Galectins are a family of proteins that bind to b-galactosidase-containing carbohydrates via a 130-140 residue carbohydrate recognition domain (CRD) (1,2). The CRD is composed of five-and six-stranded antiparallel b sheets arranged in a ''jelly roll'' fold that is highly conserved among all galectins, despite the proteins only sharing $30% sequence identity (3). In the CRD, affinity for carbohydrates is driven by a combination of polar contacts and CH-p interactions with aromatic sidechains belonging to a typical signature sequence (4-7). There are three types of design of vertebrate galectins (8). The first has CRDs arranged as a non-covalent homodimer (proto-type), e.g., galectin-1, which exists in a reversible monomer/dimer equilibrium. In this proto-type galectin structure, each CRD cooperatively influences the other's affinity for ligand binding (9)(10)(11). The second type has CRDs connected by a polypeptide linker in a heterodimeric tandem-repeat-type display. In this modular arrangement, linker length and flexibility influence galectin supramolecular assembly (12). The third arrangement, chimera-type, is only found in wild-type galectin-3 (WT Gal-3; Fig. 1). The trimodular protein consists of a $120 amino acid N-terminal domain containing nine proline-glycine-rich non-triple helix collagen-like repeat units (13) and an N-terminal peptide with two serine residues as substrates for phosphorylation. (14) The N-terminal domain is highly flexible, which altogether confers unique oligomerization capabilities to Gal-3 (15,16). The N-terminal domain is known to bind neighboring Gal-3 CRDs, allowing the formation of pentamers and higher-order supramolecular lattice structures (14,17).
Ganglioside GM1 is a glycolipid binding partner of Gal-3 found on the outer leaflet of the plasma membrane. GM1 is composed of a sphingosine core and a branched pentasaccharide motif (Galb1-3GalNAcb1-4(Neu5Aca2-3)Galb1-4GlcbCer) that extends above the surface of the cell membrane. GM1 is found in various cell types, but it is highly enriched in neuronal cell membranes in which gangliosides account for up to 10%-12% of the total lipid content (18,19). Cellular studies have shown that Gal-3 plays a role in membrane-mediated signaling through affinity to the cell surface, potentially through interactions with GM1 (20,21). Interactions between GM1 and Gal-3 mediate neuronal cell growth and adhesion as well as regulating metastasis in carcinomas (22)(23)(24); therefore, it is essential to understand how Gal-3 binds GM1 in a lipid membrane. Crystallographic studies have elucidated the atomistic structures of the Gal-3 (25,26) CRD bound to small glycan ligands; however there is limited information about these galectin structures as they interact with GM1 in a membrane context.
The study aimed to determine the structure of Gal-3 binding in situ to a lipid membrane monolayer model containing ganglioside GM1, which bridges several gaps in knowledge. First is the uncertainty regarding Gal-3's binding orientation relative to the membrane, particularly the N-terminal domain's role in membrane interactions. Another question is how the balance of protein affinity and membrane crowding determines galectin surface density, which is relevant for understanding the galectin lattices that form on the surface of cell membranes. Finally, given the importance of galectin-mediated macromolecule complexes for cell-signaling events, it is critical to understand how Gal-3 can associate with other proteins while simultaneously bound to GM1 at the membrane surface. To determine the structure of membrane-bound Gal-3, we used liquid surface X-ray reflectivity (XR) from a lipid monolayer assembled on an aqueous subphase. XR is a valuable technique for determining the electron density profile of a lipid membrane and the associated proteins bound at the membrane surface. We used this method to elucidate the orientation and surface density of Gal-3 bound to the membrane, and provide information about membrane structure.
Overall, in this work, we provide experimental evidence to support an atomistic model for the structure of Gal-3 bound in situ to a membrane containing GM1. Electron density profiles indicate that the Gal-3 CRD is closely adsorbed to the GM1 glycans, while most of the N-terminal domain extends away from the membrane. Gal-3 can bind the membrane at high density, with one molecule bound per approximately 2130 A ˚2 of the membrane. In this in vitro system, Gal-3 covered $50% of the membrane surface, reaching the limits predicted by random sequential adsorption models. These interactions between Gal-3 and GM1 observed in this study may help understand how Gal-3 can affect biologically relevant processes such as neuronal growth or adhesion (24,27).
