It is now almost redundant to state that carbohydrates are crucial molecules in nature, taking part in a wide variety of biological process:^1,2  prevalent on cell surfaces, carbohydrates are involved in cell–cell signalling events and regulate the interaction with extracellular matrix; oligosaccharides modulate glycoprotein folding in the ER and are subsequently modified to achieve fully functional mature glycoproteins;^3,4  recognition of carbohydrates by lectins forms the basis for inflammatory and immune responses, as well as for the adhesion and infection processes of pathogenic microorganisms. Surprisingly therefore, the importance of glycan structure in biology has been underestimated in the past and this class of molecule is only now gaining recognition for its significant role on the crowded stage of scientific understanding (for context, see Fig. 1).

The level of diversity⁵  present in glycan structures can be overwhelming due to the intrinsic nature of carbohydrates. A detailed theoretical examination of all the possible monosaccharide combinations that can give rise to disaccharides through hexasaccharides estimated an astonishing 10¹² possible structures.⁶  Based on biosynthesis considerations, refinement of this number arrived at a projected number of mammalian glycans of just over than 3000, to which may be added other 4000 estimated glycosaminoglycans.⁷  So, a projected ∼7000 mammalian glycans overall, requiring overall the action of ∼700 enzymes for their biosynthesis.⁷  The study of such a complex set of carbohydrates, known as the “glycome”, and its relation to biological processes is referred to as “glycomics”;⁸  the latter has seen increasing attention in recent years thanks to the development of glycan microarrays,^9–12  nanoparticle^13–15  and biosensors,^13,16–19  and improvements in glycoanalysis techniques.^20,21  However, in the absence of the biological context – conjugated to protein or lipids, for instance – the true function of glycans may be lost. There is therefore a pressing need for efficient methods to prepare natural, homogenous glycoproteins, or structural mimetics thereof.

An ideal glycoconjugation method should fulfil a series of requirements. Very often either the protein or the carbohydrate moieties are not available in large quantities, so conjugation methods should be highly efficient to avoid the loss of valuable material. Related to this issue, the possibility of recovering the starting materials in a usable form is very much desirable. For glycoconjugation applied to glycoprotein synthesis the reaction conditions should ideally involve neutral pH, performed at room temperature, and generally mild so as not to disrupt the protein structure/activity. Most of the glycoconjugation methods used to date to generate non-native glycan–protein conjugates have relied on the random modification of one or more of the amino acid side chains of lysine, cysteine or tyrosine.

It is clear that glycoconjugates have a crucial role in fundamental biology, with scope for application as therapeutics.^17,22,23  Given the complexity and variety of carbohydrates, and the difficulties in obtaining homogeneous forms of glycans and glycoproteins from natural sources, efficient methods for the preparation of homogeneous glycoconjugates are key. With regard to natural glycoproteins,²  glycans may be connected to polypeptides via the amide bond of the side chain of Asn, (N-glycans),^24,25  via glycosidic bonds to the side chain of Ser or Thr (O-glycans),^26–28  or via an ethanolamine phosphate linkage that bridges a protein C-terminus to a glycolipid, as in glycosylphosphatidylinositol (GPI) anchors.²⁹  Recent progress in the field chemical synthesis of glycoproteins has been extensively reported,^30–32  highlighting powerful approaches to accurately control the site and nature of glycosylation. Recent example of homogeneous glycoproteins obtained by chemical means are, among the others, the total synthesis of EPO,^33,34  interferon β,³⁵  MCP-3,³⁶  homogeneous antifreeze glycoprotein,³⁷  the α and β subunit of human glycoprotein hormone.^38,39  In term of non-natural glycoconjugates, linkers that permit glycan coupling to the reactive side chains lysine and cysteine have traditionally been popular.^30,40  In addition, where chemical synthesis fails to afford reasonable yields, the synthesis of natural and unnatural oligosaccharides and their conjugation to proteins can be achieved by chemo-enzymatic methods, which have become a valuable approach to tackle difficult syntheses of less straightforward oligosaccharide conjugates.⁴¹ 

As summarised above, it is evident that the structural diversity and potential heterogeneity of natural glycoproteins makes deciphering structure–function relationships a challenging task – one that may be overcome by the preparation of homogeneous glycoproteins by means of chemical or biological methods, or more often these days by a combination of both. Likewise, efficient conjugation methods that are tolerant of diverse glycan and protein structures will open new avenues for conjugate vaccines.⁴²  The conjugation of carbohydrates to proteins, in particular, may be achieved by a range of approaches that have been extensively reviewed in the past.^30,43,44  Herein we illustrate a range of newer methodologies and their applications, with reference to some of the classic methods that provided the inspiration for continued method development.

The definition of a “glycoconjugate” is rather broad and comprises classes of molecules where a carbohydrate unit is covalently linked to another molecule, generally a protein. Conjugation of carbohydrates to proteins has been a challenging task, tackled in classic work performed by Pauly, which dates back to the beginning of the twentieth century, where he was able to couple glucose and galactose to serum globulin.⁴⁵  His seminal work inspired later work by Avery and Goebel in the 1930s,⁴⁶  which described the synthesis of diazonium salt derivatives of glucose and galactose and their conjugation to serum globulin for use in a series of immunological studies. The reaction of diazonium sugar derivatives with proteins is non-specific resulting in indiscriminate modification of tyrosine, histidine, lysine and, if used in large excess, with tryptophan and arginine.^47–50  Since then, efforts at improving the production of glycoconjugates have been unceasing.

