Investigating Glycol-Split-Heparin-Derived Inhibitors of Heparanase: A Study of Synthetic Trisaccharides

Heparanase is the only known endoglycosidase able to cleave heparan sulfate. Roneparstat and necuparanib, heparanase inhibitors obtained from heparin and currently being tested in man as a potential drugs against cancer, contain in their structure glycol-split uronic acid moieties probably responsible for their strong inhibitory activity. We describe here the total chemical synthesis of the trisaccharide GlcNS6S-GlcA-1,6anGlcNS (1) and its glycol-split (gs) counterpart GlcNS6S-gsGlcA-1,6anGlcNS (2) from glucose. As expected, in a heparanase inhibition assay, compound 2 is one order of magnitude more potent than 1. Using molecular modeling techniques we have created a 3D model of 1 and 2 that has been validated by NOESY NMR experiments. The pure synthetic oligosaccharides have allowed the first in depth study of the conformation of a glycol-split glucuronic acid. Introducing a glycol-split unit in the structure of 1 increases the conformational flexibility and shortens the distance between the two glucosamine motives, thus promoting interaction with heparanase. However, comparing the relative activities of 2 and roneparstat, we can conclude that the glycol-split motive is not the only determinant of the strong inhibitory effect of roneparstat.


Introduction
Heparanase is an endo-β-D-glucuronidase that cleaves the glycosaminoglycan chains of heparan sulfate at specific sites, thus modulating the biological function of this proteoglycan expressed at the cell surface of nearly all animal species [1]. Heparan sulfate and heparanase are involved in several pathological conditions, particularly tumor development [2], but also inflammation, diabetes, and atherosclerosis [3]. Heparan sulfate glycosaminoglycan is a linear polysaccharide containing alternating units of uronic acids (D-glucuronic or L-iduronic) and glucosamine, both bearing various substituent groups at various positions [4]. Heparanase degrades heparan sulfate through hydrolysis of the glycosidic linkage between selected glucuronic acid units and glucosamine [5][6][7]. The 3D structure of human heparanase has been recently reported [8].
Drugs targeting heparanase are actively looked for [9,10]. Among the known heparanase inhibitors roneparstat (SST0001, Sigma Tau Research) [11] and necuparanib (M 402, Momenta) [12], are the most advanced in clinical development (SST0001 currently in phase I in advanced multiple myeloma and M 402 in phase II for pancreatic cancer). Roneparstat is obtained from standard porcine mucosal heparin after total N-desulfation, N-acetylation, controlled periodate oxidation, and finally borohydride reduction (the sequence of the last two steps is referred to as glycol-splitting, abbreviated gs), whereas necuparanib is a low molecular weight heparin, resulting from depolymerization of heparin and further periodate oxidation and borohydride reduction. Schematic representations of roneparstat and necuparanib, both containing glycol-split uronic acid moieties, are depicted in Figure 1. Several studies have been carried out to understand the structure-activity relationship of heparin for heparanase inhibition, including variation in the degree of 2-O-sulfation, 6-O-sulfation, N-sulfation, N-acetylation, glycol-split uronic acid [13,14]. These studies led to conclude that O-2-sulfation of IdoA and O-6 sulfation of GlcN were not essential for effective inhibition, and that one N-sulfate group per disaccharide units was in interaction with the enzyme [7]. It was also found that glycol-split of non-sulfated uronic acid dramatically increased the inhibitory activity [15]. Recent studies on the kinetics of heparanase inhibition by roneparstat suggest that it interacts with heparanase by multiple protein-ligand mode depending on the concentration of the inhibitor [16]. However the mechanism of action of both roneparstat and necuparanib at a molecular level is not clearly understood, particularly the role of the "glycol-split" uronic acid units that are key element in the structure must be clarified. Whether these units critically interfere with the active site of heparanase or merely serve for introducing conformational flexibility in the polysaccharide chains is still an open question. To document this aspect we have embarked on the chemical synthesis of homogenous heparan sulfate fragments and their "glycol-split" counterparts. Their activity as heparanase inhibitors will be tested and they will also be used for interaction studies, particularly by NMR. We report here on the synthesis, activity and conformation of a trisaccharide unit 1 and its glycol-split counterpart 2.