1,2-Dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) and ganglioside GM1 were purchased from Avanti Polar Lipids (Alabaster, AL) as a dry powder. DPPC and GM1 stock solutions were prepared at 2-3 mg/mL in 8:2 v/v chloroform:methanol and sonicated for 30 seconds. A spreading solution of 0.2 mg/mL with the desired molar ratio (80:20, 85:15, 90:10, 95:5, or 100:0) of DPPC:GM1 in 8:2 chloroform to methanol was prepared for Langmuir trough experiments. Galectin proteins were expressed and purified according to previously published methods (23,28). Lyophilized proteins were dissolved in water at 1 mg/mL. Protein aliquots were frozen and used immediately once thawed.
Experiments were performed using a Nanoscience Instruments KSV-Nima extra-small Teflon trough (50 Â 260 mm) with symmetric movable Delrin barriers and a Wilhelmy plate balance. The 46-mL subphase was pure 18 MU water at room temperature. A lipid spreading solution containing various mixtures of DPPC:GM1 with 0.5 mol % Texas Red-DHPE was deposited onto the air/water interface, 10 minutes were allowed for solvent evaporation, and the barriers were compressed until the surface pressure reached 20 mN/m. Protein was injected to reach a final concentration of FIGURE 1 Schematic of proteins constructed from the Gal-3 CRD. WT Gal-3 contains one CRD bound to an N-terminal domain that promotes oligomeric assembly. Several engineered variants were tested, including a truncated Gal-3 CRD lacking the N-terminal domain (trGal-3) and a Gal-3 CRD homodimer (Gal-3-Gal-3).To see this figure in color, go online.
12.5 mg/mL in the subphase. After protein injection, the barrier expansion was fixed to create a constant-area system where protein insertion would increase surface pressure. Fluorescence microscopy images were taken using an Olympus IX73 inverted microscope with a 10Â objective, ORCA Spark camera, and X-Cite miniþ LED light source. Images were collected and processed using Olympus cellSens Standard software. Images were collected before protein injection and 5 h after protein injection.
Liquid surface scattering experiments were performed at the Advanced Photon Source at Argonne National Labs (Sector 15 NSF's ChemMatCARS) using a 20-mL Langmuir trough (6.5 Â 6.5 cm 2 ) filled with degassed water. Surface pressure was monitored with a Wilhelmy plate balance (KSV Instruments, Finland), and the experiments were performed at room temperature (23.5 C 5 0.5 C). Langmuir trough experiments were performed under constant-area conditions, where protein binding caused an increase in surface pressure. Lipids were deposited on the air/water interface to a surface pressure of 20 mN/m, and 10 min was allowed for solvent evaporation before proteins were injected into the subphase to a final concentration of 12.5 or 60 mg/mL.
During data collection, gaseous oxygen content was maintained at < 2% to prevent background X-ray scattering and oxidative beam damage to the monolayer. This was achieved by sealing the trough in a canister and purging the system with helium gas. X-ray wavelength was 1.24 A ˚, and the incoming X-ray beam footprint's dimensions on the liquid surface was $1 Â 3-10 mm 2 for XR. As a precaution against beam damage, the trough was systematically translated by 1 mm horizontally perpendicular to the X-ray beam after each scan. A Dectris PILATUS 100 detector was used to detect X-ray scattering, and X-ray scattering images were integrated using Python software (https:// github.com/weibu/Liquid_Surface_ChemMatCARS).
XR was used to measure the electron density of materials deposited at the air/water interface on a Langmuir trough. X-ray scattering theory has been fully described previously (29)(30)(31)(32), and only a brief summary is provided here. By measuring the intensity of reflected X-rays, one can deduce detailed information on the electron density distribution normal to the interface, r(z), laterally averaged over both the ordered and disordered parts of the film. The reflectivity is defined as the ratio of reflected to incident beam intensities, in a specular geometry, as a function of the vertical momentum transfer vector q z ¼ (4p/l) sinq, where q is the incident angle of the X-ray on the surface. Finally, the reflectivity curve can be analyzed to obtain the in-plane averaged electron density distribution normal to the interface.