Synthetic carbohydrates generally have the reducing end available for conjugation; on the other hand, carbohydrates from natural sources may have the reducing end available or not, depending on the nature of the glycan concerned. There are a variety of well-established reducing end conjugations that exploit the reactivity of the exposed aldehyde of ring-open reducing sugars towards amines.⁵¹  A classic approach involves the reductive amination of a reducing sugar with the amino group of lysine with cyanoborohydride,^11,52,53  giving secondary amines that are stable to hydrolysis. The aldehyde form of reducing sugars can also react with hydrazines, hydrazides or aminooxy-functionalised proteins to give hydrazones or oximes¹¹  (Fig. 2a). Where the reducing end of a glycan is not available for coupling, internal sugar units or the non-reducing terminus may be oxidised with periodate to generate aldehydes that can undergo reductive amination (Fig. 2b).

Alternatively, the reducing end or specific functional groups, such as carboxylic acid in uronic acids or amine in amino sugars, can be selectively modified by the use of heterobifunctional cross-linkers introducing a variety of functional groups^11,51  (Fig. 2c), such as amine, hydrazine or thiol, which can react with N-hydroxy-succinimide (NHS) ester- or maleimide-functionalised proteins. In the opposite sense, the carbohydrate can be functionalised with linkers carrying NHS-ester, para-nitrophenyl ester, squarate,⁵⁴  maleimide, isocyanate, isothiocyanate or acyl azide⁵⁵  (Fig. 2c–d) to react with lysine and cysteine of the protein, or with amine, thiol, or hydrazine-functionalised proteins.⁵¹  The hydroxyl groups of carbohydrate moiety can also be traditionally activated by cyanogen bromide (CNBr)⁵⁶  or carbonyldiimidazole (CDI)⁵⁷  to react with amino groups of the proteins and generate isourea and carbamate bonds, respectively (Fig. 2e–f). If the carboxylic groups on the sugar or proteins are free and not activated by NHS-ester, they can be activated in situ with water soluble carbodiimide,⁵⁸  forming an O-acylisourea active ester that can be attacked by amino groups forming an amide bond.

Historically, some of the most important glycan–protein conjugates investigated have been non-natural materials, created as potential anti-microbial vaccines. Vaccination has had a major impact in improving human health, eliminating various diseases, such as pneumonia, influenza, typhoid fever, pertussis, measles and smallpox, which have caused millions of deaths in the past.^59,60  Glycoconjugate vaccines are particularly effective in this context; while the Haemophilus influenza type B (Hib) capsular polysaccharide alone is poor at inducing an immune response, immunogenicity was triggered by conjugation of the polysaccharide to a carrier protein.⁶¹  In this instance, protein–glycan conjugation was achieved through the bifunctional linker adipic acid dihydrazide (ADH), which provides access to a protein acyl hydrazide by EDC coupling with glutamate or aspartate through one end; CNBr activation of the hydroxyl groups of the polysaccharide then allows isourea bond formation through the other end of the adipic acid linker, as illustrated in Fig. 3.

The Hib vaccines have been further improved by Verez-Bencomo et al.⁶²  using a synthetically prepared capsular polysaccharide, obtained with a controlled degree of polymerization, which enabled the production of a homogeneous glycoconjugate in a cost efficient manner. This example paved the way for the development of glycoconjugate vaccines against Neisseria meningitides, Salmonella typhi and Streptococcus pneumoniae.⁶³ 

Pozsgay et al.⁶⁴  have applied well-known oxime formation to obtain glycoconjugates. The approach consists of a two-step functionalization of BSA (Fig. 4): first, the protein is treated with the commercially available succinimidyl 3-(bromoacetamido)propionate 2.1. The conjugation yield was estimated to be 80–90%, with MALDI-TOF analysis confirming an average of 30–35 Lys residues functionalised. In a second step, the chemically modified BSA was reacted with a bifunctional linker containing thiol and aminooxy groups. With the aminooxy functionalized protein in hand, the authors successfully incorporated a range of l-rhamnose derivatives (2.7a–2.7d) bearing either an aldehyde or a ketone, including the keto-ribitol 2.8 and the keto-tetrasaccharide 2.5 (Fig. 4), which yielded the BSA glycoconjugate 2.6.

Lysine on BSA was also targeted for glycoconjugation by Kovac et al. by using a series of squarate derivative installed on Vibrio cholerae polysaccharide fragments,⁶⁵  of which an example (2.10) is shown in Fig. 5. Also in this case, the advantage of using squarate derivatives as 2.9 is that it can be used in excess and the unreacted material can be recovered.

