N-glycan elongation, thus obtaining a high proportion of the eukaryote-common Man₈GlcNAc₂ N-glycan backbone. In order to increase the phosphorylation degree, the genes involved in mannosylphosphorylation can be overexpressed (Figure 3, panel B). In early work in the model organism S. cerevisiae, Mnn4p was identified as the yield-determining factor in mannosylphosphorylation. When Mnn4p is overexpressed, up to three times more N-glycans can be mannosylphosphorylated. These findings were confirmed for the ortholog in O. minuta (MNN4), Y. lipolytica (MPO1) and P. pastoris (PNO1) [48–50]. Optimization of the gene copy number in this approach allows for >80% of the N-glycans to carry one or two mannose-Pi-6-man glycotopes [50] (See also Table 1).

A major challenge then was the development of a methodology to remove the mannose “cap” on the phosphate to expose the Pi-6-Man ligand for the CI-M6PR. Whereas Chiba et al discovered such enzymatic activity in the secretome of a bacterium isolated from soil, we could identify, clone and recombinantly express a suitable “uncapping” enzyme from Cellulosimicrobium cellulans (Arthrobacter luteus) called CcGH92_5 [50,51]. Adventitiously, we discovered that also the terminal α-mannoses on the Pi-6-man carrying glycan branches need to be removed for efficient CI-M6PR-mediated enzyme endocytosis. This can be achieved with a second α-mannosidase from C.cellulans (CcGH92_4), or with Jack bean α-mannosidase (Figure 3, panel B). These developments now allow for the production of recombinant lysosomal enzymes with superior Pi-6-Man substitution in yeast.

O-glycosylation engineering for biotherapeutics

As fungal O-glycans are structurally different from the main human type O-glycans, there is some concern with regard to potential immunogenicity. Therefore, fungal O-glycosylation has also been an engineering target in the field of bio-therapeutic development. Both in P. pastoris and S. cerevisiae, the genes encoding protein-O-mannosyltransferases (Pmt’s) were identified, of which Pmt1p and Pmt2p are the main O-mannosyltransferases. The elimination of fungal O-glycosylation is difficult because knock-out of PMT1 or PMT2 drastically impacts viability and growth [52,53]. Eliminating the target-protein’s O-glycosylation sites is also troublesome because there is no known consensus sequence for O-glycosylation, making it difficult to predict the sites in advance [2]. Moreover, such approach entails a change in the protein sequence, again raising immunogenic potential. However, if the therapeutic application allows such O-glycosylation site elimination, it is still by far the most preferred approach. In practice, close attention is needed, as any heterologously expressed protein can be O-glycosylated. Mass spectrometric analysis of the purified proteins is currently the most informative method to detect such modifications.

O-glycosylation inhibitors

A number of small molecule inhibitors against Candida albicans Pmt’s have been designed [54]. These inhibitors were also tested successfully in P. pastoris and S. cerevisiae [55,56]. Similar to Δpmt mutants, the Pmt-inhibitors not only cause a reduction in O-glycan site occupancy but also reduce the O-glycan length [53]. In addition, yeast strains can be grown in the absence of inhibitor, only adding the compounds during the protein production phase. Therefore, Pmt-inhibitors allow to circumvent the slow growth phenotype seen in Δpmt-strains [53]. However, this strategy can only reduce O-glycosylation, it does not fully eliminate it.

Re-engineering O-glycosylation

Pichia pastoris strains devoid of β-mannosylation and phosphomannosylation have O-glycans only consisting of α-1,2 linked mannose residues. The O-glycan chain can then be trimmed down to a single O-linked mannose by in vivo expression of a secreted T. reesei α-1,2-mannosidase. Full O-glycan removal can then be achieved by in vitro digestion with Jack bean mannosidase (Figure 4, panel A). However, this requires that the single O-linked mannose is sterically accessible [57]. Unfortunately, such enzymatic downstream processing adds a lot of complexity and cost to the manufacturing process and would likely only be feasible for very high-value pharmaceuticals [58]. On the other hand, the single O-linked mannose residue could also be used for further engineering. Co-expression of a medial-(ScMnn2p) or late-Golgi (ScMnn6p)-targeted mouse protein O-linked mannose β-1,2-N-acetylglucosaminyltransferase 1 (MmPomGnT1), a N-acetyl-glucosaminyl transferase specific for O-glycans, results in a disaccharide which could be galactosylated and even sialylated to resemble the human dystroglycan-type O-mannosyl glycan [59] (Figure 4, panel A). The increased surface charge on such O-glycoproteins can also contribute to increased serum half-life [60].

