Measuring hydroxyl exchange rates in glycans using a synergistic combination of saturation transfer and relaxation dispersion NMR

Abstract

Molecular recognition events mediated by glycans play pivotal roles in controlling the fate of diverse biological processes such as cellular communication and the immune response. The affinity of glycans for their target receptors is governed primarily by the hydrogen bonds formed by hydroxyl groups decorating the glycan surface. Hydroxyl exchange rate constants are therefore vital parameters that report on glycan structure and dynamics. Here we present a strategy for characterizing hydroxyl hydrogen/deuterium (H/D) exchange in glycans that employs a synergistic combination of ¹³C chemical exchange saturation transfer (CEST) and Carr-Purcell-Meiboom-Gill relaxation dispersion (CPMG) NMR methods. We show that the combination of CEST and CPMG experiments facilitates the sensitive detection of the small (~ 0.1 ppm) two-bond deuterium isotope shift on a ¹³C nucleus when the attached hydroxyl group fluctuates between protonated and deuterated states. This shift is leveraged for measuring site-specific kinetic H/D exchange rate constants as well as thermodynamic free energies of isotope fractionation. The CEST and CPMG modules are integrated with a selective J-cross-polarization scheme that provides the flexibility for rapid characterization of H/D exchange at a specific hydroxyl site. Moreover, our approach enables the precise isothermal measurement of hydroxyl exchange rate constants without the need for cumbersome isotope labeling. The H/D exchange rate constants of three different glycans assessed using this method highlight its potential for detecting transient intra and intermolecular hydrogen bonds. In addition, the trends in H/D exchange rate constants establish site-specific steric accessibility as a key determinant of solvent exchange dynamics in glycans.

Introduction

Glycans are the most abundant biomolecules found in living organisms.¹ They are present either in isolation or conjugated to proteins and lipids, and perform a variety of structural and modulatory functions.^2,3 For example, glucose is the primary source of energy in all life forms, while glycogen and starch serve as storage reserves⁴ and glycosaminoglycans such as heparin modulate the outcome of diverse pathways such as cell proliferation and angiogenesis⁵. The information content in glycans is encrypted in both their sequence and structure, and is decoded by carbohydrate-binding proteins known as lectins. The recognition of a glycan by a lectin involves the formation of specific hydrogen bonds between the hydroxyl groups in the glycan with solvent exposed polar and charged amino acids in the lectin.^6–8 In addition, hydroxyl groups also contribute substantially to the tensile strength of structural carbohydrates like cellulose.¹ The rates of hydroxyl group solvent exchange are, therefore, key descriptors of glycan structure and dynamics.

Two broad classes of NMR methods have been developed to measure hydroxyl exchange rates in glycans. In the first class of methods, exchange of protons between the glycan and water is directly observed, either by perturbing the water through a radiofrequency field and observing its effect on hydroxyl resonances^9,10, or via a 2D ¹H-¹H magnetization exchange (EXSY) experiment in which cross-peaks between hydroxyl and water protons are quantified^11,12. The second class of methods avoids perturbation or detection of the large water magnetization, which is complicated by experimental factors like radiation damping and receiver overflow. These methods exploit the two-bond deuterium isotope shift of the attached ¹³C nucleus (²Δ¹³C(D)) when the hydroxyl group exchanges from a protonated to a deuterated state in a mixed H₂O/D₂O solvent system. Hydroxyl hydrogen/deuterium (H/D) exchange rates are then measured either via ¹³C-detection and lineshape analysis (LSA)¹³ or through the 2D INTOXSY approach, where the diagonal and cross-peaks are separated by ²Δ¹³C(D), and cross-peak build-up is dictated by the H/D exchange rate constant¹⁴. Except for LSA, all these methods involve ¹H evolution during the exchange period, in which magnetization transfer occurs not only because of chemical exchange, but also because of dipolar cross-relaxation between the network of coupled protons in the glycan. Careful separation of the effects of exchange and cross-relaxation then becomes necessary to reliably quantify exchange rate constants. On the other hand, LSA does not give an isothermal estimate of the H/D exchange rate constant and relies on ¹³C detection which is insensitive. Moreover, the range of rate constants measurable by all these methods has been limited to < 100 s^-1, necessitating the use of mixed solvent systems or supercooled aqueous solutions to slow down H/D exchange.