A remarkable example of classic methodology has been recently reported by the Davis group, which relates to a new approach in antibacterial vaccines design.⁶⁶  The Hep₂KDO₂ tetrasaccharide in Fig. 6 was obtained by a new synthetic strategy starting from mannose; it was then coupled via the well-established isothiocyanate methodology to a diphtheria toxin mutant. The tetrasaccharide mimics a common inner core of bacterial capsular polysaccharide (CPS). Access to the inner core is usually prevented by the presence of the outer core polysaccharide, but using γ-cyclodextrin as a co-administered inhibitor of the outer polysaccharide transporter, results in exposure of the inner core part on the bacteria LPS. At this stage, the bacteria can be targeted by antibodies previously raised against the Hep₂KDO₂-functionalized DT-carrier protein. This innovative approach could pave the way to a new concepts for the design of anti-bacterial vaccines, avoiding the challenging task of decorating carrier protein with limited accessibility O-antigen polysaccharides.⁶⁷ 

Several others glycoconjugate vaccines are in Phase I–III trials, including against groups A and B streptococci, breast cancer, prostate cancer and HIV-1.⁶³ 

After its discovery in 2002, the copper-catalysed azide–alkyne dipolar cycloaddition (CuAAC) click reaction^68,69  has found widespread application in glycoscience^70–73  due to its broad tolerance of functional groups, offering high specificity and yield. It has found application in glycoproteins synthesis as well, for instance, in the transformation of the amine sidechain of lysine into an azide by treatment with imidazole-1-sulfonyl azide for subsequent CuAAC modification. Such methodology have been applied by Lipinski et al.⁷⁴  en route to the preparation of a vaccine against Candida albicans. Of the 40 surface amino groups present on the chicken serum albumin (CSA) surface, about half were converted into azide form (Fig. 7). Subsequent CuAAC reaction gave an average of 17 and 19 oligosaccharides per protein from sugar alkynes 3.1 and 3.2, respectively. In a first attempt towards the generation of the vaccines, the authors analysed these two different linkers construction in order to compare the influence of an unstructured versus a structured linker: the former appears to decrease undesired immunogenic responses against the non-carbohydrate portion of the conjugate.

Later work from the same authors generated tetanus toxoid glycoconjugate vaccines bearing two different oligosaccharides installed in orthogonal method on the carrier protein (Fig. 8).⁷⁵  Firstly, they transformed the exposed amine of lysine side chains of the carrier protein into azide, not only providing a handle for the CuAAC glycoconjugation, but also preventing the cross-coupling in the second step, where the carboxylic acid on the side chain of aspartic acid and glutamic acid were activated for the formation of an amide bond with an amino terminal functionalized β-mannan trisaccharide. The last step of the protocol was the CuAAC of an alkyne-laminarin derivative, followed by the re-conversion of the unreacted azide to amine. A comparison between the two vaccine candidates with and without the laminarin hapten confirmed a better efficiency of the former.

Functionalisation of protein lysine side chains was also achieved by Crotti et al.⁷⁶  with a two-step conjugation protocol. Diphtheria toxin mutant, CRM₁₉₇, often used as a carrier in commercial conjugate vaccines, was firstly functionalized at lysine with either an alkyne or an azide, installed via the azido/alkyne N-hydroxysuccinimide derivatives in Fig. 9b. Secondly, the functionalized CRM₁₉₇ was reacted with the respective azido or alkyne sugar counterpart (Fig. 9a) by CuAAC chemistry.

The authors observed a low coupling efficiency, leaving unreacted azide or alkyne functional groups on the protein surface. By carefully mapping the functionalized lysine residues, the more exposed sites were identified, leading to the conclusion that the installation of three to six sugar haptens on the protein represented a practical maximum. This was borne out by limiting the number of functional alkyne/azide on the protein to three–six, when the efficiency of the CuAAC conjugation was raised from 5–20% to >95%, and it was possible to predict the position of the each hapten installed (Fig. 10).

In 2010, Barbas⁷⁷  group developed a new reagent, the 4-phenyl-1,2,4-triazoline-3,5-dione (PTAD), for the site-specific modification of Tyr side chains. Berti et al. optimized the application of PTAD to functionalize CRM₁₉₇ either with azide^78,79  or alkyne;⁸⁰  the latter aimed to prepare a vaccines against candidiasis, with a β-(1,3) hexasaccharide azide, yielding an average of 3.5 hexasaccharide units per protein (Fig. 11), followed by more in-depth structural/immunological studies of a series of glycoconjugates obtained by defined conjugation (PTAD) and random conjugation.⁸¹ 

The great potential of CuAAC for glycoconjugate synthesis was enhanced by Shoda's method⁸²  for the direct preparation of sugar anomeric azides directly from unprotected carbohydrates. Such methodology offers the obvious advantages of avoiding protective group manipulation of complex carbohydrates, which are often expensive or obtained in small amount by isolation from natural products. The method has been applied, among others, by Fairbanks et al.⁸³  who prepared a number of different carbohydrate azides and employed CuAAC to couple them with a synthetic peptide fragment of MUC1 obtaining the glycopeptide with high efficiency (Fig. 12).

Shoda's method was also applied by Winssinger et al.⁸⁴  who successively coupled the resulting sugar azides by CuAAC to a short linker containing a tetrazine moiety, followed by conjugation to a protein containing either a trans-cyclooctene (Fig. 13, route 1) or a bicyclononyne (Fig. 13, route 2) ring via an inverse-electron-demand Diels–Alder strain-promoted cycloaddition reaction⁸⁵  between the tetrazine/strained-ring couple, representative examples of glycans conjugated with this methodology are shown in the box in Fig. 13. The advantage of this two-step approach is that it provides an opportunity to remove the potentially toxic Cu^I well ahead of the final glycoconjugation.