The data were collected by tilting a germanium monochromator crystal to deflect the beam and change the angle of incidence on the sample. Intensities were collected over the range 0.01 < q z < 0.8 A ˚À1 , background subtracted, and normalized to incident beam flux. Data presented are divided by the Fresnel reflectivity (scattering from infinitely sharp air-water interface) (R F ) with error bars representing one standard deviation error for each data point. Division by the Fresnel reflectivity provided better visualization of the XR data.
The data were analyzed using a model-free approach based on cubic B-splines to obtain the electron density profile normal to the interface (33). The coefficients in the B-spline series were determined by constrained nonlinear least-squares methods, in which the smoothest solution with the lowest c 2 goodness of fit was chosen. Over several thousand refinements were performed within the parameter space and a family of models is presented for each reflectivity dataset, all of which satisfy c 2 % c 2 min þ 10% with typical values of c min 2 < 10. (34) The set of fits to each measurement fall within the shaded regions presented in the R/R F versus q z plots. The superposition of the profiles matching these fits yielded electron density ''ribbons'' (regions between the dotted lines on the XR graphs), which are a measure of the uncertainty in the real space structures.
A complementary model-dependent method was also used, where a ''slab'' model was used to obtain the electron density profile normal to the interface. The studied system was divided into layers, or slabs, of certain thickness, electron density, and the slabs were interconnected by interfacial roughness approximated by error functions. The parameters of this model were adjusted using program Motofit (35) to obtain lowest c 2 values and reasonable values of the parameters. Parameter uncertainties were estimated using a finite difference approach with a covariance matrix describing the concavity of c 2 with respect to each fit parameter. However, due to interdependence of the parameters, we expect that parameter errors may be underestimated in the reported data. The results are best interpreted with consideration given to both the model-independent and model-dependent fitting methods.
CHARMM-GUI (36) was used to build the pure-membrane systems containing 80:20 or 85:15 mol % DPPC:GM1. Each system was built with 50 A ˚of water on either side of the membrane and contained 0.15 M KCl. The 80:20 mol % DPPC:GM1 membrane was built using 168 DPPC and 42 GM1 lipid with 25,098 TIP3 waters into a rectangular box 80.94 Â 80.94 Â 140 A ˚in size. The 85:15 mol % DPPC:GM1 membrane was built with 204 DPPC, 36 GM1, and 29,267 TIP3 waters in an 86.6 Â 86.6 Â 170 A ˚rectangular box. For the 85:15 DPPC:GM1 system containing Gal-3 CRD, the protein coordinates were obtained from PDB (PDB: 3AYC), which contains the GM1 pentasaccharide bound to the Gal-3 CRD. The location of the crystallographically observed pentasaccharide was used to position the Gal-3 CRD onto an exposed GM1 glycan in the pure-membrane system that was assembled by CHARMM-GUI. The systems were parameterized with the CHARMM c36m forcefield (37), and simulations were performed using Gromacs 2021.2 (38). The system was simulated for 10 ns of production using 2-fs time steps, and the final 1 ns was used for analysis. The electron density of each membrane component was calculated using the Gromacs tool gmx density.
For all simulations, minimization was performed by steepest descent until the maximum force was less than 1000 kJ/(molnm). Equilibration was performed in six steps of 125, 125, 125, 250, 250, and 250 ps each, beginning with a force constant of 1000 kJ/(molnm 2 ) to restrain the lipids and gradually reducing the position restraints every step. For the systems containing protein, the protein was initially restrained with a force constant of 3000 kJ/(molnm 2 ) in the first equilibration step, with gradual reduction of the restraints in each of the following equilibration steps. The temperature was maintained at 303.15 K using the Berendsen thermostat and pressure was controlled using the Berendsen semi-isotropic barostat. Longrange electrostatics were calculated using particle-mesh Ewald (PME) and a Verlet cutoff scheme was applied with 1.2-nm distance for all nonbonding interactions. Bonds were constrained using LINCS. During production, the temperature and pressure were controlled with a Nose-Hoover and semi-isotropic Parrinello-Rahman thermostat and barostat, respectively.