The recently developed chemical exchange saturation transfer (CEST), and Carr-Purcell-Meiboom-Gill (CPMG) and rotating frame (R_1ρ) relaxation dispersion NMR methods have been very successful in thermodynamic, kinetic and structural characterization of chemical and conformational exchange in proteins and nucleic acids.^15–17 However, their potential in probing μs-ms timescale dynamics in glycans remains poorly explored.¹⁸ Here, we show that a synergistic combination of ¹³C-CEST and CPMG methods, in conjunction with a selective 1D Hartmann-Hahn excitation scheme, provides reliable estimates of glycan H/D exchange rate constants. The trends in H/D exchange rates in sucrose, maltose and cellobiose reflect the steric accessibility of the hydroxyl group and also serve as markers for the presence of transient intra- and inter-glycan hydrogen bonds.

Results and Discussion

Evaluating the applicability of CEST and CPMG methodology to measure H/D exchange rate constants

H/D exchange (Eq. 1) at hydroxyl groups in glycans results in a two-bond deuterium isotope shift of the attached carbon (²Δ¹³C(D)).

$$\begin{array}{lll}\text{C}-\text{OH}+\frac{1}{2}{\text{D}}_2\text{O}\hfill & \underset{{\text{k}}_{\text{HD}}}{\overset{{\text{k}}_{\text{DH}}}{\leftrightharpoons }}\hfill & \text{C}-\text{OD}+\frac{1}{2}{\text{H}}_2\text{O}\hfill \end{array}$$ (1)

Therefore, in samples containing mixtures of H₂O and D₂O, glycan carbons attached to hydroxyl groups fluctuate between two sites separated in chemical shift by ²Δ¹³C(D). The relative populations of the two states (p_C-OH and p_C-OD) are determined predominantly by the fractional amounts of H₂O and D₂O, while the exchange rate constant k_ex is given by k_ex = k_HD + k_DH, where k_HD and k_DH are the forward and backward rate constants. The solvent exchange rate of hydroxyl protons in 100 % H₂O, k_HH, is related to k_HD via a kinetic isotope effect.

The magnitude of ²Δ¹³C(D) in glycans is small (²Δ¹³C(D) = ϖ_C-OD - ϖ_C-OH = 0.11 ppm)^14,19, corresponding to ~ 20 Hz on a 16.45 T (700 MHz) spectrometer. Since both CEST and CPMG methods require sizeable chemical shift differences (Δϖ) to characterize μs-ms timescale dynamics, we first used simulations to assess the range of k_ex values that can be robustly measured for Δϖ = 20 Hz. Initially, CEST profiles simulated at B₁ fields of 1, 2, 3 and 5 Hz for k_ex varying from 50 – 500 s^-1 were fit to the Bloch-McConnell equations. The correlation plot in Figure S1A shows that the fit k_ex values match the inputs only for k_ex < 300 s^-1, while systematic deviations between the two are observed at larger k_ex. However, the addition of a CPMG dataset at a single static magnetic field strength substantially improves the agreement between input and fit values for k_ex up to 1000 s^-1, extending the applicability of this methodology to H/D exchange rate constants of the order of 50-1000 s^-1 expected for glycans in aqueous solution (Fig. S1B).

Exploiting a selective 1D-based strategy for glycan ¹³C-CEST and CPMG data acquisition

The principal factor modulating site-specific hydroxyl exchange rates in glycans is the presence of transient intra and intermolecular hydrogen bonds that have been detected in crystal structures as well as neutron diffraction and NMR data.^19–21 An intramolecular hydrogen bond has been observed in the crystal structure of sucrose between the hydroxyl at position 2 of the glucose ring (GOH2) and the hydroxyl at position 1 of the fructose ring (FOH1)²² (Fig. 1A). In order to gauge whether H/D exchange rate constants measured through ¹³C-CEST and CPMG methods are sensitive to the presence of transient hydrogen bonds, we chose sucrose as the initial model system. Sucrose affords the additional benefit of being a non-reducing sugar that exists as a single configuration in solution, unlike reducing sugars whose NMR spectra contain two sets of peaks because of anomerization.

Figure 1. Methodology to study hydroxyl exchange rate in glycans.