In the course of developing vaccines against Salmonella, Micoli et al.^86,87  installed the lipopolysaccharide O-antigen on CRM₁₉₇ protein carrier using different methodologies, combining new developed methods with well-established one (Fig. 14). Of the newer methods, partial disulfide bond reduction of CRM₁₉₇ with TCEP gave access to co-localised cysteine modification sites that are reactive towards 1,2-dichloracetone, resulting in the installation of a ketone for oxime ligation. In contrast, a pH-controlled transglutaminase-catalysed reaction modified only at lysine, resulting in transfer of Cbz-Gln-Gly-Peg3-azide on one/two lysine residue depending on the pH employed. Functionalisation of an average of 7 of the 39 lysine residues available on CRM₁₉₇ was achievable in this manner. Further, triazolidinone-ene-Peg4-azide was used to functionalize an average of 1.5–3.8 of the eighteen tyrosine residues in the protein. Random functionalization of glutamates (38 total residues) and aspartate (28 total residues) by conventional activation of their side chain carboxyl groups resulted in an average of 4.8 amino acids modified. The library of glycoconjugate was then subjected to systematic assessment by immunization of mice. Interestingly the highest immune response was obtained with the CRM₁₉₇ conjugates containing the highest number of antigens, regardless the amino acid connectivity.

Most of the effort in new glycoconjugation methodologies has been directed to the formation of O-linked glycosides at Ser, Thr and Tyr residue or N-glycosylation at Asn. The recent discovery of an unusual arginine GlcNAc transferase activity triggered the effort towards the generation and biological role of this type of modification.⁸⁸  Ornithine was used as a precursor for the GlcNAc-Arg and combined with chemical conjugation of a GlcNAc-based S-alkyl-isothiourea residue furnished the required adduct, which was used to generate antibodies that can specifically recognize this novel N-GlcNAc modification (Fig. 15).

Extensive work on the ‘tag-and-modify’ approach has been carried out by the Davis group,⁴⁰  spanning from the modification of a cysteine tag to the inclusion of non-proteinogenic amino acids (e.g. dehydroalanine, homoallylglycine, S-allylcysteine, azidohomoalanine) in a protein sequence. Starting from free sugars of different complexity and chain length, Davis et al. optimized the formation of a glycosyl thiol by using the Lawesson reagent (LR),⁸⁹  offering the advantage of avoiding the need for protecting group manipulation (Fig. 16a). The sugar thiol can be used as such as nucleophile to react with an electrophilic cysteine (phenylselenyl) sulfide (Fig. 16b); this principle can also be reversed, exploiting the nucleophilicity of a free cysteine thiol on a protein in reaction with an (phenylselenyl) sulfide thioglycoside (Fig. 16c–d).

By using the thioglycoside 3.3–3.9 (Fig. 17) and selected mutants of the model proteins, such as Bacillus lentus subtilisin SBL-S156C or Sulfolobus solfataricus SSbG-Cys344Cys432, a number of single and multiply glycosylated proteins were successfully obtained. Notably, all sugar thiols 3.3–3.9 were prepared directly from free sugars with LR, reacted with the cysteine (phenylselenyl)sulfide of SBL-S156C,⁸⁹  with a conversion yield exceeding 95% for all substrates, illustrating the potential of the methodology. Thioglycosides 3.3, 3.7, 3.10 and 3.11 (Fig. 17) were instead coupled by a selenylsulfide-mediated conjugation to SBL-SSePh (Fig. 16b) again with >95% conversion.⁹⁴ 

Exploiting the incorporation of the unnatural amino acid l-homoallyglycine and using a radical hydrothiolation reaction (Fig. 16e), Davis et al.⁹²  also developed a mild glycoconjugation reaction to install several thioglycoside in model proteins, such as β-glycosidase from Sulfolobus solfataricus SsbG-Hag43, Nostoc punctiforme “cuboid” protein Np276-Hag61 and on virus like particle Qβ-(Hag16)₁₈₀; also in this case, with few exceptions, the conjugation yield was >95%. For instance, using this methodology, Qb-Hag16 multimer was conjugated with a series of rhamnosyl derivatives (3.12, 3.13) including the pentasaccharide, 3.14, an O-antigen polysaccharide, creating an impressive multi-glycosylated carrier bearing 180 glycan chains per protein. The same principle was applied by Dondoni et al. who reported a complementary method where the allyl C-glycoside 3.15 was conjugated to the side chain thiol of cysteine residues contained in native BSA following the generic scheme of Fig. 16f.⁹⁰  A recent update by Davis et al.⁹¹  reported the conjugation of the free cysteine of subtisilin SBL-S156C via a selenenyl sulfide bond (Fig. 16g) obtained by reacting the protein with compounds 3.16–3.21, with a highly efficiency of conjugation yield (>95%). A great advantage offered by these approaches lies in the opportunity to site-specifically control the conjugation, in contrast to the traditional unselective modification methods that rely on natural amino acid side chain reactivity.⁹³  A further advantage of the newer methods is that the glycoconjugation protocol is generally mild, leaving the protein function intact, and it is toxic metal-free, unlike the CuAAC chemistry.