Figure 4. In vivo O-glycosylation engineering in yeast.

Biosynthesis of O-mannosyl glycans in yeasts starts in the ER with the transfer of a single mannose from a dolichol-phosphate-β-D-mannose donor to a Serine or Threonine residue (S/T) of the acceptor protein. This is catalysed by members of the PMT-family. In the Golgi, the O-glycan will be extended.

A. In Pichia pastoris, O-glycans will consist of a linear chain of α-1,2-linked mannose residues that can be capped with phospomannose (Pi-Man) or with two consecutive β-1,2-mannose residues. In strains devoid of phospomannosyltransfer (Δpno1 and Δmnn4b) and β-mannosylation (e.g. Δbmt2), the O-glycans solely consist of α-1,2-linked mannose residues. Co-secretion of T. reesei α-1,2 mannosidase reduces the O-glycan down to the proximal mannose residue. On the one hand, this mannose residue can be removed by in vitro digestion with Jack Bean mannosidase yielding an unglycosylated protein [57]. On the other hand, the single O-linked mannose can be extended by Golgi-targeted expression of the human β-1,3 N-Acetylglucosaminyltransferase (POmGnT-I), GalT and the engineering of the sialylation pathway. This allows to mimic the dystroglycan-type O-glycan, found in mammalian hosts [59].

Panel B. Saccharomyces cerevisiae O-glycans can, in addition to phosphomannose (Pi-Man), also be modified with α-1,3-mannose but not with β-1,2-mannose. In order to mimic human O-glycosylation, initiation of endogenous O-mannosylation in the ER can be suppressed using a Pmt-inhibitor such as Rhodanine-3-acetic acid derivates (R3A). To mimick mucin core 1 type O-glycosylation, expression of the B. subtilis UDP-Gal/GalNAc epimerase and the human UDP-GalNAc transporter allows UDP-Galactose (UDP-Gal) and UDP-N-Acetylgalactosamine (UDP-GalNAc) to enter the Golgi. Next, Golgi-targeted human polypetide GalNAc-transferase (ppGalNAcTs) and core β-1,3 Galactosyl transferase (core β-1,3 GalT) sequentially transfer a GalNAc and Gal residue to the Ser/Thr residue of an acceptor protein before being secreted [65].

The introduction of O-fucosylation pathway required the biosynthesis of GDP-Fucose from GDP-mannose in the cytoplasm by heterologous expression of Arabidopsis thaliana MUR1 (AtMUR1) and AtFX/GER1. The GDP-fucose in the cytoplasm can be transported in the Golgi by an endogenous transporter. Next, introduction of Golgi-targeted human O-Fucosyl Transferase-1 (O-FucT-1) and a β-1,3-GlcNAc-transferase (Fringe) resulted in a fucose-β-1,3-GlcNAc disaccharide [61].

Sugar symbol nomenclature is indicated in the figure key. Linkages of individual sugars are indicated in the figure.

O-glycosylation engineering can be useful for structural studies on proteins where specific types of O-glycosylation are functionally important. To this end, the entire O-fucosylation pathway was engineered in S. cerevisiae by the heterologous expression of Arabidopsis thaliana GDP-mannose-4,6-dehydratase (AtMUR1), the GDP-4-keto-6-deoxy-mannose-3,5-epimersae (AtFXGER1) and human O-fucosyltransferase-1 (O-FucT-1), leading to O-fucosylation of the factor-VII EGF-domain. Further engineering with a human β-1,3-N-acetyl glucosaminyl transferase (Fringe) resulted in the elongation of the O-fucose with β-1,3-GlcNAc (Figure 4, panel B)[61,62]. Moreover, mimicking mammalian O-glycosylation could be useful to produce specific antigens (e.g. to then produce mAbs for the treatment of cancer) [63,64]. Mucin-type O-glycosylation was recently engineered in S. cerevisiae [65] (Figure 4, panel B). After suppressing the endogenous O-mannosylation, the synthesis of UDP-GalNAc by expressing the Bacillus subtilis UDP-galactose 4-epimerase (GalE) was introduced and its transport across the Golgi-membrane was engineered by expressing a human UDP-GalNAc transporter (UGT2). This was followed by the introduction of human polypeptide:N-acetylgalactosaminyl transferase (ppGalNAcT) and introduction of a human core1-β-1,3-galactosyl transferase (core 1 b1-3GalT), both fused to the ScMnn9p for localization in the cis-Golgi. This allowed to generate mucin-type glycoproteins modified with core1 O-glycans (Galβ1,3GalNAc1-O-Ser). No in vivo O-glycan sialylation was reported for S. cerevisiae so far, but this must be feasible as it was already done in P. pastoris [59].

