Reversible derivatization of sugars with carbobenzyloxy groups and use of the derivatives in solution-phase enzymatic oligosaccharide synthesis

B T Simple protocols for attaching and detaching carbobenzyloxy (Cbz) groups at the reducing end of sugars was developed. Briefly, lactose was converted into its glycosylamine, which was then acylated with carbobenzyloxy chloride in high overall yield. The obtained lactose Cbz derivative was used in sequential glycosylations using glycosyltransferases and nucleotide sugars in aqueous buffers. Isolation of the reaction products after each step was by simple C-18 solid-phase extraction. The Cbz group was removed by catalytic hydrogenolysis or catalytic transfer hydrogenation followed by in situ glycosylamine hydrolysis. In this way, a trisaccharide (GlcNAc-lactose), a human milk tetrasaccharide (LNnT), and a human milk pentasaccharide (LNFPIII) were prepared in a simple and efficient way.


Introduction
Of the biopolymers, oligosaccharides and polysaccharides are the most difficult to prepare by chemical synthesis. Nucleic acids and peptides can be routinely synthesized by stepwise solid-phase chemical synthesis, but analogous chemical synthesis of oligo-and polysaccharides has not yet reached the same stage of maturation. The major reason for this is that both α and β stereoisomers can be formed in classical organic synthesis coupling reactions between monosaccharides resulting in complex product mixtures already after a few steps, and also the fact that a comparatively large effort is required to prepare all the different protected monosaccharides needed [1].
Nature uses strictly regio-and stereospecific enzymes to produce oligo-or polysaccharides [2] without the use of protecting groups, and since many enzymes are now available through genetic engineering, they have often been used in synthesis of oligosaccharides. For example [3], a sialyl lewis x tetrasaccharide derivative was synthesized starting from a solid-phase linked monosaccharide, using, sequentially, the appropriate activated monomers/enzymes. After cleavage from the solid phase, a tetrasaccharide was isolated in 57% overall yield. Thus, solid-phase enzymatic oligosaccharide synthesis is possible, and might be a viable alternative in some cases. However, large excesses of expensive reagents (enzymes and activated monosaccharides) are needed to drive the solid-phase coupling reactions to completion.
Solution phase synthesis, where much less reagent excesses are needed, is for shorter oligomers a more realistic alternative and has been used in many cases, even on a large scale. However, a simple and rapid protocol for isolation of the reaction product after each step is essential. Isolation of an oligosaccharide product from a typical enzymatic reaction mixture is not simple, unless the product bears a specific group ("tag") that enables it to be easily isolated. Boons et al. recently [4] reported use of charged tags in combination with ion-exchange resin adsorption/desorption in stepwise enzymatic synthesis of neutral oligosaccharides. Hindsgaul et al. [5] used oligosaccharides with hydrophobic tags for fast isolation of the enzymatic reaction products by reversed phase silica solid-phase extraction. Gangliosides and other glycolipids were assembled [6] using a similar approach. The hydrophobic tag principle was also used by us [7,8] in examples of enzymatic syntheses of oligosaccharides. The hydrophobic group was then attached either by reductive amination at the sugar reducing end or by conversion into the glycosylamine, acylation with diethyl squarate, and reaction with an alkylamine. The reductive amination-mediated linkage was however not easily cleaved, which was a disadvantage in some cases. The squarate-mediated linkage, on the other hand, could be cleaved to regenerate the free reducing end and was flexible in the choice of lipid. However, the overall squarate-mediated attachment-cleavage procedure was somewhat laborious.
In an effort to simplify, we now report attachment of hydrophobic carbobenzyloxy (Cbz) groups to sugar reducing ends and use of the Cbz derivatives in stepwise enzymatic synthesis of oligosaccharides in solution. Facile removal of the Cbz group/glycosylamine hydrolysis after finished synthesis produced oligosaccharides with a free reducing end. The advantage of synthesizing an oligosaccharide with a free reducing end is that it provides a choice of many subsequent modifications, such as reduction, reductive amination with different amines, direct conjugation with macromolecules, etc.