Where the availability of carbohydrates is limited by difficult synthesis or inadequate natural supply, chemo-enzymatic approaches can help synthesising complex glycans in a selective and specific way. Glycosyltransferases (GTases), glycoside hydrolases (GHs or glycosidases), glycosynthases (GSs) and glycoside phosphorylases (GPs) have been largely characterised and exploited in this area (Fig. 18).^41,95–97  In particular, GTases, thanks to their high regio- and stereo-specificity, are often the only choice to form specific glycosidic bonds.⁹⁸  Challenges in the use of GTases include their availability, the expensive sugar-nucleotide donors and the feedback inhibition. Many of these problems have been addressed over the years expanding the number of available GTases and making sugar-nucleotides more affordable by chemical synthesis⁹⁹  and by enzymatic regeneration systems.¹⁰⁰  Even with the advent of new carbohydrate-modifying enzymes, GTases are still a very valuable tool for the chemo-enzymatic synthesis of glycoproteins and their biological evaluation.

New knowledge of a variety of enzymes, their activities and substrate preferences has been deployed to address the synthesis of numerous glycoconjugate structures. Protein engineering has also increased the availability of enzymes with wider substrate tolerance, expanding the available chemo-enzymatic toolbox. Herein, we describe the application of recently developed enzymatic approaches for the production of glycoconjugates, which can be useful to mimic natural glycoproteins for structural and biological studies, such as GPI-anchored or N-linked glycoproteins, therapeutic antibody–drug conjugates (ADCs), or chemo-enzymatic labelling of glycoproteins.

Sortase enzymes were first isolated in the late ‘90s from Staphylococcus aureus. These enzymes were found to be responsible for anchoring surface proteins to the peptidoglycan cell wall of Gram-positive bacteria.^101,102  These enzymes are transpeptidases, cleaving the peptide bond between threonine and glycine of a conserved LPXTG peptide sequence to form an activated thioester. The thioester in turn reacts with the amino group of a terminal glycine of peptidoglycans, but can be used to react with the phosphoethanolamine group of GPI anchors. This enzyme has been extensively used for protein engineering and, more recently, has found applications in the synthesis of glycoproteins, in particular GPI-anchored proteins. The Guo group has studied the chemo-enzymatic synthesis of GPI-anchored proteins using the sortase reaction, initially for the synthesis of small GPI analogue–peptide conjugates,¹⁰³  moving on to more complex GPI derivatives conjugated to glycopeptides (Fig. 19).^104,105 

Finally, the full conserved core of GPI anchors was successfully coupled to intact GFP protein (4.8), and to intact polypeptide sequences of CD52 (4.9) and CD24 (4.10),¹⁰⁶  (Fig. 20).¹⁰⁷  While the total synthesis of GPI glycolipid anchors has advanced a lot in the last decade,¹⁰⁸  the synthesis of full GPI anchored proteins is still very challenging. The chemo-enzymatic approach using sortases is very promising, although the impact of introducing the unnatural LPXTGG sequence between the protein and the GPI anchor remains to be established.

Endo-β-N-acetylglucosaminidases (ENGases) are a very successful class of endo-glycosidases that have found use for the chemo-enzymatic synthesis of N-glycan conjugates.⁴¹  In contrast to exo-glycosidases, which cleave a single terminal sugar from the non-reducing end of an oligosaccharide, ENGases cleave the internal β-(1→4) glycosidic bond of the N,N′-diacetylchitobiose core of N-glycans, leaving a single GlcNAc unit bound to the asparagine residue of the protein. ENGases have different specificities; for example, Endo-A and Endo-H are specific for high mannose, Endo-M for high mannose and complex/hybrid type, while Endo-S is specific for human IgG with fucosylated core. Their reaction mechanism is substrate-assisted, where anchimeric assistance by the C-2 acetamido group of the penultimate HexNAc residue forms an oxazoline, which is subsequently opened by an activated water molecule to complete hydrolysis of the glycosidic bond. These enzymes, as well as exo-glycosidases and other endo-glycosidases, can catalyse the reverse reaction of trans-glycosylation and act as synthases, which form glycosidic bonds. Excess sugar donor, pH or a mixture of solvents can push the reaction in favour of trans-glycosylation over hydrolysis. However, a more powerful way to exploit these enzymes for synthesis is to activate the sugar donor substrate and to avoid the competing product hydrolysis. Glycoengineering can modify key residues in the enzyme, completely extinguishing the hydrolytic activity.⁴¹ 