(A) The chemical structure of sucrose. Glucose carbons are numbered in green while fructose carbons are labelled in blue. The hydrogen bond between FOH1 and GOH2 observed in crystal structures of sucrose is shown as a green dashed line. (B) The ¹H-¹³C HSQC spectrum of unlabeled sucrose at pH 5.8 and 278 K. Fructose and glucose resonances are labelled with F and G prefixes, respectively. The anomeric peak of the glucose ring is shown as an inset. (C) The selective 1D pulse sequence for recording ¹³C-CEST and CPMG of methylene and methine spin systems in glycans. In this pulse sequence, ¹H magnetization is selectively transferred from the ¹H of interest to the directly bonded ¹³C by a selective J cross polarization (SCP) module. CEST or CPMG data are acquired on this ¹³C magnetization. The ¹³C magnetization at the end of the CEST or CPMG period is then transferred back to ¹H for detection by another SCP module. (Bottom left) The pulse sequence for the ¹³C-CEST module, where ¹³C z-magnetization is subjected to a weak B₁ field during a period T_ex. In every slice of the experiment, the ¹³C offset at which the weak B₁ field is applied is varied. (Bottom right) The pulse scheme for ¹³C-CPMG, where the number of ¹³C 180° pulses (represented as wide open rectangles) applied for a constant time T_ex is varied in every slice of the experiment. Details of the pulse sequence are provided in Supporting Information. (D) Overlay of the aliphatic region of a ¹H 1D spectrum of sucrose (red) with a selective 1D spectrum of a methylene system (GC6-H6, green) and a methine system (FC3-H3, blue) acquired using the pulse sequence shown in panel (C).

Figure 1B shows the ¹H-¹³C HSQC spectrum of sucrose acquired on a 100 mM sample at natural ¹³C abundance at pH 5.8 and 278 K. The peaks in the aliphatic region are sparsely dispersed over > 20 ppm in the ¹³C dimension, which would necessitate a CEST sweep of ~ 3.5 kHz on a 16.45 T spectrometer in the conventional 2D-based CEST experiment. On the other hand, the chemical shifts of the two states in exchange differ by a mere 20 Hz for all resonances of interest. In addition, since the ¹³C chemical shift difference between the protonated and deuterated states is small, weak B₁ fields between 1-5 Hz are necessary for visualizing the two dips in intensity. Since the offset spacing in CEST experiments is typically of the order of the B₁ field, 300-3000 planes are required to cover the entire ¹³C sweep width in a conventional 2D CEST experiment, leading to unreasonably long measurement times.

We therefore adopted a selective 1D excitation scheme to interrogate peaks one at a time in the context of either a CEST or a CPMG experiment (Fig. 1C, SI Text). In this scheme, selective excitation is achieved through a Hartmann-Hahn J-cross-polarization module,²³ which employs weak radiofrequency fields with matching amplitudes on both the ¹H and ¹³C channels applied on-resonance to the respective ¹H and ¹³C chemical shifts of a ¹H-¹³C covalently bonded spin pair.²⁴ When RF field strengths of the order of ¹J_HC are used (¹J_HC ~ 145 Hz, B₁ ~ 130 Hz for methine carbons and ¹J_HC ~ 125 Hz, B₁ ~ 115 Hz for methylene carbons), magnetization is transferred from the hydrogen of interest to its covalently bonded ¹³C nucleus, resulting in a single peak in the ¹³C-edited selective 1D ¹H spectrum.^24–26 Figure 1D shows the 1D ¹H spectrum of sucrose in red, overlaid with a selective 1D spectrum exciting either the FC3-H3 methine (Fig. 1D, blue) or the GC6-H6 methylene (Fig. 1D, green) spin systems. A conventional 2D-based CEST experiment on sucrose would take 184 h for acquiring 736 planes (92 planes centred around each of the eight non-overlapping carbons attached to hydroxyls) spaced 2 Hz apart (4 transients per increment) covering the entire ¹³C sweep width. On the other hand, the selective 1D CEST will only need 45 min per peak (16 transients per increment) for the same chemical shift offset spacing, since only a sweep of 180 Hz is required around the resonance of interest.

The selective 1D scheme is then interfaced with the CEST (Fig. 1C, left) or CPMG (Fig. 1C, right) modules. In the CEST experiment, a weak RF field is applied at offsets sweeping through the chemical shifts of the states in conformational exchange for a duration T_ex. ¹³C magnetization is placed along the z-axis at the start of T_ex and the amount left over at the end of T_ex is transferred back to ¹H for detection.²⁷ In the CPMG experiment, increasing repeats (N) of the τ_CP-180°-τ_CP element are applied on in-phase transverse ¹³C magnetization for the constant exchange period T_ex. The residual transverse magnetization at the end of T_ex is estimated by transferring back to ¹H for detection.^{28,29 13}C-CEST and CPMG profiles of the selectively excited carbons are then obtained by quantifying the peak intensity in the selective 1D spectrum as function of chemical shift offset (CEST) or CPMG pulsing frequency (ν_CPMG = 1/(4τ_CP + 2pw₁₈₀), pw₁₈₀ being the duration of the 180° CPMG pulse). The covalently attached proton is decoupled during T_ex using a DIPSI-2³⁰ composite decoupling sequence in the CEST experiment³¹ and a continuous wave RF field³² in the CPMG experiment. As decoupling is uninterrupted during the exchange period, there is no interconversion between in-phase and anti-phase magnetization due to scalar coupling evolution, eliminating the need for a P-element in the sequence that would otherwise compensate for differential relaxation of the in- and anti-phase components. In addition, the CPMG module used here has been shown to self-compensate for imperfections in the CPMG 180° pulses.³² This extends the range of the CPMG pulse sequence for probing small k_ex, as both even and odd numbers of echoes (N) can be employed on either side of the 180°_ϕ4 ¹³C pulse without introducing artifacts in the CPMG dispersions.³² Since only one ¹H-¹³C spin pair is examined in each selective 1D dataset, the transmitter offsets can be individually tailored to the values for each nucleus. This is an additional advantage in selective 1D-based experiments as it eliminates off-resonance effects during CPMG pulsing and significantly improves the ¹H decoupling efficiency in both sequences.

