Protein Recognition of Glycosphingolipids in Membranes: Mechanistic and Quantitative Insights

Abstract

Recognition of glycosphingolipids (GSLs) in cell membranes by glycan-binding proteins (GBPs) is essential to diverse biological processes. However, owing to deficiencies in available analytical methods, the thermodynamics of GBP–GSL interactions remain poorly characterized. Native mass spectrometry (nMS) analysis performed using soluble GSL-containing model membranes provides a direct readout of the identity and stoichiometry of bound GSL ligands and, under certain conditions, can inform on affinity. Yet, for multivalent GBPs capable of engaging multiple model membranes simultaneously, data analysis relies on untested assumptions, which has limited adoption of the assay. Here, we apply mass photometry to quantify a series of high-affinity interactions between glycolipids in soluble model membranes (nanodiscs) and mono- and multivalent GBPs and compare with binding data acquired with nMS. Remarkably, the mass photometry results indicate that glycolipids are distributed non-statistically across the lipid bilayer and engage in clustering that is sensitive to GBP binding. Moreover, the affinities and stoichiometries (of bound nanodiscs) measured for multivalent GBPs are strongly modulated by glycolipid clustering, which can overwhelm avidity gains from multivalent binding. After normalizing for the number of GBP binding sites and glycolipid content, the affinities from mass photometry are found to be, overall, in good agreement with native nMS-derived affinities. Collectively, the findings of this study provide critically needed affinity and stoichiometry benchmarks for assay validation and significant new insights into the mechanisms of GBP recognition of GSLs in model membranes, which serve as a foundation for understanding binding in natural cellular environments.

Graphical Abstract

INTRODUCTION

Glycosphingolipids (GSLs) comprise between 2% and 20% of the total lipids in the plasma membranes of human and other vertebrate cells.^1,2 Recognition of GSLs by soluble and membrane-bound glycan-binding proteins (GBPs) is important in many physiological processes, including cellular recognition, adhesion, signal transduction, trafficking and the immunological response.^1,2 Additionally, many pathogens utilize host GSLs in the infection process. For example, GSLs serve as receptors (primary or co-receptors) for many human viruses (e.g. human immunodeficiency virus recognizes GM3 and Gb3 gangliosides,³ and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) uses sialylated GSLs for cell entry)^3,4 and microbial virulence factors (e.g. Shiga toxin exploits globosides,⁵ while GM1 ganglioside is a high-affinity ligand of cholera toxin⁶ and E. coli enterotoxin).^5–8 GSLs are also cancer-related antigens (e.g. the GD2 and GD3 gangliosides are upregulated in breast cancer and melanoma tissue),⁹ suggesting that they may serve as targets for vaccines¹⁰ or therapeutic monoclonal antibodies (mAbs).^11,12

Despite the importance of GBP–GSL interactions in human health and disease, relatively little is known about the GSL specificities of most GBPs. Moreover, the kinetic and thermodynamic parameters of the majority of identified interactions have not been quantified. The GSL binding properties of GBPs are often established from measurements performed on the carbohydrate moiety of GSLs, free in solution or immobilized on a surface,² or intact GSLs immobilized via the ceramide on a surface.^13,14 Such measurements are relatively easy to perform using a variety of established binding assays, including isothermal titration calorimetry (ITC)¹⁵ and glycan microarrays.¹⁶ However, the non-native environment and presentation and the lack of mobility and clustering may significantly affect the measured parameters.^14,17

Alternatively, measurements can be performed using GSL-containing model lipid membranes, which mimic the natural cell membrane environment.^14,17,18 Model membrane systems from all three structural classes - planar supported bilayers (supported lipid bilayers), monolayer (e.g., micelles) and bilayer vesicles (e.g., liposomes), and bilayer islands (e.g., nanodiscs (NDs) - have been used to study GBP–GSL binding using a variety of analytical techniques.^{14,19,20–27} While convenient for quantifying associations of GBPs with GSL-containing model membranes, interpretation of the resulting binding data for GBPs with multiple binding sites requires knowledge of the number and type of GSLs bound. Moreover, the model membrane binding stoichiometry must also be known when using bilayer islands or vesicles. Currently, though, there is no binding assay capable of directly measuring the stoichiometry of GSL- and model membrane-binding to multivalent GBPs. This deficiency represents a significant barrier to developing a comprehensive model of GBP recognition of GSLs in membranes.

