EASyMap-Guided Stepwise One-Pot Multienzyme (StOPMe) Synthesis and Multiplex Assays Identify Functional Tetraose-Core-Human Milk Oligosaccharides

Abstract

Carbohydrates are biologically and medicinally important molecules that are attracting growing attention to their synthesis and applications. Unlike the biosynthetic processes for nucleic acids and proteins, carbohydrate biosynthesis is not template-driven, more challenging, and often leads to product variations. In lieu of templates for carbohydrate biosynthesis, we describe herein a new concept of designing enzyme assembly synthetic maps (EASyMaps) as blueprints to guide glycosyltransferase-dependent stepwise one-pot multienzyme (StOPMe) synthesis to systematically access structurally diverse carbohydrates in a target-oriented manner. The strategy is demonstrated for the construction of a comprehensive library of tetraose-core-containing human milk oligosaccharides (HMOs) presenting diverse functional important glycan epitopes shared by more complex HMOs. The tetraose-core-HMOs are attractive candidates for large-scale production and for the development of HMO-based nutraceuticals. To achieve the preparative-scale synthesis of targets containing a Neu5Acα2–6GlcNAc component, a human α2–6-sialyltransferase hST6GALNAC5 is successfully expressed in E. coli. Neoglycoproteins with controlled glycan valencies are prepared and immobilized on fluorescent magnetic beads. Multiplex bead assays reveal ligands of glycan-binding proteins from plants, influenza viruses, human, and bacteria, identifying promising HMO targets for functional applications. The concept of designing EASyMaps as blueprints to guide StOPMe synthesis in a systematic target-oriented manner is broadly applicable beyond the synthesis of HMOs. The efficient StOPMe process is suitable for the large-scale production of complex carbohydrates and can be potentially adapted for automation.

Introduction

Carbohydrates are ubiquitously produced by all living organisms where they play major biological and pathological roles.¹ Unlike the template-driven biosynthetic processes for nucleic acids and proteins which are linear biomolecules, carbohydrate biosynthesis is not template-driven, more complicated, and often leads to product variations.² The inherent complexity and diversity of carbohydrate structures (linear or branched with variation on the stereo- and regio-specificities of glycosyl linkages) and nontemplate-driven biosynthetic processes pose challenges to not only glycan analysis,³ prediction and characterization of the glycome,⁴⁻⁶ but also carbohydrates synthesis, functional studies, and the development of diagnostics and therapeutics.⁷⁻⁹ Nature uses glycosyltransferases (GlyTs) as the key enzymes for carbohydrate assembly.¹⁰ These key glycan biocatalysts have been produced as recombinant forms in different cell types¹¹⁻¹⁵ and have been applied for the synthesis of an ever increasing number of carbohydrate targets in vitro or in metabolically engineered cells or organisms.^12,16−18 Compared to the cell-based carbohydrate production systems, the in vitro enzymatic synthesis has the advantage of a greater controllability regarding the types of the substrates and the enzymes used, their concentrations, the order of the glycosylation processes, and whether or not to purify the intermediates.¹⁹⁻²² To decrease the cost for in vitro large-scale synthesis of complex carbohydrates, it is beneficial to produce GlyTs with desired properties in low-cost E. coli expression systems.²³ Nevertheless, only a limited number of mammalian GlyTs have been successfully expressed in E. coli at levels that are sufficient for preparative-scale synthesis. Efforts for efficient expression of GlyTs, including those from mammalian sources, in E. coli are needed.²¹

Recent advances in carbohydrate synthesis and glycan microarray studies²⁴⁻²⁸ highlight the importance of the systematic access to comprehensive synthetic glycan libraries for evaluating glycan-binding properties of proteins. In lieu of templates for carbohydrate biosynthesis, we describe herein a new concept of designing Enzyme Assembly Synthetic Maps (EASyMaps) as blueprints for guiding GlyT-dependent Stepwise One-Pot Multienzyme (StOPMe) synthesis to systematically access structurally diverse carbohydrates in a target-oriented manner. A growing interest has been paid to producing human milk oligosaccharides (HMOs) due to their important roles in promoting infant health and their nutraceutical potentials.²⁹ HMOs with increased structural complexity including those with branched structures are now becoming synthetically available with enzymatic and chemoenzymatic methods.^{13,26−28,30−32} Among all HMOs, those containing a tetraose core are amidst the most abundant³³ and functionally important ones. They present a comprehensive collection of terminal epitopes shared by other more complex HMOs including the branched ones. Compared to the more complex HMOs, the tetraose-core-HMOs are more attractive candidates for industrial scale production and for the development of HMO-based nutraceuticals. So far, only a partial collection of the tetraose-core-HMOs has been obtained,^22,34−39 and the synthesis was neither systematic nor comprehensive. The synthesis of a total of twenty-one naturally existing tetraose-core-HMOs or the corresponding glycosides^29,39,40 including 14 neutral and seven sialylated structures have been reported using enzymes, engineered cells or organisms. These include recently reported four sialylated tetraose-core-HMOs containing a Neu5Acα2–6GlcNAc-component synthesized using a mammalian cell-expressed human α2–6-sialyltransferase hST6GALNAC6.²⁶ On the other hand, successful expression of hST6GALNAC5 and hST6GALNAC6 in E. coli was reported only recently but their expression levels were not sufficient for preparative-scale synthesis.⁴¹ Improvement on their E. coli expression is needed to decrease the cost for efficient enzymatic synthesis of Neu5Acα2–6GlcNAc-containing HMOs and other glycans.

To demonstrate the efficiency of EASyMaps as blueprints to guide GlyT-dependent StOPMe synthesis, we aimed to construct a comprehensive library of tetraose-core-HMOs in a systematic manner. The tetraose-core-HMOs in the library present the terminal glycan epitopes that can be found in the single branches of more complex HMOs. To achieve preparative-scale synthesis of targets containing a Neu5Acα2–6GlcNAc-component, a human α2–6-sialyltransferase hST6GALNAC5 is successfully expressed in E. coli and used for high-yield preparation of four Neu5Acα2–6GlcNAc-containing tetraose-core-HMO targets.

We also aimed to take advantage of the comprehensive epitopes presented in the tetraose-core-HMO library to construct multiplex bead glycan arrays for identifying glycan-binding preferences of important glycan-binding proteins (GBPs) from plants, influenza viruses, human, and bacteria. To enhance the bead-immobilizing efficiency and present the glycans in a controlled valency manner, neoglycoproteins with controlled glycan valencies are prepared.

The combination of the novel concept of designing EASyMaps, the straightforward StOPMe synthetic strategy with pregeneration of stable sugar nucleotides for efficient synthesis, glycan tagging for facile product purification, a systematic access to a comprehensive library of glycan epitopes, a high-throughput clinically applicable multiplex neoglycoprotein array platform developed with neoglycoproteins of controlled glycan valency, and the identification of glycan ligands for biologically important glycan-binding protein targets will facilitate the development of glycan-based nutraceuticals, diagnostics, and therapeutics.