Measuring H/D exchange rate constants in sucrose: the methodology

Figures 2 and S2 shows the CEST (Fig. 2A, Fig. S2A) and CPMG (Fig. 2B, Fig. S2B) profiles of sucrose acquired on a 100 mM unlabeled sample at pH 5.8 and 278 K. The sample was prepared in 50 % (v/v) H₂O/D₂O, which was identified through simulations to be the optimal p_C-OH/p_C-OD ratio for clearly distinguishing exchange fingerprints from CEST and CPMG profiles (Fig. S3). Since the ¹³C-OH and ¹³C-OD states are separated only by ²Δ¹³C(D) ~ 20 Hz), both these states are excited by the J-cross-polarization module. Consequently, CEST and CPMG profiles derive from the evolution of both ${\overrightarrow{M}}_{C-OH}$ and ${\overrightarrow{M}}_{C-OD}$ (the magnetization vectors of ¹³C nuclei bonded to protonated and deuterated hydroxyls respectively) during the exchange period.

Figure 2. Measuring sucrose H/D exchange rate constants in 50 % (v/v) D₂O/H₂O using selective 1D ¹³C-CEST and CPMG data.

(A) Selective ¹³C-CEST and (B) ¹³C-CPMG profiles of different carbon nuclei in sucrose. CEST data were acquired at 4 B₁ field strengths between 1 and 5 Hz. Solid lines are fits of the data (filled circles) to the Bloch-McConnell equations as described in the Main Text. The H/D exchange rate estimated at each site by fitting ¹³C-CEST and ¹³C-CPMG profiles together is indicated at the top along with the name of the site.

The fluctuation of each ¹³C nucleus between -OH and -OD bonded states is apparent in the CEST profiles, either as two distinct dips in intensity (Fig. 2A, FC3 and FC4) or as a broadening of the CEST profile (Fig. 2A, GC3 and GC6). For each ¹³C site, the width of the intensity dips increases at higher B₁ field strengths, and two distinct dips arising from protonated and deuterated hydroxyl groups are visible only at small RF amplitudes given the small chemical shift difference between the exchanging states (²Δ¹³C(D) ~ 20 Hz). It is therefore important to acquire data at B₁ fields ≤ 5 Hz in order to reliably determine the parameters quantifying the exchange event. Figure 2B shows ¹³C-CPMG profiles acquired on the same sample at a static field strength of 16.45 T. Chemical exchange between the C-OH and C-OD states separated in chemical shift by ²Δ¹³C(D) results in CPMG dispersions ranging in magnitude $\begin{array}{ll}({\text{ΔR}}_2^{eff}=\hfill & {\text{R}}_2^{eff}\left(v_{\text{CPMG}}=25\text{Hz}\right)-{\text{R}}_2^{eff}\left(v_{\text{CPMG},\max }\right))\hfill \end{array}$ from 3.5 – 5 s^-1. In contrast, the dispersions for the GC5 of glucose are flat (Fig. S4), confirming that three-bond deuterium isotope shifts are below the detection limit of our method and do not interfere with H/D exchange rate constant measurements.

The ¹³C nucleus during T_ex is an excellent approximation to an isolated spin-1/2 nucleus since covalently bonded ¹H are decoupled during T_ex. ¹³C-¹³C homonuclear scalar and dipolar couplings do not complicate data analysis as natural abundance ¹³C magnetization is used in these experiments. Moreover, dipolar cross-relaxation rate constants with remote protons are scaled by a factor of ${\left(\frac{{\gamma }_C}{{\gamma }_H}\right)}^2$ compared to rate constants at similar internuclear distances in proton-based exchange experiments.³³ Therefore, artefactual contributions of cross-relaxation to the exchange rate constant measurements are negligible. Finally, owing to the fact that small B₁ fields are employed in the ¹³C-CEST experiment, B₁ fields were calibrated using the CONDENZ method³⁴, as this allows for precise and accurate measurements of small B₁ field strengths.