Fitting XR results to determine trGal-3 binding orientation trGal-3 XR results were used to model the orientation of the Gal-3 CRD bound onto the membrane. trGal-3 electron density was calculated using PDB (PDB: 5OAX). The protein was oriented relative to the membrane using the crystal structure of Gal-3 CRD bound to the GM1 pentasaccharide. The XR results were analyzed using a B-spline or box-model scattering length density (SLD) profile fit, assuming the lipid tails contained no solvent and the interfacial roughness was equal for the heads/tails and tail/ air interfaces (3.0 A ˚).
To determine the electron density of the protein, all atoms were projected on to the y-z plane and a Voronoi tessellation algorithm was used to calculate the area of the protein on the membrane surface. For the Voronoi tessellation algorithm, the projected protein area was 966.6 A ˚2, and a grid of points with 1 A ˚spacing was used to bound the protein. Any grid point within 2.8 A ˚of a projected atom position was removed to eliminate interior grid points. The value of 2.8 A ˚approximates the atomic diameters (van der Waals radius for hydrogen ¼ 1.2 A ˚, carbon ¼ 1.7 A ˚), which was optimized to match the volume calculated using the 0.73 cm 3 /g approximation for globular soluble proteins (39). The Voronoi tessellation approach was applied on slices through the protein structure. Slices in the y-z plane were 1 A ˚thick in the x direction and any atoms with an x position within (R VDW þ slice thickness/2) ¼ 1.9 A ˚were included in the slice.
Summing Voronoi tessellations of the x slices yielded the protein area as a function of x, which was integrated to determine the overall protein volume and calculate electron density. The XR-determined electron density distributions were compared before and after protein binding to determine the experimental electron density profile of the protein.
To study Gal-3 interactions with GM1, we used a model membrane system made from a lipid monolayer deposited at the air/water interface in a Langmuir trough. This in vitro model membrane has several strengths, including the ability to finely tune membrane composition and surface pressure (density of membrane-forming components), and measure protein interactions over an extended period of time (up to 12 h). We chose a simple lipid composition containing various ratios of GM1 and DPPC to probe the impact of GM1 concentration on Gal-3 binding specifically. X-rays can be scattered from the liquid surface to perform in situ XR. In these experiments, the membrane was first compressed to a target surface pressure, where surface pressure (p) was measured using a Wilhelmy plate suspended from a balance. We chose a starting surface pressure of 20 mN/m to form a membrane containing both ordered (liquid condensed) and disordered (liquid expanded) phase lipids, confirmed by fluorescence microscopy using a phase-sensitive dye (Fig. S1). The protein was injected into the aqueous subphase below the membrane and allowed to interact with the membrane for several hours.
One challenge of in vitro galectin experiments is the protein's tendency to embed into the lipid hydrocarbon tails non-specifically (40,41), which poses a challenge in capturing biologically relevant interactions between the CRD and the GM1 glycan. We tested a range of membrane compositions (0-20 mol % GM1) to combat this, with the rationale that the bulky glycan would deter the galectin proteins from embedding into the membrane by blocking access to the hydrophobic lipid tails. Langmuir trough experiments were performed to test WT Gal-3 binding to membranes composed of mixtures of DPPC and GM1 (Fig. 2). In experiments with the highest GM1 concentration (20 mol %), the increase in surface pressure (Dp ¼ 6.5 mN/m) indicates that WT Gal-3 bound the membrane.
Interestingly, WT Gal-3 interacted with all membrane compositions, including those without GM1 (Fig. 2). Langmuir trough isotherm experiments cannot discern between protein embedding into the membrane via non-specific interactions with hydrophobic lipid tails or adsorbing to the membrane surface through specific interactions with the GM1 glycans. However, protein adsorption beneath the monolayer is expected to manifest as a smaller surface pressure increase than protein insertion into the lipid tails (42). At 20 mol % GM1, the surface pressure increase was less than that observed for interaction with a pure DPPC membrane (Dp ¼ 7.3 mN/m), suggesting that WT Gal-3 interactions were more likely to include specific contacts with the GM1 glycan when the GM1 concentration was high. As indicated by the changes in surface pressure, membrane binding was highest at 10 mol% GM1. However, this is likely due to a combination of specific interactions with GM1 and non-specific insertion into the lipid monolayer.