However, it should be noted that so far, none of these approaches offer a complete solution to O-glycosylation re-engineering as none of them results in proteins completely devoid of yeast-specific O-glycan structures. Moreover, the experience with N-glycan engineering learns that the increasing complexity of the engineered pathway may lead to additional issues such as genetic stability, incomplete substrate to product conversions and interference by endogenous glycosyltransferases.

Future perspectives

There is an increasing interest to produce ‘bio-betters’ as well as to switch production of biopharmaceuticals to more economically-scalable and less time-consuming expression systems. Bio-betters are biopharmaceuticals with enhanced properties such as a more defined structure, improved pharmacokinetics and of course enhanced efficacy. For example, a glyco-engineered mAb has been launched with improved ADCC-activity [66]. Consequently, the unprecedented N-glycan homogeneity on carefully selected targets that can sometimes be obtained in glyco-engineered yeast strains, can hold promise for yeasts to produce part of the next-generation bio-therapeutics. For example, considerable progress was made in the development of better molecules for enzyme replacement therapy [50]. In addition, yeast is becoming ever more popular as an expression platform for recombinant vaccines, where glyco-engineering is also poised to play a role [67].

However, if the target glycans are too biosynthetically remote from yeast’s own glycosylation, such that their synthesis requires more than just a few heterologous enzymes, challenges remain when translating these developments from academic exercise to a robust manufacturing technology. This is especially the case for attempts at fully humanizing the N-glycosylation pathway to bi- or multi-antennary structures. As this glycosylation-type is the one that mammalian cells produce naturally, to be compellingly competitive, a yeast technology to produce mammalian-type N-glycans needs to produce these glycans in a homogenous fashion, devoid of any yeast-specific background structures. Especially the latter is still an active area of research [21,26,28,56].

Furthermore, we strongly feel that there is probably even more value in targeting glycan structures that mammalian cells do not efficiently make, to allow for the production of truly differentiated glycoprotein drugs with customized properties.

Several commercial ventures were launched early on in this field, based on the premise that copying the standard, complex human N-glycosylation pathway in yeast would yield compelling alternatives to the mammalian cells in current use. However, these ventures seem to be struggling at this time, likely due to the mentioned technical complexities as well as improvements in yield and glycosylation control in mammalian cell culture. For example, Merck acquired the Pichia glyco-engineering company Glycofi (USA) in 2006, but recently announced a shift to producing “classical” bio-similars rather than products from the Glycofi glyco-engineered yeast platform [66]. Similarly, Glycode SAS (France), a company focusing on S. cerevisiae glyco-engineering, is no longer active. By contrast, GlycART (Switzerland), a company focusing on changing Fc glycosylation of mAbs in mammalian cells to enhance ADCC-activity was acquired by Roche and the first products are now hitting the market. Glyco-engineering to achieve differentiated and application-specific customized products is clearly key to warrant future success. In this regard, the intermediates between yeast-type glycosylation and fully human glycosylation might turn out to be of higher value. For example, a GlycoSwitch^® Pichia pastoris strain that mainly produces the Man₅GlcNAc₂ structure in a robust way is now available off the shelf and allows for the production of glycoproteins with strongly reduced heterogeneity, which has important advantages in downstream processing and product characterization. It would be compelling to have a eukaryotic platform that delivers a homogenously N-glycosylated product, equalling the product homogeneity of unglycosylated products from prokaryotic hosts (such as E. coli). Therefore, further engineering approaches of yeast production hosts to deliver a single glycoform would be of great value.

Executive summary.

Expression hosts for humanized therapeutic glycoproteins

• Glyco-engineered yeasts can cope with the drawbacks of mammalian expression hosts for therapeutic protein production.

Engineering human-like N-glycans for biotherapeutics

• Different N-glycosylation engineering strategies can be successful depending on the yeast strain.

• Recombinant therapeutics produced in N-glycosylation engineered expression systems have advanced into pre-clinical development.

• Current technology for yeast N-glycosylation engineering can improve current enzyme replacement therapy.

O-glycosylation engineering for biotherapeutics

• O-glycosylation engineering is possible to some extent to avoid immunogenicity or for specific structure-function requirements.

Future perspectives

• Glyco-engineered yeasts are a valuable tool for the production of therapeutic glycoproteins with application-customized glycan structures.

• There is still room for innovative approaches to improve glyco-engineering of yeast.