In order to extract sucrose H/D exchange rate constants from selective 1D data, we first fit the site-specific 4 B₁ field CEST data to a two-state model of the Bloch-McConnell equations³⁵. Since the thermodynamics and kinetics of H/D exchange is expected to be different at every hydroxyl site, no parameters were shared between sites. Intrinsic longitudinal (R₁) and transverse relaxation rate constants (R₂) are not expected to differ significantly for the C-OH and C-OD states (SI Text, Fig. S5) (ΔR₁, ΔR₂~ 0) and were assumed to be the same in the fitting routine. The ${\chi }_{red}^2$ surface for k_ex (Fig. 3A, green), which shows the deterioration in fit quality (estimated by ${\chi }_{red}^2$) when k_ex is fixed to the ‘incorrect value’, has a pronounced minimum for small k_ex (50-70 s^-1, FC3 and FC4), but becomes flat at higher k_ex values (> 90 s^-1). This indicates that large k_ex values cannot be reliably extracted from CEST data alone, in agreement with simulations (Fig. S1, see above). To determine why ${\chi }_{red}^2$ surfaces become flat at high k_ex, we plotted p_C-OD, ²Δ¹³C(D), R_{1, CEST} and R_{2, CEST} at every k_ex value for GC6 (Fig. S6A-D, yellow filled circles). While p_C-OD and R_{1, CEST} do not vary significantly when k_ex is altered (Fig. S6A, S6C), we see systematic increases in ²Δ¹³C(D) (Fig. S6B) and decreases in R_{2, CEST} (Fig. S6D) with increasing k_ex, which compensate for changes occurring in k_ex to maintain the same quality of fit.

Figure 3. H/D exchange rate constants can be reliably estimated from a global analysis of ¹³C-CEST and ¹³C-CPMG data.

(A) ${\chi }_{red}^2$ plots of k_ex for different carbon nuclei in sucrose showing the increase in ${\chi }_{red}^2$ from the best-fit value caused by fixing k_ex to a particular value during the fit. ${\chi }_{red}^2$ curves were generated by fitting ¹³C-CEST data alone (filled green circles), ¹³C-CEST data bounding the R₂ value (open red circles), or ¹³C-CEST and ¹³C-CPMG data together with bounds on R₂ (filled blue triangles). The H/D exchange rates estimated by fitting ¹³C-CEST and ¹³C-CPMG data together with bounded R₂ are indicated at the top of every ${\chi }_{red}^2$ surface along with the site name. (B, C) k_ex distributions obtained by Monte Carlo analysis of ¹³C-CEST and ¹³C-CPMG datasets for methine (B) and methylene (C) systems in sucrose.

The correlation between R₂ and k_ex has previously been observed for protein CEST profiles in fast exchange³⁶ and can be rationalized based on the equation for the CEST profile that is approximately given by:³⁷

$$\text{I}/{\text{I}}_0={\cos }^2\theta {\text{e}}^{-{\text{R}}_{1\rho }{\text{T}}_{\text{ex}}}$$ (2)

where

$$R_{1\rho }=R_1{\cos }^2\theta +\left(R_2+R_{ex}\right){\sin }^2\theta$$ (3)

Here, R_1ρ is the rotating frame relaxation rate constant and tan θ = ω₁/Ω, where θ is the angle made by the effective field with the z-axis, ω₁ is the RF field in rad/s and Ω is the chemical shift of each state in slow exchange (or the population weighted average chemical shift in fast exchange). When the dips from the two states overlap significantly in the CEST profiles, as for GC3 and GC6, the contributions to transverse relaxation from intrinsic R₂, which is encoded in the width of the dip, cannot be clearly distinguished from exchange broadening of each state (R_ex). Since R_ex depends on the exact value of k_ex, fixing k_ex to different values during the fitting procedure results in changes to R_ex. These changes are compensated by the intrinsic R₂ in order to keep R₂+R_ex constant and maintain the same fit quality at different k_ex values. In order to break the correlation between R₂ and k_ex, we measured exchange-free R₂s of ¹³C nuclei (Fig. S7) in sucrose from the time-dependent decay of transverse magnetization in a variable time CPMG experiment at constant ν_CPMG (1 kHz), where exchange broadening is completely quenched by CPMG pulsing (Fig. 2B). The R₂ values deviate from the asymptotes (R_2,∞ = R_2,eff(ν_CPMG = 1000Hz)) of the respective CPMG profiles by up to 10 %. Therefore, we used a conservative bound of R_2,∞ ± 25% on the CEST-derived R₂s and fit the 4 B₁ field CEST profiles together. Now, the ${\chi }_{red}^2$ surfaces of FC3, FC4 and GC3 (k_ex ≤ 150 s^-1) show steeper minima compared to fitting CEST data without bounded R₂s (Fig. 3A, green vs red).