Chemistry
It should be underlined first that whereas all glucosamine units in roneparstat are N-acetylated, we decided to synthesize oligosaccharides containing N-sulfated-glucosamine, reflecting the situation in M402, with whom we expect a stronger interaction with heparanase. Indeed a stronger interaction with proteins is expected with N-sulfated glucosamine-containing oligosaccharides compared to their N-acetylated counterparts [7] (N-acetyl groups were introduced in roneparstat with a view to restrict undesired interactions with growth factors which is not an issue with short oligosaccharides, particularly if they do not contain 3-O-sulfated glucosamine units). From a synthesis standpoint, the amino group substituent is introduced during the last step of the synthesis, which makes the preparation of both N-acetylated and N-sulfated compounds very similar although more care must be taken for N-sulfated derivatives due to their higher sensitivity to acidic media. Thus, our primary target molecules were trisaccharide 1 (GlcNS6S-GlcA-1,6anGlcNS) and its derivative 2 (GlcNS6S-gsGlcA-1,6anGlcNS) that contains one single glycol-split glucuronic acid (gsGlcA) unit (Scheme 1). Note that a methyl substituent is present at position 4 of the non-reducing glucosamine in 1 to prevent cleavage of this unit by periodate during the conversion of 1 into 2. The methyl groups also mimics the anomeric carbon of the next unit in a polysaccharide chain. In the framework of our whole project, fully protected 1,6-anhydro containing trisaccharides will be used as synthetic intermediates and, in passing, Several studies have been carried out to understand the structure-activity relationship of heparin for heparanase inhibition, including variation in the degree of 2-O-sulfation, 6-O-sulfation, N-sulfation, N-acetylation, glycol-split uronic acid [13,14]. These studies led to conclude that O-2-sulfation of IdoA and O-6 sulfation of GlcN were not essential for effective inhibition, and that one N-sulfate group per disaccharide units was in interaction with the enzyme [7]. It was also found that glycol-split of non-sulfated uronic acid dramatically increased the inhibitory activity [15]. Recent studies on the kinetics of heparanase inhibition by roneparstat suggest that it interacts with heparanase by multiple protein-ligand mode depending on the concentration of the inhibitor [16]. However the mechanism of action of both roneparstat and necuparanib at a molecular level is not clearly understood, particularly the role of the "glycol-split" uronic acid units that are key element in the structure must be clarified. Whether these units critically interfere with the active site of heparanase or merely serve for introducing conformational flexibility in the polysaccharide chains is still an open question. To document this aspect we have embarked on the chemical synthesis of homogenous heparan sulfate fragments and their "glycol-split" counterparts. Their activity as heparanase inhibitors will be tested and they will also be used for interaction studies, particularly by NMR. We report here on the synthesis, activity and conformation of a trisaccharide unit 1 and its glycol-split counterpart 2.

Chemistry
It should be underlined first that whereas all glucosamine units in roneparstat are N-acetylated, we decided to synthesize oligosaccharides containing N-sulfated-glucosamine, reflecting the situation in M402, with whom we expect a stronger interaction with heparanase. Indeed a stronger interaction with proteins is expected with N-sulfated glucosamine-containing oligosaccharides compared to their N-acetylated counterparts [7] (N-acetyl groups were introduced in roneparstat with a view to restrict undesired interactions with growth factors which is not an issue with short oligosaccharides, particularly if they do not contain 3-O-sulfated glucosamine units). From a synthesis standpoint, the amino group substituent is introduced during the last step of the synthesis, which makes the preparation of both N-acetylated and N-sulfated compounds very similar although more care must be taken for N-sulfated derivatives due to their higher sensitivity to acidic media. Thus, our primary target molecules were trisaccharide 1 (GlcNS6S-GlcA-1,6anGlcNS) and its derivative 2 (GlcNS6S-gsGlcA-1,6anGlcNS) that contains one single glycol-split glucuronic acid (gsGlcA) unit (Scheme 1). Note that a methyl substituent is present at position 4 of the non-reducing glucosamine in 1 to prevent cleavage of this unit by periodate during the conversion of 1 into 2. The methyl groups also mimics