Experimental Section

Multigram-Scale Synthesis of LacβNHCbz (1) with an Improved Procedure

D-Lactose monohydrate (10.0 g, 27.8 mmol) was dissolved in 60 mL NH₄OH in a 500 mL round-bottom flask, NH₄HCO₃ (3 g, 38 mmol) was then added. The reaction was heated to 45–47 °C in an oil bath using a water condenser for 20 h to form LacβNH₂. The solvent was removed by rotary evaporation, and LacβNH₂ was dissolved in 20% Na₂CO₃ solution (65 mL) in a 1 L round-bottom flask. The solution was incubated in an ice bath. Benzyl chloroformate (9.86 mL, 69.4 mmol, 2.5 equiv) was dissolved in ethanol (EtOH) (15 mL) in a 50 mL centrifuge tube, and the mixture was added drop-wisely to the LacβNH₂ in aqueous Na₂CO₃ solution. Another 70 mL of EtOH was added to the reaction. The reaction pH was adjusted by adding 20% Na₂CO₃ (35 mL) solution and was kept at pH 8–9. The reaction was then removed from the ice bath and kept at room temperature for 4 h. Furthermore, another batch of benzyl chloroformate (11.8 mL, 83.2 mmol, 3.0 equiv) was dissolved in EtOH (20 mL) and added to the reaction mixture, followed by addition of EtOH (100 mL) to prevent precipitation. Again, to maintain the pH, 20% sodium carbonate (∼50 mL) was added drop wisely. Once the pH was adjusted, the reaction was stirred at room temperature. After the reaction was completed (20 h) as monitored by TLC with EtOAc:MeOH:H₂O = 7:2:1 (by volume) as the developing solvent, EtOH was removed from the reaction by rotary evaporation. To the resulting residue in a 1 L round-bottom flask, 600 mL of a mixed solvent ethyl acetate and H₂O (1:1 by volume) was added. The mixture was transferred to a separation funnel to obtain the crude product in the aqueous layer. The ethyl acetate (EtOAc) layer was washed with 60 mL of water each time for three times. The aqueous layer samples were combined, the solvent was removed by rotary evaporation, and the crude product (55 g) was collected. For crystallization, 275 mL of a mixed solvent (5 mL for each gram of crude) of 1-butanol and H₂O (1:1 by volume) was transferred to a 1 L Erlenmeyer flask. It was heated to almost boiling. To the solvent, 55 g crude was added and completely dissolved. After cooling the mixture to room temperature, a small amount of pure compound was added for seeding. The mixture was incubated at 4 °C for 24 h and the crystals were filtered and collected. The flow-through was collected, dried and used for recrystallization by following a procedure similar to that described above. A total of 9.6 g (73% yield) pure product crystals were obtained from two rounds of crystallization procedures.

C18-Cartridge Purification Process for βNHCbz-Tagged HMO Products

After the reaction was completed, the reaction mixture was incubated in a boiling water bath for 5 min to denature enzymes. The mixture was then cooled down to room temperature and centrifuged at 9016×g at 4 °C for 30 min. The supernatant was collected. The precipitate was washed twice, each time with H₂O (3 mL), and the supernatants were combined. The combined supernatant was concentrated by rotavap (25–40 min) to reduce the volume to about 3–5 mL which was then purified by passing through a ODS-SM column (51 g, 50 μm, 120 Å, Yamazen) pre-equilibrated with three column volumes of mobile phase A (water) on a CombiFlash Rf 200i system and monitored at 214 nm. The product was eluted with a mixed solvent of acetonitrile and water with a flow rate of 20 mL min^–1. The eluting program used was: Mobile phase A: water; Mobile phase B: acetonitrile; 0% B for 8 min followed by gradient 0 to 40% B over 25 min, gradient 40 to 100% B over 3 min, 100% B for 2 min, then 100 to 80% B over 2 min.

Gram-Scale StOPMe Synthesis of LSTaβNHCbz (3S2)

GlcNAc (390 mg, 1.76 mmol), Gal (318 mg, 1.76 mmol), ATP (1.71 g, 3.10 mmol), and UTP (1.64 g, 3.10 mmol) were dissolved in water in a 250 mL plastic bottle containing Tris-HCl buffer (100 mM, pH 7.5), MgCl₂ (20 mM), BLNahK (12 mg), PmGlmU (9 mg), SpGalK (12 mg), BLUSP (12 mg), and PmPpA (10 mg). The reaction mixture (44 mL) was incubated at 30 °C for 20 h with agitation at 180 rpm. LacβNHCbz (550 mg, 1.16 mmol) and NmLgtA (18 mg) were then added to the reaction mixture to bring the concentration of LacβNHCbz to around 25 mM. The reaction mixture was incubated at 30 °C with agitation at 180 rpm. The product formation was monitored by HRMS. After the trisaccharide formation was completed (20 h), the reaction mixture was incubated in a boiling water bath for 5 min and cooled down to room temperature. Cvβ3GalT (25 mg) was added and the acceptor concentration was around 20 mM. The reaction mixture was incubated at 30 °C with agitation at 180 rpm. The reaction was monitored by HRMS. After the formation of tetrasaccharide was completed (10 h), the reaction mixture was incubated in a boiling water bath for 5 min and cooled down to room temperature. Neu5Ac (1.76 mmol), CTP (2.30 mmol), NmCSS (6 mg) and PmST1_M144D (14 mg) were added. After adjusting the reaction mixture to pH 8.0 by adding 4 M NaOH, the reaction mixture was incubated at 30 °C with agitation at 180 rpm. The reaction was monitored by HRMS. After the formation of pentasaccharide was completed (16 h), the reaction was incubated in a boiling water bath for 5 min and cooled down to room temperature. The mixture was centrifuged at 9016×g at 4 °C for 30 min. The supernatant was concentrated and purified by a C18 cartridge to obtain pure LSTaβNHCbz (3S2, 1.17 g, 88%).

Results and Discussion

Defining Tetraose-Core-HMO Synthetic Targets

The first step for the systematic synthesis is to define the scope of the targets and analyze their structural features. Tetraose-core-HMOs are formed from lactose by extending its nonreducing end with a β1–3-linked N-acetylglucosamine (GlcNAc) followed by a β1–3/4-linked galactose (Gal) to form lacto-N-tetraose (LNT) and lacto-N-neotetraose (LNnT), two of the top 15 most abundant HMOs in pooled human milk.³³ The LNT (Figure 1a) and LNnT (Figure 1b) are further diversified by adding one or more α1–2/3/4-linked L-fucose (Fuc) and/or one or more α2–3/6-linked N-acetylneuraminic acid (Neu5Ac, the most abundant sialic acid form in nature).³⁶

Figure 1.

Possible fucosylation and sialylation sites of tetraose-core-HMOs and HMO-like glycans based on (a) lacto-N-tetraose (LNT) or (b) lacto-N-neotetraose (LNnT) define the scope of thirty-three tetraose-core-HMO synthetic targets (c–f) including 16 neutral glycans with 0–3 L-fucose residues (c), eight monosialylated glycans with a terminal Neu5Acα2–3Gal component (d), four monosialylated glycans with a terminal Neu5Acα2–6Gal component (e), and five mono- or disialylated glycans with a Neu5Acα2–6GlcNAc component highlighted with a red rectangle (f) that can be synthesized using (g) glycosyltransferases (GlyTs) and the corresponding sugar activation (SA) systems. Sialyltransferases and abbreviations: PmST1_M144D,⁵³Pasteurella multocida α2–3-sialyltransferase 1 M144D mutant (GlyT4a); PmST3,⁵⁴Pasteurella multocida α2–3-sialyltransferase 3 (GlyT4b); Pd2,6ST,⁵⁵Photobacterium damselae α2–6-sialyltransferase (can add Neu5Ac to both terminal and internal Gal residues) (GlyT4c); Pd2,6ST_A200Y/S232Y,^22,56Photobacterium damselae α2–6-sialyltransferase A200Y/S232Y double mutant (a regioselective α2–6-sialyltransferase for adding Neu5Ac selectively to terminal Gal only) (GlyT4d); hST6GALNAC5, recombinant human ST6GALNAC5 expressed in E. coli (GlyT4e). Adapted with permission from refs (22, 45, and 57). Copyright 2017⁵⁷ and 2019⁴⁵ the Royal Society of Chemistry, 2022²² John Wiley and Sons.

A total of 16 neutral HMOs or HMO-like glycans containing a tetraose core of either LNT or LNnT (Figure 1c) are possible. These neutral glycans contain 0–3 fucose residues with the consideration for the possibility of the presence and the absence of an α1–3-linked fucose at the d-glucose (Glc) on the reducing end, an α1–4-linked (for LNT-based HMOs) or an α1–3-linked (for LNnT-based HMOs) fucose at the internal GlcNAc, and an α1–2-linked Fuc at the Gal on the nonreducing end. Among these, the LNnT-based glycans with an α1–2-Fuc linked to the terminal Gal are HMO-like glycans and their presences in human milk have not been reported. They are nonetheless possible products of known fucosyltransferases. Accessing to these glycans synthetically will present additional opportunities for exploring the applications of functional HMOs and HMO-like glycans.