The first use of the native trans-glycosidase activity of ENGases dates back to early ‘90s, when Takegawa described the use of Endo-A from Arthrobacter protophormiae for the synthesis of homogeneous Man₆GlcNAc₂ ribonuclease B (RNase B) (Fig. 21, A).¹⁰⁹  Heterogeneous RNase B was treated with Endo-A to give homogeneous GlcNAc-RNase B (4.11); then, Man₆GlcNAc₂-Asn (4.12) was added together with Endo-A, which catalysed the trans-glycosylation with formation of homogeneous Man₆GlcNAc₂ RNase B (4.13) in 2% yield and release of GlcNAc-Asn (Fig. 21, A). Similar studies were carried out by Inazu et al., where homogeneous NeuAc₂Gal₂GlcNAc₂Man₃GlcNAc₂ eel calcitonin (4.16) was obtained in 8.5% yield by trans-glycosylation of synthetic GlcNAc-calcitonin (4.14) with Endo-M (Fig. 21, B).¹¹⁰ 

With better understanding of the reaction mechanism of ENGases, an oxazoline oligosaccharide mimicking the natural intermediate in the enzymatic reaction was proposed to act as a reactive substrate for the trans-glycosylation reaction. In 2005, Wang et al. synthesised di- and tetra-saccharide oxazolines and used them as substrates for ENGases;¹¹¹  they obtained the trans-glycosylation products in ca. 75% yield, showing the benefits of oxazoline substrates for ENGase trans-glycosylation. With the advent of synthetic oxazoline substrates, ENGases have received increased attention and they have become a powerful tool for the chemo-enzymatic synthesis of N-linked glycoproteins. Wang et al. explored the substrate specificity of Endo-A in the trans-glycosylation of RNase B (Fig. 22), accessing structures with additional sugars on the outer mannoses of the Man₃GlcNAc₂ core (4.18), such as Gal (4.19) or lactose (4.20) units, and also changing the core itself, such as in Man₄GlcNAc₂ (4.21) or introducing azide group in C-6 of mannose unit (4.22).¹¹² 

Endo-A showed very relaxed substrate specificity, making it suitable for production of a large variety of different N-linked proteins. Moreover, Endo-A was found useful for glycosylation of complex proteins, such as the hydrophobic saposin C.¹¹³  In 2008, Wang et al. engineered Pichia pastoris to produce human IgG-Fc in high yield, which was subsequently deglycosylated by Endo-H to obtain homogeneous GlcNAc-IgG-Fc. This intermediate was then successfully glycosylated by Endo-A with sugar oxazoline donors to obtain homogeneous glycosylated IgG-Fc.¹¹⁴  This experiment showed that Endo-A was able to access GlcNAc on IgG-Fc without the need to denature the protein, which made N-glycosylation more straightforward and quick to perform.

Davis et al. investigated the use of unnatural linkers between GlcNAc and the peptide/protein to enhance the stability of the N-glycan conjugates toward chemical and enzymatic hydrolysis (Fig. 23).¹¹⁵ 

First, they compared the natural GlcNAc-Asn linkage (4.23a) with a series of GlcNAc-linker-amino acids (4.23b–f); they also tested the ability of Endo-A to trans-glycosylate these materials using sugar oxazoline donor 4.24 (Fig. 23, A). Surprisingly, the natural GlcNAc-Asn linkage gave the poorest yield toward trans-glycosylation compared to triazole, disulfide, selenylsulfide and thiol GlcNAc linkages. When the protein SBL-S156C was modified with GlcNAc-disulfide (4.26a), selenylsulfide (4.26b) and thiol-linkages (4.26c), the GlcNAc-thiol SBL-S156C gave the best trans-glycosylation yield (Fig. 23, B). However, when the GlcNAc-thiol was in a hindered position on the protein, the trans-glycosylation yield dropped, suggesting that the GlcNAc position is critical for enzyme access to the glycosylation site.

In 2012, Davis et al. identified the gene of a new ENGase, Endo-S, from the genome sequence of Streptococcus pyogenes and expressed it in E. coli. This enzyme was found to be active on N-glycans with a fucosylated core, in contrast to Endo-A and Endo-H which were completely inactive on such substrates.¹¹⁶  Endo-S allowed remodelling of the glycans present on human IgG-Fc (Fig. 24), giving homogeneous glycoforms for use in antibody-dependent cell-mediated cytotoxicity (ADCC) studies.

The enzymatic remodelling of N-glycans on mammalian IgG-Fc was also used to install an azide for subsequent coupling by strain-promoted azide–alkyne cycloaddition (SPAAC) with a drug in order to generate antibody–drug conjugates (ADCs) (Fig. 25).¹¹⁷  In this approach, the heterogeneous N-glycans were fully trimmed with Endo-S2, followed by transfer of azide-modified GalNAc (GalNAz, F₂-GalNAz and F₂-GalNBAz), by β-(1→4)-Gal transferase. The final step involved the SPAAC coupling with the drug-linker. A comparative study showed that bicyclo[6.1.0]nonine (BCN) was superior to dibenzoazacyclooctyne (DBCO) for the SPAAC coupling and that the ideal length of PEG linker was between 4 and 8 ethylene units.

Another approach to the synthesis of homogeneous glycoproteins is based on oxime conjugation, wherein aldehydes that are chemically or enzymatically introduced onto proteins or into sugar units can be readily coupled to aminooxy groups to generate oxime-linked products. Herein, we present few different examples of this methodology, which is useful to for extending and remodelling the glycans on intact proteins, or for labelling or drug conjugation applications.

