Expeditious chemical synthesis of xylomannans disproves the proposed antifreeze activities

ABSTRACT Cold-adapted species are able to generate cryoprotective proteins and glycoproteins to prevent freezing damage. The [→4)-β-D-Manp-(1→4)-β-D-Xylp-(1→]n xylomannan from the Alaska beetle Upis ceramboides was disclosed by Walters and co-workers in 2009 as the first glycan-based antifreeze agent, which was later reported to be found in diverse taxa. Here, we report the rapid synthesis of four types of xylomannans, including the proposed antifreeze xylomannan up to a 64-mer (Type I), the regioisomeric [→3)-β-D-Manp-(1→4)-β-D-Xylp-(1→]n 16-mer (Type II), the diastereomeric [→4)-β-L-Manp-(1→4)-β-D-Xylp-(1→]n 16-mer (Type III) and the block-wise [→4)-β-D-Manp-(1→]m[→4)-β-D-Xylp-(1→]n 32-mer (Type IV), by employing a strategic iterative exponential glycan growth (IEGG) process. The nuclear magnetic resonance spectral data of the alleged natural xylomannan are in accordance only to those of the block-wise Type IV glycan and none of these synthetic xylomannans has been found to be capable of inducing thermal hysteresis. These results disprove the previous reports about the natural occurrence of antifreeze xylomannans.


INTRODUCTION
Subzero temperatures induce the formation of ice crystals in living organisms, causing cell membrane rupture, organelle damage and cell death.Therefore, a variety of psychrotolerant species have evolved with cryoprotective agents to survive freezing environments.The production of antifreeze proteins (AFPs) and glycoproteins (AFGPs) plays a pivotal role in the adaptation to subzero environments [1 -3 ].The AF(G)Ps are able to not only lower the freezing point of body fluid non-colligatively, but also inhibit the growth of small ice crystals into deleterious large ones [4 ].The difference between the colligative melting and hysteretic freezing points, termed thermal hysteresis (TH), is diagnostic for the presence of macromolecular antifreeze substances.In 2009, Walters and co-workers identified an antifreeze xylomannan-the first non-protein THproducing biomolecule-from the freeze-tolerant Alaskan beetle Upis ceramboides by means of iceaffinity purification [5 ].This xylomannan exhibited a remarkable TH of 3.7 ± 0.3°C at 5 mg/mL, which is comparable to that of the most active insect AFPs, as well as potent inhibition of ice recrystallization [5 ,6 ].Intriguingly, this type of TH-producing xylomannan was later found in various organisms across diverse taxa, including plants, frogs and insects, indicating its significance in cold tolerance although the exact function remains elusive (Fig. 1 a) [6 -8 ].
Cryopreservation of cells and biologics is essential in all biomedical research from routine sample storage to cell-based therapies, making the development of biocompatible cryoprotectants such as polysaccharides highly desirable [9 ].Considerable efforts have been devoted to delineating the natural antifreeze xylomannan [10 -13 ].The chemical structure of the antifreeze xylomannan was initially characterized to be [→ 4)-β-D-Man p -(1→ 4)-β-D-Xyl p -(1→ ] n by Walters and co-workers on the basis of nuclear magnetic resonance (NMR) and mass spectrometry (MS) analysis combined with enzymatic degradative studies [5 ].However, small possibilities of block-wise and/or branched arrangement of the Man p and Xyl p units were not excluded [5 ,10 ].Although accompanying lipid components were detected in the ice-purified sample, they have not been proven to be covalently linked to the glycan.In 2011, the Crich group synthesized three [→ 4)- Comparison of the NMR spectroscopic data supported the initially proposed xylomannan structure [10 ].At around the same time, Ito and collaborators described the chemical synthesis of a similar xylomannan tetrasaccharide; NMR analysis and molecular modeling indicated a helical amphiphilic nature of the xylomannan, which could possibly confer the ice-binding capability [11 ].In 2019, Serianni and co-workers synthesized five 13 C-labelled xylomannan glycans; conformational analysis based on NMR experiments, density functional theory calculations, coupled with molecular dynamic simulations also suggested the helical amphiphilic topography [12 ].
Notwithstanding, the structure of the antifreeze xylomannan remained not conclusive in view of the suspicious discrepancies in the NMR data between the synthetic oligosaccharides and the authentic xylomannan, which were ascribed to the short length of the synthetic fragments.In addition, no TH property was detected for the short synthetic xylomannan [11 ].As such, the synthesis of longer xylomannans has become necessary for the structural confirmation of the natural xylomannan as well as the verification of its antifreeze activity.In fact, the synthesis of glycans that are longer than 20-mer has only been performed sporadically, and most of those syntheses relied on linear or multiplicative assembly tactics [14 ].In 2020, we developed an iterative exponential glycan growth (IEGG) strategy based on the gold(I)-catalysed glycosylation method for the synthesis of a 128-mer rhamnomannan relevant to the O -antigen of Bacteroides vulgatus [15 ].This IEGG process entailed a parallel synthesis of glycosyl donor and acceptor from the same precursor glycan and subsequent glycosylation to produce the double-sized glycan, which can be invoked to the next cycle of assembly (Fig. 1