For negatively charged tetraose-core-HMOs, 12 monosialylated HMOs or HMO-like glycans including eight with a Neu5Ac α2–3 (Figure 1d) and four with a Neu5Ac α2–6 (Figure 1e)-linked to the terminal Gal of the tetraose core are possible. In addition, four reported LNT-derived HMOs containing a Neu5Acα2–6GlcNAc component (Figure 1f) including DSLNT, a naturally occurring HMO that has shown potential in treating necrotizing enterocolitis in a neonatal rat model,⁴² are part of our synthetic targets. We are also interested in synthesizing FDS-LNT-I (Figure 1f),⁴³ a complex fucose-containing HMO resembling the glycan portion of the disialyl Lewis a (DSLe^a) antigen that is mainly found on normal colonic epithelial cells but is under-expressed in human colon cancers.⁴⁴ Of the thirty-three tetraose-core-HMOs or HMO-like glycans,⁴⁵ only 13 LNT-based and eight LNnT-based HMO structures have been reported to have been found in human milk so far.³⁶

Selecting Sugar Activation (SA) Systems and GlyTs for Synthesizing Target HMOs

The second step for the systematic synthesis is to identify the GlyTs and the corresponding sugar activation (SA) systems and the enzymes needed. To synthesize the tetraose-core-HMOs, four types of GlyTs are needed and the HMO-production cost can be decreased by in situ generation of the corresponding sugar nucleotide donors from inexpensive commercially available simple monosaccharides which can be achieved by four sugar activation (SA) systems (Figure 1g).^22,45 Uridine 5‘-diphosphate N-acetylglucosamine (UDP-GlcNAc) can be formed from GlcNAc, adenosine 5‘-triphosphate (ATP), and uridine 5‘-triphosphate (UTP) via SA1 which contains three enzymes including Bifidobacterium longumN-acetylhexosamine-1-kinase (BLNahK),⁴⁶Pasteurella multocidaN-acetylglucosamine 1-phosphate uridylyltransferase (PmGlmU),⁴⁷ and Pasteurella multocida inorganic pyrophosphatase (PmPpA).⁴⁸ UDP-Gal can be formed from Gal, ATP, and UTP via SA2 which contains three enzymes including Streptococcus pneumoniae galactokinase (SpGalK),⁴⁹Bifidobacterium longum UDP-sugar synthase (BLUSP),⁵⁰ and PmPpA. Guanosine 5′-diphosphate-L-fucose (GDP-Fuc) can be formed from L-fucose (Fuc), ATP, and guanosine 5‘-triphosphate (GTP) via SA3 which contains two enzymes including a bifunctional enzyme from Bacteroides fragilis that has both L-fucokinase and GDP-fucose pyrophosphorylase activities (BfFKP),⁵¹ and PmPpA. Cytidine 5′-monophosphate-N-acetylneuraminic acid (CMP-Neu5Ac) can be formed from Neu5Ac and cytidine 5′-triphosphate (CTP) via SA4 with a single enzyme Neisseria meningitidis CMP-sialic acid synthetase (NmCSS).⁵² GlyTs needed (Figure 1g) include a β1–3-N-acetylglucosaminyltransferase (β3GlcNAcT, GlyT1a); a β1–3-galactosyltransferase (β3GalT, GlyT2a) and a β1–4-galactosyltransferase (β4GalT, GlyT2b); an α1–2-fucosyltransferase (α2FucT, GlyT3a) and an α1–3/4-fucosyltransferase (α3/4FucT, GlyT3b); as well as α2–3-sialyltransferases (α3SiaT, GlyT4a–GlyT4b) and α2–6-sialyltransferases (α6SiaTs, GlyT4c–GlyT4e).

EASyMaps and StOPMe Synthesis of Neutral Tetraose-Core-HMOs and HMO-like Glycans from a Hydrophobic Tag-Derived Lactoside LacβNHCbz (1)

The tetraose-core-HMO synthetic targets (Figure 1c–f) can be produced by glycosyltransferase (GlyT)-catalyzed extension of commercially available inexpensive lactose. However, the product purification would be challenging. We reported previously the design of a carboxybenzyl (Cbz)-derivatized β-lactosylamine (LacβNHCbz, 1)^22,58 as an important intermediate to facilitate the simple purification of HMOβNHCbz formed by enzymes in aqueous solution using a single C18-cartridge. The HMOβNHCbz was readily converted to the target HMO by catalytic hydrogenation and hydrolysis.²² The use of LacβNHCbz (1) is thus a method of choice to simplify the product purification process. To produce LacβNHCbz (1) in large quantities and at a low cost, we improved its two-step protecting-group-free synthetic process from commercially available inexpensive lactose. Briefly, after the preparation of lactosylamine (LacβNH₂) from lactose by treating it with NH₄HCO₃ in NH₄OH, instead of using methanol, we found that a mixed solvent of ethanol and water (Scheme S1, see ESI) significantly improved the solubility of the crude LacβNH₂, allowing us to reduce the solvent volume by 6-fold for the acylation of LacβNH₂ with benzyl chloroformate.²² In addition, once the LacβNHCbz formation was completed, a mixed solvent of ethyl acetate and water (1:1 by volume) was used to extract the product. The crude product was readily collected from the water layer, dried, and crystallized in 1-butanol and water (1:1 by volume). The product crystallization allowed us to obtain pure LacβNHCbz (1) in multigram scales efficiently omitting column purification procedures, a practice preferred by industrial-scale production.

With LacβNHCbz (1) prepared efficiently in large amounts, we envisioned a systematic target-oriented chemoenzymatic synthetic process for the formation of any given HMOβNHCbz target from LacβNHCbz (1) in a single pot using a Stepwise One-Pot Multienzyme (StOPMe) strategy, followed by a simple C18-cartridge purification process (40 min). The construction of the 16 neutral tetraose-core-HMOβNHCbz compound library (Figure 1c) is greatly facilitated by the design of enzymatic assembly synthetic maps (EASyMaps) for those containing the LNT (Figure 2a) or the LNnT (Figure 2b) core with 0–3 fucose residues. To simplify the one-pot StOPMe synthetic process to allow its easy conversion to future automation and to minimize the reaction times for glycosyltransferase-catalyzed steps, we envisioned that relatively stable sugar nucleotides including UDP-GlcNAc, UDP-Gal, and GDP-Fuc can be pregenerated (indicated by an asterisk such as SA1*, SA2*, SA3* on the maps) at the beginning before the first glycosylation step and be stable throughout the whole StOPMe reaction process. On the other hand, due to the relative instability of CMP-Neu5Ac,⁵⁹ it is preferable to be generated by SA4 at the same time when a sialyltransferase-catalyzed reaction is performed. After the completion of each glycosyltransferase-catalyzed reaction, the enzyme is deactivated by incubating the reaction mixture in a boiling water bath for 5 min before being cooled down to room temperature, followed by the addition of a glycosyltransferase to initiate the next glycosylation step. This ensures that the glycosylation reaction is stopped at the end of the desired step (“Stop me”) without complicating future glycosylation steps.

Figure 2.

EASyMaps for StOPMe systematic synthesis of neutral HMOs contain 0–3 fucose residues (2–17) on the (a) LNT or (b) LNnT core structure and three exemplary NHCbz-tagged sialylated glycosides 3S2, DSLNT (20), and 3S10. (c) Symbol representations of the intermediate glycosides (IM1–11) which are not part of the targets. The HMO-like targets having not been identified from human milk samples are underlined. Targets marked with a caged or uncaged red star (8 out of 16 structures) without a terminal α1–2-linked L-fucose can be further sialylated with a Neu5Ac α2–3-linked to the terminal Gal by a sialyltransferase. Those marked with a caged red star (4 out of 16 structures) without a terminal α1–2-linked Fuc or α1–3/4-fucosylated GlcNAc can be further sialylated with a Neu5Ac α2–6-linked to the terminal Gal by a sialyltransferase. SA4 is designed to be carried out together with a sialyltransferase in the same step. SA1*–SA3* are designed to be performed before the first glycosylation step of the overall StOPMe process, which is the NmLgtA (GlyT1a)-catalyzed step for all targets shown. For easy planning/viewing purpose, each SA* is shown with the corresponding glycosyltransferase in the EASyMaps. For LNTFHepβNHCbz (4) and LNnTFHepβNHCbz (12) preparations, the “2×” shown in front of the “SA3*” in the last step indicates the equivalents of the donor needed for this glycosylation step.