Recently, native mass spectrometry (nMS) has emerged as a promising method for probing GBP interactions with GSLs in soluble model membranes.^14,28 Importantly, nMS analysis, which typically involves electrospray ionization (ESI)-MS performed using native-like solution conditions and gentle sampling conditions that preserve non-covalent interactions, enables the identity and stoichiometry of bound GSLs to be directly measured. Using direct (slow mixing mode (SLOMO))²⁹ or competitive binding strategies,^14,28 such as the proxy ligand assay, the strength of binding can also be assessed by nMS. But because the model membrane-associated GBP–GSL complexes are kinetically unstable in ESI, and dissociate predominantly via loss of intact GBP-GSL complexes,^14,28,29 nMS analysis doesn’t provide a measure of the GBP-to-model membrane stoichiometries that exist in solution. As a result, interpretation of nMS-derived affinity data often relies on untested assumptions of the GBP-to-model membrane stoichiometries. This limitation has hindered the widespread adoption of nMS for studying GBP–GSL interactions.

In this work, we seek to gain a deeper understanding of GBP recognition of GSLs in a membrane environment and, in so doing, advance the development of assays capable of reliably quantifying GBP binding to GSL-containing membranes. To this end, we applied nMS and mass photometry (MP) to quantify a series of high-affinity interactions between glycolipids in soluble model membranes (NDs) and soluble mono- and multivalent GBPs. While MP has previously been used to quantify biomolecular interactions, this is the first demonstration of the assay for the quantification of GBP–GSL binding.^30–32 The MP data provide valuable insights into the mechanisms of GBP recognition of GSLs in model membranes, which serve as a foundation for understanding binding in natural cellular environments. They also yield affinity and stoichiometry benchmarks critically needed for assay validation and, importantly, establish the reliability of nMS for quantifying GBP binding to GSL-containing model membranes.

EXPERIMENTAL

Materials and Methods

Proteins, carbohydrates, lipids and glycolipids.

Details of the proteins, (glyco)lipids, oligosaccharides and other reagents used in this study are given as Supporting Information. Structures of the (glyco)lipids and oligosaccharides are shown in Figure S1.

Preparation of model membranes.

A detailed description of the procedures used to prepare and characterize empty and glycolipid-containing ND: 0.5%, 1.0%, 5.0% and 10% GD2-ND; 0.5%, 1.0%, 5.0% and 10% A1_NGL-ND; 0.5%, 10.5%, 1.0%, 2.0% and 5.0% GM1-ND prepared with 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) or 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) is given as Supporting Information. The molecular weights (MWs) and glycolipid content of the NDs are listed in Table S1.

Isothermal titration calorimetry (ITC).

The affinity of the GD2 hexasaccharide (GD2_os) for the antigen-binding fragment (Fab) was measured by ITC performed using a MicroCal PEAQ instrument (Malvern Panalytical, Worcestershire, United Kingdom). Experimental details are given as Supporting Information.

Mass photometry.

The MP measurements were performed using a Refeyn One^MP mass photometer (Refeyn Ltd, Oxford, UK). Data acquisition and processing was performed using AcquireMP (Refeyn Ltd, v2.4).^30–32 Details of experimental and instrumental parameters used and data analysis procedures are provided as Supporting Information.

Mass spectrometry.

All MS measurements were carried out using a Synapt G2S ESI quadrupole-ion mobility separation-time-of-flight (Q-IMS-TOF) mass spectrometer (Waters, Manchester, UK), a Q Exactive Classic Orbitrap or a Q Exactive Ultra High Mass Range Orbitrap mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). All three instruments were equipped with a nanoflow ESI (nanoESI) source. Details of the experimental and instrumental parameters used and data analysis procedures are provided as Supporting Information.

RESULTS AND DISCUSSION

Model systems.

The stoichiometry and affinity (apparent and intrinsic) of interactions between soluble GBPs, possessing one, two or five high-affinity glycan binding sites, and glycolipid-containing NDs were measured using MP and nMS (Figure S2). Suitable test systems are limited by the disparate concentration and affinity requirements of MP (≤1 μM) and nMS (≥0.1 μM). The Fab fragments of humanized 3F8 anti-GD2 monoclonal antibody (hu3F8) and the related double point mutant E1K/D32H (hu3F8 E1K/D32H), which specifically recognize the ganglioside GD2, served as model systems for monovalent GBPs.^33,34 In the present work, the K_d of E1K/D32H Fab binding to GD2_os was determined to be 0.62 ± 0.22 μM by ITC (Figure S3) and 0.5 ± 0.1 μM by nMS (Figure S4a–c). Due to limited protein availability, the K_d for hu3F8 Fab binding could not be measured by ITC, but a value of 2.3 ± 0.1 μM was determined by nMS (Figure S4d–f). R-phycoerythrin (PE) mouse (IgG3, κ isotype) anti-human blood group A antibody clone NaM87–1F6 (NaM87–1F6 mAb), which possesses two equivalent and independent binding sites for blood group A antigens,³⁵ served as a model for GBPs capable of forming divalent interactions with membrane GSLs. Because there are no commercially available GSLs corresponding to any of the six A subtypes, a blood group A type 1 neoglycolipid (A1_NGL) was used. There is no reported affinity of A type 1 oligosaccharide (or A1_NGL) for NaM87–1F6 mAb. However, an intrinsic (per binding site) K_d (K_d,int) of 0.60 ± 0.14 μM was measured by nMS for the A1 tetrasaccharide. Cholera toxin B subunit homopentamer (CTB₅), which has five equivalent and dependent, high-affinity GM1 ganglioside binding sites, was used as a model pentavalent GBP. Intrinsic K_d of 0.16 ± 0.01 μM¹⁶ and 0.31 ± 0.02 μM,³⁶ measured by ITC and nMS, respectively, have been reported for the GM1 pentasaccharide (GM1_os). Occupancy of one and two neighbouring subunits strengthens binding by factors of 1.7 and 2.9, respectively.³⁶ The K_d,int for interactions between CTB₅ and 1%, 2% and 5% GM1-ND, measured using the proxy ligand nMS method, are 0.53 ± 0.11 μM, 1.32 ± 0.19 μM, and 1.79 ± 0.26 μM, respectively.³⁷