the anomeric carbon of the next unit in a polysaccharide chain. In the framework of our whole project, fully protected 1,6-anhydro containing trisaccharides will be used as synthetic intermediates and, in passing, to settle the experimental conditions for periodate oxidation we decided to use a trisaccharide with a 1,6-anhydro-glucose ring at the reducing end (compound 1). Synthesis of the target trisaccharide 1 is depicted in Schemes 2 and 3. They were obtained by an adaptation of methods already used for this type of oligosaccharide [17]. A fully protected equivalent of 1 was prepared first with full orthogonal protection (15, Scheme 2) and then sequentially de-protected and substituted. We prepared first the known key derivative 3 [18] from commercially available levoglucosan (overall yield 21%), using the method of Oikawa et al. [19]. A first attempt to glycosylate 3 using tetraacetyl glucopyranose trichloroacetimidate [20] (8) in the presence of TMSOTf gave very little desired product 9, together with a large quantity of a side product identified as 1,6-anhydro-2-azido-3-O-benzyl-4-O-acetyl-2-deoxy-β-D-glucopyranose. Its structure was confirmed by NMR and MS. Using BF3·Et2O as Lewis acid we could get a much improved yield of 9 (76%) that gave 11 after acetyl group removal and benzylidenation. Benzylation followed by cleavage of the benzylidene protection gave 13 [21] ready for selective oxidation of the primary alcohol by dibromantin-TEMPO. Methylation gave 14 (82% from 11). The known imidate 7 [22] was obtained (Scheme 3) from 3 in 60% yield after methylation followed by acetolysis, selective anomeric deacetylation and treatment with trichloroacetonitrile in the Synthesis of the target trisaccharide 1 is depicted in Schemes 2 and 3. They were obtained by an adaptation of methods already used for this type of oligosaccharide [17]. A fully protected equivalent of 1 was prepared first with full orthogonal protection (15, Scheme 2) and then sequentially de-protected and substituted. We prepared first the known key derivative 3 [18] from commercially available levoglucosan (overall yield 21%), using the method of Oikawa et al. [19]. A first attempt to glycosylate 3 using tetraacetyl glucopyranose trichloroacetimidate [20] (8) in the presence of TMSOTf gave very little desired product 9, together with a large quantity of a side product identified as 1,6-anhydro-2-azido-3-O-benzyl-4-O-acetyl-2-deoxy-β-D-glucopyranose. Its structure was confirmed by NMR and MS. Using BF 3 ·Et 2 O as Lewis acid we could get a much improved yield of 9 (76%) that gave 11 after acetyl group removal and benzylidenation. Benzylation followed by cleavage of the benzylidene protection gave 13 [21] ready for selective oxidation of the primary alcohol by dibromantin-TEMPO. Methylation gave 14 (82% from 11). Synthesis of the target trisaccharide 1 is depicted in Schemes 2 and 3. They were obtained by an adaptation of methods already used for this type of oligosaccharide [17]. A fully protected equivalent of 1 was prepared first with full orthogonal protection (15, Scheme 2) and then sequentially de-protected and substituted. We prepared first the known key derivative 3 [18] from commercially available levoglucosan (overall yield 21%), using the method of Oikawa et al. [19]. A first attempt to glycosylate 3 using tetraacetyl glucopyranose trichloroacetimidate [20] (8) in the presence of TMSOTf gave very little desired product 9, together with a large quantity of a side product identified as 1,6-anhydro-2-azido-3-O-benzyl-4-O-acetyl-2-deoxy-β-D-glucopyranose. Its structure was confirmed by NMR and MS. Using BF3·Et2O as Lewis acid we could get a much improved yield of 9 (76%) that gave 11 after acetyl group removal and benzylidenation. Benzylation followed by cleavage of the benzylidene protection gave 13 [21] ready for selective oxidation of the primary alcohol by dibromantin-TEMPO. Methylation gave 14 (82% from 11). The known imidate 7 [22] was obtained (Scheme 3) from 3 in 60% yield after methylation followed by acetolysis, selective anomeric deacetylation and treatment with trichloroacetonitrile in the  Periodate oxidation of 1 followed by borohydride reduction gave 2 (Scheme 1, 70.4%, 2 steps). The oxidation reaction was monitored by NMR spectroscopy (Supplementary Materials Figure S1), observing the disappearance of H-2 of GlcA at 3.44 ppm. We tested up to 20 equivalents of sodium periodate with respect to the diol. Finally, using four equivalents a 96% conversion from diol to dialdehyde was obtained within 4 hours and the formation of side products was avoided.