Guided by the HMO EASyMaps (Figure 2a,b), the synthesis of target HMOs was straightforward. The synthesis of the LNT-core-containing neutral HMOβNHCbz targets (2–9) was achieved by following the EASyMap shown in Figure 2a. For example, LNTβNHCbz (2) was obtained by pregenerating UDP-GlcNAc (SA1*) and UDP-Gal (SA2*), followed by the addition of LacβNHCbz (1, 100 mg) and NmLgtA (GlyT1a). Once the reaction was completed as determined by high-resolution mass spectrometry (HRMS) analysis and facilitated by thin-layer chromatography (TLC) analysis, the enzymes in the reaction mixture were deactivated and Cvβ3GalT (GlyT2a) was added. Once the reaction was completed, the enzyme was deactivated and pure (>95% purity) LNTβNHCbz (2) product (162 mg) was obtained in 92% yield after a single C18-cartridge purification process. LNFP-IβNHCbz (3) was obtained by pregenerating UDP-GlcNAc (SA1*), UDP-Gal (SA2*), and GDP-Fuc (SA3*), followed by the addition of the LacβNHCbz (1) and NmLgtA (GlyT1a) for the first glycosylation step. Enzyme deactivation, addition of Cvβ3GalT (GlyT2a) for the second glycosylation step, enzyme deactivation, addition of Hm2FT (GlyT3a) for the third glycosylation step, followed by enzyme deactivation and product purification by a C18-cartidge produced pure (>95% purity) LNFP-IβNHCbz (3) (176 mg) in 85% yield. LNTFHepβNHCbz (4) was obtained similarly as LNFP-IβNHCbz (3) except for two extra equivalents of GDP-Fuc was pregenerated before the first glycosylation reaction and an additional glycosylation reaction process with Hp3/4FT (GlyT3b) was included before the C18-cartridge purification process to obtain pure (>95% purity) LNTFHepβNHCbz (4) (224 mg) product in a total yield of 84%. Hp3/4FT was responsible for adding both α1–3- and α1–4-fucose residues in LNTFHepβNHCbz (4) in a single glycosylation step.

Regioselective fucosylation for the synthesis of targets LNFP-VβNHCbz (5) (183 mg, 88%) and LNDFH-IIIβNHCbz (7) (220 mg, 92%) containing an α1–3-fucose at Glc but not an α1–4-fucose at GlcNAc was achieved by changing the order of Hp3/4FT (GlyT3b)-catalyzed α1–3-fucosylation and Cvβ3GalT (GlyT2a)-catalyzed β1–3-galactosylation steps by following the procedure illustrated in the EASyMap shown in Figure 2a. LNDFH-IIβNHCbz (6, 210 mg, 88%) was readily obtained in a procedure similar to that for LNFP-VβNHCbz (5) except for the pregeneration of an additional equivalent of GDP-Fuc before the first glycosylation step and an additional Hp3/4FT (GlyT3b)-glycosylation before the C18-cartridge purification process.

To synthesize targets LNFP-IIβNHCbz (8) and LNDFH-IβNHCbz (9) containing an α1–4-fucose at GlcNAc but not an α1–3-fucose at Glc, regioselective fucosylation was achieved using an α2–6-sialylation protection strategy similar to that described previously by the Cao group.⁶⁰ For example, LNFP-IIβNHCbz (8) (203 mg, 85%) was synthesized by pregeneration of UDP-GlcNAc (SA1*), UDP-Gal (SA2*), and GDP-Fuc (SA3*) followed by NmLgtA (GlyT1a)-catalyzed glycosylation reaction, enzyme deactivation, Photobacterium damselae α2–6-sialyltransferase (Pd2,6ST, GlyT4c)-catalyzed glycosylation reaction with in situ generation of CMP-Neu5Ac (SA4) to add an Neu5Ac α2–6-linked to the internal Gal on the GlcNAcβ3LacβNHCbz (LNT-IIβNHCbz, IM1) trisaccharide intermediate to form the tetrasaccharide intermediate IM3 to protect Glc from future α1–3/4-fucosylation by Hp3/4FT (GlyT3b), enzyme deactivation, Cvβ3GalT (GlyT2a)-catalyzed glycosylation, enzyme deactivation, and Hp3/4FT (GlyT3b)-catalyzed glycosylation, enzyme deactivation, and a Streptococcus pneumoniae sialidase SpNanA⁶¹-catalyzed reaction to remove the α2–6-linked Neu5Ac which served as a protecting group for Glc from Hp3/4FT-catalyzed α1–3-fucosylation, enzyme deactivation, and C18 cartridge purification. LNDFH-IβNHCbz (9) (193 mg, 80%) was synthesized similarly as LNFP-IIβNHCbz (8) except for the inclusion of the pregeneration of an additional equivalent of GDP-Fuc and an additional Hm2FT (GlyT3a)-catalyzed glycosylation step before the Hp3/4FT (GlyT3b)-catalyzed glycosylation step.

The synthesis of LNnT-core-containing neutral HMOβNHCbz targets (10–17) followed the EASyMap shown in Figure 2b which was similar to Figure 2a. The only difference was replacing Cvβ3GalT (GlyT2a) in Figure 2a designed for targets containing the LNT core with NmLgtB (GlyT2b) in Figure 2b as indicated by a gray round-corner rectangle background of the enzyme names on the maps. Pure (>95% purity) HMOβNHCbz products (10–17) were obtained in yields ranging from 82–91%.

Key considerations for high-yield production and easy purification of the product by the StOPMe strategy are to ensure the completion of each glycosylation reaction, meaning quantitative conversion of the acceptor to product, by (i) providing an excess amount (e.g., 1.2–1.5 equiv) of each sugar nucleotide donor by pregeneration (for UDP-sugars and GDP-Fuc) or in situ generation (for CMP-Neu5Ac); (ii) providing sufficient amounts of enzymes determined by test-run of small-scale (10–20 μL) reactions to make sure that the enzymes work as expected and have not lost significant activities during storage; and (iii) monitoring reaction progresses (mainly by HRMS analysis and facilitated by TLC analysis) to make sure that the acceptor substrate is completely consumed before deactivating the enzyme to prepare for the next step.

The StOPMe strategy described here shares similarities to the multistep one-pot multienzyme (MSOPME) strategy that we reported previously.²² Both involve multiple enzyme-catalyzed steps and include a process to heat deactivate the glycosyltransferase after the completion of each glycosylation step. The difference is that the MSOPME method involves in situ generation of the sugar nucleotide donor for the corresponding glycosyltransferase in the same glycosylation step while the StOPMe strategy described here involves the pregeneration of relatively stable sugar nucleotide donors including UDP-GlcNAc, UDP-Gal, and/or GDP-Fuc before all glycosylation steps. It is worth pointing out that a heat deactivation step after the pregeneration of UDP-sugars and/or GDP-Fuc in the StOPMe method is not necessary as it does not impact the first glycosylation step. Compared to the MSOPME method, the StOPMe strategy described here is more efficient by reducing the overall reaction time and simplifying the reagent addition processes in glycosylation steps. For example, the total reaction times for synthesizing LNFP-VβNHCbz (5) were 43 and 48 h using StOPMe and MSOPME, respectively. For synthesizing LNDFH-IIβNHCbz (6) which required more steps, the total reaction time using the StOPMe method was 48 h, demonstrating a significant time saving compared to the total reaction time of 62 h using the MSOPME method. To better clarify the similarities and the differences of the two methods, we rename the previous MSOPME strategy as the “StOPMe 1.0” and the current StOPMe strategy as the “StOPMe 2.0”.

The EASyMaps can be efficient tools to guide nonspecialists in performing enzymatic synthesis of complex glycans. In addition, we successfully demonstrate here that the α2–6-sialyltransferase-catalyzed enzymatic protection strategy for site-specific fucosylation⁶⁰ and the downstream sialidase-catalyzed deprotection strategy can be conveniently carried out in one-pot with the StOPMe strategy, which has not been achieved previously.