Synthesis of the proposed natural xylomannans (Type I)
The initially proposed antifreeze xylomannan contains two types of glycosidic connections, i.e.Man-(1 β→ 4)-Xyl and Xyl-(1 β→ 4)-Man.Stereoselective formation of the β-mannopyranoside linkages is known to be difficult, whereas the βxylopyranoside linkages can be robustly constructed with the assistance of neighboring group participation [16 ,17 ].Therefore, we decided to fix the βmannopyranoside linkage at a disaccharide level and to extend the sugar chain resorting to the β-specific xylopyranosylation.
Our synthesis commenced with a large-scale preparation of the ManXyl disaccharide building blocks (Fig. 2 a).Activation of mannopyranosyl sulfoxide donor 1 with trifluoromethanesulfonic anhydride in the presence of 2,4,6-tri-tertbutylpyrimidine followed by the addition of xylosederived acceptor 2 effectively provided disaccharide 3 (83%, β/ α > 19 : 1) [10 ,18 ].The 2-O -acetyl group in 3 was changed to the benzoyl group to prevent the possible generation of orthoesters during the planned IEGG assembly [19 ].Regioselective reductive opening of the benzylidene acetal with hydrogen chloride and sodium cyanoborohydride delivered disaccharide acceptor 4A in 87% yield [10 ,20 ].Capping of the free hydroxyl group in 4A (TBSOTf, 2,6-lutidine) followed by oxidative cleavage of the anomeric 4-methoxyphenyl (MP) group with ceric ammonium nitrate (CAN) and subsequent esterification with o -hexynylbenzoic acid furnished disaccharide donor 4D smoothly (69% over three steps).Coupling of disaccharide acceptor 4A and donor 4D under the Au(I)-catalysed conditions (0.1 equiv.Ph 3 PAuNTf 2 , 5 Å MS, CH 2 Cl 2 ) proceeded well, leading to tetrasaccharide 5 in a high 87% yield [21 ].Treatment of 5 with tetrabutylammonium fluoride (TBAF) and acetic acid gave rise to tetrasaccharide acceptor 5A nearly quantitatively (95%).Oxidative cleavage of the anomeric MP group with CAN followed by condensation with o -hexynylbenzoic acid produced tetrasaccharide 5D in a satisfactory 75% yield.These robust transformations laid the foundation for the subsequent IEGG assembly of the desired xylomannans, which involved the removal of tert -buty ldimethy lsily l (TBS) ether with TBAF in the presence of acetic acid, cleavage of the anomeric MP group under the oxidation of CAN and subsequent condensation of the hemiacetal with o -hexynylbenzoic acid using 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide hydrochloride and 4-dimethylaminopyridine, and Au(I)-catalysed glycosylation with the acceptor and donor.To our delight, this IEGG assembly cycle could be consecutively applied as many as six times, providing fully protected xylomannans up to a 128-mer ( 6 -10 ) in 73%-95% yields.Further implementation of the IEGG cycle led to 256-mer 11 successfully, albeit in only 25% yield.Remarkably, the present IEGG strategy offered a pragmatic approach to polysaccharides, with 64-mer 9 and 128-mer 10 being prepared on scales of 1.1 and 0.6 g, respectively.It is noteworthy that gel permeation chromatography was proven to be viable to purify the long glycans such as 64-mer 9 , 128-mer 10 and 256-mer 11 [15 ].
Next, we turned to the global cleavage of the protecting groups.The deprotection process required three essential operations: (i) removal of the terminal silyl group with TBAF and acetic acid that had been carried out for the preparation of glycosyl acceptors; (ii) removal of the benzyl groups via hydrogenolysis catalysed by the combination of 10% Pd/C and 20% Pd(OH) 2 /C [22 ,23 ]; (iii) removal of the acyl groups via transesterification with methanol in the presence of NaOMe.A redundant acetylation step after hydrogenolysis was found to be necessary for the 8-mer and longer glycans, as the release of numerous hydroxyl groups significantly depressed their solubilities in organic solvents.In addition, repetitive hydrogenolysis was required for 16-mer and longer glycans, because a single hydrogenolysis step was not able to cleave all the benzyl C-O bonds.Thus, five free glycans ( 5F -9F ) up to 64-mer were prepared in 50%-89% yields.We noticed that these synthetic xylomannans tended to aggregate in water as the sugar length increased.Particularly, 16-mer 7F showed a poor solubility of ∼3.5 mg/mL in water, which was lower than that used for the assessment of the TH-producing activity for the natural xylomannan [5 ].Unexpectedly, the NMR spectroscopic data of 16-mer 7F differed apparently from those of the natural isolate ( vide infra ; for details, see Table S1).