Heparanase Inhibition Assay In Vitro
Heparanase inhibitory activity of compounds 1 and 2 in comparison with roneparstat was determined in vitro according to Hammond et al. [23] using fondaparinux as a substrate. Upon heparanase cleavage fondaparinux yields a trisaccharide and a disaccharide that can be quantitatively assessed using the tetrazolium salt WST-1.
The results are shown in Figure 2. As expected, compound 2 (IC50 2 µg/mL) was found to be one order of magnitude more potent than its analog 1 (IC50 30 µg/mL), with both compounds being much less potent than roneparstat (IC50 6 ng/mL).

Conformation Characterization
To understand the role of glycol-split units in the interaction with heparanase we performed molecular modeling of 1 and 2 and compared the models to experimental NMR data. The models were obtained using a molecular dynamic approach including a variable temperature "enhanced sampling" procedure, no constraint was applied during the simulation.
Periodate oxidation of 1 followed by borohydride reduction gave 2 (Scheme 1, 70.4%, 2 steps). The oxidation reaction was monitored by NMR spectroscopy (Supplementary Materials Figure S1), observing the disappearance of H-2 of GlcA at 3.44 ppm. We tested up to 20 equivalents of sodium periodate with respect to the diol. Finally, using four equivalents a 96% conversion from diol to dialdehyde was obtained within 4 hours and the formation of side products was avoided.

Heparanase Inhibition Assay In Vitro
Heparanase inhibitory activity of compounds 1 and 2 in comparison with roneparstat was determined in vitro according to Hammond et al. [23] using fondaparinux as a substrate. Upon heparanase cleavage fondaparinux yields a trisaccharide and a disaccharide that can be quantitatively assessed using the tetrazolium salt WST-1.
The results are shown in Figure 2. As expected, compound 2 (IC 50 2 µg/mL) was found to be one order of magnitude more potent than its analog 1 (IC 50 30 µg/mL), with both compounds being much less potent than roneparstat (IC 50 6 ng/mL). Periodate oxidation of 1 followed by borohydride reduction gave 2 (Scheme 1, 70.4%, 2 steps). The oxidation reaction was monitored by NMR spectroscopy (Supplementary Materials Figure S1), observing the disappearance of H-2 of GlcA at 3.44 ppm. We tested up to 20 equivalents of sodium periodate with respect to the diol. Finally, using four equivalents a 96% conversion from diol to dialdehyde was obtained within 4 hours and the formation of side products was avoided.

Heparanase Inhibition Assay In Vitro
Heparanase inhibitory activity of compounds 1 and 2 in comparison with roneparstat was determined in vitro according to Hammond et al. [23] using fondaparinux as a substrate. Upon heparanase cleavage fondaparinux yields a trisaccharide and a disaccharide that can be quantitatively assessed using the tetrazolium salt WST-1.
The results are shown in Figure 2. As expected, compound 2 (IC50 2 µg/mL) was found to be one order of magnitude more potent than its analog 1 (IC50 30 µg/mL), with both compounds being much less potent than roneparstat (IC50 6 ng/mL).

Conformation Characterization
To understand the role of glycol-split units in the interaction with heparanase we performed molecular modeling of 1 and 2 and compared the models to experimental NMR data. The models were obtained using a molecular dynamic approach including a variable temperature "enhanced sampling" procedure, no constraint was applied during the simulation.
Considering the monosaccharide residues involved in 1 and 2, while the conformation of pyranose rings in glycosaminoglycans has been largely studied [24,25], the conformation of glycol-split units

Conformation Characterization
To understand the role of glycol-split units in the interaction with heparanase we performed molecular modeling of 1 and 2 and compared the models to experimental NMR data. The models were obtained using a molecular dynamic approach including a variable temperature "enhanced sampling" procedure, no constraint was applied during the simulation.