OP2E or StOPMe Synthesis of Sialyl Tetraose-Core-HMOs

With the access to all 16 neutral tetraose-core-HMOs, α2–3-sialylation was carried out for those without an α1–2-linked fucose at the terminal Gal (eight HMOβNHCbz targets marked with a caged or uncaged red star in Figure 2 including compounds 2, 5, 6, 8, 10, 13, 14, and 16). This was achieved by α2–3-sialylation of the neutral HMOβNHCbz using a single one-pot two-enzyme (OP2E) sialic-acid-activation-and-transfer system containing NmCSS and Pasteurella multocida α2–3-sialyltransferase 1 M144D mutant (PmST1_M144D, GlyT4a).⁵³ Adding a Neu5Ac α2–3-linked to the terminal galactose in LNTβNHCbz (2), LNFP-VβNHCbz (5), LNDFH-IIβNHCbz (6), LNFP-IIβNHCbz (8), LNnDFH-IIβNHCbz (14), and LNFP-IIIβNHCbz (16) was achieved in 65–95% yields (Table 1). An OP2E α2–3-sialylation system containing NmCSS and Pasteurella multocida α2–3-sialyltransferase 3 (PmST3, GlyT4b)⁵⁴ was used to carry out α2–3-sialylation of the terminal β1–4-linked galactose in LNnTβNHCbz (10) and LNnFP-VβNHCbz (13). Both targets were obtained in excellent 92–96% yields. It was shown previously that while PmST3 did not have undesired sialidase activity,⁵⁴ it was selective toward β1–4-galactoside acceptors⁶² and thus would be ideal for sialylating LNnTβNHCbz (10) and LNnFP-VβNHCbz (13) with high yields. On the other hand, while PmST1_M144D was active toward both β1–3- and β1–4-galactoside acceptors,⁶² it retained residue sialidase activity although its sialidase activity was significantly reduced compared to the wild-type PmST1.⁶³ The M144D mutation of PmST1 also allowed direct α2–3-sialylation of fucosylated glycans terminated with a Lewis x or a Lewis y structure at the nonreducing end⁵³ such as LNDFH-IIβNHCbz (6), LNFP-IIβNHCbz (8), LNnDFH-IIβNHCbz (14), and LNFP-IIIβNHCbz (16).

Table 1. Yields and the Amounts of the Synthetic Sialyl HMOβNHCbz Targets^a.

HMOβNHCbz	% yield (mg)	HMOβNHCbz	% yield (mg)
LSTaβNHCbz, 3S2	95% (130 mg)b; 89% (211 mg)c; 88% (1.17 g)d	6S2	92% (62 mg)b
3S5	91% (59 mg)b	6S5	93% (60 mg)b
3S6	65% (41 mg)b
3S8	86% (56 mg)b
3S10	96% (132 mg)b; 92% (219 mg)c	LSTcβNHCbz, 6S10	94% (63 mg)b
3S13	92% (60 mg)b	6S13	92% (60 mg)b
3S14	91% (57 mg)b
3S16	90% (59 mg)b

^aThe structures that have not been identified from human milk yet (8 out of 12) are in italics.

^bSynthesized by a single one-pot two-enzyme (OP2E) sialylation reaction from neutral HMOβNHCbz products.

^cStOPMe synthesis from LacβNHCbz (1) (100 mg).

^dGram-scale StOPMe synthesis from LacβNHCbz (1) (550 mg).

Similarly, α2–6-sialylation using OP2E containing NmCSS and Pd2,6ST A200Y/S232Y double mutant (Pd2,6ST_A200Y/S232Y, which adds α2–6-linked Neu5Ac regio-selectively to the terminal Gal only, GlyT4d)^22,56 was carried out for those not containing an α1–2-linked fucose at the terminal Gal and without a fucosylated GlcNAc (four HMOβNHCbz targets marked with a caged red star in Figure 2) including LNTβNHCbz (2), LNFP-VβNHCbz (5), LNnTβNHCbz (10), and LNnFP-VβNHCbz (13). The resulting α2–6-sialosides were obtained in excellent yields (92–94%) (Table 1).

The StOPMe procedure was also applied successfully to the synthesis of Neu5Acα3LNTβNHCbz (LSTaβNHCbz) (3S2, 211 mg, 89% yield) and Neu5Acα3LNnTβNHCbz (LSTcβNHCbz) (3S10, 219 mg, 92% yield) (Table 1) from LacβNHCbz (1, 100 mg) with very good yields (89–92%) by following the EASyMaps shown in Figure 2. The procedures were similar to the synthesis of LNTβNHCbz (2) and LNnTβNHCbz (10), respectively, but including an additional OP2E α2–3-sialylation step with NmCSS and PmST1_M144D (GlyT4a for 3S2) or PmST3 (GlyT4b for 3S10) as the last step before the single C18-cartridge purification process. Gram-scale StOPMe synthesis of LSTaβNHCbz (3S2, 1.17 g, 88% yield) was also demonstrated successfully (Table 1) by simply increasing the volume of the reaction mixture and the amounts of the reagents and enzymes used accordingly while retaining their concentrations.

Recombinant hST6GALNAC5 and Its Application in Synthesizing Neu5Acα2–6GlcNAc-Containing HMOs

The library of HMOs with an LNT core also includes five compounds containing a quite unique Neu5Acα2–6GlcNAc linkage that has been found in several colon cancer cell lines.⁶⁴ The sialyltransferase that is responsible for forming the Neu5Acα2–6GlcNAc linkage in HMOs is unclear. Nevertheless, a recombinant hST6GALNAC5 expressed in human embryonic kidney HEK293 cells⁶⁵ has been used for catalyzing the addition of a Neu5Ac α2–6-linked to the GlcNAc or the GalNAc in the acceptor substrate for the synthesis of a DSLNT derivative³⁰ and a disialyl Gb5 (DSGb5) ganglioside glycan, respectively.⁶⁶ More recently, GFP-hST6GALNAC6 expressed in HEK293-F cells as an N-terminal green fluorescence protein (GFP)-fused recombinant protein was shown to be a more efficient biocatalyst than GFP-hST6GALNAC4 and GFP-hST6GALNAC5 expressed in HEK293 cells⁶⁵ purchased from Glyco Expression Technologies, Inc. (Athens, GA) in synthesizing the Neu5Acα2–6GlcNAc linkage in HMOs.²⁶ The recombinant GFP-hST6GALNAC6 was used in chemoenzymatic synthesis of several targets but only in low milligram amounts. Previous attempts to express N-terminal truncated hST6GALNAC5 and hST6GALNAC6 in E. coli resulted in active enzymes but with very low yields⁴¹ that were not applicable for preparative-scale synthesis.

In order to synthesize Neu5Acα2–6GlcNAc-containing HMO targets, a plasmid for expressing an N-terminal 49-amino-acid-residue truncated recombinant hST6GALNAC5 fused with an N-terminal maltose binding protein (MBP) and a C-terminal His₆-tag (Figure S1, see ESI) was constructed using a codon optimized synthetic gene for E. coli expression. The MBP-Δ49hST6GALNAC5-His₆ was successfully expressed as a soluble and active fusion protein in E. coli Origami B (DE3) cells harboring pGro7 with an expression level up to 3.37 U per liter culture when Neu5Acα2–3LNTβNHCbz (3S2) was used as the acceptor (1 U = 1 μmol min^–1 at 37 °C, pH 7.5). Compared to the recent report for lower activity level of hST6GALNAC5 expressed in E. coli,⁴¹ the MBP-tag in our construct may have contributed to the improved expression and stability of the enzyme. Acceptor substrate specificity studies (Table S1, see ESI) showed that the recombinant MBP-Δ49hST6GALNAC5-His₆ was capable of using different HMOβNHCbz as acceptor substrates. In addition to Neu5Acα3LNTβNHCbz (3S2), Neu5Acα3LNFP-VβNHCbz (3S5) with an additional fucose α1–3-linked to the glucose residue at the reducing end of 3S2 was identified as a well-suited acceptor substrate. Furthermore, weak activity was observed when nonsialylated LNTβNHCbz (2) was used as the acceptor. On the other hand, no significant activity was observed when Neu5Acα3LNFP-IIβNHCbz (3S8) or LNFP-IβNHCbz (3) was tested as the acceptor.