XR experiments were performed to resolve the position of Gal-3 bound to the lipid monolayer and confirm specific interactions with the GM1 glycans. XR can resolve layers of electron-dense material present at the air/water interface, including differences between protein and lipids. This is done by analyzing deviations of the measured reflectivity (R) from the Fresnel reflectivity of an ideal interface (R F ). The membrane can be described by electron density distribution (r) as a function of depth along the z axis, normal to the air/water interface. For ease of comparison between experimental and computational methods, the measured electron density was plotted after normalization to the solvent electron density (r/r water ). r/r water can be used to resolve the position of distinct chemical moieties in the monolayer due to the electron density difference between the solvent and hydrocarbon lipid tails (r/r water ¼ 0.97), phospholipid heads (r/r water ¼ 1.32), and glycans (r/r water ¼ 1.16). For a homogenous protein layer completely excluding solvent, a r/r water of 1.23 is expected (43). Protein adsorption beneath the monolayer rarely results in complete membrane coverage. The resulting r/r water is expected to range between 1.0 and 1.23 depending on the density by which the protein covers the membrane and excludes solvent. The length of this r/r water increase along the z axis will indicate the length of the protein layer adsorbed to the membrane. Furthermore, if the protein embeds into the lipids, it will change the average r/r water in that region based on the molar ratio of protein to lipid components. r/r water was determined using model-independent fitting of normalized reflectivity (R/R F ) as a function of the scattering vector (q z )(Fig. 3 A). Reflectivity data were collected from a pure 80:20 mol % DPPC:GM1 membrane, and 6 h after addition of 12.5 or 60.0 mg/mL WT Gal-3. There is agreement between the data and the fit for each dataset, indicating confidence in the resulting r/r water models.
From the r/r water distribution of a pure 80:20 DPPC:GM1 membrane (Fig. 3 B and C, black trace), the hydrocarbon tails can be observed from approximately À20 to À5A ˚, phospholipid heads from À5t oþ3A ˚, and glycans from þ3toþ16 A ˚. In these plots, z ¼ 0 is defined as the center of the lipid heads. WT Gal-3 was observed adsorbing to the 80:20 DPPC:GM1 membrane, shown by the increased r/r water at z distances of $16-65 A ˚(Fig. 3 B andC). As expected, the protein appears to bind the membrane through interfacial interactions with the GM1 glycans, not through embedding into membrane lipids. The protein r/r water indicates the amount of protein adsorbed to the membrane surface. Comparing the concentrations of 12.5 mg/mL with 60 mg/mL WT Gal-3 in the water subphase, there was no significant change to the r/r water profile, suggesting both concentrations resulted in similar amount of protein adsorbed to the membrane and binding was saturated at 12.5 mg/mL.
Model-dependent fitting was used for each dataset to quantitate the length of the protein layer (Fig. S2 and Table S1). Model-dependent fitting is performed by building layers of electron density to represent each unique chemical composition in the membrane, one layer each for the hydrocarbon tails, phospholipid heads, glycans, and protein. Each layer is parameterized with a length, electron density, and interfacial roughness to account for interfaces between layers.
Results from the model-dependent fitting show the WT Gal-3 protein was best fit with two separate layers of distinct electron density, totaling $50 A ˚in length altogether. The first layer, likely representing the Gal-3 CRD, extends 26.9 5 0.4 A ˚away from the glycans, while the second layer, likely representing Gal-3 N-terminal domain, extends 22.5 5 0.9 A ˚away from the CRD and into the water subphase. Crystal structures of isolated Gal-3 CRD (lacking the chimeric domain) show a maximum expected length of $30 A ˚(PDB: 3AYC) (25), supporting our observation of distinct regions occupied by the Gal-3 CRD and the N-terminal domain.
As a result of Gal-3 binding, the glycans extend toward the water subphase. In the pure membrane, the glycan layer has a length of 12.8 5 0.1 A ˚; however, upon protein binding, this increases to 13.5 5 0.1 A ˚. This glycan extension is also clearly observed in the SLD results from model-independent fitting (Fig. 3 C). The implication is that, in the pure-membrane system, the glycans have a compact configuration and are ''matted down'' on the surface of the membrane to optimize hydrogen bonding with themselves and the phosphocholine headgroups. In contrast, when Gal-3 binds, the glycans extend away from the monolayer and form favorable contacts with the Gal-3 CRD.