Considering the monosaccharide residues involved in 1 and 2, while the conformation of pyranose rings in glycosaminoglycans has been largely studied [24,25], the conformation of glycol-split units particularly glycol-split glucuronic acid has been until now poorly documented. In one report [26] the authors applied a semi-quantitative interpretation of NOESY data obtained on glycol-split partially O-desulfated pig mucosal heparin (the uronic acid content of which was: glucuronic 20%, unsulfated iduronic 30%, sulfated iduronic 50%). A second report dealt with the conformation glycol-split iduronic acid (gsIdoA) in a heparin-like hexasaccharide bound to Fibroblast Growth Factor 2 (FGF-2). Greater flexibility of the molecule was observed, in comparison to a hexasaccharide containing intact iduronic acid [27], suggesting that glycol splitting induces a drastic divergence of the GAG chain from the propagation required for setting in the basic canyon generated by FGF/FGF-receptor assemblies. However an in depth study of the conformation of glycol-split glucuronic acid (gsGlcA) has not been reported.
We thus paid special attention to the backbone conformation of 2 that can be described by eight consecutive dihedral angles, from the non-reducing to the reducing end: τ, α/β, γ, δ/ε, ω/w. For 1 only five torsional angles are involved: τ, α/β, ω/w (Figure 3). For both 1 and 2, α/β and ω/w characterize the classical glycosidic linkages, while τ describes the methoxy group conformation. The angles γ, and δ/ε, characterize additional degree of freedom that appear upon cleavage of the glucuronate ring. particularly glycol-split glucuronic acid has been until now poorly documented. In one report [26] the authors applied a semi-quantitative interpretation of NOESY data obtained on glycol-split partially O-desulfated pig mucosal heparin (the uronic acid content of which was: glucuronic 20%, unsulfated iduronic 30%, sulfated iduronic 50%). A second report dealt with the conformation glycol-split iduronic acid (gsIdoA) in a heparin-like hexasaccharide bound to Fibroblast Growth Factor 2 (FGF-2). Greater flexibility of the molecule was observed, in comparison to a hexasaccharide containing intact iduronic acid [27], suggesting that glycol splitting induces a drastic divergence of the GAG chain from the propagation required for setting in the basic canyon generated by FGF/FGF-receptor assemblies.
However an in depth study of the conformation of glycol-split glucuronic acid (gsGlcA) has not been reported. We thus paid special attention to the backbone conformation of 2 that can be described by eight consecutive dihedral angles, from the non-reducing to the reducing end: τ, α/β, γ, δ/ε, ω/w. For 1 only five torsional angles are involved: τ, α/β, ω/w (Figure 3). For both 1 and 2, α/β and ω/w characterize the classical glycosidic linkages, while τ describes the methoxy group conformation. The angles γ, and δ/ε, characterize additional degree of freedom that appear upon cleavage of the glucuronate ring. Figure 3. Trisaccharide predicted conformation for compound 2 and compound 1 adjusting the backbone dihedral angles: τ, α/β, γ, δ/ε, ω/w for the former and τ, α/β, ω/w for the latter, in accord to the average values estimated using the MD simulation sampling procedure (material and methods). Possible extra-residue (GlcA) hydrogen bonds, contributing to 2 and 1 conformation stability, are underlined by dashed lines joining donor and acceptors atoms.
Conformational sampling was done with temperature progressively increasing from 300 to 400 K, and decreasing from 400 to 300 K (Supplementary Materials Table S1); the last MD simulation step at 300 K allowed average dihedral angle calculation. Two initial conformations for each glycan were used which lead to the same final conformation showing the reliability of the method (Supplementary  Materials Table S2). The dihedrals angle pairs: α/β, δ/ε, ω/w are reported in Ramachandran plots (Supplementary Materials Figure S2), while γ is reported as a function of time as it resembles an aliphatic chain torsional degree of freedom (H-C(R,R′)-C(R′′,R′′′)-H). The yellow stars in Ramachandran plots (Supplementary Materials Figure S2) indicate the estimated most probable states.
As shown in Figure 3, at the end of the modeling process the pyranose rings in 1 and 2 maintain the expected classical conformations, namely 4 C1 for GlcNS6S and GlcA and 1 C4 for 1,6anGlcNS, which is also in agreement with the 1 H-NMR 3 J inter-proton coupling constants observed (see the Materials and Methods section).