Synthesis of 16-mer xylomannans 13F and 15F corresponding to the possible regioisomeric (Type II) and diastereoisomeric (Type III) xylomannans
Considering that the endo β-(1→ 4)-xylosidase used in the original structural characterization might also be able to cleave 1,3-β-xylopyranosidic linkage [24 ], the regioisomeric [→ 3)-β-D-Man p -(1→ 4)-β-D-Xyl p -(1→ ] n could not be ruled out from being the authentic natural xylomannan.Indeed, Crich and co-workers once excluded this alternative Xyl-(1 β→ 3)-Man linkage via NMR comparison with a synthetic tetrasaccharide [10 ].The preparation of a longer glycan, such as 16-mer 13F , for comparison with the natural and synthetic Type II xylomannan became imperative to solve the present structural puzzle (Fig. 2 b).Thus, starting with the coupling of disaccharide donor 12D and acceptor 12A , three rounds of the IEGG process smoothly afforded the fully protected 16-mer 13 on a scale of 2.2 g, which was then transformed into free 16mer 13F in a good 75% yield by using the aforementioned deprotection procedure (for details, see Figs S15-S18).
Another possible structure for the natural xylomannan was the diastereomeric Type III in which Lmannose was incorporated instead of the previously assigned D-mannose.Although scarce, L-mannose has been found in natural glycans [25 -27 ].Thus, we also set out to synthesize the relevant 16-mer 15F (Fig. 2 c).Subjection of disaccharide donor 14D and acceptor 14A to three rounds of IEGG assembly provided the fully protected 16-mer 15 on a scale of 1.8 g.Subsequent removal of the TBS ether, benzyl groups and benzoyl groups furnished 15F in 71% yield (for details, see Figs S19-S22).
The synthetic regioisomeric (Type II) and diastereoisomeric (Type III) xylomannan 16-mers 13F and 15F showed better water solubility than the Type I 16-mer 7F .Unfortunately, discernible deviations were observed for their NMR spectroscopic data in comparison with those of the isolated sample, although their water solubility was enhanced ( vide infra ; for details, see Tables S2 and S3).