Gal-3 interactions with a membrane containing 5% GM1 led to a sparsely populated protein layer extended beneath the glycans (Figs. 3 D andS4; Table S2). From the model-dependent fitting, there was 58% less electron density associated with the protein compared with Gal-3 bound to a membrane containing 20% GM1 (Table S1; Fig. S2). Gal-3 binding to a 10% GM1 membrane was similar to the 5% GM1 results (Fig. S4 Table S2). Finally, in experiments with a pure DPPC membrane, no protein was adsorbed to the membrane surface (Fig. 3 E), as indicated by small changes in the electron density profiles. However, a slight decrease in lipid headgroup r/r water indicates Gal-3 was embedded into the membrane and formed non-specific contacts with the DPPC lipids. Overall, this suggests that Gal-3 forms specific interactions with the GM1 glycans, and membrane surface coverage is dependent on the amount of GM1 in the membrane.
As mentioned in the introduction, Gal-3 is a chimera-type galectin with a C-terminal CRD and an N-terminal domain that promotes oligomerization (17). To discern the role of the N-terminal domain on membrane binding, we tested a series of engineered variants (Fig. 1) including a truncated Gal-3 containing only the CRD (trGal-3) and a covalently linked Gal-3 CRD homodimer (Gal-3-Gal-3).
Langmuir trough assay experiments with an 80:20 DPPC:GM1 membrane indicate strong interaction for all constructs, suggesting that the N-terminal domain is not essential for membrane binding (Fig. 4 A). However, the N-terminal domain seems to contribute to membrane interactions because trGal-3 required more time to reach its surface pressure plateau than WT Gal-3.This supports previous findings that removing the N-terminal domain reduces Gal-3 affinity to the cell surface, increasing the K d from 0.74 to 3.1 mM ( 22). However, this delay in membrane binding was not observed for Gal-3-Gal-3. We interpret these binding kinetics to indicate additional Gal-3 CRDs drive rapid saturation on the membrane. However, the N-terminal domain likely contributes to a more dynamic and complex Gal-3 oligomerization equilibrium with a slower on-rate.
XR was used to determine the structure of each construct bound in situ to the membrane surface. trGal-3 formed a shorter protein layer (27.6 5 0.3 A ˚) than the WT Gal-3, shown by model-independent and -dependent XR fitting results (Fig. 4 B andC). The Gal-3-Gal-3 homodimer accommodated both CRDs oriented in close contact with the membrane (Fig. 4 B-D), forming a protein layer only 31 5 0.7 A ˚in length. Compared with the $50 A ˚protein layer observed for WT Gal-3, this finding supports the interpretation that the CRD contacts GM1, while the N-terminal domain extends away from the membrane. Furthermore, this supports the interpretation that the second electron-dense layer of 23 A ˚can be attributed to the N-terminal domain, and is corroborated by models constructed from SAXS measurements of WT Gal-3 (44). We suggest this supports a model for a compacted N-terminal domain configuration and WT Gal-3 oligomerization on the membrane surface. If the N-terminal domain was behaving as a fully extended Flory chain in a good solvent, it would extend $60 A ˚in length (45). The XR model only provides strong evidence for $23 A ˚of welldefined electron density. However, given that XR data represent an average of many spatial/temporal states, we expect the model represents an ensemble containing a range of extended to compact N-terminal domain configurations present at the membrane.
Interestingly, trGal-3 has 25% less protein electron density in the region attributed to the CRD compared with WT Gal-3 (Fig. 4 B; Table S1). We suggest two possible reasons for this. The first is that the N-terminal domain may increase Gal-3's affinity for the membrane without directly interacting with GM1 (22). Higher membrane affinity would increase WT Gal-3 binding, causing higher averaged electron density where the CRD is bound. The second reason is that small portions of the N-terminal domain may be stably associated with the CRD, increasing the electron density in that region. A crystal structure (PDB: 6FOF) shows 14 amino acids from the N-terminal domain form strong contacts with the Gal-3 CRD (44). NMR studies indicate that most of the N-terminal domain is disordered but that some residues have significantly reduced mobility in the region proximal to the CRD (46)(47)(48). Furthermore, interactions between the N-terminal domain and the CRD are observed to modulate Gal-3 oligomerization and glycan interactions (15).