Bertozzi et al. engineered IgG-Fc (4.36) in order to introduce a LCTPSR consensus sequence for the formylglycine generating enzyme (FGE), which converted Cys into formylglycine (fGly) by oxidation (4.37), thus introducing an aldehyde functional group for conjugation to aminooxy-modified sugar (Fig. 26).¹¹⁸  That is, a synthetic aminooxy-GlcNAc (4.38) was conjugated to fGly and the glycan subsequently extended by Endo-S-D233Q and sugar oxazoline donor 4.40 to obtain a homogeneous glycan-oxime-linked IgG (4.41, Fig. 26).

Zhou et al. treated IgG-Fc with galactosyltransferase and sialyltransferase to obtain homogeneous N-glycans with a terminal sialic acid (4.43), whose C-7/8 were chemically oxidised to aldehyde¹¹⁹  allowing its conjugation with aminooxy drug-linkers to produce ADCs (Fig. 27). Alternatively, the oxime conjugation could be used for introducing labels or extra sugars.¹²⁰ 

The enzymatic introduction of synthetic 2-keto-galactose using galactosyltransferase onto the N-glycans of IgG-Fc allowed subsequent oxime conjugation chemistry for coupling drugs, labels or extra sugars on the protein skeleton (Fig. 28).¹²¹ 

The introduction of an aldehyde onto a sugar for oxime conjugation can also be achieved by galactose oxidase (GO) oxidation, which oxidises C-6 of Gal to an aldehyde group.¹²²  Engineering of GO has allowed the development of enzyme mutants which are able to oxidise C-6 of Man or GlcNAc, expanding the potential of this class of enzymes for labelling and conjugation.¹²³ 

N-Glycosylation is a post-translational protein modification present in all kingdoms of life. The bacterial N-glycosylation system has been recently disclosed and the key enzymes have been studied and characterised for their application in the chemo-enzymatic synthesis of N-glycoproteins.^124,125  The main bacterial N-glycosylation system involves the transfer of an oligosaccharide, synthesised in the cytoplasm, from a lipid anchor to the asparagine side chain of a protein. The transfer, which takes place into the periplasm, is catalysed by a membrane-bound oligosaccharyltransferase (OST) to a consensus peptide sequence, D/EXNXS/T, of the protein. The identification of an oligosaccharyltransferase, PglB, from Campylobacter jejuni, and its expression in E. coli has opened up new possibilities for glycoengineering.^126–128  Currently, the most popular methodology to produce glycoconjugated vaccines is based on the protein glycan coupling technology (PGCT) (Fig. 29, A).¹²⁹  With this technique, glycoconjugates can be efficiently obtained by engineering E. coli with three plasmids (1): the first is dedicated to the synthesis of the desired glycan/polysaccharide on a pyrophosphate lipid anchor; a second encodes the carrier protein, CP, containing the necessary D/EYNXS/T consensus sequence for the PglB; and a third plasmid for PglB expression. Once the N-glycan is synthesized and extended onto a lipid anchor by the action of glycosyltransferases in the cytoplasm (2), a flippase can flip it into the periplasm (3), where PglB will transfer the N-glycan onto the carrier protein (4). With this system it is possible to obtain glycoconjugates in a single fermentation reaction, minimizing the purification steps and resulting in the production of homogeneous products.^130,131  This powerful technique has been employed to produce a candidate vaccine against Shigella dysenteriae, which has now completed phase I clinical trials in human.¹³² 

However, PGCT has its limitations, the main one being the necessity for a consensus sequence in the carrier protein, even though recent findings have demonstrated that this sequence can be introduced at the N- or C-terminus of the protein minimizing the disruption of its 3D structure. In addition, PglB can only transfer oligosaccharides containing, at the reducing end, carbohydrates with an acetamido group at the C2 position; moreover, it is able to transfer them only onto flexible regions of folded proteins. While the PglB enzyme has quite a broad substrate specificity, it will not accept mammalian glycan substrates. At present, different PglB enzymes with broader substrates specificities are under investigation.¹³³  Wang et al. have engineered the pgl locus of C. jejuni (Fig. 29, B) in order to install mammalian N-glycans onto proteins by PGCT instead of C. jejuni unique heptasaccharide, Glc₁GalNAc₅diNAcBac₁.¹³⁴  Firstly, the genes responsible for the synthesis (pglDEF) and transfer (pglC) of the first bacillosamine (diNAcBac) onto the undecaprenyl pyrophosphate (Und-PP) anchor, and for the Glc branch (pglI) in C. jejuni were deleted (1) from the pgl operon, which was renamed pgl2. In the absence of PglC, E. coli WecA transferred GlcNAc-1-P onto the lipid anchor; then, GalNAc units were successively transferred by the remaining PglAJH glycosyltransferases (2). The oligosaccharide was then flipped into the periplasm by PglK flippase (3) and transferred by PglB onto the AcrA protein (4); once isolated, the resulting glycoprotein was trimmed by exo-α-N-acetylgalactosaminidase to give a homogeneous GlcNAc-Acra (5). Finally, endo-α-N-acetylglucosaminidase (Endo-A) catalysed the trans-glycosylation between GlcNAc-Acra and Man₃GlcNAc oxazoline 4.17 to give a homogeneous glycoprotein carrying a mammalian N-glycan (6).