Interpretation of measured K_d.

For each GBP, the complexes formed with glycolipid-containing NDs, their equilibrium concentrations and corresponding apparent (macroscopic) K_d (K_d,app) were measured using MP. For interactions between monovalent GBP (single binding site) and ND containing a single glycolipid, K_d,app is equivalent to K_d,int. Extracting K_d,int from K_d,app for multivalent interactions (GBP with multiple binding sites and ND containing multiple glycolipid ligands) is not straightforward, as the magnitude of K_d,app is influenced by statistical factors, multivalent and cooperative binding, and glycolipid clustering. Uncertainty in the distribution of glycolipids within the NDs further complicates interpretation. While it is possible to construct theoretical binding models that take into account these various effects, it is generally not possible to use them to extract K_d,int from MP data. Instead, in an effort to capture effects related to statistical factors, glycolipid distribution and clustering, we report $K_{\text{d,app}}^{\text{norm}}$, which corresponds to K_d,app normalized for number of glycolipids (per ND) and number of GBP binding sites.³⁸ Direct nMS, implemented with SLOMO, served to inform on glycolipid binding stoichiometry, the fractional abundances of GBP-glycolipid complexes and establish K_d,int. Two different indirect nMS assays, utilizing a proxy ligand (L_proxy) strategy, were also used to assess K_d,int.

Apparent affinities of GBP–glycolipid interactions from MP measurements

Monovalent GBP.

Signal corresponding to free Fab and GD2-ND and the 1:1 Fab–GD2-ND complex was detected by MP for the 0.5% (1.2 GD2 per ND), 1% (1.9), 5% (8.4) and 10% (17.4) GD2-ND (Figures 1a–h). A 2:1 Fab–GD2-ND complex was also detected for all but the 0.5% GD2-ND (Table S2). Plots of the concentration-dependent fractions of the 1:1 and 2:1 complexes are shown in Figures 1i–n. Apparent affinities for the 1:1 (K_d,app,Fab-ND) and 2:1 Fab-GD2-ND complexes (K_{d,app,2Fab-ND}) were calculated (eqs S1a–g) at each Fab concentration and average values are listed in Table S3.

Figure 1. Binding of monovalent hu3F8 and E1K/D32H Fabs to GD2-ND quantified by MP.

(a-h) Representative mass distributions measured for (a, b) 0.5% GD2-ND (50 nM) with (a) hu3F8 Fab (480 nM) or (b) E1K/D32H Fab (480 nM); (c, d) 1% GD2-ND (50 nM) with (c) hu3F8 Fab (480 nM) or (d) E1K/D32H Fab (480 nM); (e, f) 5% GD2-ND (50 nM) with (e) hu3F8 Fab (100 nM) or (f) E1K/D32H Fab (100 nM); (g, h) 10% GD2-ND (50 nM) with (g) hu3F8 Fab (100 nM) or (h) E1K/D32H Fab (100 nM). (i, j) Fractions of 1:1 Fab–GD2-ND complex measured for solutions containing 50 nM of (i) 0.5% and (j) 1% GD2-ND and different concentrations of hu3F8 (green) and E1K/D32H (red) Fab. (k-n) Fractions of 1:1 Fab–GD2-ND (brown) and 2:1 Fab–GD2-ND (orange) complexes measured for solutions of 5% GD2-ND (50 nM) and (k) hu3F8 or (l) E1K/D32H Fab or solutions of 10% GD2-ND (50 nM) and (m) hu3F8 or (n) E1K/D32H Fab. (o) Plot of $K_{\text{d,app,Fab-GD2}}^{\text{norm}}$ (μM) versus average number of GD2 in the ND for the interactions between hu3F8 Fab (green) or E1K/D32H Fab (red) and 0.5%–10% GD2-ND. The solid curves are best linear fits. All measurements were performed in ammonium acetate solutions (200 mM, pH 6.8, 22 °C). Each affinity value is the average of n ≥3 replicates. Errors correspond to one standard deviation.