Results and discussion
The conversion of reducing sugars into glycosylamines is a common way to create an N-nucleophilic center for site-specific functionalization (e.g. acylation). Reaction of sugars with either methanolic ammonia [9], aqueous ammonia/ammonium hydrogen carbonate [10,11], ammonium carbamate/methanol [12], or saturated aqueous ammonium hydrogen carbonate [13] have been reported to give glycosylamines or the corresponding carbamates in good yields. We used a slight modification (24 h reaction of a reducing sugar with saturated aqueous ammonium bicarbonate at 40 • C and pH 9) of the Likoshersterov procedure [13] to obtain a crude glycosylamine preparation suitable for acylation. In previous work, acryloyl chloride [14], fluorenylmethyloxycarbonyl chloride [11,15], acylthiones [9], or fatty acid chlorides [10] have been reacted with glycosylamines to obtain the corresponding glycosylamides. We sought a reasonably cheap acylating agent that could be used in excess and that introduced a hydrophobic group which could be removed later in a simple way. Carbobenzyloxy chloride fulfilled these criteria. Acylation of crude lactose glycosylamine with an excess of carbobenzyloxy chloride in aq sodium carbonate/dioxane gave, after purification by solid-phase extraction with C-18 silica, the lactose-Cbz derivative 3 (72% overall yield from lactose (Scheme 1). Cbz-glycosylamides have been described before [9,16]. The H NMR spectrum of 3 in D 2 O at 25 • C showed similar broadening of the H-1 signal as was noted for analogous compounds [8]. However, raising the temperature to 80 • C gave a sharp H-1 doublet with J = 9.2 Hz, indicative of the expected β configuration.
The lactose Cbz-derivative was converted back to lactose either by catalytic hydrogenolysis (H 2 /Pd/C) in slightly acidic aqueous media or [17] by catalytic transfer hydrogenolysis (ammonium formate/Pd/C), demonstrating the reversibility of the Cbz derivatization.
In analogy with earlier work [7,8] we subjected the lactose-Cbz derivative 3 to consecutive enzymatic glycosylations, with isolation of the product after each step by reversed-phase C-18 solid-phase extraction. Products other than the desired products were seldom observed in the enzymatic reaction mixtures. Minor amounts of unreacted starting oligosaccharide or glycolysis products (with crude enzyme preparations) were occasionally seen. However, the choice of fucosyltransferase in the last step was crucial. The 1-3/4 fucosyltransferase from Helicobacter mustela gave more than 90% specific 4-alpha-monofucosylation and no detectable difucosylation product (TLC and H NMR). In contrast, preliminary attempts with recombinant Fuc-T5 gave substantial amounts of difucosylation product in addition to monofucosylated products (TLC and LC-MS, data not shown).
After the desired oligosaccharide Cbz-derivatives had been assembled, treatment with hydrogen/Pd/C or ammonium formate/Pd/C in weakly acidic aqueous medium yielded the free oligosaccharides 5, 7, and 9, which were finally purified by gel filtration. Scheme 2 shows the overall yields (from 3) of the pure oligosaccharides. Two occur [18,19] in human milk (7 = LNnT and 9 = LNFPIII). So far, we did not encounter Cbz-oligosaccharides with insufficient adsorption to C-18 cartridges. However, we observed slightly weaker adsorptions when the sugar chains became longer. We are currently investigating C-18 silica adsorptions of neutral Cbz-oligosaccharides higher than pentasaccharides. Recently, Chen et al. reported [16] gram-scale isolation of Cbz-derivatized sialic acid oligosaccharides on C-18 columns. Thus, even some negatively charged oligosaccharides could be within the scope of the present approach, as well as larger scale purifications.

Conclusion
Reversible derivatization of lactose with a Cbz group enabled efficient and simple stepwise synthesis of neutral reducing oligosaccharides, utilizing the appropriate glycosyltransferases and nucleotide sugars. Isolation of glycosylation products after each step was by simple C-18 solid-phase extraction, and the final products after removal of the Cbz group were isolated in good yields.
Synthesis and purification of the pentasaccharide 9 (three coupling steps) starting from the lactose-Cbz derivative 3 required 4 days, most of this time being overnight coupling time. Compared to "traditional" stepwise chemical synthesis this is much faster and requires less laboratory skills. A future increased availability of nucleotide sugars and enzymes would make this type of enzymatic approach even more attractive.