The activity of PglB from C. jejuni has been investigated not only in vivo with its application in the PGCT, but also in vitro for chemo-enzymatic synthesis of glycoproteins using natural or unnatural substrates. A key factor in these studies was the opportunity to explore the enzyme tolerance to sugar or lipid modifications (Fig. 30).

Ito et al. have carried out in vitro N-glycosylation of peptides with chemically synthesised donors to explore the chemo-enzymatic route to glycoproteins (Fig. 30, A).¹³⁵  PglB was able to transfer its natural biosynthetic intermediates, Und-PP-(diNAcBAc)₁ (4.50a), Und-PP-(GalNAc₂diNAcBac₁) (4.50b) and Und-PP-(Glc₁GalNAc₅diNAcBac₁) (4.50c), to TAMRA-modified peptides (4.49) suggesting relaxed specificity towards the carbohydrate moiety. Due to the in vitro ability of PglB to transfer any intermediate onto Und-PP independent of their natural biosynthetic order, the experiment also demonstrated the necessity for in vivo compartmentalization of oligosaccharide synthesis into the cytoplasm and oligosaccharyltransferase into the periplasm, in order to obtain glycoproteins carrying a precise oligosaccharide.

The Imperiali group carried out the chemo-enzymatic synthesis of PglB substrates (4.53a–c), introducing azide-modified diNAcBAc or GalNAc onto Und-PP by the action of glycosyltransferases PglCAJ.¹³⁶  Then, they showed the acceptance of these substrates by PglB, which transferred the glycans onto a FITC-modified peptide (4.52) with yields increasing from 4.54a, with azide-modified diNAcBac, to 4.54c, with the distal GalNAz (Fig. 30, B).

The natural donor substrate of PglB, containing the rare sugar bacillosamine and C55-Und-PP, represents a substantial limitation to the use of PglB for in vitro chemo-enzymatic synthesis of N-glycoproteins. To overcome this limitation, Davis et al. tested PglB activity using the common GlcNAc instead of diNAcBac, anchored to a library of synthetic short polyisoprenols (C10 to C40 polyisoprenol GlcNAc-PP), containing mixed trans/cis-prenyl units instead of the exclusively trans-prenyl units present in the natural C55-Und-PP.¹³⁷  They found out that PglB was able to use short polyisoprenol GlcNAc-PP down to C20 with high yields (90% for 4.57c) (Fig. 30, C). However, PglB did not tolerate polyisoprenol GlcNAc-PP shorter than C20, or C20 polyisoprenol with azide-modified GlcNAc on C-2 and C-6 (Fig. 30, D). This last result together with Imperiali's demonstrates that PglB tolerates azide-modification as long as the structure of the substrate is not too different from the natural one.

Another bacterial N-glycosylation pathway has been discovered wherein a cytoplasmic N-glycosyltransferase (NGT) transfers Glc from UDP-Glc directly to the asparagine side chain of a NXS/T consensus sequence.¹³⁸  NGT isolated from Actinobacillus pleuropneumoniae is an inverting glycosyltransferase that exhibits a preference for similar consensus sequence to PglB OST from C. jejuni but, in contrast to PglB, it uses nucleotide-activated monosaccharides instead of lipid-bound oligosaccharides as donor substrates. While NGT is able to transfer a single Glc onto the protein, an α-(1→6)glucosyltransferase present in A. pleuropneumoniae can extend the nascent oligosaccharide in the cytoplasm. NGT, even if strictly specific for the formation of β-glucosidic linkages (very low activity with Gal), can be useful for the chemo-enzymatic direct introduction of Glc units onto proteins, which can be further decorated with complex oligosaccharides by the use of endo-glycosidases and oxazoline donors. Wang et al. synthesised a glycoform of polypeptide C34, a potent HIV inhibitor, by trans-glycosylation with Endo-M N175A of Glc-C34 4.63, obtained by NGT glucosylation of the synthetic peptide 4.61, or of the fully synthetic GlcNAc-C34 4.64, in comparable high yields (Fig. 31).¹³⁹  Moreover, the Glc-bound glycoform (4.66) was completely resistant to PNGase F hydrolysis and 10 times more resistant to Endo-M hydrolysis compared to the GlcNAc-bound glycoform (4.67). The introduction of Glc-bound versus GlcNAc-bound oligosaccharides onto proteins could be a substantial advantage in the synthesis of inhibitors or therapeutics, which need to resist in vivo degradation.

Rapid progress in biology and biomedicine places increasing demands on methods for the construction of precise and often elaborate glycoconjugate structures with which to explore and explain biological phenomena. In this article, we have provided an overview of the increasing range of chemical and chemo-enzymatic methodologies used for glycoconjugation reactions in recent years. Even for the chemist, biochemistry and molecular biology approaches are becoming routine tools to complement the traditional chemical synthesis toolbox. Indeed, the powerful integration of chemical and chemo-enzymatic methods offers many and varied new ways to address future questions in this vast field that is glycobiology.

Studies at the JIC were supported by the UK BBSRC Institute Strategic Programme on Understanding and Exploiting Metabolism (MET) [BB/J004561/1] and the John Innes Foundation.