For the 0.5% GD2-ND, the K_d,app,Fab-ND (≡ K_{d,int,Fab-GD2}) are similar to the K_d measured for GD2_os, indicating that, in the absence of other effects, the ceramide moiety and ND environment have little effect on Fab recognition of GD2. Notably, K_d,app,Fab-ND systematically decreases with increasing number of GD2 in the ND. However, when normalized for GD2 content, the resulting $K_{\text{d,app,Fab-GD2}}^{\text{norm}}$ (Table S4) increases linearly (~0.25 μM/GD2) with number of GD2 (Figure 1o). This increase is attributed to clustering of the GD2, similar to what has been reported for GM1.^13,17,37,39 Clustering, which is believed driven by intermolecular H-bonding between carbohydrate moieties,¹⁷ impedes binding by reducing the number of accessible GD2. Analysis of the corresponding values of binding free energies (-ΔG°), which were calculated from $K_{\text{d,app,Fab-GD2}}^{\text{norm}}$ (Table S5), reveals that the per GD2 free energy penalty (ΔΔG°) resulting from clustering is in the 0.2 kJ mol⁻¹ to 0.5 kJ mol⁻¹ range (Table S6). These data demonstrate that clustering strongly regulates the thermodynamics of GBP–GSL binding and, possibly, suggest that disease-associated upregulation of GSLs in cells may result in attenuated endogenous GBP binding.

From a consideration of statistical factors, K_{d,app,2Fab-ND} is expected to be 2- to 4-times larger than K_d,app,Fab-ND if all GD2 are equally accessible (Figure M1). Instead, K_{d,app,2Fab-ND} is 2- to 4-times smaller for the 5% and 10% GD2-ND (Table S3). This surprising finding, indicative of positive cooperativity, may result from the disruption of GD2 clustering upon binding of the first Fab, resulting in more accessible GD2.⁴⁰ Indeed, GBP-driven dispersal of GSL-rich ordered domains in supported lipid bilayers has been postulated previously.^41,42 However, to our knowledge, no connection to binding strength has yet been described. Alternatively, positive cooperativity could result from favourable Fab-Fab interactions, assuming each leaflet can accommodate multiple Fab molecules (Figure S5).⁴³ For the 1% GD2-ND, precise values of K_{d,app,2Fab-ND} could not be obtained because of low signal. Nevertheless, the estimated values are ~4K_d,Fab-ND (Figure M1), which suggests binding-induced GD2 declustering as the source of positive cooperativity observed for the 5% and 10% GD2-ND. Evidence of GBP-induced GSL (glycolipid) declustering is also found for the other model systems investigated, vide infra, pointing to this being a general phenomenon. However, the importance of this effect in the context of cell recognition remains to be demonstrated.

Bivalent GBP (with equivalent and independent binding sites).

Representative MP histograms for solutions of NaM87–1F6 mAb and A1_NGL-ND are shown in Figures S6a and 2a,b. Due to spectral overlap the equilibrium concentrations of free mAb and A1_NGL-ND were determined from the initial concentrations, the relative abundances of the complexes and mass balance considerations (eqs S2a–i). The 1:1 mAb-A1_NGL-ND complex was detected in solutions of 0.5% (1.1 A1_NGL per ND) and 1% A1_NGL-ND (2.1) (Figure 2a, Table S2). For the 5% (9.1) and 10% A1_NGL-ND (18.9), signals centered at MW 478 ± 8 and 485 ± 3 kDa, respectively, were also detected, albeit at low abundance (Figure 2b). These signals could, in principle, arise from either or both of the 2:1 and 1:2 mAb-A1_NGL-ND complexes, which have similar MWs (Table S2). To clarify, protein A, which consists of a mixture of proteins with MWs ranging from 21.5 to 28.5 kDa (Figure S6b), that binds strongly to the Fc region of mouse IgG3,⁴⁴ was added. According to MP data, protein A forms 2:1 complexes with the mAb, with MWs of 194 ± 2 kDa (Figure S6c). Measurements performed on solutions of 5% or 10% A1_NGL-ND and mAb, in the absence and presence of protein A (100 nM), revealed that the addition of protein A results in ~100 kDa increase in the MW of the complexes. This increase is consistent with 4 protein A bound to 2 mAb. Based on these results, it is concluded that the signal measured at ~500 kDa corresponds to the 2:1 mAb-ND complex. The absence of 1:2 mAb-ND complex is, however, surprising (Figures S6d,e) and could be due to unfavourable interactions arising from two A1_NGL-ND bound simultaneously (Figure 2e). Such steric effects have been previously observed for protein-lipid membrane interactions.⁴⁵ For the higher percentage A1_NGL-ND, it is also possible that the two mAb binding sites engage two A1_NGL present on the same ND leaflet, which precludes mAb binding to another ND (Figure 2j). As described below, the results of nMS measurements provide support for both explanations.