Regarding the interglycosidic angles, similar α/β distributions were found for 1 and 2 (Supplementary Materials Figure S2) but they have different ω/w distribution, 2 showing two accessible states (A and B) as opposed to a single state in 1. As discussed below, only conformation A fits the experimental NOE data. The estimated backbone torsional angles of 1 and 2 were used to refine the models initially built. The refined conformations are displayed (Figure 3) where only the A conformer . Trisaccharide predicted conformation for compound 2 and compound 1 adjusting the backbone dihedral angles: τ, α/β, γ, δ/ε, ω/w for the former and τ, α/β, ω/w for the latter, in accord to the average values estimated using the MD simulation sampling procedure (material and methods). Possible extra-residue (GlcA) hydrogen bonds, contributing to 2 and 1 conformation stability, are underlined by dashed lines joining donor and acceptors atoms.
Conformational sampling was done with temperature progressively increasing from 300 to 400 K, and decreasing from 400 to 300 K (Supplementary Materials Table S1); the last MD simulation step at 300 K allowed average dihedral angle calculation. Two initial conformations for each glycan were used which lead to the same final conformation showing the reliability of the method (Supplementary Materials Table S2). The dihedrals angle pairs: α/β, δ/ε, ω/w are reported in Ramachandran plots (Supplementary Materials Figure S2), while γ is reported as a function of time as it resembles an aliphatic chain torsional degree of freedom (H-C(R,R )-C(R ,R )-H). The yellow stars in Ramachandran plots (Supplementary Materials Figure S2) indicate the estimated most probable states.
As shown in Figure 3, at the end of the modeling process the pyranose rings in 1 and 2 maintain the expected classical conformations, namely 4 C 1 for GlcNS6S and GlcA and 1 C 4 for 1,6anGlcNS, which is also in agreement with the 1 H-NMR 3 J inter-proton coupling constants observed (see the Materials and Methods section).
Regarding the interglycosidic angles, similar α/β distributions were found for 1 and 2 (Supplementary Materials Figure S2) but they have different ω/w distribution, 2 showing two accessible states (A and B) as opposed to a single state in 1. As discussed below, only conformation A fits the experimental NOE data. The estimated backbone torsional angles of 1 and 2 were used to refine the models initially built. The refined conformations are displayed (Figure 3) where only the A conformer of 2 is shown. In contrast to GlcA in 1, where OH-2 and OH-3 are in trans diequatorial orientation, in compound 2 cleavage of the C-2/C-3 bond allows OH-2 and OH-3 to be oriented at the opposite side of the molecule backbone ( Figure 3). Interestingly, in 2 the most populated γ angle (approx. −60 • ) allows the highest distances between the carboxylate of gsGlcA and the two sulfate groups of the next GlcNS6S residue, while the greater flexibility of gsGlcA, allows a smaller trisaccharide "end-to-end" distance (between C-4 of GlcNS6S and C-1 of 1,6anGlcNS) 10.4 Å in 2, vs. 12.4 Å in 1 (Figure 3).
The model (Maestro graphical interface, see material and methods) allows prediction of intra-molecular hydrogen bonds (Figure 3). In 1, two bonds are predicted, the first one between an oxygen atom of the N-sulfate of GlcNS6S (acceptor) and OH-3 of GlcA (donor), and the second one between the GlcA-1,6anGlcNS interglycosidic oxygen (acceptor) and the NH of 1,6anGlcNS (donor). In 2, three bonds are predicted: first between NH of GlcNS6S (donor) and OH-3 of gsGlcA (acceptor), second between OH-3 of gsGlcA (donor) and O-5 of gsGlcA (acceptor) and third between OH-2 of gsGlcA (donor) and the interglycosidic GlcNS6S-gsGlcA oxygen (acceptor).
The models obtained were then confronted to experimental NMR data. Regarding chemical shifts, signals of the HSQC spectra of 1 and 2 (Supplementary Materials Figure S3) support their modelled conformation. Thus, the anomeric proton of GlcNS6S in 1 is more de-shielded than in 2 (5.61 vs. 5.34 ppm) which can be explained by the shorter distance between this proton and the carboxyl group of the adjacent uronic acid, smaller in 1 (4.6 Å) than in 2 (5.1 Å). NMR HSQC experiment data also support the above H-bond network in 2, showing H-2a and H-2b resonances as superimposed signals centered at 3.66 ppm, in accord with a partially impaired rotation of the CH 2 OH group around the C-1/C-2 bond. In contrast, the H-3a and H-3b (3.92 and 3.80 ppm) show diastereotopic effect, being separated by more than 0.1 ppm, suggesting a higher impairment in the rotation of CH 2 OH around C-3/C-4 in gsGlcA.