With the active MBP-Δ49hST6GALNAC5-His₆ in hand and an understanding of its acceptor substrate specificity, the synthesis of Neu5Acα2–6GlcNAc-containing HMOβNHCbz targets (18–21) was readily accomplished (Figure 3). As shown in Figure 3a, LSTbβNHCbz (18) (18 mg) was synthesized from LNTβNHCbz (2) in 89% yield using an OP2E α2–6-sialylation reaction containing NmCSS and MBP-Δ49hST6GALNAC5-His₆ (GlyT4e). Similarly, OP2E α2–6-sialylation reaction of LSTaβNHCbz (3S2) and S-LNF-IIβNHCbz (3S5) led to the formation of DSLNTβNHCbz (20) (124 mg) (Figure 3b) and FDS-LNT-IIβNHCbz (21) (115 mg) (Figure 3c) with 98% and 93% yields, respectively. As LNFP-IβNHCbz (3) was not a suitable acceptor substrate for MBP-Δ49hST6GALNAC5-His₆ (Table S1), F-LSTbβNHCbz (19) could not be obtained by α2–6-sialylation of LNFP-IβNHCbz (3). To our delight, LSTbβNHCbz (18) was a suitable acceptor substrate for Hm2FT, and F-LSTbβNHCbz (19) (33 mg) was readily obtained from LSTbβNHCbz (18) in 98% yield using a one-pot three-enzyme α1–2-fucosylation system containing BfFKP, PmPpA, and Hm2FT (Figure 3a).

Figure 3.

OPME synthesis of HMOβNHCbz targets containing a Neu5Acα2–6GlcNAc component (18–21). (a) one-pot two-enzyme (OP2E) α2–6-sialylation of LNTβNHCbz (2) for the formation of LSTβNHCbz (18) followed by one-pot three-enzyme α1–2-fucosylation for the synthesis of F-LSTbβNHCbz (19); (b) OP2E α2–6-sialylation of LSTaβNHCbz (3S2) for the synthesis of DSLNTβNHCbz (20); and (c) OP2E α2–6-sialylation of S-LNF-IIβNHCbz (3S5) for the synthesis of FDS-LNT-IIβNHCbz (21).

StOPMe synthesis of DSLNTβNHCbz (20, 275 mg, 92% yield) (Figures 2a and S2, see ESI) from LacβNHCbz (1, 100 mg) was also successfully accomplished by following a procedure similar to the synthesis of LSTaβNHCbz (3S2) but including an additional OP2E α2–6-sialylation step with NmCSS and MBP-Δ49hST6GALNAC5-His₆ (GlyT4e) as the last glycosylation step before the single C18-cartridge purification process.

As compound 3S8 was a poor acceptor substrate for MBP-Δ49hST6GALNAC5-His₆ (Table S1) and DSLNTβNHCbz (20) was also a poor acceptor substrate for Hp3/4FT (data not shown), the synthesis of structurally complex FDS-LNT-IβNHCbz (Figure 1f) was not achieved, an outcome similar to that reported recently with hST6GALNAC4, 5, or 6 expressed in mammalian cells.²⁶

HMO-BSA Synthesis and Immobilization to Fluorescent Magnetic Beads

Each pure native HMO with a free reducing end was obtained readily from the corresponding synthetic HMOβNHCbz by catalytic hydrogenation to remove the Cbz tag which was released as toluene and CO₂, resulting in the formation of glycosylamine, which went through hydrolysis spontaneously to form the HMO target with the release of NH₃.²² The powder of the pure HMO product was obtained by lyophilization without the need of any purification. The solvents and the byproducts including toluene, CO₂, and NH₃ were removed by evaporation during the filtrate drying and lyophilization processes.

The obtained synthetic tetraose-core-HMOs and HMO-like glycans, presenting diverse terminal human glycan epitopes in a systematic manner, were conjugated to bovine serum albumin (BSA) and used to investigate glycan-binding proteins (GBPs) from different sources in a high-throughput neoglycoprotein-immobilized fluorescent magnetic bead-based multiplex assay. Compared to the more commonly used printed flat-surface glycan microarrays,⁶⁷⁻⁷³ the neoglycoprotein bead multiplex assay platform has the advantages of glycan valency tunability and controllability, flexibility, and easiness for handling.⁷⁴ It is better suited to analyze large numbers of samples and it can be easily adapted to clinical settings.⁷⁵⁻⁸⁰ The conjugation of the aldehyde group in the open-chain glucose at the free reducing end of HMOs to the lysine residues on BSA (59 lysine residues in the mature BSA of 583 amino acids, UniProt ID P02769) was achieved according to the reported reductive amination conditions⁸¹ with optimization (see ESI). While high glycan valencies (up to 35) could be achieved for the HMOs without a fucose α1–3-linked to the glucose by using high ratios (e.g., 146 equiv) of the HMO versus BSA in the conjugation reactions, the maximal glycan valencies obtained for the HMOs containing a fucosylated glucose at the reducing end were 5–9. In order to better compare the glycan-binding preferences of various GBPs, HMO-BSA neoglycoproteins with glycan valency controlled in a range of 5–8 (Table S2, see ESI) were prepared. HMO-BSA neoglycoproteins with simpler lactose-core-containing HMOs including LNT-II, 2′-fucosyllactose (2’FL),⁸² lactodifucotetraose (LDFT),⁸² 3′-sialyllactose (3′SL),⁶³ and 6′-sialyllactose (6’SL)⁵⁵ were also prepared for binding assays.

The glycan valencies of the HMO-BSA neoglycoproteins were determined by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) analysis. While direct MALDI-TOF MS analysis worked well for HMO-BSA neoglycoproteins containing neutral glycans as we reported previously,⁷⁴ it was challenging for those containing sialylated HMOs due to the cleavage of some of the sialyl linkages which are labile during the MALDI-TOF MS analysis process.^83,84 To accurately determine the sialylated HMO valency, the neoglycoprotein was treated with a linkage-nonspecific sialidase such as SpNanA to completely cleave sialic acid before the MALDI-TOF MS analysis.

It is worth mentioning that the HMO-BSA neoglycoproteins prepared here are different from our previously reported glycan-BSA neoglycoprotein preparation⁷⁴ which used neutral glycosides containing a propylamine aglycon that was derivatized and conjugated to BSA with a squarate linker with varied glycan valencies without the complication of instability issue for determining glycan valency by MALDI-TOF MS analysis.

Multiplex HMO-BSA Bead Assays Revealing the Glycan-Binding Profiles of Plant Lectins, Influenza Virus Hemagglutinins, Human Proteins, and Bacterial Toxins

Each HMO-BSA neoglycoprotein sample was immobilized via the primary amino groups of the remaining unreacted lysine residues on BSA to carboxyl-coated MagPlex Beads similar to what we reported previously.⁷⁴ The assay platform was evaluated with plant lectins with known glycan-binding preferences and then applied to investigating GBPs from influenza A viruses (IAVs), human, and bacteria (Figure 4).

Figure 4.

Heatmap representation of the glycan-binding results.