However, the ${\chi }_{red}^2$ surface for GC6 remains flat even with bounded R₂ (Fig. 3A, red), because of the tight correlation between ²Δ¹³C(D) and k_ex (Fig. S6B, cyan). We next tested whether ¹³C-CPMG profiles could help in breaking the correlation between ²Δ¹³C(D) and k_ex, since CPMG data are sensitive to both k_ex and |²Δ¹³C(D)|. CPMG profiles simulated (Fig. S8) using three correlated (k_ex, ²Δ¹³C(D)) pairs (Fig. S6B, yellow) are significantly different, suggesting that CPMG data may help in resolving the k_ex-²Δ¹³C(D) correlation. Indeed, when 4 B₁ field CEST profiles are fit globally with a single CPMG dataset at the same B₀ field strength with bounded R₂, ${\chi }_{red}^2$ surfaces for all sites including GC6 show pronounced minima (Fig. 3A, blue and Fig. S9). The variations in ²Δ¹³C(D) and R₂ from CEST profiles are marginal over the tested range, indicating that different exchange parameters do not adjust in concert with k_ex (Fig. S6B, S6D, pink). Distributions of k_ex generated from a Monte Carlo analysis are narrow (Figs. 3B, 3C, SI Text) with standard deviations ranging from 0.3 – 2 s^-1 (Table S5). Moreover, ²Δ¹³C(D) of different sites are very similar (Fig. S10), as expected from relative insensitivity of deuterium isotope shifts on the chemical environment.¹⁴ Taken together, our analysis establishes that glycan H/D exchange rate constants in the range 50-200 s^-1 can be reliably measured from a global modelling of experimental CPMG and CEST profiles.

Measuring H/D exchange rate constants in sucrose: evidence for steric and hydrogen bonding effects

The site-specific H/D exchange rate constants obtained by fitting ¹³C-CEST and CPMG profiles along with R₂ bounds are indicated at the top of Figure 2. The trends in k_ex values are reflected in the shapes of the CEST profiles, providing confidence in the k_ex estimates generated by the fitting routine. For example, as k_ex increases from FC3 through FC4 and GC3 to GC6, k_ex/²Δ¹³C(D) also increases (since ²Δ¹³C(D) is virtually invariant with site) and the exchange process goes from slow (k_ex/²Δ¹³C(D) = 0.49 for FC3) to intermediate exchange (k_ex/²Δ¹³C(D) = 1.56 for GC6). In parallel, CEST profiles move from slow exchange (FC3 and FC4) where two dips are clearly discernible, through intermediate exchange (GC3), where the two dips coalesce into a single broad dip, to faster exchange, where the single dip begins narrowing (GC6). Similarly, the $\text{Δ}{R}_2^{eff}$ for GC6, which is close to intermediate exchange, is 5 s^-1, which is larger than the $\text{Δ}{R}_2^{eff}$ values for sites in slower exchange (3.5 s^-1 for FC3, FC4 and GC3).

Figure 4 shows the H/D exchange rate constants for all hydroxyls in sucrose. CH₂OH groups (open bars, k_ex ≥ 100 s^-1) typically undergo faster exchange than CHOH moieties (filled bars, k_ex ≤ 100 s^-1); this is most likely a direct consequence of the higher steric accessibility of CH₂OH groups compared to CHOHs. Interestingly, the H/D exchange rate constants in sucrose also appear to reflect the presence of transient hydrogen bonds in solution. The two hydroxyl groups that form an intra-glycan hydrogen bond in the crystal structure of sucrose, FOH1 and GOH2, have ~ 33% slower exchange rate constants than their CH₂OH (FOH6, GOH6) and CHOH (GOH3) counterparts respectively. In addition, slower H/D exchange rate constants are observed for FOH3 and GOH4, which form an inter-glycan hydrogen bond seen in NMR spectra and molecular dynamics simulations²⁰. It must be noted that there is no evidence for intra- or inter-glycan hydrogen bonds formed by FOH4, suggesting that there may be other subtle physicochemical factors like site-specific pKas governing the differences in H/D exchange rates.