Synthesis of block-wise xylomannan (Type IV) 32-mer 23F
Upon carefully examining the structural data of the natural xylomannan reported by Walters and co-workers [5 ], we found the sequential loss of a similar monosaccharide unit in the mass spectrum, implying the existence of oligomeric mannan and xylan segments.Thus, we devised a block-wise 32-mer 23F with a Man/Xyl molar ratio of 9 : 7, so as to match the reported Man/Xyl ratio for the natural xylomannan.The synthesis of 32-mer 23F is depicted in Fig. 3 .The D-mannuronate derivative was selected as the precursor of the D-mannose unit due to its β-directing capacity in glycosylation [28 -31 ], thereby enabling convergent construction of the embedded β-1,4-mannan substructure without resorting to 4,6-O -benzylidene protection.Thus, condensation of tetramannuronate imidate 16 with disaccharide acceptor 4A under the action of triflic acid delivered hexasaccharide 17 in 43% yield, in addition to 25% of the α-anomer [31 ,32 ].Removal of the terminal levulinoyl group with hydrazine acetate followed by glycosylation with 16 appended another tetramannuronate segment, giving decasaccharide 19 in a good 87% yield and satisfactory 8 : 1 β/ α selectivity.Treatment of 19 with hydrazine acetate produced acceptor 20 smoothly.Next, 20 was coupled with o -hexynylbenzoate 21 under the promotion of Ph 3 PAuNTf 2 to afford 16-mer 22 in an excellent 95% yield.Subjection of 22 to one round of the IEGG process doubled the length of the sugar chain effectively, furnishing the fully protected 32-mer 23 on a scale of 0.5 g.Finally, an array of steps including (i) cleavage of the TBS group with TBAF, (ii) reduction of the D-mannuronate residues to the corresponding D-mannose units with lithium triethylborohydride [30 ], (iii) hydrogenolysis under the catalysis of 10% Pd/C and 20% Pd(OH) 2 /C and (iv) saponification with 0.5 N NaOH solution led to the desired block-wise 32-mer 23F in an appreciable 22% yield.