Figure 2. Binding of bivalent NaM87–1F6 mAb to A1_NGL-ND quantified by MP.

(a-c) Representative mass distributions measured for solutions of NaM87–1F6 mAb (20 nM) and (a) 0.5% A1NGL-ND (17 nM) or (b) 10% A1NGL-ND (6 nM) or (c) protein A (100 nM) and 10% A1NGL-ND (6 nM). (d) Plot of $K_{\text{d,app,mAb-A1NGL}}^{\text{norm}}$ (μM) versus average number of A1NGL in ND for interactions between NaM87–1F6 mAb and 0.5%–10% A1NGL-ND. Dashed line is best linear fit. (e, f) Proposed geometric orientation for the simultaneous binding of (e) mAb (red) to 2 A1NGL-ND (grey), which leads to steric hindrance, and (f) mAb to 2 A1_NGL located on the same ND leaflet. All measurements were performed in ammonium acetate solutions (200 mM, pH 6.8, 22 °C). Each measurement was performed with n ≥ 3 replicates. Errors correspond to one standard deviation.

Listed in Table S3 are the average K_d,app,mAb-ND (1:1 mAb-ND complex) and K_{d,app,2mAb-ND} (2:1 complex) for all four A1_NGL-ND calculated from the concentration-dependent fractional abundances of free and bound mAb (Figure S7). As with the Fab interactions with GD2-ND, K_d,app,mAb-ND decreases with the increasing A1_NGL content but $K_{\text{d,app,mAb-A1NGL}}^{\text{norm}}$ increases slightly (slope ~0.08 μM/A1_NGL), which is consistent with A1_NGL clustering in the ND. Notably, $K_{\text{d,app,mAb-A1NGL}}^{\text{norm}}$ measured for the 0.5% A1_NGL-ND is similar to the K_d measured for A1 oligosaccharide, suggesting that the lipid moiety doesn’t significantly affect mAb binding. For the 5% and 10% A1_NGL-ND, the values of K_{d,app,2mAb-ND} are approximately equal to K_d,app,mAb-ND (Table S3). Based on statistical factors alone, K_{d,app,2mAb-ND} is expected to be 2K_d,app,mAb-ND or 4K_d,app.mAb-ND depending on whether mAb can bind to one or two A1_NGL on the same ND leaflet (Figure M2). The similarity of K_{d,app,2mAb-ND} and K_d,app,mAb-ND values indicates (like for the Fab-GD2-ND system) that sequential mAb binding exhibits positive cooperativity, presumably due to binding-induced declustering of A1_NGL.

Pentavalent GBP (five equivalent and dependent binding sites).

Representative MP histograms measured for solutions of CTB₅ and 0.5% (1.1 GM1 per ND), 1% (2.0), 2% (3.8), 5% (5.7) and 10% (15.4) GM1-ND are shown in Figures 3a–d, S8 and S9. The concentration-dependent fractions (f_{i_j}) of each species are shown in Figures 3e–h. The values of K_d,app for the stepwise binding of GM1-ND to CTB₅ are summarized in Table S3. The measurements performed on the solutions containing 0.5% GM1-ND enable K_d,int to be directly determined, as well as the number of GM1-ND that can bind simultaneously to CTB₅. The higher percentage GM1-ND, which possess multiple GM1 distributed across both leaflets, enable the contributions of multivalent binding and clustering to K_d,app to be assessed.

Figure 3. Binding of pentavalent CTB₅ to GM1-ND quantified by MP.