All three galactoside-binding plant lectins tested including Erythrina cristagalli lectin (ECL), Wisteria floribunda lectin (WFL), and Ricinus communis agglutinin I (RCA-I) (Figures 4 and S3a–c, see ESI) preferred binding to LNnT (10)⁷⁴ with a varied degree of tolerance toward its derivatization, agreeing well with that reported previously.^13,85 ECL additionally bound well to LNnFP-I (11) with a fucose α1–2-linked to the terminal Gal of LNnT (10). Its binding to LNnFP-V (13) with a fucose α1–3-linked to the Glc at the reducing end of LNnT (10) and to LNnDFH-III (15) with both an α1–2-linked fucose to the terminal Gal and an α1–3-linked fucose to the Glc was reduced gradually. In comparison, WFL did not show binding to LNnFP-I (11) or LNnDFH-III (15), and only weak binding to LNnFP-V (13) was observed. Interestingly, its binding to LSTb (18) which had a Neu5Ac α2–6-linked to the internal GlcNAc in LNT (2) was also observed. WFL additionally bound well to LNT (2) and weakly to LNFP-V (5) with a fucose α1–3-linked to the Glc at the reducing end of LNT (2). Compared to ECL, RCA-I tolerated LNnFP-V (13) with a fucose α1–3-linked to the Glc at the reducing end of LNnT (10) better, but bound more weakly to LNnFP-I (11) with a fucose α1–2-linked to the terminal Gal of LNnT (10). Different from ECL, RCA-I bound weakly to LNnDFH-II (14) but not to LNnDFH-III (15). In addition, RCA-I tolerated 6S10 with an α2–6-linked Neu5Ac on the terminal Gal of LNnT (10). RCA-I also bound weakly to LNT (2) and Lac (1). None of the terminal α2–3-sialylated structures were recognized by these three lectins under the experimental conditions.

For fucose-binding plant lectins including Ulex europeus lectin/agglutinin I (UEA-I) and Aleuria aurantia lectin (AAL) (Figures 4 and S3d,e, see ESI), UEA-I preferred terminal α1–2-fucosylated lactose-core- or LNnT-core-containing glycans, 2’FL and LNnFP-I (11), respectively. The results agreed well to that reported previously.⁸⁶ Additional α1–3-fucosylation of LNnFP-I (11) at either Glc (15) or GlcNAc (17), or both (12) decreased UEA-I binding. On the other hand, AAL bound to all fucose-containing structures, irrespective of the core structure or the type and the location of the fucosyl linkage.

For sialic acid-binding plant lectins including SNA-I and MAL-II (Figures 4 and S3f,g, see ESI), SNA-I preferred binding to α2–6-sialylated LNnT-core-containing structures without (6S10) or with (6S13) a fucose α1–3-linked to the Glc at the reducing end and had weaker binding to their LNT-core containing counterparts 6S2 and 6S5. It also showed binding to 6’SL and weaker binding to LNnT-core-containing LNnFP-I (11) with a terminal α1–2-linked fucose. MAL-II preferred binding to α2–3-sialylated LNnT-core-containing structures without (3S10) or with (3S13) a fucose α1–3-linked to the Glc at the reducing end.

After validating the multiplex glycan-binding assay platform with the HMO-BSA neoglycoprotein-immobilized beads using plant lectins, the glycan-binding assays were carried out for influenza A virus (IAV) hemagglutinins (HAs), recombinant human glycan-binding proteins (GBPs) including human E-selectin, and bacterial toxins, due to their importance in glycan-mediated pathogenesis, virulence, and immune responses.⁸⁷⁻⁸⁹

As shown in Figures 4 and S4a–d (see ESI), the preferred binding of four IAV HAs toward terminal α2–6- or α2–3-linked sialic acid were clearly observed. HAs⁹⁰ from A/New Caledonia/20/1999 (H1N1)⁹¹ and A/Wisconsin/67/2005 (H3N2)⁹² bound to both terminal α2–6-sialylated LNnT (6S10) and α2–6-sialylated LNT (6S2) with a preference toward the former. On the contrary, HAs from A/Guangdong/17SF003/2016 (H7N9)⁹³ and A/harbor seal/Germany/1/2014 (H10N7)⁹⁴⁻⁹⁶ preferred binding to the terminal α2–3-sialylated LNT (3S2) and α2–3-sialylated LNnT (3S10). Their derivatives with an additional α2–6-Neu5Ac or α1–3-Fuc linked to the internal GlcNAc and/or with an α1–3-fucose linked to the Glc at the reducing end were tolerated at varied degrees. The α2–3-sialoside-binding HAs also recognized 3′SL. The results indicated that both H1N1 and H3N2 HAs tested retained their specificity toward “human-type” receptors⁹⁷ while the H7N9 and H10N7 HAs tested preferred “avian-type” receptors indicating the avian origins of the corresponding IAVs and their potential inefficiency in human-to-human transmission.⁹⁷ The results demonstrated that the set of HMO-BSA neoglycoproteins and the multiplex bead glycan-binding platform are valuable tools for monitoring receptor-binding adaptations of IAV HAs during interspecies transmissions, although the sensitivity for some IAV HAs can be further improved by using sialosides with longer polyLacNAc sequences.⁹⁸

For human GBPs tested, recombinant human dendritic cell-specific intercellular adhesion molecule-3-grabbing nonintegrin (DC-SIGN)^99,100 (Figures 4 and S5a, see ESI) bound very well to 2′FL and tetraose-core-HMOs presenting the terminal Lewis a (8 and 6), Lewis b (9 and 4), Lewis x (16 and 14), Lewis y (17 and 12), sialyl Lewis a (3S8 and 3S6), and sialyl Lewis x (3S16 and 3S14) structures without or with an α1–3-fucose linked to the Glc at the reducing end. It also bound well to LNnT containing a fucose α1–2-linked to the terminal Gal (type II blood group H-antigens) without (11) or with (15) an α1–3-fucose linked to the Glc at the reducing end but not to their LNT-core-containing counterparts LNFP-I (3) or LNDFH-III (7). The results agreed well with a previous report using glycan microarrays printed with HMOs purified from human milk.¹⁰⁰ The study reported here has the advantage of including all possible terminal glycan epitopes of HMOs in the assays in a systematic manner and is better suited for guiding GBP ligand specificity investigations facilitated by machine learning.⁸⁵

Agreeing with a previous report,¹⁰¹ binding results of two human recombinant galectins¹⁰² (Figures 4 and S5b,c, see ESI) showed that Galectin-4¹⁰³ bound to both LNT (2) and LNnT (10) and their derivatives with a fucose α1–2-linked to the terminal Gal (3 and 11). Adding an additional α1–3-fucose linked to the Glc at the reducing end of 3 and 11 in 7 and 15, respectively, decreased but did not block the binding. Quite interestingly, while adding a Neu5Ac α2–3-linked to the terminal Gal of LNT (2) in 3S2, but not to the terminal Gal of LNnT (10) in 3S10, was tolerated well. On the contrary, adding a Neu5Ac α2–6-linked to the terminal Gal of LNnT (10) in 6S10, but not to the terminal Gal of LNT (2) in 6S2, was tolerated well. Furthermore, the addition of a Neu5Ac α2–6-linked to the internal GlcNAc of LNT (2) without (18) or with (19) an additional fucose α1–2-linked to the terminal Gal was tolerated well. In comparison, Galectin-8¹⁰⁴ had preference toward modified LNT with a terminal α1–2-fucose (3) or α2–3-Neu5Ac (3S2). It had decreased binding to 3S5 with both a terminal α2–3-Neu5Ac and an α1–3-fucose linked to the Glc at the reducing end of the LNT core; and to 19 with both a terminal α1–2-fucose and a Neu5Ac α2–6-linked to the internal GlcNAc of LNT (2). On the other hand, Galectin-8 bound to nonmodified LNnT (10) and the additional of an α1–3-fucose linked to the internal GlcNAc (16) or a terminal α2–6-linked Neu5Ac (6S10) decreased, but did not block, the binding. It also bound to 3′SL weakly.