The strength of various hydrogen bonds can be compared using 1 H-NMR at variable temperature. The chemical shift of the exchangeable protons in 2: NH of the two glucosamines, OH-2 and OH-3 of gsGlcA, were measured at temperature from 5 to 35 • C. The chemical shifts (ppm) were fitted to the temperature (K) using a linear regression model and the temperature coefficient was calculated (see experimental part). The temperature coefficient reflects the extent to which NH and OH protons are protected from exchange through hydrogen bonding. Temperature coefficients showed different behavior between NHSO 3 and OH protons, the former had a value of −5.7/−8.1 ppb/K while the second had a value of −9.8/13.0 ppb/K (Supplementary Materials Figure S4 and Table S3). We can conclude from the present experiment that the hydrogen bonds involving NH of the glucosamines are stronger than those involving the OH groups. Additionally, the bond involving OH-3 appears slightly stronger than that involving OH-2 , in agreement with the intra-molecular hydrogen bond network predicted, and the interpretation of the HSQC spectra.
The models were also used to simulate selected intra and inter-residue 2D NOE signals to be compared with the corresponding experimental values. These latter were determined as build-up functions of the mixing time between 0.2 s to 1.0 s for the trisaccharide 1 or 0.2 s to 1.5 s, for the glycol-split 2. Comparison of experimental and calculated NOEs for the two possible models obtained for 2 (Supplementary Materials Figure S2A Finally, the model predicts that 1 and 2 having the same total charge, differ in charge distribution, highlighted by dissimilar electric dipole moment (Supplementary Materials Figure S7). A possible explanation could be the higher flexibility of the gsGlcA residue allowing a better neutralization of the atomically distributed charges, resulting in a smaller dipole moment vector length in 2 which could be the basis for a significantly different intermolecular force behaviour.

General Information
All reagents were purchased from Sigma-Aldrich (Milan, Italy) and Iris Biotech GmbH (Marktredwitz, Germany). Solvents were freshly distilled before to use. All reactions were carried out under a nitrogen atmosphere if necessary. Flash chromatography was performed on silica gel (230-400 mesh, Merck, Darmstadt, Germany). TLC was carried out on silica gel plate (Merck 60, F254) and detected visually by ultraviolet irradiation (254 nm) or by detected with spray (10% conc. sulphuric acid, heating at 130 °C. NMR spectra were recorded on a Bruker Avance 500 MHz or Avance 600 MHz spectrometer (Bruker, Karlsruhe, Germany). All values were reported in ppm (δ) downfield from solvent. HRMS was obtained with MicrOTOF-Q (Bruker Daltonics, Brema, Germany). LC-MS was performed on an Ultimate 3000 HPLC-UV system (Dionex, Rodano, Italy) coupled to an Esi-Q-TOF-MS MicrOTOF-Q (Bruker Daltonics). Ion-pair reversed-phase separation was carried out on a Kinetex C18 column (2.1 mm × 100 mm, 2.6 µm, 100 Å, Phenomenex, Bologna, Italy). A binary solvent system plus 10 mM DBU and AcOH was used for gradient elution. (3). Compound 3 was synthesized in 7 steps (21% overall yield) from commercially available 1,6-anhydro-β-D-glucopyranose following the method of Oikawa et al. [19]. (4). Methyl iodide (0.5 mL, 8.6 mmol) was added to a cooled (0 °C) solution of 3 (2 g, 7.2 mmol) in dry DMF (32 mL). Sodium hydride 50% in oil (0.5 g, 10.8 mmol) was then introduced in portions and the reaction mixture was stirred at RT for 2 h. After cooling to 0 °C NaH in excess was destroyed by methanol (20 mL). After  Figure S6). Due to several signal overlap and strong couplings, no other NOE build-up curves could be undoubtedly measured for 1.