(a-d) Representative mass distributions measured for solutions of CTB₅ (50 nM) and (a) 0.5% GM1-ND (200 nM), (b) 1% GM1-ND (200 nM), (c) 2% GM1-ND (150 nM) or (d) 5% GM1-ND (100 nM). (e-h) Plots of concentration-dependent fractions (f_i__j) of free and ND-bound CTB₅ species (where i and j are number of CTB₅ and ND in the complex: f_{1_0}, blue; f_{1_1}, red; f_{1_2}, green; f_{1_3}, orange and f_{2_1}, purple). Solid curves in (e) represent the theoretical fractions calculated for 0.5% GM1-ND considering five equivalent and independent binding sites and a K_d,int of 150 nM. Solid curves in (f-h) represent the theoretical fractions calculated for (f) 1%, (g) 2% and (h) 5% GM1-ND using experimental values of K_{d,app,CTB5-ND}, K_{d,app,CTB5–2ND}, K_{d,app,CTB5–3ND} and K_{d,app,2CTB5-ND} (if applicable) in Table S3. (i) Plot of $K_{\text{d,app,CTB5-ND}}^{\text{norm}}$ (red), $K_{\text{d,app,CTB5–2ND}}^{\text{norm}}$ (green) and $K_{\text{d,app,2CTB5-ND}}^{\text{norm}}$ (purple) versus average number of GM1 in the ND. All measurements were performed in ammonium acetate solutions (200 mM, pH 6.8, 22 °C). Each measurement was performed with n ≥3 replicates. Errors correspond to one standard deviation.

At the 0.5% GM1-ND concentrations tested, free GM1-ND and CTB₅ bound to up to 3 GM1-ND were detected (Figures 3a and S8a–d). Free CTB₅ was also observed at concentrations <100 nM. It was previously suggested that CTB₅ binds, at most, two GM1-containing ND.²⁸ However, the current results establish that CTB₅ can simultaneously engage up to three GM1-ND. Indeed, treating the GM1-ND and CTB₅ as rigid particles reveals that CTB₅ can accommodate three, but not four GM1-ND (Figures S10 and S11). Along with the species described above, GM1-ND bound to 2 CTB₅ was detected with 1% GM1-ND (Figure 3b). This complex presumably involves a CTB₅ binding to each ND leaflet. For 2% GM1-ND, with the exception of an absence of signal corresponding to CTB₅ bound to 3 GM1-ND, the complexes are the same as for 1% GM1-ND (Figure 3c). For the 5% GM1-ND, the only complexes detected are GM1-ND bound to 1 and 2 CTB₅ (Figure 3d), while for 10% GM1-ND, it was not possible to unambiguously identify any complexes formed (Figure S12).

Sequential GM1-ND binding.

The trend of increasing K_{d,app,CTB5-jND} for the 0.5% – 2% GM1-ND samples is expected based on increasing occupancy of CTB₅ binding sites. The 0.5% GM1-ND possesses, on average, a single GM1 and the affinity data are, therefore, free of contributions from multivalent binding and clustering. By treating the five CTB₅ binding sites as equivalent and independent (Figure M3a), the measured K_{d,app,CTB5-ND} corresponds to a K_{d,int,CTB5-ND} of 150 ± 35 nM. This value agrees with the K_d,int measured by ITC for GM1_os (156 ± 7 nM)¹⁵ and is within a factor of 3 of values reported for CTB₅ binding to GM1-ND using the proxy ligand nMS assay.^14,37 Moreover, the concentration dependence of f_{1_j} is reasonably well described using this K_{d,int,CTB5-ND} value and a binding model (Figure M3b) that assumes five equivalent and independent binding sites, but precludes three adjacent binding sites being occupied at the same time (Figure 3f). Notably, explicit consideration of cooperativity factors did not lead to a significant improvement in fitting (data not shown). Taken together, these results indicate that there are no significant steric effects associated with CTB₅ interacting with up to three 0.5% GM1-ND. This suprising finding, in turn, suggests that the three GM1-ND are likely not co-planar with CTB₅, but instead are able to adopt orientations more favourable to GM1 binding.^46,47

The trend of K_{d,app,CTB5-jND} for the 1% GM1-ND and their differences relative to those found for 0.5% GM1-ND provide important clues into how GM1 molecules are distributed within NDs and how they interact with CTB₅. Assuming that GM1 molecules are distributed equally between each leaflet, the expected K_{d,app,CTB5-ND} ratio (1.0% to 0.5% GM1-ND) is 0.5. However, the measured ratio (0.13 ± 0.03) is indicative of multivalent binding. This is only possible if a fraction of the 1% GM1-ND has two GM1 on the same leaflet or if unbound GM1 species (on the opposite leaflet) flips to the other face upon CTB₅ binding. Support for the latter explanation can be found in results of molecular dynamics (MD) simulations, which suggest that lipids can undergo frequent flip-flopping between the two leaflets.⁴⁸ Further evidence of an asymmetric distribution of GM1 in the ND is found in the K_{d,app,CTB5–2ND}/K_{d,app,CTB5-ND} ratio (18.8 ± 3.8), which is significantly larger than the value of 2.5 expected based on the assumption that the two GM1 are located in different leaflets (Figure M4).