For human E-selectin¹⁰⁵ (Figures 4 and S5d, see ESI),¹⁰⁶ in addition to binding to sialyl Lewis a-terminated structures without (3S8) or with (3S6) an α1–3-fucose linked to the Glc at the reducing end, its binding to a structure presenting sialyl Lewis x (3S16) was observed, aligning well with the result from a previous report.¹⁰⁷

Nonganglioside glycan-binding preferences of the B subunits of several AB₅ toxins¹⁰⁸ including Vibrio cholera toxin subunit B (CtxB),¹⁰⁹ extraintestinal E. coli toxin (EcxAB), Shiga toxigenic E. coli subtilase cytotoxin (SubAB), and typhoid fever-causing Salmonella enterica serovar Typhi pertussis-like toxin subunit B (PltB)^110,111 were readily revealed (Figures 4 and S6a–d, see ESI). The reported preference of CtxB binding to glycans presenting a terminal Lewis y-type structure¹¹² was clearly observed for LNnDFH-I (17). Weaker bindings to 3S16 presenting a terminal sialyl Lewis x structure and to LNFP-III (16) presenting a terminal Lewis x structure were also observed. The previously reported binding to lactodifucotetraose (LDFT)¹¹³ was not observed, indicating its weaker binding compared to 17 with a longer-chain structure. In comparison, EcxAB bound to LNnT (10) preferably. It also bound to its difucosyl Lewis y-type derivative LNnDFH-I (17) and monofucosyl Lewis x-type derivative LNFP-III (16). Weak binding of EcxAB to sialyl Lewis x-type structure 3S16 and to LNnT (2) was also observed. SubAB bound well to LNFP-I (3), the α1–2-fucosylated LNT presenting a terminal type-I H-blood group antigen. SubAB did not show significant binding to other glycans tested. PltB showed preferred binding to α2–3-sialylated LNT (3S2) and weaker binding to α2–3-sialylated LNnT (3S10) or the terminal α1–2-fucosylated LNT (3). It also bound weakly to 3′SL. The binding of these toxins to selective HMOs indicated the potential of applying HMOs and/or their related glycoconjugates as detoxification reagents.

Analysis of the GBP glycan-binding results shown in the heatmap in Figure 4 revealed the identities of HMOs that can be potentially used for inhibiting influenza hemagglutinins (e.g., 3S2, 6S10), regulating human lectin functions (e.g., 3, 8, 16, 3S2, 3S8), and protecting against bacterial toxins (e.g., 3, 10, 17, 3S2). Compared to the lactose-core-HMOs, the richer epitopes presented in tetraose-core-HMOs and HMO-like glycans are clearly observed and their higher application potentials are indicated. On the other hand, compared to more complex HMOs, these epitope-rich tetraose-core-HMOs are more practical targets for large-scale production to explore their potentials as food supplements and/or therapeutics.

Conclusions

We describe a new concept of designing EASyMaps to guide the systematic synthesis of glycans using a glycosyltransferase-based Stepwise One-Pot Multienzyme (StOPMe) strategy with pregeneration of relatively stable glycosyltransferase donors (StOPMe 2.0). The EASyMaps illustrating optimized routes to access any given targets are highly useful blueprints to guide even nonspecialists for the synthesis of complex glycans. The shared steps and the common glycan intermediates of different targets are also clearly revealed by the EASyMaps. The EASyMap concept and the StOPMe synthetic method are broadly applicable to diverse carbohydrate targets and are demonstrated here for the systematic construction of a comprehensive library of NHCbz-tagged tetraose-core-HMOs and HMO-like glycans in a target-orientated manner. Each HMOβNHCbz target was produced from an easily prepared disaccharide derivative LacβNHCbz in a single vessel via several enzyme-catalyzed glycosylation steps and purified via a single C18-cartridge purification process. The enzymatic regio-selective fucosylation strategy⁶⁰ using sialyltransferase-catalyzed site protection with downstream sialidase-catalyzed deprotection was successfully established in the one-pot strategy (StOPMe) for the first time. In addition to the previously reported enzymes, a recombinant human ST6GALNAC5 was successfully expressed in E. coli as a soluble and active MBP-fusion protein which was effective in catalyzing the preparative-scale synthesis of disialyl LNT and other Neu5Acα2–6GlcNAc-containing glycans. The thirty-two tetraose-core-HMOs and HMO-like glycans systematically synthesized in this study include 13 structures that have not been found from human milk samples. Among these, only twenty-one structures have been synthesized previously from the combined results reported from several groups including us. These structurally pure compounds are valuable standards for qualitative and quantitative analyses of HMOs in human milk samples. The process is suitable for large-scale synthesis and has a potential for automation, adding to the advancements in developing machine-driven and automated chemoenzymatic synthetic platforms.^114−116 The advantage of constructing a comprehensive library of tetraose-core-HMOs and HMO-like glycans presenting diverse terminal glycan-epitopes in HMOs is evident in the results obtained from the multiplex glycan-binding assays using HMO-BSA neoglycoproteins with controlled valencies. The high-throughput multiplex neoglycoprotein bead assays have revealed binding preferences of bacterial toxins and glycan-binding proteins from plants, influenza viruses, and human, and have identified HMO targets for the development of potential HMO-based diagnostics, prebiotics,^117,118 immunoregulators,¹¹⁹ as well as antiviral,^120−123 antibacterial,¹²⁴ and detoxification reagents.¹²⁵ The tetraose-core-HMOs and HMO-like glycans present a wide-range of glycan epitopes resembling those in the more complex HMOs and are richer than those in the simpler lactose-core-HMOs. They are attractive candidates for large-scale production and for the development of HMO-based nutraceuticals, diagnostics, and therapeutics.

Glossary

Abbreviations

ATP

adenosine 5′-triphosphate

BSA

bovine serum albumin

Cbz

carboxybenzyl

CMP-Neu5Ac

cytidine 5′-monophosphate-N-acetylneuraminic acid

CTP

cytidine 5′-triphosphate

DSLe^a

disialyl Lewis a

EASyMap

enzyme assembly synthetic maps

Fuc

L-fucose

GBP

glycan-binding protein

Gal

galactose

GDP-Fuc

guanosine 5′-diphosphate-L-fucose

GFP

green fluorescence protein

Glc

glucose

GlcNAc

N-acetylglucosamine

GlyT

glycosyltransferase

GTP

guanosine 5′-triphosphate

HMO

human milk oligosaccharide

IAV

influenza A virus

Lac

lactose

LNT

lacto-N-tetraose

LNnT

lacto-N-neotetraose

MALDI-TOF MS

matrix-assisted laser desorption/ionization time-of-flight mass spectrometry

Neu5Ac

N-acetylneuraminic acid

SA

sugar activation

StOPMe

stepwise one-pot multienzyme

TLC

thin-layer chromatography

UDP-GlcNAc

uridine 5′-diphosphate N-acetylglucosamine

UDP-Gal

uridine 5′-diphosphate galactose

UTP

uridine 5′-triphosphate

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.4c01094.

• Additional experimental details, materials, and methods, including high-resolution mass spectrometry (HRMS) results and spectra for final glycan products, as well as ¹H and ¹³C nuclear magnetic resonance (NMR) results and spectra for new final glycan products (PDF)

Author Present Address

^‡ Present Address: State Key Laboratory of Green Pesticide, Key Laboratory of Green Pesticide and Agricultural Bioengineering, Ministry of Education, Center for Research and Development of Fine Chemicals of Guizhou University, Guiyang, Guizhou 550025, China

Author Contributions

¹ Y.B. and A.K.A. contributed equally to this work. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. CRediT: Yuanyuan Bai conceptualization, data curation, formal analysis, investigation, methodology, writing - original draft, writing - review & editing; Anand Kumar Agrahari conceptualization, data curation, formal analysis, investigation, methodology, writing - original draft, writing - review & editing; Libo Zhang conceptualization, data curation, formal analysis, investigation, methodology, writing - original draft, writing - review & editing; Hai Yu conceptualization, data curation, formal analysis, investigation, methodology, writing - original draft, writing - review & editing; Xiaoxiao Yang data curation, formal analysis, investigation, methodology, writing - original draft, writing - review & editing; Zimin Zheng data curation, investigation, methodology, writing - original draft, writing - review & editing; William Su data curation, investigation, methodology, writing - original draft, writing - review & editing; Jingxin Fu data curation, investigation, methodology, writing - original draft, writing - review & editing; Xi Chen conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, writing - original draft, writing - review & editing.

This work was supported by the United States (U.S.) National Institutes of Health (NIH) grant under the award number R01GM148568. The Thermo Scientific Q Exactive HF Orbitrap Mass Spectrometer was purchased with a U.S. NIH Shared Instrumentation Grant under the award number S10OD025271. The Bruker UltrafleXtreme MALDI-TOF/TOF Mass Spectrometer was purchased with a U.S. NIH Shared Instrumentation Grant under the award number S10OD18913. The Bruker AVANCE-800 NMR spectrometer was purchased with a grant funded by the U.S. National Science Foundation under the award number DBI-0722538.

The authors declare no competing financial interest.