To determine the electron density contributed by the N-terminal domain, we compared our XR results with a calculated electron density for a published Gal-3 CRD crystal structure (PDB: 3AYC) (25)(Fig. S5). Modeling results indicate that the N-terminal domain contributes electron density from 50-100 A ˚(reported in terms of the ''Depth'' scale in Fig. 4 B). This suggests the N-terminal domain in-teracts with the Gal-3 CRD distal to the membrane and does not interact with GM1 or the membrane surface.
To correlate an atomistic model with the XR results, we performed a preliminary molecular dynamics simulation of the Gal-3 CRD bound to GM1 in a membrane. The starting structure of the simulation was based on the published crystal structure of the Gal-3 CRD bound to the GM1 pentasaccharide (PDB: 3AYC) (25). Guided by this structure, the Gal-3 CRD was positioned onto a GM1 molecule embedded into an 85:15 DPPC:GM1 membrane built with 204 DPPC, 36 GM1, and 29,267 TIP3 waters in an 86.6 Â 86.6 Â 170 A ˚rectangular box using CHARMM-GUI (36)(Fig. 5 A and B). The simulation was built with 15 mol % GM1 in the membrane because the protein was more easily accommodated among the crowded glycan environment on the membrane surface. However, a pure 80:20 DPPC:GM1 membrane (without protein) was also simulated and compared with experimental XR results.
The electron density of the simulated membrane showed consensus with the profiles derived from XR results (Fig. 5 C and D). However, the simulated membranes had slightly lower electron density for the phospholipid heads and more roughness between the head and glycan layers. These slight differences may be due to the XR experiments being performed with a lipid monolayer, while the simulations were performed with a lipid bilayer (49). The electron density was calculated for each chemical moiety (hydrocarbon tails, phospholipid heads, glycans, and protein), and the results generally align with the layers derived from modeldependent XR fitting.
Using the simulation, the electron density was calculated for a single Gal-3 CRD molecule in the predicted orientation for GM1 binding (Fig. 5 D). The electron density profile generally matches the protein layer observed from XR modeling of trGal-3 bound to 80:20 DPPC:GM1. The magnitude of the simulation profile is lower because the system was constructed with lower density of bound protein compared with the experimental system. The shape of the XR profile is a highly accurate representation of protein electron density averaged from billions of molecules in the path of the X-ray beam. In contrast, this simulationderived profile is based on the position of a single molecule averaged over a short simulation time.
We also performed a Voronoi tessellation algorithm on slices through the three-dimensional protein structure to calculate electron density of the membrane-bound Gal-3 CRD (Fig. S5). The calculated Gal-3 CRD volume was 18,756.5 A ˚3, which is within 1% of the predicted volume based on amino acid sequence and the standard density of globular proteins (0.73 cm 3 /g) (39). This corresponds to an electron density of 0.44 e À /A ˚3. Overall, the calculated protein electron density was a good match to the experimental results, supporting an atomistic model describing the Gal-3 CRD binding the GM1 in a membrane.
In addition to Gal-3 interactions with GM1 glycans, we also observed a few close contacts between the Gal-3 CRD and DPPC headgroups, which may help to stabilize protein interactions with the membrane (Fig. S6). However, this will need further investigation with more extended molecular dynamics simulations to determine if the contacts persist over time. The binding orientation of the Gal-3 CRD predicted from crystallography indicates that the loops containing asparagines 141, 166, and 167 would be in proximity to form energetically favorable polar contacts with the lipid headgroups.
The surface density of membrane-bound Gal-3 likely influences cell signaling through Gal-3 oligomerization and supramolecular organization. Previous isothermal titration calorimetry and surface plasmon resonance experiments showed that Gal-3 binds free GM1 with a K d of $57-62 mM ( 25). However, we were curious to understand the interplay between GM1 membrane content and Gal-3 surface coverage, especially considering the limited space on the surface of the membrane. Increasing the protein concentration $5Â (from 12.5 to 60 mg/mL) did not increase Gal-3 surface density (Fig. 3), suggesting that protein concentration was not a limiting factor in these experiments.