Comparison of NMR spectroscopic data and TH-producing activities
With the synthetic xylomannans in hand, we were able to systematically compare their NMR spectroscopic data with those presented in the literature (Fig. 4 ; for details, see Figs S27-S30).Notably, the resonances of the internal monosaccharide residues were coincident and they dominated the spectra when the length of the sugar chain reached 16 mer, thus simplifying the interpretation of these signals.
In line with the previous reports, the 1 H and 13 C spectra of 16-mer 7F resembled those of the natural isolate, except the variances of 1.9 and 1.6 ppm for the anomeric carbons of the mannose and xylose residues, respectively [10 ,11 ].The 13 C NMR data of 13F differed greatly from those of the natural xylomannan, possessing significant discrepancies of 9.0 and 10.9 ppm for the C3 and C4 of the mannose residue, respectively, therefore strongly ruling out the regioisomeric Xyl-(1 β→ 3)-Man linkage.Not surprisingly, the 1 H and 13 C NMR spectroscopic data of 15F did not agree with those of the natural xylomannan, indicating that the rare L-mannose was not the component.To our delight, signals that showed in the NMR spectra of 23F were virtually identical to those reported by Walters and co-workers [5 ], with the deviations in the 1 H and 13 C chemical shifts not exceeding 0.03 and 0.2 ppm, respectively.It is of note that the small discrepancies in the 1 H NMR spectrum between synthetic 23F and natural isolate can be ascribed to the trace amounts of impurities in the natural isolate.Moreover, the 1 H- 13 C hetereonuclear single quantum coherence (HSQC) spectrum of block-wise 23F closely resembled that of the natural xylomannan (Fig. 4  The antifreeze properties of the synthetic xylomannans were tested.Alternating xylomannans 16-mer 7F , 32-mer 8F and 64-mer 9F exhibited only marginal TH ranging from 0.01 to 0.03°C at their saturated concentrations.Meanwhile, isomeric xylomannan 16-mer 13F and 15F were found to be not capable of inducing TH at a concentration of 10 mg/mL.The block-wise 32-mer 23F , whose spectral data matched well with those of the natural isolate, did not produce any detectable TH at 10 mg/mL.Matrix-assisted laser-desorption ionization mass analysis of the natural xylomannan indicated an average molecular weight of 1.0-2.4kDa [5 ], while the molecular weight of 23F is 4.9 kDa.Considering that the superior ionization capacity of short glycans over longer ones usually led to underestimation of the average molecular weight of polydisperse polysaccharides [33 ,34 ], we believed that monodisperse 23F could serve as a representative of the natural xylomannan.Therefore, the fact that block-wise 32-mer 23F did not show any TH-producing activity was sufficient  to negate the antifreeze property of the naturally occurring xylomannan.We surmised that the icepurified substance that was studied by Walters and co-workers might have contained some hyperactive components that conferred the antifreeze activity [7 ,35 ,36 ].The synthesis features an IEGG strategy that permits extremely rapid extension of the glycan sizes [15 ,37 -39 ].In fact, a fully protected Type I xylomannan 256-mer ( 11 ) has been obtained from disaccharide building blocks 4A and 4D through only seven rounds of assemblies.The successful synthesis also testifies to the robustness of the Au(I)-catalysed glycosylation with glycosyl orthoalkynylbenzoates [21 ], which have been applied in coupling with high-molecular-weight glycan fragments, as showcased by the semi-gram-scale construction of 128-mer 10 via a [64-mer + 64mer] glycosylation.It is noteworthy that a 100-mer linear mannan and a 1080-mer linear arabinan have recently been synthesized using stepwise solid-phase synthesis and multiplicative liquid-phase assembly, respectively [40 -43 ].The present IEGG strategy offers an expeditious and scalable solution to the synthesis of structurally complex polysaccharides and thus facilitates ensuing functional studies.

DISCUSSION
The availability of the homogeneous xylomannans via chemical synthesis has allowed us to examine carefully the structure and activity of the natural antifreeze xylomannan, which was reported to occur in diverse cold-adapted species.In contrast to the proposed alternating structure of [→ 4)-β-D-Man p -(1→ 4)-β-D-Xyl p -(1→ ] n , comparison of the NMR spectroscopic data with those of our synthetic xylomannans strongly supports a hybrid of β-1,4-mannan and β-1,4-x ylan, w hich might result from occasional incorporation of xylose and mannose units during the enzymatic synthesis of mannan and xylan, respectively.All the synthetic xylomannans did not show any TH inducing activity, including the block-wise 32-mer xylomannan 23F , which matched well with the reported antifreeze isolates in NMR data.Xylan and mannan are frequently found in plants and fungi but rarely in animals [44 ].From the viewpoint of glycan biosynthesis, which requires specific glycosyltransferases, synthesis of a same type of xylomannan in diverse species, as reported in the literature [5 -8 ], is questionable.Taken together, our study disproves the occurrence of TH-active xylomannans in nature.The non-protein antifreeze substance in the natural isolates and the mechanism of action await further investigation.

Figure 1 .
Figure 1.The originally proposed and presently synthesized antifreeze xylomannans and the general synthetic strategy.(a) The alleged antifreeze xylomannan (Type I).(b) Schematic illustration of the iterative exponential glycan growth (IEGG) strategy for glycan synthesis.(c) Synthetic xylomannans (Type I-IV) in this work.PG, protecting group; LG, leaving group; MP, 4-methoxyphenyl.

Figure 4 .
Figure 4. NMR spectral comparison of the natural xylomannan and synthetic xylomannans.(a) Overlaid 1 H NMR spectra of the natural xylomannan and the synthetic xylomannans 7F , 13F , 15F and 23F .(b) 1 H-13 C HSQC spectra of the natural xylomannan and block-wise 23F .The 1 H NMR and HSQC spectra of the natural xylomannan was reproduced from [5 ].