That the maximum number of GM1-ND bound to CTB₅ decreases with increasing GM1 percentage (from 1% to 5%) can be explained by the greater number of GM1 per disc. Curiously though, while K_{d,app,CTB5-ND} systematically decreases with increasing GM1 percentage (Figure 3i), the $K_{\text{d,app,CTB5-ND}}^{\text{norm}}$ for 1% and 2% GM1-ND are similar in magnitude and smaller than that of 5%. These seemingly paradoxical results suggest that measured avidity of ND containing multiple GM1 results from not only multivalent binding (which increases avidity), but from GM1 clustering (which reduces avidity). Moreover, for the 5% GM1-ND, the effect of clustering appears to dominate over multivalent binding.

Sequential CTB₅ binding.

The observation that binding of a second CTB₅ to 1% and 2% GM1-ND is weaker (larger K_d,app) than weaker than binding of the first CTB₅ is consistent with a decrease in available GM1. However, based on consideration of statistical factors, a ratio of 4 is expected, assuming a symmetrical distribution of GM1 (Figures M4-M6). That the experimental K_{d,app,2CTB5-ND}/K_{d,app,CTB5-ND} ratios are much larger (~17 (1%) and ~16 (2%)) is further evidence of a non-symmetric distribution of GM1 in the ND leaflets. For the 5% GM1-ND, however, the measured ratio (~0.6) is smaller than expected. This result can be explained in terms of declustering of GM1 upon binding of the first CTB₅. Together, these findings point to the important interplay between GM1 clustering and binding-induced declustering in regulating CTB₅ binding to GM1 in membranes.

Affinity benchmarks for nMS

Native MS analysis of each model system detected gaseous ions corresponding to the expected GBP-glycolipid complexes (Figures S13–S16). And, in all cases, the maximum number of bound glycolipids is the same as number of binding sites. As discussed above and elsewhere,²⁸ the GBP-glycolipid complexes are thought to spontaneously detach from the ND during the ESI process. Because the GBP–glycolipid complexes are initially associated with the ND, their gas-phase ions are expected to exhibit nMS response factors (RFs) that are substantially different from those of the free GBP ions (which come directly from solution).²⁹ Consequently, the relative abundances of free and glycolipid-bound GBP ions measured by nMS will not necessarily reflect the relative solution concentrations. Therefore, to comprehensively evaluate K_d for the model GBP–glycolipid systems, three different nMS-based assays were used: (i) SLOMO-nMS,²⁹ (ii) proxy ligand nMS,⁴⁹ and (iii) the newly introduced CUPRA-proxy ligand nMS.⁵⁰

Fab-GD2-ND interactions.

The binding data (K_{d,int,Fab-GD2}) acquired with three nMS-based approaches are consistent and are within a factor of 2 of affinities measured by MP (Table S4). Moreover, K_{d,int,Fab-GD2} increases linearly with GD2 content (0.44 – 0.66 μM/GD2 for hu3F8 Fab (Figure S17a) and 0.21 – 0.37 μM/GD2 for hu3F8 E1K/D32H Fab (Figure S17b)), which is consistent with GD2 clustering in the ND. Moreover, the −ΔG° (Table S5) and corresponding ΔΔG° (Table S6) values are in good agreement with those measured by MP (Table S6). Taken together, these data demonstrate that the nMS-based assays, either direct (but accounting for relative RF) or indirect, can reliably quantify monovalent interactions involving GBPs and GSL-containing ND.

mAb-A1_NGL-ND interactions.

Direct nMS performed on solutions of mAb and A1_NGL-ND confirmed the binding site occupancy inferred from MP data. Specifically, for the 0.5% A1_NGL-ND, only 1:1 mAb-A1_NGL complex ions were observed, which suggests that binding of the 1^st ND sterically hinders binding of the 2^nd ND to the other binding site of the same mAb (Figures S15c,d). For 1%, 5% and 10% A1_NGL-ND, both 1:1 and 1:2 mAb-A1_NGL complex ions were detected. These results support the hypothesis that the 1:2 mAb-A1_NGL complexes involve engagement of two A1_NGL molecules on the same ND leaflet. Furthermore, the observation of the 1:2 mAb-A1_NGL from the 1% A1_NGL-ND implies asymmetric A1_NGL distribution across the leaflets (Figure S15e–g). Due to spectral overlap, only the CUPRA-proxy ligand nMS method was used to estimate K_{d,int,mAb-A1NGL}. Overall, the values agree (within a factor of ~3) with the affinities measured by MP (Figure S21 and Table S4). Additionally, K_{d,int,mAb-A1NGL} is seen to increase linearly with A1_NGL content (Figure S17c), which is consistent with glycolipid clustering.

CTB₅-GM1-ND interactions.