Figure 4.

Hydroxyl H/D exchange rate constants of sucrose at pH 5.8 and 278 K measured using selective ¹³C-CEST and CPMG experiments. Blue and green bars represent exchange at fructose and glucose hydroxyls respectively. Open bars show exchange rate constants for hydroxyls attached to methylene carbons, while filled bars depict exchange at methine carbons.

Similar trends in hydroxyl H/D exchange rate constants have been observed in literature. Specifically, saturation transfer and 2D ¹H-¹H EXSY measurements on sucrose and glucose in acetone/water mixtures revealed that CH₂OH groups exchange more rapidly with water than CHOHs^10,11 (Fig. S11). The saturation transfer-based H/D exchange rate constant for GOH3 in sucrose is also noticeably higher than for other CHOH¹⁰, agreeing well with our observations (Fig. S11A). Interestingly, all the CHOHs in the monosaccharide glucose were observed to exchange at comparable rates in 2D EXSY data¹¹, and the uniformity in glucose H/D exchange rate constants supports our observation that the smaller k_ex seen in FOH3, FOH4, GOH2 and GOH4 of sucrose indeed originate from transient intra- and inter-glycan hydrogen bonds in sucrose (Fig. S11B). However, the sensitivity of our method for larger k_ex values makes the patterns in H/D exchange rate constants more noticeable, and isothermal rate constant measurements can be made without taking recourse to mixed solvent systems or supercooled water.

The thermodynamics underlying hydrogen exchange can be probed using the fractionation factor (ϕ), which is given by the equation:

$$\varphi =\frac{{\text{k}}_{\text{HD}}}{{\text{k}}_{\text{DH}}}=\frac{{\text{p}}_{\text{C}-\text{OD}}\left[{\text{H}}_2\text{O}\right]}{{\text{p}}_{\text{C}-\text{OH}}\left[{\text{D}}_2\text{O}\right]}$$

where [H₂O] and [D₂O] define the solvent composition.^14,38,39 The fractionation factor therefore is the relative occupancy of the deuterated (C-OD) over the protonated (C-OH) states when the concentrations of H₂O and D₂O in the solvent are equal. Fractionation factors are related to the differences between the zero-point energies of a proton and deuteron at a hydroxyl site in the glycan versus in water.³⁹ ϕ values deviate from unity for strongly hydrogen-bonded systems.^40,41 The site-specific ϕ values for sucrose can be directly obtained from p_C-OD and p_C-OH derived from fits of CEST and CPMG data to the Bloch-McConnell equations and are shown in Figure S12. ϕ values for sucrose cluster closely around 1.0, confirming that the hydrogen bonds formed by hydroxyl groups in glycans are very similar in nature to those found in water.

Detecting transient hydrogen bonds in maltose and cellobiose

Since hydroxyl groups of sucrose engaged in transient hydrogen bonds show significantly slower H/D exchange, we sought to verify this trend in other glycans that are known to form intra-glycan hydrogen bonds. Accordingly, we chose maltose and cellobiose as model systems, because they are both disaccharides of glucose that have the same 1-4 regiochemistry at the glycosidic linkage but differ in the stereochemistry (maltose: α(1-4), cellobiose: β(1-4)) (Fig. 5A, 5B). In addition, X-ray and neutron diffraction data have reported the existence of hydrogen bonds in both maltose^42,43 and cellobiose^44,45. The hydroxyl attached to C3 in the glucose ring at the reducing end (G1) is hydrogen bonded in both glycans, but to G2OH2 in maltose^46,47 and to the G2 ring oxygen in cellobiose⁴⁸. Based on the H/D exchange trends seen in sucrose, we predicted that G1OH3 in both maltose and cellobiose should undergo slower exchange than a reference CHOH group like G1OH2 that is not involved in hydrogen bonds.

Figure 5.

H/D exchange rate constants in cellobiose and maltose indicate the existence of transient hydrogen bonds. Chemical structures of maltose (A) and cellobiose (B). In each structure, the glucose unit at the reducing end is labelled as G1 and the glucose ring at the non-reducing end is labelled as G2. The glycosidic linkage is highlighted in red and the hydrogen bond reported in literature is denoted as a green dashed line. Carbons attached to hydrogen bonded hydroxyls (for which ¹³C-CEST and ¹³C-CPMG profiles are shown in panels (E) and (F)) are labelled in green and carbons attached to non-hydrogen bonded hydroxyls (for which ¹³C-CEST and ¹³C-CPMG profiles are shown in panels (C) and (D)) are labelled in blue. (C-F) ¹³C-CEST profiles (left) at four B₁ fields and the ¹³C-CPMG profile (right) for non-hydrogen bonded hydroxyls attached to βG1C2 of maltose (C) and αG1C2 of cellobiose (D), and hydrogen bonded hydroxyls attached to βG1C3 of maltose (E) and αG1C3 of cellobiose (F). Solid lines are global fits of the data (solid circles) to the Bloch-McConnell equations. The H/D exchange rate constant at 278 K and pH 6.2 estimated by fitting ¹³C-CEST and ¹³C-CPMG together is indicated at the top of each panel along with the name of the site.