1,6-Anhydro-2-azido-3-O-benzyl-2-deoxy-β-D-glucopyranose
Finally, the model predicts that 1 and 2 having the same total charge, differ in charge distribution, highlighted by dissimilar electric dipole moment (Supplementary Materials Figure S7). A possible explanation could be the higher flexibility of the gsGlcA residue allowing a better neutralization of the atomically distributed charges, resulting in a smaller dipole moment vector length in 2 which could be the basis for a significantly different intermolecular force behaviour.

General Information
All reagents were purchased from Sigma-Aldrich (Milan, Italy) and Iris Biotech GmbH (Marktredwitz, Germany). Solvents were freshly distilled before to use. All reactions were carried out under a nitrogen atmosphere if necessary. Flash chromatography was performed on silica gel (230-400 mesh, Merck, Darmstadt, Germany). TLC was carried out on silica gel plate (Merck 60, F 254 ) and detected visually by ultraviolet irradiation (254 nm) or by detected with spray (10% conc. sulphuric acid, heating at 130 • C. NMR spectra were recorded on a Bruker Avance 500 MHz or Avance 600 MHz spectrometer (Bruker, Karlsruhe, Germany). All values were reported in ppm (δ) downfield from solvent. HRMS was obtained with MicrOTOF-Q (Bruker Daltonics, Brema, Germany). LC-MS was performed on an Ultimate 3000 HPLC-UV system (Dionex, Rodano, Italy) coupled to an Esi-Q-TOF-MS MicrOTOF-Q (Bruker Daltonics). Ion-pair reversed-phase separation was carried out on a Kinetex C18
Models for 1 and 2 are built by Maestro/Macromodel 9.8 software where GlcNS6S and GlcA are drawn in 4 C 1 conformation while the 1,6anGlcNS is built in 1 C 4 conformation, as previously determined [24,25]. The conformation of the glycol-split GlcA residue is initially guessed by visual inspection. Two different initial conformations for both 1 and 2 are built, characterized by two sets of dihedral angles values. In 2 in going from non reducing to the reducing end, the dihedrals are defined as follows: τ (C-O4-C4-H4 between MeO and GlcNS6S), α/β (H1-C1-O4-C4/C1-O4-C4-H4 between GlcNS6S and gsGlcA), γ (H4-C4-C5-H5 in gsGlcA), δ/ε (H5-C5-O5-C1/C5-O5-C1-H1 in gsGlcA), ω/w (H1-C1-O4-C4/C1-O4-C4-H4 between gsGlcA and 1,6anGlcNS). In 1 only five dihedral angles are defined: τ, α/β, and ω/w. The Amber* force field as implemented in Maestro/Macromodel 9.8 software is used, the non-bonded cut-off are set to 20.0, 8.0 and 4.0 Å for electrostatic, Van der Waals and hydrogen bond interactions respectively. The solvent description involve the Generalized Born Implicit solvent method. The glycan models after building are energy minimized (bmin procedure) setting: Max number of steps = 10 K, and Gradient Threshold = 10 −3 KJ·mol −1 ·Å −1 . Each of the four glycan models is then submitted to a sequence of eleven MD simulation run with temperature progressively increasing from 300 to the highest temperature value of 400 K, using steps of 20 K, to decrease again to the final value of 300 K. The time length of each fixed temperature MD run is 5 ns for a whole duration of 55 ns. The MD simulation thermal history is summarized in Table S1. The glycan backbone torsional angles at the beginning and at the end of the MD sampling method, averaging for a suitable amount of time in the final MD run, are reported in Table S2

Conclusions
To investigate the role of glycol-split uronic acid units in glycol-split-heparin-derived inhibitors of heparanase we have synthesized trisaccharide 1 and its glycol-split counterpart 2. In an heparanase inhibition assay 2 was found one order magnitude more active than the ring closed 1. The activity of 1 and 2 is however low (IC 50 in the µM range) in comparison to roneparstat. In parallel a conformational characterization of compound 1 and 2 is reported using NOESY NMR and MD simulations. The GlcA ring opening introduces three additional dihedral angles: γ, δ, ε, improving the backbone degree of freedom. The most populated backbone dihedral angle states for both compound 1 and 2 are estimated, using gradient temperature MD simulation, and selected inter-glycosidic NOEs enhancements as constraint. The glycol-split trisaccharide is characterised by a greater flexibility, a smaller end-to-end distance, and a different electric charge distribution.