Compound 3 [16]: A solution of ammonium bicarbonate (8.0 mL, saturated during 2 h stirring, then pH-adjusted to 9 with ammonia) was used to dissolve lactose monohydrate (180 mg, 0.5 mmol). After being kept for 24 h at 40 • C, the solution was evaporated and the residue was coevaporated x 4 with water (30 mL). Uptake in water (30 mL) and lyophilization gave a residue (192 mg), presumably consisting mainly of the oligosaccharide β -glycosylamine or its corresponding carbamate [14]. A TLC analysis at this stage showed a very weak rf 0.2 spot (sulfuric acid detection) and a much slower, strong rf 0.05 spot (sulfuric acid and ninhydrin detection). The entire batch was taken up in 0.5 M aq sodium carbonate/dioxane (10/5 mL, 5.0 mmol sodium carbonate), then stirred at r. t. while a solution of fresh carbobenzyloxy chloride (0.60 mL, 4.2 mmol) was added. Analysis by TLC showed that, after 3 h, all starting material had disappeared. The mixture was concentrated to about half the volume (to remove most of the dioxane), diluted with water (10 mL) then washed with diethyl ether (20 mL). The etheral layer was washed with water (5 mL). The combined aq layers were concentrated to approx. 15 mL, then applied to a C-18 column (15 g, wetted with methanol, then equilibrated with water). The column was eluted first with water (30 mL), then with methanol-water mixtures (20 mL each, from 10 to 60% methanol, 10% increments). The fractions were checked by TLC, appropriate fractions were pooled, evaporated to a small volume, and lyophilized to give 3 as a colorless solid (172 mg, 72%). Crystals were obtained from cold (+4  (7) and pentasaccharide LNFPIII (9). All yields are calculated from compound 3. from a baloon-syringe combination, then flushed with hydrogen and left stirring under hydrogen at rt for 2 h, after which TLC showed complete disappearance of starting material, a major lactose spot, and a weak slow-moving spot. After flushing the reaction vessel with nitrogen, the mixture was filtered (5 μm disposable filter), and the filter was washed with water and methanol. The filtrate was evaporated, coevaporated x 1 with water, then taken up in water. This solution, according to TLC, was almost pure lactose, however, some slow-moving impurities remained. The solution was evaporated and the residue (26 mg) was purified by gel filtration to give pure lactose (15 mg, 70%), identical to an authentic sample.
Deprotection of 3 to give lactose (1), method 2 (with ammonium formate/Pd/C): A solution of 3 (30 mg, 0.063 mmol) in water (3 mL) containing conc. aqueous ammonia (110 microL, 2.0 mmol) and formic acid (80 microL, 2.0 mmol) was mixed with a suspension of Pd/C (10%, 15 mg) in methanol (2 mL) and then stirred at r. t. while being monitored by TLC. The starting material had disappeared in less than 10 min, and there was a major glycosylamine spot (slow) and a minor lactose spot (faster). Another portion of formic acid (40 microL) was added, after which there was a gradual (over several hours) transformation of glycosylamine into lactose. The mixture was left overnight at r. t., after which TLC (charring with sulfuric acid) detected a single spot, corresponding to lactose. The mixture was filtered (5 μm disposable filter), and the solids were washed with water and methanol. The filtrate was evaporated to dryness, coevaporated once with water and then lyophilized. The residue (33 mg) was purified by gel filtration, which gave pure lactose (21 mg, 95%), indistinguishable from an authentic sample.
Compound 5 (GlcNAc-lactose): A solution of 3 (60 mg, 126 micromol), UDP-GlcNAc (disodium salt, Mw 651.3, 150 mg, 230 micromol, and BSA (50 mg) in sodium cacodylate buffer (0.25 M, with 0.015 M MnCl2, pH 7.3, 12 mL) was mixed with GlcNAc-T (1.0 U, 2.5 mL of a 0.4U/mL stock solution), the mixture was slowly stirred at 35 • C for 24 h. Analysis by TLC showed emergence of a slower, sulfuric acid and ninhydrin-positive spot. At this point, the reaction appeared to be 75-80% finished, with remaining starting material 3 present. More UDP-GlcNAc (50 mg) and GlcNAc-T (1.0 mL) was added, and the mixture was stirred for another 16 h at 35 • C, after which a TLC detected only traces of starting material 3. The mixture was diluted with water (5 mL) and then applied onto a C-18 column (3.0 g, wetted with methanol, then equilibrated with water). The column was eluted first with water (10 mL), then with methanol-water mixtures (10 mL each, from 10 to 60% methanol, 10% increments). The fractions were checked by TLC (ethyl acetate-methanol-acetic acid-water 12:3:3:2), fractions containing the major product 4 were pooled, evaporated to a small volume, and lyophilized to give a white solid (72 mg). Analysis by LC-MS showed a major peak with a strong positive ion at m/z 678.2 (M + H) + , and a strong negative ion at 677.1 (M − H) -. The H-and 13C NMR spectra showed the expected signals. A solution of this material (30 mg) in water (4 mL) was mixed with a suspension of Pd/C (10%, 25 mg) in methanol (4 mL) containing acetic acid (40 microL). The 100 mL round-bottom flask was equipped with a stirrer