From the XR results (Figs. 3, S2, and S4; Tables S1 andS2), it is evident that WT Gal-3 membrane coverage is dependent on GM1 concentration. A similar trend was observed for trGal-3 binding membranes with various GM1 compositions (Fig. S4; Table S2). In our Langmuir trough experiments with 12.5 mg/mL WT Gal-3 and a membrane containing 20% GM1, there is $9.5 nmol of Gal-3 CRD and $3.5 nmol of GM1 available for interactions. Therefore, the amount of GM1 is likely a limiting factor for Gal-3 binding.
Using protein electron density values determined from XR measurements, we can estimate Gal-3 surface coverage on a membrane containing 20% GM1. We used the electron density profile of the simulated 80:20 DPPC:GM1 membrane and added incrementally increasing quantities of calculated Gal-3 CRD electron density (Fig. 6). Comparing the simulated data with the experimental, we predict that about three protein molecules are bound for the area of membrane in the simulation (6400 A ˚2, $11 Â 11 lipids in size), which equates to one protein bound per $2130 A ˚2. Surface coverage estimations were also performed based on fitting the XR data using calculated protein density and area from Voronoi tessellation calculations (Table S3). Our estimates range from 45%-56% coverage, which generally agrees with a simple division of the observed WT Gal-3 protein electron density by the theoretical maximum protein electron density that could be observed with 100% membrane coverage (0.12/ 0.23 ¼ 52% coverage).
Comparing this result with the literature, we find that Gal-3 membrane coverage matches the theoretical maximum of 54.67% predicted for random sequential adsorption of disks onto a two-dimensional surface (50). This adsorption limit is due to steric overlaps between neighboring adsorbed particles that are randomly distributed on the surface. We estimate that the disk shape is a reasonable approximation for the Gal-3 CRD. Disk dimers, such as the Gal-3-Gal-3 homodimer, also exhibit a maximum theoretical adsorption of 54.7% (51).
The theory of random sequential adsorption has two key assumptions that are relevant to consider in our system. The first assumption is that the particles can adsorb anywhere on the surface; in other words, the surface is homogeneous (52). Gal-3 did not reach the maximum theoretical adsorption when binding a membrane containing 5 or 10 mol % GM1, suggesting these concentrations are insufficient for the membrane to appear homogeneous to interacting Gal-3 molecules. The second assumption is that particles cannot rearrange once adsorbed onto the surface, meaning they hold a fixed location on the membrane once bound (52). However, in a theoretical system where diskshaped particles can rearrange after adsorption, the maximum surface coverage may approach up to 85%, although the surface will likely have crystalline characteristics at this coverage density (53). At the low membrane surface pressures of 20-30 mN/m used in these experiments, the lipids exist in a mixture of liquid condensed and liquid expanded states (Fig. S1). This suggests that the GM1 molecules should have free two-dimensional diffusion around the membrane via the liquid expanded state. Interestingly, for all GM1 concentrations tested, Gal-3 adsorption never went above the estimated 54.7% for random sequential adsorption, suggesting that Gal-3 is not moving laterally around the membrane once bound. Furthermore, this indicates that Gal-3 does not assemble into a densely packed in-plane ordered conformation while forming supramolecular assemblies on the membrane surface. Higher GM1 concentrations could possibly induce a crystalline assembly, but higher GM1 concentrations are highly unlikely to occur in a biological system. Overall, the observed Gal-3 packing density is generally consistent with prevailing models for Gal-3 oligomerization and amorphous lattice formation (54), but does not uniquely support a specific assembly model.
Biophysical Journal 122, 1926-1937, June 6, 2023 1927
Biophysical Journal 122, 1926-1937, June 6, 2023 1929
Biophysical Journal 122, 1926-1937, June 6, 2023 1931
Biophysical Journal 122, 1926-1937,June 6, 2023June 6, 1933
Biophysical Journal 122, 1926-1937, June 6, 2023 1937