Signal corresponding to CTB₅ bound to up to three GM1 was detected by nMS analysis of the 0.5% GM1-ND sample, which supports the conclusion that CTB₅ can bind simultaneously to a maximum of three ND (Figure S16a). For 1% GM1-ND, up to four bound GM1 were detected, a result consistent with the ascertion that the two GM1 are asymmetrically distributed between the leaflets (Figure S16b). For the 2% and 5% GM1-ND, up to 5 GM1 bound to CTB₅ were detected (Figures S16c–f).

As SLOMO-nMS is not readily applied to measure binding of CTB₅ and GM1 in multiple ND, the proxy ligand nMS assay was implemented using GM1_os at a concentration that leaves, on average, a single binding site available. The assay was previously used to estimate K_{d,int,CTB5-GM1} for 0.5% and 1% GM1-ND and, in the present work, the measurements were extended to the 2% and 5% GM1-ND. Notably, the K_{d,int,CTB5-GM1} for the 0.5% GM1-ND is in reasonable agreement (within a factor of 4) with the $K_{\text{d,app,CTB5-GM1}}^{\text{norm}}$ measured by MP, indicating that, in the absence of other effects, the competitive nMS assay provides a reliable measure of intrinsic GSL affinities for GBP with multiple binding sites. In contrast, the nMS-derived values measured for the 1% and 2% GM1-ND are approximately 20-fold larger than the MP affinities. These differences likely reflect the design of the proxy ligand nMS experiments, wherein the contribution from multivalent binding is minimized. Finally, there is good agreement between the nMS- and MP-derived affinities for the 5% GM1-ND. However, this agreement appears to be largely coincidental. As noted above, the affinity for the 5% GM1-ND measured by MP is strongly influenced by clustering, which effectively offsets avidity gain through multivalent binding. In contrast, the nMS-derived affinity is only sensitive to the effect of clustering on the monovalent interaction.

CONCLUSIONS

The results of this work, the first comprehensive, quantitative investigation of the interactions between soluble GBPs and high-affinity glycolipid ligands incorporated into soluble model membranes, provide affinity and stoichiometry benchmarks critically needed for the development and validation of assays for characterizing GBP recognition of GSLs in membrane environments. They also yield important new insights into the factors that control the strength of binding. Key findings of this work are summarized below.

1. Rapid translocation of glycolipids occurs between leaflets. The binding data generated in this study point to a non-homogenous distribution of glycolipids between ND leaflets. While some heterogeneity may arise at the time of preparation, the present data suggests that glycolipids translocate (flip-flop) between leaflets on a timescale that is sufficiently short to influence GBP binding. This finding has important consequences for the interpretation of binding data acquired for multivalent GBPs using GSL-containing bilayer islands, as the asymmetric distribution will, together with other factors, modulate the contribution of multivalent binding to the apparent affinity (avidity).

2. Glycolipids engage in clustering that is sensitive to GBP binding. The affinities and stoichiometries of GBP complexes with glycolipid-ND are strongly modulated by glycolipid content. Clustering generally weakens GBP binding and, at high glycolipid content, the inhibitory effect is of comparable magnitude to avidity gains achieved from multivalent binding. Moreover, sequential GBP binding to NDs is found to exhibit positive cooperativity, suggestive of binding-induced glycolipid declustering. These findings have important implications for cell recognition, such as in the context of pathogens that exploit GSLs for infection and the development of protein therapeutics that target GSLs.

3. GBP–GSL affinities are reliably quantified by nMS. Overall, there is good agreement between the MP-derived affinities, normalized for binding sites and number of glycolipids in the ND, and affinities measured by direct and indirect nMS assays. This finding is significant as nMS analysis is not limited to high-affinity interactions and can be applied generally to GBP–GSL interactions, which are typically of low affinity. However, challenges associated with the quantification of coupled equilibria remain to be addressed for nMS-based analysis to reach its full potential for quantifying GBP–GSL interactions in membrane environments.

This study has uncovered GBP–GSL binding details on a level not previously achieved, and yielded important new insights into the mechanisms that govern GBP interactions with GSLs in model membranes. These findings will serve not only as a guide for experimental design and data interpretation of binding data acquired using model membranes, but also provide a critical foundation for understanding GBP–GSL binding in natural cellular environments. These insights are of fundamental importance and may aid the development of new anti-GSL antibodies for the treatment and diagnosis of human diseases, including neurological disorders and cancer.^51,52

Supplementary Material

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

• Additional details on the preparation of the nanodisc samples and quantification of glycolipid content; experimental details and data analysis procedures for ITC, MP, and nMS binding measurements; structures of the gangliosides, oligosaccharides and lipids; geometrical illustrations of interactions between GBPs and NDs; representative MP mass histograms and ESI mass spectra; and titration plots (Tables S1–S9 and Figures S1–S24, PDF)