magnet and a septum and was stirred while flushed with nitrogen, then the flask was flushed with hydrogen and left stirring at rt with a slight overpressure of hydrogen for 2 h, after which TLC showed complete disappearance of starting material. After flushing of the reaction vessel with nitrogen, the mixture was filtered (5 μm disposable filter), and the filter was washed with water and methanol. Since TLC detected traces of a glycosylamine-like spot, the filtrate was mixed with another portion of acetic acid (20 microL), and was left at rt overnight in a glass vessel after which TLC indicated disappearance of this spot. The solution was partially evaporated and applied onto a gel filtration column. Appropriate fractions were pooled, partially evaporated and then lyophilized to give pure (by TLC and H NMR) 5 as a colorless powder (17.5 mg, 60%, calculated from 3). Physical constants (NMR and MS) were as reported [8]. micromol), α-lactalbumin (10 mg) and then Gal-T (approx. 0.5-1.0 U) was added. The mixture was slowly stirred overnight at 30 • C, after which TLC indicated a complete reaction. The reaction mixture was diluted with water (20 mL) and applied to a C-18 column (1.2 g, wetted with methanol, then equilibrated with water). The column was eluted first with water (10 mL), then with methanol-water mixtures (12 mL each, from 10 to 60% methanol, 10% increments). The fractions were checked by TLC. Fractions containing product were pooled, evaporated to a small volume, and lyophilized to give a white solid (67 mg) containing compound 6 and minor impurities. Analysis by LC-MS showed a major peak with a strong positive ion at m/z 841.2 (M + H) + , and a strong negative ion at 839.2 (M − H) -. The H-and 13C NMR spectra showed the expected signals. A solution of this material (18 mg) in water (2.5 mL) was mixed with a suspension of Pd/C (10%, 15 mg) in methanol (2 mL) containing acetic acid (20 microL), all in a 50 mL roundbottom flask, which was then equipped with a stirrer magnet and a septum, and stirred while flushed with first nitrogen, then hydrogen. The mixture was left stirring under a slight hydrogen overpressure at rt for 2 h, after which TLC showed complete disappearance of starting material and appearance of a major, slower spot and a minor, even slower spot (presumably the intermediate glycosylamine). After flushing with nitrogen, the mixture was filtered (5 microM disposable filter), and the filter was washed with water and methanol. The filtrate was mixed with another portion of acetic acid (20 microL), and was left at rt overnight in a glass vessel after which TLC indicated disappearance of the presumed glycosylamine spot. The solution was partially evaporated and applied onto a gel filtration column. Appropriate fractions were pooled, partially evaporated and then lyophilized to give pure (by TLC and NMR) 7 as a colorless powder (10 mg, 50%, calculated from 3) identical to an authentic sample of LNnT. Physical constants (NMR and MS) were as reported [18].
Compound 9 (LNFPIII): Compound 6 from the prep above (12 mg, 14.3 micromol) and GDP-fucose diammonium salt (13 mg, approx. 21 micromol) in sodium cacodylate buffer (0.10 M, with 0.015 M MnCl2, pH 7.5, 3.0 mL) was mixed with Fuc-T (100 mU, 0.10 mL of a 1 U/mL stock solution) and the mixture was stirred gently overnight at 35 • C, after which TLC indicated a complete reaction. The mixture was diluted with water (3 mL) and applied to a C-18 column (0.33 g, wetted with methanol, then equilibrated with water). The column was eluted first with water (4 mL), then with methanol-water mixtures (4 mL each, from 5 to 50% methanol, 5% increments). The fractions were checked by TLC. Appropriate fractions were pooled, partially evaporated, and lyophilized to give a white solid (18 mg) containing compound 7 and minor impurities. Analysis by LC-MS showed a peak with a strong positive ion at m/z 987.3 (M + H) + , and the H-and 13C NMR spectra showed the expected signals. This material (18 mg) in water (2.5 mL) was mixed with a suspension of PdC (10%, 20 mg) in methanol (2 mL) containing acetic acid (20 microL). After stirring under hydrogen at room temperature and atmospheric pressure for 1 h, analysis by TLC (ethyl acetatemethanol-acetic acid-water, 6:3:3:2) showed complete disappearance of starting material. The mixture was filtered with a 5 μm disposable filter, and the filter was washed with water and methanol. Since TLC detected traces of a glycosylamine-like spot, the filtrate was mixed with another portion of acetic acid (20 microL), and was left at rt overnight after which TLC indicated disappearance of this spot. The mixture was applied onto a gel filtration column. Appropriate fractions were pooled, partially evaporated and then lyophilized to give a residue (11 mg), which, according to TLC (ethyl acetate-methanol-acetic acid-water, 6:3:3:2) and H NMR in D 2 O, contained LNFPIII [18) and traces of unreacted LNnT (<10%). The estimated yield of pure LNFPIII was 41% calculated from compound 3, and 81% calculated from compound 6.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.