Figures S13 and S14 show the ¹H-¹³C HSQC spectra of maltose and cellobiose, while the ¹³C-CEST and CPMG profiles of G1C2 and G1C3 of maltose and cellobiose are depicted in Figures 5C-5F and Figure S15. The H/D exchange rate constant for the hydroxyl group G1OH2 which is not hydrogen-bonded in either sugar is 114 s^-1 (maltose) or 106 s^-1 (cellobiose) (Fig. 5 and S16). These values compare favourably with the free CHOH in sucrose (94 s^-1). In contrast, the hydroxyl group G1OH3 which is predicted to be hydrogen bonded in both sugars has much smaller H/D exchange rate constants of 76 s^-1 in maltose and 61 s^-1 in cellobiose (Fig. 5, S15 and S16), comparable to the hydrogen-bonded hydroxyls like GOH2 in sucrose. It must be emphasized that the difference in k_ex values between G1OH2 and G1OH3 in maltose cellobiose is not merely an output of the fitting routine, but can be directly visualized from the CEST and CPMG profiles for each site; faster H/D exchange in G1OH2 results in a coalescence of the dips from C-OH and C-OD that can be differentiated in CEST profiles of G1C3, and larger dispersions in CPMG profiles of G1C2 arise from faster H/D exchange in G1OH2 over G1OH3 (Fig. 5). In summary, the data on maltose and cellobiose provide support for the role of intra-glycan hydrogen bonds in slowing down H/D exchange as well as for the ability of a combined ¹³C-CEST and ¹³C-CPMG analysis in detecting these transient hydrogen bonds.

Concluding remarks

Our perception of protein molecular dynamics has been considerably enriched by NMR experiments like chemical exchange saturation transfer (CEST) and Carr–Purcell–Meiboom–Gill (CPMG) relaxation dispersion.^17,37,49 These experiments have enabled the structural characterization of ‘invisible’ states⁵⁰ and the measurement of molecular properties like diffusion coefficients⁵¹, paramagnetic relaxation enhancement⁵² and hydrogen exchange rates⁵³ in these elusive states. Here we demonstrate the potential of a synergistic combination of ¹³C-CEST and CPMG methods in measuring isothermal H/D exchange rate constants in glycans. Since exchange is studied while magnetization is present on ¹³C nuclei, rate constants can be quantified robustly without interference from dipolar cross-relaxation. ¹³C-¹³C homonuclear scalar couplings also do not complicate the measurement of H/D exchange rates because natural abundance ¹³C magnetization is leveraged in these experiments. Moreover, the use of unlabeled glycan samples eliminates the need for glycan isotope labelling, which is challenging because glycan synthesis is not templated, unlike the cellular production of proteins and nucleic acids. The selective 1D excitation provides flexibility for probing nuclei of interest one at a time, and high-sensitivity CPMG and CEST profiles for a target nucleus can be collected on glycan samples down to 25 mM concentration within a few hours (Fig. S17). We anticipate that precise estimates of hydroxyl exchange rate constants available from our method will be beneficial for detecting transient hydrogen bonds in glycans, which can be used as structural restraints in defining glycan ensembles and for locating hotspots involved in glycan-protein and glycan-glycan recognition.

Associated content

Supporting Information

Supporting Text including a Materials and Methods section describing sample preparation, details of the pulse sequence, ¹³C-CEST and CPMG data processing and analysis, quantitative analysis of differential carbon relaxation in two states; Supporting Figures highlighting the results of simulations, additional glycan ¹³C-CEST and CPMG profiles, ${\chi }_{red}^2$ surfaces, variations in fit parameters at different k_ex values in the ${\chi }_{red}^2$ surface of GC6 in sucrose, intrinsic R₂ measurements from ¹³C-CPMG data, site-specific fractionation factors and ²Δ¹³C(D), results of bootstrap analysis for error estimation, and HSQC spectra of cellobiose and maltose; Supporting Tables of glycan chemical shifts, fit parameters and error analysis from ¹³C-CEST and CPMG profiles.